Add files using upload-large-folder tool

ckpts/universal/global_step20/zero/26.post_attention_layernorm.weight/exp_avg.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3599b233683cf5c07d6f254c52c98de7eaa4f5e3539175700226804e72b3359e
+size 9372

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/context/context.h

@@ -0,0 +1,174 @@
+#pragma once
+
+#include <cstdint>
+#include <functional>
+
+#include <ATen/core/Dict.h>
+#include <torch/csrc/autograd/engine.h>
+#include <torch/csrc/distributed/autograd/functions/recvrpc_backward.h>
+#include <torch/csrc/distributed/autograd/functions/sendrpc_backward.h>
+#include <torch/csrc/distributed/rpc/rpc_agent.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+class RecvRpcBackward;
+
+// DistAutogradContext which stores information for a single distributed
+// autograd pass on a worker.
+class TORCH_API DistAutogradContext {
+ public:
+  using GradCallback = std::function<bool(torch::Tensor&)>;
+
+  explicit DistAutogradContext(int64_t contextId);
+
+  // Retrieves the autograd context id for this context.
+  int64_t contextId() const;
+
+  // Records a 'send' autograd function for this context with the provided
+  // message id.
+  void addSendFunction(
+      const std::shared_ptr<SendRpcBackward>& func,
+      int64_t autograd_message_id);
+
+  // Records a 'recv' autograd function for this context with the provided
+  // message id.
+  void addRecvFunction(
+      std::shared_ptr<RecvRpcBackward>& func,
+      int64_t autograd_message_id);
+
+  // Given an autograd_message_id, retrieve the appropriate send function.
+  std::shared_ptr<SendRpcBackward> retrieveSendFunction(
+      int64_t autograd_message_id);
+
+  // Return all send functions for this context.
+  std::unordered_map<int64_t, std::shared_ptr<SendRpcBackward>> sendFunctions()
+      const;
+
+  // Return all recv functions for this context.
+  std::unordered_map<int64_t, std::shared_ptr<RecvRpcBackward>> recvFunctions()
+      const;
+
+  // Adds a future message recording an outstanding RPC.
+  void addOutstandingRpc(const c10::intrusive_ptr<rpc::JitFuture>& jitFuture);
+
+  // Returns all gradients.
+  const c10::Dict<torch::Tensor, torch::Tensor> getGradients() const;
+
+  // This function gives a mutable grad reference to the callback.
+  // If the callback returns true, it means the grad in the context
+  // needs to be updated.
+  void runGradCallbackForVariable(
+      const torch::autograd::Variable& variable,
+      GradCallback&& cb);
+
+  DistAutogradContext(const DistAutogradContext&) = delete;
+  DistAutogradContext& operator=(const DistAutogradContext&) = delete;
+  DistAutogradContext(DistAutogradContext&&) = delete;
+  DistAutogradContext& operator=(DistAutogradContext&&) = delete;
+
+  // records the workerID of a node that we sent an RPC to.
+  // workerIDs are added here when we attach a send function to this autograd
+  // context
+  void addKnownWorkerId(const rpc::worker_id_t workerId);
+
+  // Retrieves a set containing the known workerIds for this context
+  // These are the different workers that this context has sent RPCs to.
+  std::unordered_set<rpc::worker_id_t> getKnownWorkerIds() const;
+
+ private:
+  friend class BackwardPassCleanupGuard;
+  friend class DistEngine;
+  friend class RecvRpcBackward;
+  friend class DistAccumulateGradCaptureHook;
+
+  // Record that we would like to accumulate the provided gradient on the given
+  // variable.
+  void accumulateGrad(
+      const torch::autograd::Variable& variable,
+      const torch::Tensor& grad,
+      size_t num_expected_refs);
+
+  // Retrieve the GraphTask.
+  std::shared_ptr<torch::autograd::GraphTask> retrieveGraphTask();
+
+  // Set the appropriate graph task for the backward pass. Can be called only
+  // once.
+  void setGraphTask(std::shared_ptr<torch::autograd::GraphTask> graphTask);
+
+  // Resets the graph task to ensure we can run another distributed backward
+  // pass for the same autograd context.
+  void resetGraphTask();
+
+  // Waits for all outstanding RPCs for this context to finish and clears all
+  // outstanding rpcs held in this context. This should be called only once.
+  c10::intrusive_ptr<c10::ivalue::Future> clearAndWaitForOutstandingRpcsAsync();
+
+  void clearOutstandingRpcs();
+
+  // Record an event to mark the completion of gradient computation. These
+  // events will later help to properly synchronize gradients consumptions
+  // in getGradients(). We need these events because backward and
+  // optimizer.step are separate RPC calls, and will occur on different CUDA
+  // streams. Without synchronization, it is possible that gradients are
+  // consumed before they are ready.
+  void recordGradEvent(c10::Device device);
+
+  const int64_t contextId_;
+
+  // Set containing known worker IDs, used in cleaning up autograd context.
+  // Whenever a sendRpcBackward is attached to the autograd graph for this
+  // context, the destination is added here.
+  std::unordered_set<rpc::worker_id_t> knownWorkerIds_;
+
+  // Map from autograd_message_id to appropriate 'send' autograd function.
+  std::unordered_map<int64_t, std::shared_ptr<SendRpcBackward>>
+      sendAutogradFunctions_;
+
+  // Map from autograd_message_id to appropriate 'recv' autograd function.
+  std::unordered_map<int64_t, std::shared_ptr<RecvRpcBackward>>
+      recvAutogradFunctions_;
+
+  // Gradients accumulated in this context so far. The key is the variable on
+  // which the gradient needs to be accumulated and the value is the gradient
+  // that needs to be accumulated on that variable..
+  c10::Dict<torch::Tensor, torch::Tensor> accumulatedGrads_;
+
+  // See comments for recordGradEvent(c10::Device device);
+  std::unordered_map<c10::Device, c10::Event> gradReadyEvents_;
+  const c10::impl::VirtualGuardImpl impl_;
+
+  // The autograd GraphTask for the backward pass on this node for this context.
+  std::shared_ptr<torch::autograd::GraphTask> graphTask_;
+
+  // List of futures for RPCs initiated by this node to propagate gradients to
+  // other nodes. The distributed autograd engine on this node can return
+  // successfully only if all these futures are done and are successful.
+  std::vector<c10::intrusive_ptr<rpc::JitFuture>> outStandingRpcs_;
+
+  // Lock to protect concurrent modification of the context.
+  mutable std::mutex lock_;
+};
+
+using ContextPtr = std::shared_ptr<DistAutogradContext>;
+
+// This class stores a shared_ptr to a DistAutogradContext instance in a
+// thread local variable. The instance is given by the call site. The class
+// doesn't know the current context. It's just a util class.
+class TORCH_API ThreadLocalDistAutogradContext {
+ public:
+  // Store 'new_context' to the thread local variable maintained by this class.
+  explicit ThreadLocalDistAutogradContext(ContextPtr&& new_context);
+  ~ThreadLocalDistAutogradContext();
+
+  // Retrieve the stored DistAutogradContext instance.
+  static ContextPtr getContextPtr();
+
+ private:
+  ContextPtr prev_context_ptr_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/functions/recvrpc_backward.h

@@ -0,0 +1,49 @@
+#pragma once
+
+#include <torch/csrc/autograd/function.h>
+#include <torch/csrc/distributed/autograd/context/context.h>
+#include <torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h>
+#include <torch/csrc/distributed/rpc/rpc_agent.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Forward declarations.
+class DistAutogradContext;
+
+// As part of our distributed autograd implementation, whenever we receive an
+// RPC from a node, we add a 'RecvRpcBackward' autograd function to the
+// autograd graph. This is more or less a placeholder function that is used to
+// pass gradients to the remote host during the backward pass. The inputs to the
+// RPC function are the inputs to this autograd function.
+class TORCH_API RecvRpcBackward : public torch::autograd::Node {
+ public:
+  explicit RecvRpcBackward(
+      const AutogradMetadata& autogradMetadata,
+      std::shared_ptr<DistAutogradContext> autogradContext,
+      rpc::worker_id_t fromWorkerId,
+      rpc::DeviceMap deviceMap);
+
+  torch::autograd::variable_list apply(
+      torch::autograd::variable_list&& grads) override;
+
+ private:
+  const AutogradMetadata autogradMetadata_;
+
+  // Hold a weak reference to the autograd context to avoid circular
+  // dependencies with the context (since it holds a reference to
+  // RecvRpcBackward).
+  std::weak_ptr<DistAutogradContext> autogradContext_;
+
+  // The worker id from which the RPC was received. During the backward pass,
+  // we need to propagate the gradients to this workerId.
+  rpc::worker_id_t fromWorkerId_;
+
+  // Device mapping for tensors sent over RPC.
+  const rpc::DeviceMap deviceMap_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/functions/sendrpc_backward.h

@@ -0,0 +1,37 @@
+#pragma once
+
+#include <torch/csrc/autograd/function.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// As part of our distributed autograd implementation, whenever we send an RPC
+// from one node to another, we add a 'SendRpcBackward' autograd function to the
+// autograd graph. This is more or less a placeholder function that is used to
+// kickoff the autograd engine on the current worker on the backward pass. The
+// edges for this autograd function are the inputs to the RPC method.
+//
+// During the backward pass, this function is queued for execution in the
+// autograd engine which eventually runs the rest of the autograd graph.
+struct TORCH_API SendRpcBackward : public torch::autograd::Node {
+ public:
+  torch::autograd::variable_list apply(
+      torch::autograd::variable_list&& inputs) override;
+
+  // SendRpcBackward is actually the root of an autograd graph on the local
+  // node. As a result, it doesn't receive any 'inputs', but rather the RPC
+  // framework passes gradients over to this function to kickoff local autograd
+  // computation.
+  void setGrads(const torch::autograd::variable_list& grads);
+
+  // Retrieve the grads for the function.
+  const torch::autograd::variable_list& getGrads() const;
+
+ private:
+  torch::autograd::variable_list grads_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h

@@ -0,0 +1,25 @@
+#pragma once
+
+#include <torch/csrc/Export.h>
+#include <cstdint>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// This structure represents autograd metadata that we need to pass across
+// different nodes when we call an RPC which needs autograd computation.
+struct TORCH_API AutogradMetadata {
+  AutogradMetadata(int64_t autogradContextId, int64_t autogradMessageId);
+
+  // autogradContextId_ is a globally unique integer that identifies a
+  // particular distributed autograd pass.
+  int64_t autogradContextId;
+  // autogradMessageId_ is a globally unique integer that identifies a pair
+  // of send/recv autograd functions.
+  int64_t autogradMessageId;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/cleanup_autograd_context_req.h

@@ -0,0 +1,29 @@
+#pragma once
+
+#include <torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h>
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Used to request other workers to clean up their autograd context.
+class TORCH_API CleanupAutogradContextReq : public rpc::RpcCommandBase {
+ public:
+  explicit CleanupAutogradContextReq(int64_t context_id);
+  // Serialization and deserialization methods.
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<CleanupAutogradContextReq> fromMessage(
+      const rpc::Message& message);
+
+  // Retrieve the context id we are cleaning up with this message.
+  int64_t getContextId();
+
+ private:
+  int64_t context_id_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/cleanup_autograd_context_resp.h

@@ -0,0 +1,23 @@
+#pragma once
+
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Empty response for CleanupAutogradContextReq. Send to acknowledge receipt of
+// a CleanupAutogradContextReq.
+class TORCH_API CleanupAutogradContextResp : public rpc::RpcCommandBase {
+ public:
+  CleanupAutogradContextResp() = default;
+  // Serialization and deserialization methods.
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<CleanupAutogradContextResp> fromMessage(
+      const rpc::Message& message);
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/propagate_gradients_req.h

@@ -0,0 +1,42 @@
+#pragma once
+
+#include <torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h>
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+#include <vector>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Used to propagate gradients from one node to another during a distributed
+// backwards pass. This RPC call is invoked when we hit a `recv` autograd
+// function during backward pass execution.
+class TORCH_API PropagateGradientsReq : public rpc::RpcCommandBase {
+ public:
+  PropagateGradientsReq(
+      const AutogradMetadata& autogradMetadata,
+      std::vector<torch::autograd::Variable> grads,
+      bool retainGraph = false);
+
+  const AutogradMetadata& getAutogradMetadata();
+
+  const std::vector<torch::autograd::Variable>& getGrads();
+
+  // Serialization and deserialization methods.
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<PropagateGradientsReq> fromMessage(
+      const rpc::Message& message);
+
+  // Whether or not to retain the autograd graph.
+  bool retainGraph();
+
+ private:
+  AutogradMetadata autogradMetadata_;
+  std::vector<torch::autograd::Variable> grads_;
+  bool retainGraph_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/propagate_gradients_resp.h

@@ -0,0 +1,24 @@
+#pragma once
+
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Response for the PropagateGradients call. Currently, this class is mostly
+// just a placeholder and sends an empty message over the wire. The purpose of
+// this RPC command is to indicate whether or not the PropagateGradientsReq call
+// was successfully or not.
+class TORCH_API PropagateGradientsResp : public rpc::RpcCommandBase {
+ public:
+  PropagateGradientsResp() = default;
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<PropagateGradientsResp> fromMessage(
+      const rpc::Message& message);
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/rpc_with_autograd.h

@@ -0,0 +1,98 @@
+#pragma once
+
+#include <torch/csrc/distributed/autograd/rpc_messages/autograd_metadata.h>
+#include <torch/csrc/distributed/rpc/rpc_agent.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Represents an RPC that includes autograd information. This class basically
+// wraps another `RpcCommandBase` object which represents the actual RPC and has
+// additional autograd information associated with that RPC.
+class TORCH_API RpcWithAutograd final : public rpc::RpcCommandBase {
+ public:
+  // Used when we are sending an RPC over the wire.
+  RpcWithAutograd(
+      rpc::worker_id_t fromWorkerId,
+      rpc::MessageType messageType,
+      const AutogradMetadata& autogradMetadata,
+      c10::intrusive_ptr<rpc::Message> wrappedMessage,
+      rpc::DeviceMap deviceMap = {});
+
+  // Used when receiving an RPC over the wire.
+  RpcWithAutograd(
+      rpc::worker_id_t fromWorkerId,
+      rpc::MessageType messageType,
+      const AutogradMetadata& autogradMetadata,
+      std::unique_ptr<rpc::RpcCommandBase> wrappedRpc,
+      rpc::MessageType wrappedMessageType,
+      std::vector<torch::Tensor> tensors,
+      rpc::DeviceMap deviceMap = {});
+
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+
+  static std::unique_ptr<RpcWithAutograd> fromMessage(
+      const rpc::Message& message);
+
+  // Retrieves tensors as part of this RPC, which need to be considered for
+  // autograd computations.
+  std::vector<torch::Tensor>& tensors();
+
+  const AutogradMetadata& autogradMetadata() const;
+
+  RpcCommandBase& wrappedRpc();
+
+  void setWrappedRpc(std::unique_ptr<RpcCommandBase> wrappedRpc);
+
+  std::unique_ptr<RpcCommandBase> moveWrappedRpc() &&;
+
+  // Message type of the wrapped RPC.
+  rpc::MessageType wrappedMessageType() const;
+
+  // Retrieve the worker id from which the RPC originated.
+  rpc::worker_id_t fromWorkerId() const;
+
+  // Retrieve the device map.
+  const rpc::DeviceMap& deviceMap();
+
+ private:
+  // WorkerId from which this RPC originated. This is necessary for knowing
+  // which worker we need to contact during the backward pass.
+  rpc::worker_id_t fromWorkerId_;
+
+  // Message type for this call.
+  rpc::MessageType messageType_;
+
+  AutogradMetadata autogradMetadata_;
+
+  // Since wrappedMessage_ is destructively constructed from wrappedRpc_,
+  // they are valid exclusively. They are used for different purpose.
+  // wrappedRpc_ is used while constructing receive rpcWithAutograd;
+  // wrappedMessage_ is used while constructing send rpcWithAutograd;
+
+  // When receive rpcWithAutograd is constructed fromMessage, it is valid;
+  // When send rpcWithAutograd is constructed before toMessage, it is nullptr;
+  std::unique_ptr<RpcCommandBase> wrappedRpc_;
+
+  // Serialized message representing wrappedRpc_. Used mostly as a cache to
+  // avoid serializing the request twice.
+  // When receive rpcWithAutograd is constructed fromMessage, it is nullptr;
+  // When send rpcWithAutograd is constructed before toMessage, it is valid;
+  c10::intrusive_ptr<rpc::Message> wrappedMessage_;
+
+  // message type of the wrappedMessage, this is stored separately since
+  // wrappedMessage_ is not always guaranteed to be populated.
+  rpc::MessageType wrappedMessageType_;
+
+  // Tensors part of the wrappedRpc that need to be considered for autograd.
+  std::vector<torch::Tensor> tensors_;
+
+  // Device mapping for tensors that are sent across an RPC to another node.
+  rpc::DeviceMap deviceMap_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/rpc_with_profiling_req.h

@@ -0,0 +1,62 @@
+#pragma once
+
+#include <torch/csrc/autograd/profiler.h>
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_agent.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+#include <torch/csrc/distributed/rpc/types.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+class TORCH_API RpcWithProfilingReq : public rpc::RpcCommandBase {
+ public:
+  // For sending RPCs, invoked when client is creating this RPC command.
+  RpcWithProfilingReq(
+      rpc::MessageType messageType,
+      c10::intrusive_ptr<rpc::Message> wrappedMessage,
+      torch::autograd::profiler::ProfilerConfig&& profilerConfig,
+      rpc::ProfilingId profilingKeyId);
+
+  // For receiving an RPC
+  // Used in fromMessage.
+  RpcWithProfilingReq(
+      rpc::MessageType messageType,
+      std::unique_ptr<rpc::RpcCommandBase> wrappedRpc,
+      rpc::MessageType wrappedMessageType,
+      std::vector<torch::Tensor> tensors,
+      torch::autograd::profiler::ProfilerConfig&& profilerConfig,
+      rpc::ProfilingId profilingKeyId);
+
+  // Convert this RPC Command to a Message that can be sent over the wire.
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<RpcWithProfilingReq> fromMessage(
+      const rpc::Message& message);
+
+  // Retrieve the profiling data that is associated with this command.
+  torch::autograd::profiler::ProfilerConfig getProfilingConfig() const;
+  // Retrieve the globally unique profiling ID corresponding to this command.
+  const rpc::ProfilingId& getProfilingId() const;
+  // Retrieve the original RPC which this ProfilingRPC wraps.
+  RpcCommandBase& wrappedRpc();
+  // Destructively move the wrapped RPC.
+  std::unique_ptr<RpcCommandBase> moveWrappedRpc() &&;
+  // Message type of the wrapped RPC
+  rpc::MessageType wrappedMessageType() const;
+  void setWrappedRpc(std::unique_ptr<RpcCommandBase> wrappedRpc);
+
+ private:
+  // message type
+  const rpc::MessageType messageType_;
+  // wrapped message
+  c10::intrusive_ptr<rpc::Message> wrappedMessage_;
+  std::unique_ptr<RpcCommandBase> wrappedRpc_;
+  rpc::MessageType wrappedMessageType_;
+  std::vector<torch::Tensor> tensors_;
+  const torch::autograd::profiler::ProfilerConfig profilerConfig_;
+  const rpc::ProfilingId profilingKeyId_;
+};
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/rpc_with_profiling_resp.h

@@ -0,0 +1,59 @@
+#pragma once
+
+#include <torch/csrc/autograd/profiler.h>
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_agent.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+#include <torch/csrc/distributed/rpc/types.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+class TORCH_API RpcWithProfilingResp : public rpc::RpcCommandBase {
+ public:
+  // For sending RPCs over the wire
+  RpcWithProfilingResp(
+      rpc::MessageType messageType,
+      c10::intrusive_ptr<rpc::Message> wrappedMessage,
+      std::vector<torch::autograd::profiler::LegacyEvent> profiledEvents,
+      rpc::ProfilingId profilingId);
+
+  // For receiving RPCs. Used in from message when converting a message received
+  // over the wire.
+  RpcWithProfilingResp(
+      rpc::MessageType messageType,
+      std::unique_ptr<rpc::RpcCommandBase> wrappedRpc,
+      rpc::MessageType wrappedMessageType,
+      std::vector<torch::Tensor> tensors,
+      std::vector<torch::autograd::profiler::LegacyEvent> profiledEvents,
+      rpc::ProfilingId profilingId);
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<RpcWithProfilingResp> fromMessage(
+      const rpc::Message& message);
+  // Retrieve remote Events
+  std::vector<torch::autograd::profiler::LegacyEvent> getProfiledEvents() const;
+  // Retrieve the globally unique profiling ID corresponding to this command.
+  const rpc::ProfilingId& getProfilingId() const;
+  // Retrieve the original RPC which this ProfilingRPC wraps.
+  RpcCommandBase& wrappedRpc();
+  // Destructively move the wrapped RPC.
+  std::unique_ptr<RpcCommandBase> moveWrappedRpc() &&;
+  // Message type of the wrapped RPC
+  rpc::MessageType wrappedMessageType() const;
+  // Set the wrapped RPC for this RPC.
+  void setWrappedRpc(std::unique_ptr<RpcCommandBase> wrappedRpc);
+
+ private:
+  // message type
+  const rpc::MessageType messageType_;
+  // wrapped message
+  c10::intrusive_ptr<rpc::Message> wrappedMessage_;
+  std::unique_ptr<RpcCommandBase> wrappedRpc_;
+  rpc::MessageType wrappedMessageType_;
+  std::vector<torch::Tensor> tensors_;
+  const std::vector<torch::autograd::profiler::LegacyEvent> profiledEvents_;
+  const rpc::ProfilingId profilingId_;
+};
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/rref_backward_req.h

@@ -0,0 +1,39 @@
+#pragma once
+
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+#include <torch/csrc/distributed/rpc/types.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Internal system RPC to invoke distributed backward pass on remote nodes when
+// 'rref.backward()' is invoked.
+class TORCH_API RRefBackwardReq : public rpc::RpcCommandBase {
+ public:
+  RRefBackwardReq(
+      const rpc::RRefId& rrefId,
+      int64_t autogradContextId,
+      bool retainGraph = false);
+
+  const rpc::RRefId& getRRefId() const;
+
+  int64_t getAutogradContextId() const;
+
+  bool retainGraph() const;
+
+  // Serialization and deserialization methods.
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<RRefBackwardReq> fromMessage(
+      const rpc::Message& message);
+
+ private:
+  const rpc::RRefId rrefId_;
+  const int64_t autogradContextId_;
+  const bool retainGraph_;
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/autograd/rpc_messages/rref_backward_resp.h

@@ -0,0 +1,21 @@
+#pragma once
+
+#include <torch/csrc/distributed/rpc/message.h>
+#include <torch/csrc/distributed/rpc/rpc_command_base.h>
+
+namespace torch {
+namespace distributed {
+namespace autograd {
+
+// Response for the RRefBackwardReq.
+class TORCH_API RRefBackwardResp : public rpc::RpcCommandBase {
+ public:
+  RRefBackwardResp() = default;
+  c10::intrusive_ptr<rpc::Message> toMessageImpl() && override;
+  static std::unique_ptr<RRefBackwardResp> fromMessage(
+      const rpc::Message& message);
+};
+
+} // namespace autograd
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/Backend.hpp

@@ -0,0 +1,408 @@
+#pragma once
+
+#include <condition_variable>
+#include <memory>
+#include <mutex>
+#include <stdexcept>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include <ATen/ATen.h>
+#include <c10/macros/Macros.h>
+
+#include <torch/csrc/distributed/c10d/Types.hpp>
+#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <torch/csrc/distributed/c10d/Work.hpp>
+#include <torch/csrc/distributed/c10d/debug.h>
+
+constexpr auto kBackendDefaultTimeout =
+    std::chrono::milliseconds(30 * 60 * 1000);
+
+namespace c10d {
+
+class TORCH_API Backend : public torch::CustomClassHolder {
+ public:
+  // Backend Options is a base struct that defines the basic options
+  // when constructing a Backend. Each Backend subclass should
+  // extend this struct and define its options if it wants to provide more
+  // config options (beyond basic ones defined here) to end user.
+  struct TORCH_API Options : torch::CustomClassHolder {
+    explicit Options(
+        std::string backend,
+        std::chrono::milliseconds timeout = kBackendDefaultTimeout)
+        : timeout(timeout), backend(std::move(backend)) {}
+    ~Options() override = default;
+
+    std::chrono::milliseconds timeout;
+
+    // backend name
+    const std::string backend;
+  };
+
+  explicit Backend(int rank, int size);
+  ~Backend() override = 0;
+
+  int getRank() const {
+    return rank_;
+  }
+
+  int getSize() const {
+    return size_;
+  }
+
+  // Returns an unique opaque ID of this backend that can be used to correlate
+  // with its collectives.
+  int64_t getID() const {
+    return reinterpret_cast<std::intptr_t>(this);
+  }
+
+  virtual bool supportsSplitting() const {
+    return false;
+  }
+
+  virtual void startCoalescing() {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not implement startCoalescing"));
+  }
+
+  virtual c10::intrusive_ptr<Work> endCoalescing() {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ", getBackendName(), " does not implement endCoalescing"));
+  }
+
+  // Subclasses must override this method to return the backend name
+  virtual const std::string getBackendName() const {
+    TORCH_INTERNAL_ASSERT(false, "getBackendName is not implemented.");
+  };
+
+  virtual c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& /* tensors */,
+      const BroadcastOptions& /* opts */ = BroadcastOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support broadcast"));
+  }
+
+  virtual c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& /* tensors */,
+      const AllreduceOptions& /* opts */ = AllreduceOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support allreduce"));
+  }
+
+  virtual c10::intrusive_ptr<Work> allreduce_sparse(
+      std::vector<at::Tensor>& /* tensors */,
+      const AllreduceOptions& /* opts */ = AllreduceOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not support allreduce sparse"));
+  }
+
+  virtual c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& /* tensors */,
+      const AllreduceCoalescedOptions& /* opts */ =
+          AllreduceCoalescedOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not support allreduce_coalesced"));
+  }
+
+  virtual c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& /* tensors */,
+      const ReduceOptions& /* opts */ = ReduceOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support reduce"));
+  }
+
+  virtual c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& /* outputTensors */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support allgather"));
+  }
+
+  // Gathers a single tensor inputBuffer into a single buffer outputBuffer that
+  // is interpreted as a contiguous collection of size inputBuffer * WORLD_SIZE.
+  // For implementers of ProcessGroup API and advanced users only.
+  // Note: this function will be deprecated in near future.
+  virtual c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& /* outputBuffer */,
+      at::Tensor& /* inputBuffer */,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ", getBackendName(), " does not support _allgather_base"));
+  }
+
+  // This function is deprecated and will be moved out of Backend to comms:
+  // * do not add dependencies on this function,
+  // * do not implement it in your Backend, implement _allgather_base
+  //   instead.
+  virtual c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& /* outputTensorLists */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not support allgather_coalesced"));
+  }
+
+  // This function is a coalesced version of `allgather_into_tensor` (currently
+  // still named as `_allgather_base`). Each tensor in the vector corresponds to
+  // an input/output of one `allgather_into_tensor` operation.
+  virtual c10::intrusive_ptr<Work> allgather_into_tensor_coalesced(
+      std::vector<at::Tensor>& /* outputs */,
+      std::vector<at::Tensor>& /* inputs */,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not support allgather_into_tensor_coalesced"));
+  }
+
+  virtual c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& /* outputTensors */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const GatherOptions& /* opts */ = GatherOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support gather"));
+  }
+
+  virtual c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& /* outputTensors */,
+      std::vector<std::vector<at::Tensor>>& /* inputTensors */,
+      const ScatterOptions& /* opts */ = ScatterOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support scatter"));
+  }
+
+  virtual c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& /* outputTensors */,
+      std::vector<std::vector<at::Tensor>>& /* inputTensors */,
+      const ReduceScatterOptions& /* opts */ = ReduceScatterOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ", getBackendName(), " does not support reduce_scatter"));
+  }
+
+  virtual c10::intrusive_ptr<Work> _reduce_scatter_base(
+      at::Tensor& /* outputBuffer */,
+      at::Tensor& /* inputBuffer */,
+      const ReduceScatterOptions& /* opts */ = ReduceScatterOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not support _reduce_scatter_base"));
+  }
+
+  // This function is a coalesced version of `reduce_scatter_tensor` (currently
+  // still named as `_reduce_scatter_base`). Each tensor in the vector
+  // corresponds to an input/output of one `reduce_scatter_tensor` operation.
+  virtual c10::intrusive_ptr<Work> reduce_scatter_tensor_coalesced(
+      std::vector<at::Tensor>& /* outputs */,
+      std::vector<at::Tensor>& /* inputs */,
+      const ReduceScatterOptions& /* opts */ = ReduceScatterOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            getBackendName(),
+            " does not support reduce_scatter_tensor_coalesced"));
+  }
+
+  virtual c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& /* outputBuffer */,
+      at::Tensor& /* inputBuffer */,
+      std::vector<int64_t>& /* outputSplitSizes */,
+      std::vector<int64_t>& /* inputSplitSizes */,
+      const AllToAllOptions& /* opts */ = AllToAllOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ", getBackendName(), " does not support alltoall_base"));
+  }
+
+  virtual c10::intrusive_ptr<Work> alltoall(
+      std::vector<at::Tensor>& /* outputTensors */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const AllToAllOptions& opts = AllToAllOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support alltoall"));
+  }
+
+  virtual void monitoredBarrier(
+      const BarrierOptions& /* unused */,
+      bool /* unused */ = false) {
+    auto backendName = getBackendName();
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            backendName,
+            " does not support monitoredBarrier, only GLOO supports monitored barrier."));
+  }
+
+  // Agrees on an initial sequence number for the whole group by having rank 0
+  // create it and broadcast it to other ranks using the store. Only implemented
+  // for GLOO and NCCL backends currently.
+  virtual void setSequenceNumberForGroup() {
+    auto backendName = getBackendName();
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            backendName,
+            " does not yet support sequence numbers."));
+  }
+
+  // Retrieves the current sequence number for the whole group, which should be
+  // in sync. If the returned number is not consistent across the group, it
+  // may indicate that there is some sort of collective desynchronization.
+  virtual uint64_t getSequenceNumberForGroup() {
+    auto backendName = getBackendName();
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ",
+            backendName,
+            " does not yet support sequence numbers."));
+  }
+
+  virtual c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& /* tensors */,
+      int /* dstRank */,
+      int /* tag */) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support send"));
+  }
+
+  virtual c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& /* tensors */,
+      int /* srcRank */,
+      int /* tag */) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support recv"));
+  }
+
+  virtual c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& /* tensors */,
+      int /* tag */) {
+    TORCH_CHECK(
+        false,
+        c10::str(
+            "Backend ", getBackendName(), " does not support recvAnysource"));
+  }
+
+  virtual c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& /* opts */ = BarrierOptions()) {
+    TORCH_CHECK(
+        false,
+        c10::str("Backend ", getBackendName(), " does not support barrier"));
+  }
+
+  virtual void registerOnCompletionHook(
+      std::function<void(std::shared_ptr<WorkInfo>)>&& hook) {
+    TORCH_CHECK(
+        false,
+        "Only ProcessGrouppNCCL supports onCompletion hook, but got ",
+        getBackendName(),
+        " backend.");
+  }
+
+  virtual void waitForPendingWorks() {
+    TORCH_CHECK(
+        false,
+        "Only ProcessGrouppNCCL supports waitForPendingWorks, but got ",
+        getBackendName(),
+        " backend.");
+  }
+
+  virtual void enableCollectivesTiming() {
+    TORCH_CHECK(
+        false,
+        "Backend ",
+        getBackendName(),
+        " is missing implementation of enableCollectivesTiming.");
+  }
+
+  bool hasHooks() const {
+    return onCompletionHook_ != nullptr;
+  }
+
+  // Do not call this directly, use ProcessGroup::setGroupName instead.
+  void setGroupName(const std::string& name) {
+    pg_name_ = name;
+  }
+
+  const std::string& getGroupName() const {
+    return pg_name_;
+  }
+
+  // See similar functions in ProcessGroup.hpp for context.
+  c10::optional<at::Device> getBoundDeviceId() const {
+    return bound_device_id_;
+  }
+
+  // Perform an eager connect to the specified device if the backend supports
+  // it.
+  virtual void eagerConnectSingleDevice(at::Device device) {
+    // no-op in the default case; this is an optimization some
+    // backends may perform
+  }
+
+  void setBoundDeviceId(c10::optional<at::Device> device) {
+    if (device) {
+      TORCH_CHECK(device->has_index(), "setBoundDeviceId must have an index");
+    }
+    bound_device_id_ = device;
+  }
+
+ protected:
+  // Implementations of this interface need to call this to setup
+  // appropriate logging etc.
+  void init();
+
+  const int rank_;
+  const int size_;
+  // Debug level setting. It is parsed once when ProcessGroup is constructed and
+  // remains the same across use of this process group.
+  DebugLevel dist_debug_level_;
+  std::string pg_name_;
+
+  std::function<void(std::shared_ptr<WorkInfo>)> onCompletionHook_;
+
+  c10::optional<at::Device> bound_device_id_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/FakeProcessGroup.hpp

@@ -0,0 +1,186 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/Backend.hpp>
+
+namespace c10d {
+
+class FakeWork : public Work {
+ public:
+  bool wait(std::chrono::milliseconds timeout) override {
+    return true;
+  }
+
+  c10::intrusive_ptr<c10::ivalue::Future> getFuture() override {
+    auto fut = c10::make_intrusive<c10::ivalue::Future>(c10::NoneType::get());
+    fut->markCompleted();
+    return fut;
+  }
+};
+
+class FakeProcessGroup : public Backend {
+ public:
+  FakeProcessGroup(int rank, int size) : Backend(rank, size) {}
+
+  c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& /* tensors */,
+      const BroadcastOptions& /* opts */ = BroadcastOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& /* tensors */,
+      const AllreduceOptions& /* opts */ = AllreduceOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> allreduce_sparse(
+      std::vector<at::Tensor>& /* tensors */,
+      const AllreduceOptions& /* opts */ = AllreduceOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& /* tensors */,
+      const AllreduceCoalescedOptions& /* opts */ =
+          AllreduceCoalescedOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& /* tensors */,
+      const ReduceOptions& /* opts */ = ReduceOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  // NOTE [allgather on FakeProcessGroup]
+  // Assume each rank have the same input tensor so we just copy to the results
+  // since it's not a real allgather, we simply make this copying logic to let
+  // some simple validation works (i.e. calling allgather to see if each rank
+  // have the same tensor or not).
+  //
+  // NOTE: in general it's not good form to try to make FakeProcessGroup work
+  // with real data, but the reasoning here is that we want FakeProcessGroup to
+  // work with DeviceMesh's init code that have the data validation, which
+  // makes it worth the tradeoff.
+  c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) override {
+    for (auto& tensor : outputTensors[0]) {
+      tensor.copy_(inputTensors[0]);
+    }
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) override {
+    auto chunks = outputBuffer.chunk(size_);
+    for (auto& tensor : chunks) {
+      tensor.copy_(inputBuffer);
+    }
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& /* outputTensorLists */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> allgather_into_tensor_coalesced(
+      std::vector<at::Tensor>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const AllgatherOptions& /* opts */ = AllgatherOptions()) override {
+    for (size_t i = 0; i < outputs.size(); ++i) {
+      auto chunks = outputs[i].chunk(size_);
+      for (auto& chunk : chunks) {
+        chunk.copy_(inputs[i]);
+      }
+    }
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& /* outputTensors */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const GatherOptions& /* opts */ = GatherOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& /* outputTensors */,
+      std::vector<std::vector<at::Tensor>>& /* inputTensors */,
+      const ScatterOptions& /* opts */ = ScatterOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& /* outputTensors */,
+      std::vector<std::vector<at::Tensor>>& /* inputTensors */,
+      const ReduceScatterOptions& /* opts */ =
+          ReduceScatterOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> _reduce_scatter_base(
+      at::Tensor& /* outputBuffer */,
+      at::Tensor& /* inputBuffer */,
+      const ReduceScatterOptions& /* opts */ =
+          ReduceScatterOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> reduce_scatter_tensor_coalesced(
+      std::vector<at::Tensor>& /* outputs */,
+      std::vector<at::Tensor>& /* inputs */,
+      const ReduceScatterOptions& /* opts */ =
+          ReduceScatterOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& /* outputBuffer */,
+      at::Tensor& /* inputBuffer */,
+      std::vector<int64_t>& /* outputSplitSizes */,
+      std::vector<int64_t>& /* inputSplitSizes */,
+      const AllToAllOptions& /* opts */ = AllToAllOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> alltoall(
+      std::vector<at::Tensor>& /* outputTensors */,
+      std::vector<at::Tensor>& /* inputTensors */,
+      const AllToAllOptions& opts = AllToAllOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& /* tensors */,
+      int /* dstRank */,
+      int /* tag */) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& /* tensors */,
+      int /* srcRank */,
+      int /* tag */) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& /* tensors */,
+      int /* tag */) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+
+  c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& /* opts */ = BarrierOptions()) override {
+    return c10::make_intrusive<FakeWork>();
+  }
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/FileStore.hpp

@@ -0,0 +1,63 @@
+#pragma once
+
+#include <sys/types.h>
+
+#include <mutex>
+#include <unordered_map>
+
+#include <torch/csrc/distributed/c10d/Store.hpp>
+
+namespace c10d {
+
+class TORCH_API FileStore : public Store {
+ public:
+  explicit FileStore(std::string path, int numWorkers);
+
+  ~FileStore() override;
+
+  void set(const std::string& key, const std::vector<uint8_t>& value) override;
+
+  std::vector<uint8_t> compareSet(
+      const std::string& key,
+      const std::vector<uint8_t>& expectedValue,
+      const std::vector<uint8_t>& desiredValue) override;
+
+  std::vector<uint8_t> get(const std::string& key) override;
+
+  int64_t add(const std::string& key, int64_t value) override;
+
+  int64_t getNumKeys() override;
+
+  bool deleteKey(const std::string& key) override;
+
+  bool check(const std::vector<std::string>& keys) override;
+
+  void wait(const std::vector<std::string>& keys) override;
+
+  void wait(
+      const std::vector<std::string>& keys,
+      const std::chrono::milliseconds& timeout) override;
+
+  // Returns the path used by the FileStore.
+  const std::string& getPath() const noexcept {
+    return path_;
+  }
+
+ protected:
+  int64_t addHelper(const std::string& key, int64_t i);
+
+  std::string path_;
+  off_t pos_{0};
+
+  int numWorkers_;
+  const std::string cleanupKey_;
+  const std::string refCountKey_;
+  const std::string regularPrefix_;
+  const std::string deletePrefix_;
+
+  std::unordered_map<std::string, std::vector<uint8_t>> cache_;
+
+  std::mutex activeFileOpLock_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/GroupRegistry.hpp

@@ -0,0 +1,22 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+
+namespace c10d {
+
+C10_EXPORT void set_thread_isolation_mode(bool enable);
+
+bool get_thread_isolation_mode();
+
+C10_EXPORT void register_process_group(
+    const std::string& group_name,
+    c10::intrusive_ptr<c10d::ProcessGroup> group);
+
+C10_EXPORT c10::intrusive_ptr<c10d::ProcessGroup> resolve_process_group(
+    const std::string& group_name);
+
+C10_EXPORT void unregister_process_group(const std::string& group_name);
+
+C10_EXPORT void unregister_all_process_groups();
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/HashStore.hpp

@@ -0,0 +1,61 @@
+#pragma once
+
+#include <sys/types.h>
+
+#include <condition_variable>
+#include <mutex>
+#include <unordered_map>
+
+#include <torch/csrc/distributed/c10d/Store.hpp>
+
+namespace c10d {
+
+class TORCH_API HashStore : public Store {
+ public:
+  ~HashStore() override = default;
+
+  void set(const std::string& key, const std::vector<uint8_t>& data) override;
+
+  std::vector<uint8_t> compareSet(
+      const std::string& key,
+      const std::vector<uint8_t>& expectedValue,
+      const std::vector<uint8_t>& desiredValue) override;
+
+  std::vector<uint8_t> get(const std::string& key) override;
+
+  void wait(const std::vector<std::string>& keys) override {
+    wait(keys, Store::kDefaultTimeout);
+  }
+
+  void wait(
+      const std::vector<std::string>& keys,
+      const std::chrono::milliseconds& timeout) override;
+
+  int64_t add(const std::string& key, int64_t value) override;
+
+  int64_t getNumKeys() override;
+
+  bool check(const std::vector<std::string>& keys) override;
+
+  bool deleteKey(const std::string& key) override;
+
+  void append(const std::string& key, const std::vector<uint8_t>& value)
+      override;
+
+  std::vector<std::vector<uint8_t>> multiGet(
+      const std::vector<std::string>& keys) override;
+
+  void multiSet(
+      const std::vector<std::string>& keys,
+      const std::vector<std::vector<uint8_t>>& values) override;
+
+  // Returns true if this store support append, multiGet and multiSet
+  bool hasExtendedApi() const override;
+
+ protected:
+  std::unordered_map<std::string, std::vector<uint8_t>> map_;
+  std::mutex m_;
+  std::condition_variable cv_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/NCCLUtils.hpp

@@ -0,0 +1,529 @@
+#pragma once
+
+#ifdef USE_C10D_NCCL
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <memory>
+#include <mutex>
+#include <thread>
+
+#include <ATen/ATen.h>
+#include <c10/util/Exception.h>
+#include <c10/util/Optional.h>
+#include <nccl.h>
+
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 14)
+#define NCCL_HAS_COMM_NONBLOCKING
+#endif
+
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 18)
+#define NCCL_HAS_COMM_SPLIT
+#endif
+
+// ncclGetLastError() is enabled only for NCCL versions 2.13+
+// ncclRemoteError only exists in NCCL versions 2.13+
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 13)
+#define ENABLE_NCCL_GET_LAST_ERROR
+#define NCCL_REMOTE_ERROR
+#elif defined(NCCL_MAJOR) && (NCCL_MAJOR >= 3)
+#define ENABLE_NCCL_GET_LAST_ERROR
+#define NCCL_REMOTE_ERROR
+#endif
+
+// Error checking is enabled only for NCCL versions 2.4+ since ncclCommAbort()
+// and ncclCommGetAsyncError() are not supported in earlier versions.
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 4)
+#define ENABLE_NCCL_ERROR_CHECKING
+#elif defined(NCCL_MAJOR) && (NCCL_MAJOR >= 3)
+#define ENABLE_NCCL_ERROR_CHECKING
+#endif
+
+// P2P is enabled only for NCCL versions 2.7+ since ncclSend()
+// and ncclRecv() are not supported in earlier versions.
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 7)
+#define ENABLE_NCCL_P2P_SUPPORT
+#elif defined(NCCL_MAJOR) && (NCCL_MAJOR >= 3)
+#define ENABLE_NCCL_P2P_SUPPORT
+#endif
+
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 11)
+#define ENABLE_NCCL_PREMUL_SUM_SUPPORT
+#elif defined(NCCL_MAJOR) && (NCCL_MAJOR >= 3)
+#define ENABLE_NCCL_PREMUL_SUM_SUPPORT
+#endif
+
+#if defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+    (NCCL_MINOR >= 17)
+#define NCCL_HAS_COMM_CTA_CGA
+#elif defined(NCCL_MAJOR) && (NCCL_MAJOR >= 3)
+#define NCCL_HAS_COMM_CTA_CGA
+#endif
+
+#if defined(NCCL_REGISTRATION_SUPPORTED) ||                              \
+    ((defined(NCCL_MAJOR) && (NCCL_MAJOR == 2) && defined(NCCL_MINOR) && \
+      (NCCL_MINOR >= 19)))
+#define NCCL_HAS_COMM_REGISTER
+#elif defined(NCCL_MAJOR) && (NCCL_MAJOR >= 3)
+#define NCCL_HAS_COMM_REGISTER
+#endif
+
+// Macro to throw on a non-successful NCCL return value.
+#define C10D_NCCL_CHECK(cmd, failureReason)                                   \
+  do {                                                                        \
+    ncclResult_t result = cmd;                                                \
+    if (result != ncclSuccess) {                                              \
+      std::string err = "NCCL error in: " + std::string(__FILE__) + ":" +     \
+          std::to_string(__LINE__) + ", " + ncclGetErrorWithVersion(result) + \
+          "\n" + getNcclErrorDetailStr(result, failureReason);                \
+      TORCH_CHECK_WITH(DistBackendError, false, err);                         \
+    }                                                                         \
+  } while (0)
+
+// Macro to throw on a non-successful NCCL return value for NONBLOCKING calls.
+#define C10D_NCCL_CHECK_NONBLOCKING(cmd, failureReason)                       \
+  do {                                                                        \
+    ncclResult_t result = cmd;                                                \
+    if (result != ncclSuccess && result != ncclInProgress) {                  \
+      std::string err = "NCCL error in: " + std::string(__FILE__) + ":" +     \
+          std::to_string(__LINE__) + ", " + ncclGetErrorWithVersion(result) + \
+          "\n" + getNcclErrorDetailStr(result, failureReason);                \
+      TORCH_CHECK_WITH(DistBackendError, false, err);                         \
+    }                                                                         \
+  } while (0)
+
+// Macro to throw on a non-successful NCCL return value, non-blocking.
+#define C10D_NCCL_CHECK_TIMEOUT(cmd, comm, failureReason)                     \
+  ncclResult_t result = cmd;                                                  \
+  auto startTimepoint = std::chrono::steady_clock::now();                     \
+  while (result == ncclInProgress) {                                          \
+    if (nccl_nonblocking_timeout() > 0) {                                     \
+      auto currentTimepoint = std::chrono::steady_clock::now();               \
+      auto timeElapsed = std::chrono::duration_cast<std::chrono::seconds>(    \
+                             currentTimepoint - startTimepoint)               \
+                             .count();                                        \
+      if (timeElapsed > nccl_nonblocking_timeout()) {                         \
+        std::string err = "NCCL timeout in: " + std::string(__FILE__) + ":" + \
+            std::to_string(__LINE__) + ", " +                                 \
+            ncclGetErrorWithVersion(result) + "\n" +                          \
+            getNcclErrorDetailStr(result, failureReason);                     \
+        TORCH_CHECK_WITH(DistBackendError, false, err);                       \
+      }                                                                       \
+    }                                                                         \
+    ncclCommGetAsyncError(comm, &result);                                     \
+  }                                                                           \
+  if (result != ncclSuccess) {                                                \
+    std::string err = "NCCL error in: " + std::string(__FILE__) + ":" +       \
+        std::to_string(__LINE__) + ", " + ncclGetErrorWithVersion(result) +   \
+        "\n" + getNcclErrorDetailStr(result, failureReason);                  \
+    TORCH_CHECK_WITH(DistBackendError, false, err);                           \
+  }
+
+#define C10D_NCCL_CHECK_TIMEOUT_GROUPEND(cmd, comm, failureReason)           \
+  ncclResult_t state = cmd;                                                  \
+  auto startTimepoint = std::chrono::steady_clock::now();                    \
+  if (state == ncclInProgress) {                                             \
+    do {                                                                     \
+      if (nccl_nonblocking_timeout() > 0) {                                  \
+        auto currentTimepoint = std::chrono::steady_clock::now();            \
+        auto timeElapsed = std::chrono::duration_cast<std::chrono::seconds>( \
+                               currentTimepoint - startTimepoint)            \
+                               .count();                                     \
+        if (timeElapsed > nccl_nonblocking_timeout()) {                      \
+          std::string err = "NCCL timeout in: " + std::string(__FILE__) +    \
+              ":" + std::to_string(__LINE__) + ", " +                        \
+              ncclGetErrorWithVersion(state) + "\n" +                        \
+              getNcclErrorDetailStr(state, failureReason);                   \
+          TORCH_CHECK_WITH(DistBackendError, false, err);                    \
+        }                                                                    \
+      }                                                                      \
+      ncclCommGetAsyncError(comm->getNcclComm(), &state);                    \
+    } while (state == ncclInProgress);                                       \
+  }                                                                          \
+  if (state != ncclSuccess) {                                                \
+    std::string err = "NCCL error in: " + std::string(__FILE__) + ":" +      \
+        std::to_string(__LINE__) + ", " + ncclGetErrorWithVersion(state) +   \
+        "\n" + getNcclErrorDetailStr(state, failureReason);                  \
+    TORCH_CHECK_WITH(DistBackendError, false, err);                          \
+  }
+
+// Macro to print and abort on a non-successful NCCL return value.
+#define C10D_NCCL_ASSERT(cmd)                            \
+  do {                                                   \
+    ncclResult_t result = cmd;                           \
+    if (result != ncclSuccess) {                         \
+      std::string err = ncclGetErrorWithVersion(result); \
+      fprintf(                                           \
+          stderr,                                        \
+          "NCCL error in: %s:%d, %s\n",                  \
+          __FILE__,                                      \
+          __LINE__,                                      \
+          err.c_str());                                  \
+      abort();                                           \
+    }                                                    \
+  } while (0)
+
+namespace c10d {
+
+TORCH_API size_t hashTensors(const std::vector<at::Tensor>& tensors);
+std::string getNcclVersion();
+std::string ncclGetErrorWithVersion(ncclResult_t error);
+bool nccl_use_nonblocking();
+int nccl_nonblocking_timeout();
+
+// Provides additional detail into NCCL error codes based on when these are
+// thrown in the NCCL codebase.
+std::string getNcclErrorDetailStr(
+    ncclResult_t error,
+    c10::optional<std::string> processGroupFailureReason = c10::nullopt);
+
+// Write NCCL debug info to local disk or any storage users define.
+// There are some constrains we set for the debug info writer:
+// 1. The writer should only be registered once.
+// 2. Once registered, users cannot change it including un-register.
+// 3. It is recommended to register the customized writer in the trainer setup,
+//    If users don't register before calling launchAsyncDebugDump, then users
+//    lose the chance to register (and the default writer will be
+//    auto-registered).
+class TORCH_API DebugInfoWriter {
+ public:
+  virtual ~DebugInfoWriter();
+  virtual void write(const std::string& ncclTrace);
+  static DebugInfoWriter& getWriter(int rank);
+  static void registerWriter(std::unique_ptr<DebugInfoWriter> writer);
+
+ protected:
+  DebugInfoWriter(std::string namePrefix, int rank) {
+    filename_ = c10::str(namePrefix, rank);
+  }
+  std::string filename_;
+
+ private:
+  static std::unique_ptr<DebugInfoWriter> writer_;
+  static std::atomic<bool> hasWriterRegistered_;
+};
+
+// RAII wrapper for NCCL communicator
+class NCCLComm {
+ public:
+  explicit NCCLComm(ncclComm_t ncclComm)
+      : ncclComm_(ncclComm),
+        aborted_(false),
+        ncclAsyncErr_(ncclSuccess),
+        commFailureReason_(c10::nullopt),
+        initialized_(false) {}
+
+  NCCLComm() : NCCLComm(nullptr) {}
+
+  ~NCCLComm() noexcept {
+    // Add lock in this destructor, as aborted_ needs to be read after memory
+    // barrier here.
+    std::unique_lock<std::mutex> lock(mutex_);
+    if (ncclComm_ && !aborted_) {
+#ifdef ENABLE_NCCL_ERROR_CHECKING
+      // Use ncclCommAbort instead of ncclCommDestroy here since
+      // ncclCommDestroy could block forever waiting for work to complete on
+      // the communicator.
+      C10D_NCCL_ASSERT(::ncclCommAbort(ncclComm_));
+#else
+      C10D_NCCL_ASSERT(::ncclCommDestroy(ncclComm_));
+#endif
+    }
+  }
+
+  static std::shared_ptr<NCCLComm> create(
+      int numRanks,
+      int rank,
+      ncclUniqueId commId) {
+    auto comm = std::make_shared<NCCLComm>();
+    C10D_NCCL_CHECK(
+        ncclCommInitRank(&(comm->ncclComm_), numRanks, commId, rank),
+        c10::nullopt);
+    comm->ncclId_ = commId;
+    comm->rank_ = rank;
+    comm->initialized_ = true;
+    return comm;
+  }
+
+#ifdef NCCL_HAS_COMM_NONBLOCKING
+  static std::shared_ptr<NCCLComm> create(
+      int numRanks,
+      int rank,
+      ncclUniqueId commId,
+      ncclConfig_t& config) {
+    auto comm = std::make_shared<NCCLComm>();
+    bool isInitialized = false;
+    if (nccl_use_nonblocking()) {
+      config.blocking = 0;
+      LOG(INFO) << "Rank " << rank
+                << ": creating NCCL communicator in nonblocking mode";
+      C10D_NCCL_CHECK_NONBLOCKING(
+          ncclCommInitRankConfig(
+              &(comm->ncclComm_), numRanks, commId, rank, &config),
+          c10::nullopt);
+    } else {
+      C10D_NCCL_CHECK(
+          ncclCommInitRankConfig(
+              &(comm->ncclComm_), numRanks, commId, rank, &config),
+          c10::nullopt);
+      // under blocking mode, comm is initialized after NCCL CHECK
+      isInitialized = true;
+    }
+    comm->ncclId_ = commId;
+    comm->rank_ = rank;
+    comm->initialized_ = isInitialized;
+    return comm;
+  }
+#endif
+
+#ifdef NCCL_HAS_COMM_SPLIT
+  static std::shared_ptr<NCCLComm> split(
+      NCCLComm* source,
+      int color_id,
+      int rank,
+      ncclConfig_t& config) {
+    auto comm = std::make_shared<NCCLComm>();
+    C10D_NCCL_CHECK(
+        ncclCommSplit(
+            source->ncclComm_, color_id, rank, &(comm->ncclComm_), &config),
+        c10::nullopt);
+    ++source->ncclCommSplitCounter_;
+    comm->rank_ = rank;
+    return comm;
+  }
+#endif
+
+#if defined(IS_NCCL_EXP) && defined(NCCL_COMM_DUMP)
+  std::unordered_map<std::string, std::string> ncclCommDump() {
+    std::unordered_map<std::string, std::string> dump;
+    if (isAborted()) {
+      LOG(INFO) << "Communicator was aborted before trying to dump its state.";
+      return dump;
+    }
+    C10D_NCCL_CHECK(::ncclCommDump(ncclComm_, dump), c10::nullopt);
+    return dump;
+  }
+#endif
+
+  ncclUniqueId getNcclId() {
+    return ncclId_;
+  }
+
+  // Must not be copyable
+  NCCLComm(const NCCLComm&) = delete;
+  NCCLComm& operator=(const NCCLComm&) = delete;
+
+  // Do not support move assignment as there is no valid use case
+  NCCLComm& operator=(NCCLComm&& other) = delete;
+
+  // Move constructable
+  NCCLComm(NCCLComm&& other) {
+    // Using other's lock, as it reads other's states
+    // Can not use this.mutex_, as this object is being constructed.
+    std::unique_lock<std::mutex> lock(other.mutex_);
+    std::swap(ncclComm_, other.ncclComm_);
+    std::swap(aborted_, other.aborted_);
+    std::swap(ncclAsyncErr_, other.ncclAsyncErr_);
+    std::swap(initialized_, other.initialized_);
+  }
+
+  ncclComm_t getNcclComm();
+
+  c10::optional<std::string> getNcclCommFailureReason() const {
+    std::unique_lock<std::mutex> lock(mutex_);
+    return commFailureReason_;
+  }
+
+  void ncclCommAbort(
+      c10::optional<std::string> commFailureReason = c10::nullopt) {
+    std::unique_lock<std::mutex> lock(mutex_);
+#ifdef ENABLE_NCCL_ERROR_CHECKING
+    if (aborted_) {
+      // Should not abort twice.
+      return;
+    }
+
+#ifdef NCCL_HAS_COMM_REGISTER
+    // Deregister all registered segments before aborting.
+    for (auto& it : registeredSegmentHandles_) {
+      void* handle = it.second;
+      C10D_NCCL_CHECK(
+          ::ncclCommDeregister(ncclComm_, handle),
+          c10::str(
+              "Failed to deregister segment handle ",
+              handle,
+              " on ncclComm_ ",
+              ncclComm_));
+    }
+    registeredSegmentHandles_.clear();
+#endif
+
+    // Set true failure reason if provided by ProcessGroupNCCL (e.g. work
+    // timeout)
+    commFailureReason_ = commFailureReason;
+    LOG(INFO) << "Aborting ncclComm_ " << ncclComm_ << " with reason: "
+              << (commFailureReason ? *commFailureReason
+                                    : "No abort reason provided.");
+#ifndef NCCL_HAS_COMM_NONBLOCKING
+    C10D_NCCL_CHECK(::ncclCommAbort(ncclComm_), commFailureReason_);
+#else
+    C10D_NCCL_CHECK_TIMEOUT(
+        ::ncclCommAbort(ncclComm_), ncclComm_, commFailureReason_);
+#endif
+    aborted_ = true;
+    ncclComm_ = nullptr;
+
+    // Set an appropriate error so that we avoid using the communicator.
+    if (ncclAsyncErr_ == ncclSuccess) {
+      ncclAsyncErr_ = ncclSystemError;
+    }
+#else
+    // This is a NOOP, if error checks are disabled.
+    return;
+#endif
+  }
+
+  bool isAborted() const {
+    std::unique_lock<std::mutex> lock(mutex_);
+    return aborted_;
+  }
+
+  uint64_t getCommSplitCounter() const {
+    return ncclCommSplitCounter_;
+  }
+
+  ncclResult_t checkForNcclError() {
+    std::unique_lock<std::mutex> lock(mutex_);
+#ifdef ENABLE_NCCL_ERROR_CHECKING
+    if (ncclAsyncErr_ != ncclSuccess) {
+      return ncclAsyncErr_;
+    }
+    C10D_NCCL_CHECK(
+        ncclCommGetAsyncError(ncclComm_, &ncclAsyncErr_), commFailureReason_);
+    return ncclAsyncErr_;
+#else
+    // Always return success, if error checks are disabled.
+    return ncclSuccess;
+#endif
+  }
+
+  ncclResult_t registerSegment(void* ptr, size_t size) {
+    std::unique_lock<std::mutex> lock(mutex_);
+#ifdef NCCL_HAS_COMM_REGISTER
+    // We register only segments from cache allocator
+    // which are guaranteed to be with disjoint addr ranges. Thus, a ptr always
+    // maps to a unique handle and should not be registered before the current
+    // ptr is deregistered and freed.
+    TORCH_CHECK(
+        registeredSegmentHandles_.count(ptr) == 0,
+        "Segment with ptr ",
+        ptr,
+        " has already been registered on ncclComm_ ",
+        ncclComm_);
+
+    void* handle;
+    C10D_NCCL_CHECK(
+        ncclCommRegister(ncclComm_, ptr, size, &handle),
+        c10::str(
+            "Failed to register segment with ptr ",
+            ptr,
+            ", size ",
+            size,
+            " on ncclComm_ ",
+            ncclComm_));
+    registeredSegmentHandles_[ptr] = handle;
+    return ncclSuccess;
+#else
+    return ncclInvalidUsage;
+#endif
+  }
+
+  ncclResult_t deregisterSegment(void* ptr) {
+    std::unique_lock<std::mutex> lock(mutex_);
+#ifdef NCCL_HAS_COMM_REGISTER
+    TORCH_CHECK(
+        registeredSegmentHandles_.count(ptr) == 1,
+        "Segment with ptr ",
+        ptr,
+        " is not registered on ncclComm_ ",
+        ncclComm_);
+
+    void* handle = registeredSegmentHandles_[ptr];
+    C10D_NCCL_CHECK(
+        ncclCommDeregister(ncclComm_, handle),
+        c10::str(
+            "Failed to deregister segment handle ",
+            handle,
+            ", with ptr ",
+            ptr,
+            " on ncclComm_ ",
+            ncclComm_));
+    registeredSegmentHandles_.erase(ptr);
+    return ncclSuccess;
+#else
+    return ncclInvalidUsage;
+#endif
+  }
+
+  friend class ProcessGroupNCCL;
+
+ protected:
+  // a helper function to wait until the communicator is initialized;
+  void waitUntilInitialized(int timeoutSecs);
+  ncclComm_t ncclComm_;
+  // Unique nccl_id for this communicator.
+  ncclUniqueId ncclId_;
+  bool aborted_;
+  uint64_t ncclCommSplitCounter_{0};
+  ncclResult_t ncclAsyncErr_;
+  mutable std::mutex mutex_;
+  // Rank that this communicator corresponds to.
+  int rank_;
+  // Optional reason for communicator failure, provided by ProcessGroupNCCL for
+  // better error messaging.
+  c10::optional<std::string> commFailureReason_;
+  bool initialized_{false};
+#ifdef NCCL_HAS_COMM_REGISTER
+  // Stores handlers for tensors registered by NCCL
+  std::unordered_map<void*, void*> registeredSegmentHandles_;
+#endif
+};
+
+// Helper that automatically cleans up premul sums.
+struct ncclRedOpRAII {
+  ncclRedOpRAII() = default;
+  ncclRedOpRAII(ncclRedOp_t op) : op_(op) {}
+  ncclRedOpRAII(ncclRedOp_t op, ncclComm_t comm)
+      : op_(op), comm_(comm), premul_sum_(true) {}
+  ncclRedOpRAII(const ncclRedOpRAII&) = delete;
+  ncclRedOpRAII& operator=(const ncclRedOpRAII&) = delete;
+  ncclRedOpRAII(ncclRedOpRAII&& tmp) : ncclRedOpRAII() {
+    std::swap(tmp.op_, this->op_);
+    std::swap(tmp.comm_, this->comm_);
+    std::swap(tmp.premul_sum_, this->premul_sum_);
+  }
+#if defined(ENABLE_NCCL_PREMUL_SUM_SUPPORT)
+  ~ncclRedOpRAII() {
+    if (premul_sum_) {
+      ncclRedOpDestroy(op_, comm_);
+    }
+  }
+#endif
+  operator ncclRedOp_t() const {
+    return op_;
+  }
+  ncclRedOp_t op_;
+  ncclComm_t comm_;
+  bool premul_sum_ = false;
+};
+
+} // namespace c10d
+
+#endif // USE_C10D_NCCL

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/ParamCommsUtils.hpp

@@ -0,0 +1,176 @@
+#pragma once
+
+#include <ATen/core/ivalue.h>
+#include <ATen/record_function.h>
+#include <c10/macros/Macros.h>
+#include <c10/util/ThreadLocalDebugInfo.h>
+#include <string>
+#include <vector>
+
+namespace torch {
+
+class TORCH_API ParamCommsDebugInfo : public c10::DebugInfoBase {
+ public:
+  ParamCommsDebugInfo() = default;
+  ParamCommsDebugInfo(
+      int pgId,
+      int rank,
+      std::string&& colName,
+      int inNelems,
+      int outNelems,
+      at::ScalarType dType,
+      std::vector<int64_t> inSplitSizes,
+      std::vector<int64_t> outSplitSizes,
+      int globalRankStart,
+      int globalRankStride,
+      int worldSize);
+
+  ~ParamCommsDebugInfo() override = default;
+
+  int getProcessGroupId() const {
+    return pgId_;
+  }
+
+  int getRank() const {
+    return rank_;
+  }
+
+  int getWorldSize() const {
+    return worldSize_;
+  }
+
+  int getGlobalRankStart() const {
+    return globalRankStart_;
+  }
+
+  int getGlobalRankStride() const {
+    return globalRankStride_;
+  }
+
+  const std::string getColumnName() const {
+    return columnName_;
+  }
+
+  int getInMessageNelems() const {
+    return inMessageNelems_;
+  }
+
+  int getOutMessageNelems() const {
+    return outMessageNelems_;
+  }
+
+  at::ScalarType getDType() const {
+    return dType_;
+  }
+
+  const std::vector<int64_t>& getInputSplitSizes() const {
+    return inputSplitSizes_;
+  }
+
+  const std::vector<int64_t>& getOutputSplitSizes() const {
+    return outputSplitSizes_;
+  }
+
+  const std::vector<int64_t>& getGroupRanks() const {
+    return groupRanks_;
+  }
+
+ private:
+  int pgId_{};
+  int rank_{};
+  int worldSize_{};
+  std::string columnName_;
+  int inMessageNelems_{};
+  int outMessageNelems_{};
+  at::ScalarType dType_ = at::kByte;
+  std::vector<int64_t> inputSplitSizes_;
+  std::vector<int64_t> outputSplitSizes_;
+  int globalRankStart_;
+  int globalRankStride_;
+  std::vector<int64_t> groupRanks_{};
+};
+
+#define RECORD_PARAM_COMMS(                                                    \
+    seq,                                                                       \
+    pgId,                                                                      \
+    rank,                                                                      \
+    colName,                                                                   \
+    inNelems,                                                                  \
+    outNelems,                                                                 \
+    dType,                                                                     \
+    inSplitSizes,                                                              \
+    outSplitSizes,                                                             \
+    globalRankStart,                                                           \
+    globalRankStride,                                                          \
+    worldSize)                                                                 \
+  auto paramCommsInfo = std::make_shared<torch::ParamCommsDebugInfo>(          \
+      pgId,                                                                    \
+      rank,                                                                    \
+      colName,                                                                 \
+      inNelems,                                                                \
+      outNelems,                                                               \
+      dType,                                                                   \
+      inSplitSizes,                                                            \
+      outSplitSizes,                                                           \
+      globalRankStart,                                                         \
+      globalRankStride,                                                        \
+      worldSize);                                                              \
+  c10::DebugInfoGuard g(c10::DebugInfoKind::PARAM_COMMS_INFO, paramCommsInfo); \
+  std::initializer_list<const c10::IValue> paramList = {                       \
+      c10::IValue(seq),                                                        \
+      pgId,                                                                    \
+      rank,                                                                    \
+      colName,                                                                 \
+      inSplitSizes,                                                            \
+      outSplitSizes,                                                           \
+      globalRankStart,                                                         \
+      globalRankStride,                                                        \
+      worldSize};                                                              \
+  c10::ArrayRef<const c10::IValue> paramInputs(paramList);                     \
+  RECORD_FUNCTION(at::kParamCommsCallName, paramInputs);
+
+#define RECORD_PARAM_COMMS_DATA(                                               \
+    seq,                                                                       \
+    pgId,                                                                      \
+    InputTensors,                                                              \
+    OutputTensors,                                                             \
+    rank,                                                                      \
+    colName,                                                                   \
+    inNelems,                                                                  \
+    outNelems,                                                                 \
+    dType,                                                                     \
+    inSplitSizes,                                                              \
+    outSplitSizes,                                                             \
+    globalRankStart,                                                           \
+    globalRankStride,                                                          \
+    worldSize)                                                                 \
+  auto paramCommsInfo = std::make_shared<torch::ParamCommsDebugInfo>(          \
+      pgId,                                                                    \
+      rank,                                                                    \
+      colName,                                                                 \
+      inNelems,                                                                \
+      outNelems,                                                               \
+      dType,                                                                   \
+      inSplitSizes,                                                            \
+      outSplitSizes,                                                           \
+      globalRankStart,                                                         \
+      globalRankStride,                                                        \
+      worldSize);                                                              \
+  c10::DebugInfoGuard g(c10::DebugInfoKind::PARAM_COMMS_INFO, paramCommsInfo); \
+  std::initializer_list<const c10::IValue> paramList = {                       \
+      c10::IValue(InputTensors),                                               \
+      c10::IValue(seq),                                                        \
+      pgId,                                                                    \
+      rank,                                                                    \
+      colName,                                                                 \
+      inSplitSizes,                                                            \
+      outSplitSizes,                                                           \
+      globalRankStart,                                                         \
+      globalRankStride,                                                        \
+      worldSize};                                                              \
+  c10::ArrayRef<const c10::IValue> paramInputs(paramList);                     \
+  RECORD_FUNCTION_WITH_INPUTS_OUTPUTS(                                         \
+      at::kParamCommsCallName,                                                 \
+      paramInputs,                                                             \
+      std::vector<c10::IValue>(1, c10::IValue(OutputTensors)));
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/ProcessGroup.hpp

@@ -0,0 +1,743 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/Backend.hpp>
+#include <condition_variable>
+#include <memory>
+#include <mutex>
+#include <stdexcept>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include <ATen/ATen.h>
+#include <ATen/core/dispatch/Dispatcher.h>
+#include <c10/macros/Macros.h>
+
+#include <torch/csrc/distributed/c10d/Work.hpp>
+// *************************************************************************
+// PROCESS GROUP collective communication API IS BEING CHANGED BETWEEN
+// versions 1.7 and 1.8.
+// PLEASE DO NOT ADD ANY DEPENDENCIES.
+// SEE RFC: https://github.com/pytorch/pytorch/issues/39662
+// *************************************************************************
+
+constexpr auto kProcessGroupDefaultTimeout =
+    std::chrono::milliseconds(30 * 60 * 1000);
+
+namespace c10d {
+
+// ProcessGroup is a base class that captures collective and point to
+// point communication in a fixed set of processes.
+//
+// The functions specified in the class below describe the API alone;
+// implementations are provided in subclasses.
+//
+// Every function that performs I/O is executed asynchronously by a
+// thread pool owned by the ProcessGroup (by default). They return an
+// object that can be used to wait for completion or error.
+//
+// The ProcessGroup can instantiate subgroups with fewer or an equal
+// number of members. Implementations must take care that multiple
+// process groups can be used in parallel and synchronize accordingly.
+//
+// The ProcessGroup assumes a fixed set of processes. If the set
+// changes, existing instances must be destructed and instantiation
+// and initialization must start from scratch. For members of the
+// process group to find each other (referred to as rendezvous from
+// hereon)
+//
+class TORCH_API ProcessGroup : public torch::CustomClassHolder {
+ public:
+  // ProcessGroup Options is a base struct that defines the basic options
+  // when constructing a ProcessGroup. Each ProcessGroup subclass should
+  // extend this struct and define its options if it wants to provide more
+  // config options (beyond basic ones defined here) to end user.
+  struct TORCH_API Options : torch::CustomClassHolder {
+    explicit Options(
+        std::string backend,
+        std::chrono::milliseconds timeout = kProcessGroupDefaultTimeout)
+        : timeout(timeout), backend(std::move(backend)) {}
+    ~Options() override = default;
+
+    std::chrono::milliseconds timeout;
+
+    // backend name
+    const std::string backend;
+  };
+
+  enum BackendType {
+    UNDEFINED = 0,
+    GLOO = 1,
+    NCCL = 2,
+    UCC = 3,
+    MPI = 4,
+    CUSTOM = 5,
+  };
+
+  // Not used, set for backwards compatibility and only used for TypeDef in
+  // Ops.cpp
+  explicit ProcessGroup(int rank, int size);
+
+  explicit ProcessGroup(
+      const c10::intrusive_ptr<::c10d::Store>& store,
+      int rank,
+      int size,
+      c10::intrusive_ptr<Options> options);
+  ~ProcessGroup() override;
+
+  int getRank() const {
+    return rank_;
+  }
+
+  int getSize() const {
+    return size_;
+  }
+
+  // Returns an unique opaque ID of this process group object.
+  int64_t getID() const {
+    return reinterpret_cast<std::intptr_t>(this);
+  }
+
+  // Returns an unique opaque ID of a backend for the specific backend type
+  // that can correlate with this process group's collectives.
+  int64_t getBackendID(BackendType backend_type) const {
+    return reinterpret_cast<std::intptr_t>(getBackend(backend_type).get());
+  }
+
+  virtual const std::string getBackendName() const {
+    return options_->backend;
+  };
+
+  BackendType getBackendType() const {
+    return backendType_;
+  };
+
+  virtual void startCoalescing(c10::DeviceType deviceType) {
+    // only nccl has implemented startCoalescing so only execute for nccl
+    // backends
+    auto backend = getBackend(deviceType);
+    backend->startCoalescing();
+  }
+
+  virtual c10::intrusive_ptr<Work> endCoalescing(c10::DeviceType deviceType) {
+    // only nccl has implemented endCoalescing so only execute for nccl
+    // backends
+    auto backend = getBackend(deviceType);
+    auto work = backend->endCoalescing();
+    return work;
+  }
+
+  virtual c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& tensors,
+      const BroadcastOptions& opts = BroadcastOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::broadcast_", "")
+            .typed<
+                std::tuple<std::vector<at::Tensor>, c10::intrusive_ptr<Work>>(
+                    at::TensorList,
+                    const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                    int64_t,
+                    int64_t,
+                    bool,
+                    int64_t)>();
+    // It's awakward to unbox the opts here and box them again in the custom C++
+    // op. But it's also complicated to make opts as a CustomClassHolder. Leave
+    // it as it is now.
+    return std::get<1>(op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.rootRank,
+        opts.rootTensor,
+        opts.asyncOp,
+        opts.timeout.count()));
+  }
+
+  virtual c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::allreduce_", "")
+            .typed<
+                std::tuple<std::vector<at::Tensor>, c10::intrusive_ptr<Work>>(
+                    at::TensorList,
+                    const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                    const c10::intrusive_ptr<::c10d::ReduceOp>&,
+                    const c10::optional<at::Tensor>& sparse_indices,
+                    int64_t)>();
+
+    return std::get<1>(op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        c10::make_intrusive<ReduceOp>(opts.reduceOp),
+        opts.sparseIndices,
+        opts.timeout.count()));
+  }
+
+  virtual c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceCoalescedOptions& opts = AllreduceCoalescedOptions()) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::allreduce_coalesced_", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::TensorList,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             const c10::intrusive_ptr<::c10d::ReduceOp>&,
+                             int64_t)>();
+
+    return op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        c10::make_intrusive<ReduceOp>(opts.reduceOp),
+        opts.timeout.count());
+  }
+
+  virtual c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& tensors,
+      const ReduceOptions& opts = ReduceOptions()) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::reduce_", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::TensorList,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             const c10::intrusive_ptr<::c10d::ReduceOp>&,
+                             int64_t,
+                             int64_t,
+                             int64_t)>();
+    return op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        c10::make_intrusive<ReduceOp>(opts.reduceOp),
+        opts.rootRank,
+        opts.rootTensor,
+        opts.timeout.count());
+  }
+
+  virtual c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::allgather_", "")
+                         .typed<std::tuple<
+                             std::vector<std::vector<at::Tensor>>,
+                             c10::intrusive_ptr<Work>>(
+                             const std::vector<std::vector<at::Tensor>>&,
+                             at::TensorList,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             int64_t)>();
+
+    return std::get<1>(op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.timeout.count()));
+  }
+
+  // Gathers a single tensor inputBuffer into a single buffer outputBuffer that
+  // is interpreted as a contiguous collection of size inputBuffer * WORLD_SIZE.
+  // For implementers of ProcessGroup API and advanced users only.
+  // Note: this function will be deprecated in near future.
+  virtual c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      const AllgatherOptions& opts = AllgatherOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::_allgather_base_", "")
+            .typed<std::tuple<at::Tensor, c10::intrusive_ptr<Work>>(
+                at::Tensor&,
+                at::Tensor&,
+                const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                bool,
+                int64_t)>();
+
+    return std::get<1>(op.call(
+        outputBuffer,
+        inputBuffer,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.asyncOp,
+        opts.timeout.count()));
+  }
+
+  // This function is deprecated and will be moved out of ProcessGroup to comms:
+  // * do not add dependencies on this function,
+  // * do not implement it in your ProcessGroup, implement _allgather_base
+  //   instead.
+  virtual c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& outputTensorLists,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::allgather_coalesced_", "")
+            .typed<c10::intrusive_ptr<Work>(
+                const std::vector<std::vector<at::Tensor>>&,
+                const at::TensorList&,
+                const c10::intrusive_ptr<::c10d::ProcessGroup>&)>();
+
+    return op.call(
+        outputTensorLists,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this));
+  }
+
+  // This function is a coalesced version of `allgather_into_tensor` (currently
+  // still named as `_allgather_base`). Each tensor in the vector corresponds to
+  // an input/output of one `allgather_into_tensor` operation.
+  virtual c10::intrusive_ptr<Work> allgather_into_tensor_coalesced(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::allgather_into_tensor_coalesced_", "")
+            .typed<c10::intrusive_ptr<Work>(
+                const at::TensorList,
+                const at::TensorList,
+                const c10::intrusive_ptr<::c10d::ProcessGroup>&)>();
+
+    return op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this));
+  }
+
+  virtual c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const GatherOptions& opts = GatherOptions()) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::gather_", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             const std::vector<std::vector<at::Tensor>>&,
+                             const at::TensorList&,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             int64_t,
+                             int64_t)>();
+    return op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.rootRank,
+        opts.timeout.count());
+  }
+
+  virtual c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ScatterOptions& opts = ScatterOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::scatter_", "")
+            .typed<
+                std::tuple<std::vector<at::Tensor>, c10::intrusive_ptr<Work>>(
+                    const at::TensorList&,
+                    const std::vector<std::vector<at::Tensor>>&,
+                    const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                    int64_t,
+                    bool,
+                    int64_t)>();
+    return std::get<1>(op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.rootRank,
+        opts.asyncOp,
+        opts.timeout.count()));
+  }
+
+  virtual c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::reduce_scatter_", "")
+            .typed<
+                std::tuple<std::vector<at::Tensor>, c10::intrusive_ptr<Work>>(
+                    const at::TensorList&,
+                    const std::vector<std::vector<at::Tensor>>&,
+                    const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                    const c10::intrusive_ptr<::c10d::ReduceOp>&,
+                    int64_t)>();
+    return std::get<1>(op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        c10::make_intrusive<::c10d::ReduceOp>(opts.reduceOp),
+        opts.timeout.count()));
+  }
+
+  virtual c10::intrusive_ptr<Work> _reduce_scatter_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::_reduce_scatter_base_", "")
+            .typed<std::tuple<at::Tensor, c10::intrusive_ptr<Work>>(
+                at::Tensor&,
+                at::Tensor&,
+                const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                const c10::intrusive_ptr<::c10d::ReduceOp>&,
+                bool,
+                int64_t)>();
+    return std::get<1>(op.call(
+        outputBuffer,
+        inputBuffer,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        c10::make_intrusive<::c10d::ReduceOp>(opts.reduceOp),
+        opts.asyncOp,
+        opts.timeout.count()));
+  }
+
+  // This function is a coalesced version of `reduce_scatter_tensor` (currently
+  // still named as `_reduce_scatter_base`). Each tensor in the vector
+  // corresponds to an input/output of one `reduce_scatter_tensor` operation.
+  virtual c10::intrusive_ptr<Work> reduce_scatter_tensor_coalesced(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::reduce_scatter_tensor_coalesced_", "")
+            .typed<c10::intrusive_ptr<Work>(
+                const at::TensorList,
+                const at::TensorList,
+                const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                const c10::intrusive_ptr<::c10d::ReduceOp>&,
+                int64_t)>();
+
+    return op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        c10::make_intrusive<::c10d::ReduceOp>(opts.reduceOp),
+        opts.timeout.count());
+  }
+
+  virtual c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      std::vector<int64_t>& outputSplitSizes,
+      std::vector<int64_t>& inputSplitSizes,
+      const AllToAllOptions& opts = AllToAllOptions()) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::alltoall_base_", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::Tensor&,
+                             at::Tensor&,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             std::vector<int64_t>,
+                             std::vector<int64_t>,
+                             int64_t)>();
+    return op.call(
+        outputBuffer,
+        inputBuffer,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        outputSplitSizes,
+        inputSplitSizes,
+        opts.timeout.count());
+  }
+
+  virtual c10::intrusive_ptr<Work> alltoall(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllToAllOptions& opts = AllToAllOptions()) {
+    static auto op =
+        c10::Dispatcher::singleton()
+            .findSchemaOrThrow("c10d::alltoall_", "")
+            .typed<
+                std::tuple<std::vector<at::Tensor>, c10::intrusive_ptr<Work>>(
+                    const at::TensorList&,
+                    const at::TensorList&,
+                    const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                    int64_t)>();
+    return std::get<1>(op.call(
+        outputTensors,
+        inputTensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.timeout.count()));
+  }
+
+  virtual void monitoredBarrier(
+      const BarrierOptions& opts,
+      bool wait_all_ranks = false) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::monitored_barrier_", "")
+                         .typed<void(
+                             at::Tensor,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             const std::vector<int64_t>&,
+                             int64_t,
+                             bool)>();
+    // Default to using cpu implementation, monitored barrier is only for GLOO
+    at::Tensor tensor = at::empty({0}, at::TensorOptions().device(at::kCPU));
+    op.call(
+        tensor,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.device_ids,
+        opts.timeout.count(),
+        wait_all_ranks);
+  }
+
+  // Agrees on an initial sequence number for the whole group by having rank 0
+  // create it and broadcast it to other ranks using the store. Only implemented
+  // for GLOO and NCCL backends currently.
+  virtual void setSequenceNumberForGroup() {
+    auto backendType = getBackendType();
+    // TODO: HACK for backend name to get sequence number for that backend.
+    if (backendType == ProcessGroup::BackendType::GLOO ||
+        backendType == ProcessGroup::BackendType::NCCL ||
+        backendType == ProcessGroup::BackendType::UCC) {
+      getDefaultBackend()->setSequenceNumberForGroup();
+    } else {
+      TORCH_CHECK(
+          false,
+          c10::str(
+              "ProcessGroup ",
+              getBackendName(),
+              " does not yet support sequence numbers."));
+    }
+  }
+
+  // Retrieves the current sequence number for the whole group, which should be
+  // in sync. If the returned number is not consistent across the group, it
+  // may indicate that there is some sort of collective desynchronization.
+  virtual uint64_t getSequenceNumberForGroup() {
+    auto backendType = getBackendType();
+
+    // TODO: HACK for backend name to get sequence number for that backend.
+    if (backendType == ProcessGroup::BackendType::GLOO ||
+        backendType == ProcessGroup::BackendType::NCCL ||
+        backendType == ProcessGroup::BackendType::UCC) {
+      return getDefaultBackend()->getSequenceNumberForGroup();
+    } else {
+      TORCH_CHECK(
+          false,
+          c10::str(
+              "ProcessGroup ",
+              getBackendName(),
+              " does not yet support sequence numbers."));
+    }
+  }
+
+  virtual c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& tensors,
+      int dstRank,
+      int tag) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::send", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::TensorList,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             int64_t,
+                             int64_t)>();
+    return op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        dstRank,
+        tag);
+  }
+
+  virtual c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& tensors,
+      int srcRank,
+      int tag) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::recv_", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::TensorList,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             int64_t,
+                             int64_t)>();
+    return op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        srcRank,
+        tag);
+  }
+
+  virtual c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& tensors,
+      int tag) {
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::recv_any_source_", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::TensorList,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             int64_t)>();
+    return op.call(
+        tensors,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        tag);
+  }
+
+  virtual c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& opts = BarrierOptions()) {
+    static at::Tensor tensor;
+    // TODO: if nccl was specified then use it
+    auto device = opts.device;
+    if (device.has_value()) {
+      // set device tensor from argument
+      tensor = at::empty(
+          {1}, at::TensorOptions().device(device.value()).dtype(at::kByte));
+    } else if (backendType_ == c10d::ProcessGroup::BackendType::NCCL) {
+      // set cuda tensor
+      tensor = at::empty(
+          {1},
+          at::TensorOptions().device(at::DeviceType::CUDA).dtype(at::kByte));
+    } else {
+      // Default to using cpu implementation
+      tensor = at::empty(
+          {1},
+          at::TensorOptions().device(at::DeviceType::CPU).dtype(at::kByte));
+    }
+
+    static auto op = c10::Dispatcher::singleton()
+                         .findSchemaOrThrow("c10d::barrier", "")
+                         .typed<c10::intrusive_ptr<::c10d::Work>(
+                             at::Tensor,
+                             const c10::intrusive_ptr<::c10d::ProcessGroup>&,
+                             const std::vector<int64_t>&,
+                             int64_t)>();
+
+    return op.call(
+        tensor,
+        c10::intrusive_ptr<ProcessGroup>::unsafe_reclaim_from_nonowning(this),
+        opts.device_ids,
+        opts.timeout.count());
+  }
+
+  c10::intrusive_ptr<Options> getOptions() {
+    return options_;
+  }
+
+  bool hasBackends() {
+    return !deviceTypeToBackendType_.empty();
+  }
+
+  void setBackend(
+      c10::DeviceType deviceType,
+      BackendType backendType,
+      const c10::optional<c10::intrusive_ptr<Backend>>& backend) {
+    // TODO: should we add these entries after the backend setting succeeds?
+    deviceTypeToBackendType_[deviceType] = backendType;
+    deviceTypes_.insert(deviceType);
+    // if the backendType is already set then reuse it for this device
+    if (backendTypeToBackend_.find(backendType) !=
+        backendTypeToBackend_.end()) {
+      auto existingBackend = backendTypeToBackend_.at(backendType);
+      deviceTypeToBackend_[deviceType] = existingBackend;
+      TORCH_CHECK(
+          existingBackend->getBoundDeviceId() ==
+          (*backend)->getBoundDeviceId());
+    } else {
+      // check if backend has value
+      if (backend.has_value()) {
+        deviceTypeToBackend_[deviceType] = backend.value();
+        backendTypeToBackend_[backendType] = backend.value();
+        (*backend)->setBoundDeviceId(bound_device_id_);
+      }
+    }
+  }
+
+  c10::intrusive_ptr<Backend> getDefaultBackend() const {
+    TORCH_CHECK(
+        backendTypeToBackend_.find(backendType_) != backendTypeToBackend_.end(),
+        "Could not find the default backend type ",
+        backendType_,
+        " for Process Group with name ",
+        getBackendName(),
+        ".");
+    return backendTypeToBackend_.at(backendType_);
+  }
+
+  c10::intrusive_ptr<Backend> getBackend(c10::DeviceType deviceType);
+
+  c10::intrusive_ptr<Backend> getBackend(BackendType backendType) const {
+    TORCH_CHECK(
+        backendTypeToBackend_.find(backendType) != backendTypeToBackend_.end(),
+        "Could not find backend type ",
+        backendType,
+        ".");
+    return backendTypeToBackend_.at(backendType);
+  }
+
+  // Return device types supported by this ProcessGroup.
+  // Note: the return type is `Device` rather than `DeviceType` for the purpose
+  // of easy comparison at Python level. The `Device` will have default index
+  // (-1).
+  std::vector<c10::Device> getDeviceTypes() const {
+    std::vector<c10::Device> devices;
+    devices.reserve(deviceTypes_.size());
+    for (auto& dt : deviceTypes_) {
+      devices.push_back(c10::Device(dt));
+    }
+    return devices;
+  }
+
+  void registerOnCompletionHook(
+      std::function<void(std::shared_ptr<WorkInfo>)>&& hook) {
+    getDefaultBackend()->registerOnCompletionHook(std::move(hook));
+  }
+
+  void waitForPendingWorks() {
+    getDefaultBackend()->waitForPendingWorks();
+  }
+
+  bool hasHooks() const {
+    return getDefaultBackend()->hasHooks();
+  }
+
+  const std::string& getGroupName() const;
+  void setGroupName(const std::string& name);
+  void enableCollectivesTiming();
+
+  void release_resources() override;
+
+  // ProcessGroups optionally can be "bound" to a specific device.
+  // Currently this is only for nccl and allows for some opt-in
+  // optimizations such as automatic use of ncclCommSplit.  The device
+  // is specified in `init_process_group` and eventually makes it
+  // here and then down into the actual backend instances.
+  c10::optional<at::Device> getBoundDeviceId() const {
+    return bound_device_id_;
+  }
+
+  void setBoundDeviceId(c10::optional<at::Device> device) {
+    if (device) {
+      TORCH_CHECK(device->has_index(), "setBoundDeviceId must have an index");
+    }
+    bound_device_id_ = device;
+  }
+
+ protected:
+  // Implementations of this interface need to call this to setup
+  // appropriate logging etc.
+  void init();
+
+  c10::intrusive_ptr<c10d::Store> store_;
+  const int rank_;
+  const int size_;
+  const c10::intrusive_ptr<Options> options_;
+  const BackendType backendType_;
+
+  // Debug level setting. It is parsed once when ProcessGroup is constructed and
+  // remains the same across use of this process group.
+  DebugLevel dist_debug_level_;
+
+  // Backend classes for this ProcessGroup
+  std::unordered_set<c10::DeviceType> deviceTypes_;
+  std::unordered_map<c10::DeviceType, BackendType> deviceTypeToBackendType_;
+  std::unordered_map<c10::DeviceType, c10::intrusive_ptr<Backend>>
+      deviceTypeToBackend_;
+  std::unordered_map<BackendType, c10::intrusive_ptr<Backend>>
+      backendTypeToBackend_;
+
+  c10::optional<at::Device> bound_device_id_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/ProcessGroupGloo.hpp

@@ -0,0 +1,448 @@
+#pragma once
+
+#ifdef USE_C10D_GLOO
+
+#include <condition_variable>
+#include <deque>
+#include <mutex>
+#include <thread>
+#include <unordered_map>
+#include <vector>
+
+#include <gloo/algorithm.h>
+#include <gloo/common/error.h>
+#include <gloo/context.h>
+#include <gloo/rendezvous/store.h>
+#include <gloo/transport/device.h>
+
+#include <c10/util/hash.h>
+
+#include <torch/csrc/distributed/c10d/Backend.hpp>
+#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <torch/csrc/distributed/c10d/Types.hpp>
+#include <torch/csrc/distributed/c10d/Utils.hpp>
+
+namespace c10d {
+
+constexpr const char* GLOO_BACKEND_NAME = "gloo";
+
+// ProcessGroupGloo implements Gloo bindings for c10d.
+//
+// All functions on this class are expected to be called in the same
+// order across processes in the group. This is the only way that we
+// can guarantee to match up the same calls across processes. For
+// multi-threaded usage of process groups, you can use consider using
+// multiple process group instances.
+//
+// The Gloo algorithms that this class calls into are cached by their
+// signature (see description of AlgorithmKey above). This cache works
+// as follows: every function call instantiates an AlgorithmKey and
+// looks in the cache for existing entries. If there is one, it is
+// removed from the cache and returned to the caller. If there are
+// none, a new entry is created and returned. If an entry was created
+// before, but is still in use, the call will block and wait until the
+// entry is returned to the cache.
+//
+// In the future, we hope to extend this to allow multiple entries per
+// key, to enable parallelism for a single key. The number of entries
+// per key must always be identical for all processes. This maximum
+// number can be automatically tuned, but only if we let a single
+// process take charge, and have it broadcast the limits.
+//
+class TORCH_API ProcessGroupGloo : public Backend {
+ public:
+  // AsyncWork is the Gloo specific superclass for asynchronous work items.
+  // We can split asynchronous work into 3 phases:
+  // 1) Sanity checks and prepare input (e.g. memcpy)
+  // 2) Run operation on background thread
+  // 3) Synchronize with completion on foreground thread
+  //
+  // There is state to be shared between these 3 phases and all of this state
+  // is captured in the AsyncWork class and its derivatives.
+  //
+  // Note: while we are porting operations to use new style collectives, there
+  // is a split between operations using the existing caching approach and
+  // operations using the new AsyncWork base class. Over time we will port
+  // all operations and perform needed cleanup.
+  //
+  // FIXME: This probably should be called WorkGloo since the work is executed
+  // in sync mode by a background thread.
+  class TORCH_API AsyncWork : public Work {
+   public:
+    explicit AsyncWork(
+        std::vector<std::vector<at::Tensor>> outputTensors,
+        OpType opType,
+        uint64_t seq,
+        const char* profilingTitle = nullptr,
+        const c10::optional<std::vector<at::Tensor>>& inputTensors =
+            c10::nullopt);
+
+    ~AsyncWork() override = default;
+
+    static void execute(c10::intrusive_ptr<AsyncWork> work);
+
+    virtual void run() = 0;
+
+    std::vector<at::Tensor> result() override;
+
+    c10::intrusive_ptr<c10::ivalue::Future> getFuture() override;
+    uint64_t getSequencenumber() const override;
+
+   protected:
+    friend class ProcessGroupGloo;
+
+   private:
+    void finishWorkGloo();
+    void finishWorkGlooError(std::exception_ptr eptr);
+    inline void recordAsyncWorkProfilingInfo(
+        const char* profilingTitle,
+        const c10::optional<std::vector<at::Tensor>>& inputTensors);
+
+    const std::vector<std::vector<at::Tensor>> outputTensors_;
+    c10::intrusive_ptr<at::ivalue::Future> future_;
+    std::function<void()> recordFunctionBeforeCallback_;
+    const uint64_t seq_;
+  };
+
+  // Wrap c10d store as Gloo store
+  class TORCH_API GlooStore : public ::gloo::rendezvous::Store {
+   public:
+    GlooStore(const c10::intrusive_ptr<::c10d::Store>& store) : store_(store) {}
+
+    void setUint(const std::string& key, const std::vector<uint8_t>& value) {
+      store_->set(key, value);
+    }
+
+    void set(const std::string& key, const std::vector<char>& value) override {
+      std::vector<uint8_t> tmp(value.begin(), value.end());
+      store_->set(key, tmp);
+    }
+
+    std::vector<uint8_t> getUint(const std::string& key) {
+      auto value = store_->get(key);
+      return value;
+    }
+
+    std::vector<char> get(const std::string& key) override {
+      auto value = store_->get(key);
+      return std::vector<char>(value.begin(), value.end());
+    }
+
+    void wait(const std::vector<std::string>& keys) override {
+      store_->wait(keys, ::c10d::Store::kDefaultTimeout);
+    }
+
+    void wait(
+        const std::vector<std::string>& keys,
+        const std::chrono::milliseconds& timeout) override {
+      store_->wait(keys, timeout);
+    }
+
+#ifdef GLOO_STORE_HAS_STORE_V2
+    bool has_v2_support() override {
+      return store_->hasExtendedApi();
+    }
+
+    std::vector<std::vector<char>> multi_get(
+        const std::vector<std::string>& keys) override {
+      std::vector<std::vector<char>> res;
+      for (auto& value : store_->multiGet(keys)) {
+        res.emplace_back(std::vector<char>(value.begin(), value.end()));
+      }
+      return res;
+    }
+
+    void multi_set(
+        const std::vector<std::string>& keys,
+        const std::vector<std::vector<char>>& values) override {
+      std::vector<std::vector<uint8_t>> u_values;
+      for (auto& value : values) {
+        u_values.emplace_back(std::vector<uint8_t>(value.begin(), value.end()));
+      }
+      store_->multiSet(keys, u_values);
+    }
+
+    void append(const std::string& key, const std::vector<char>& value)
+        override {
+      std::vector<uint8_t> tmp(value.begin(), value.end());
+      return store_->append(key, tmp);
+    }
+
+    int64_t add(const std::string& key, int64_t value) override {
+      return store_->add(key, value);
+    }
+#endif
+
+   protected:
+    c10::intrusive_ptr<::c10d::Store> store_;
+  };
+
+  // For send and recv operations there is no need to pass them to the
+  // thread pool as they are entirely completed by the device thread.
+  // This work object is used to synchronize completion of the send or
+  // recv operation. It keeps a reference to the tensor it is
+  // operating on to prevent it from being deallocated while the
+  // operation is still in flight.
+  class TORCH_API SendWork : public Work {
+   public:
+    explicit SendWork(
+        at::Tensor& tensor,
+        std::unique_ptr<::gloo::transport::UnboundBuffer> buffer,
+        uint64_t seq);
+
+    bool wait(std::chrono::milliseconds timeout = kNoTimeout) override;
+
+    void abort() override;
+
+    uint64_t getSequencenumber() const override;
+
+   protected:
+    at::Tensor tensor_;
+    std::unique_ptr<::gloo::transport::UnboundBuffer> buffer_;
+    const uint64_t seq_;
+  };
+
+  class TORCH_API RecvWork : public Work {
+   public:
+    explicit RecvWork(
+        at::Tensor& tensor,
+        std::unique_ptr<::gloo::transport::UnboundBuffer> buffer,
+        OpType opType,
+        uint64_t seq,
+        const char* profilingTitle = nullptr);
+
+    int sourceRank() const override;
+
+    bool wait(std::chrono::milliseconds timeout = kNoTimeout) override;
+
+    void abort() override;
+
+    uint64_t getSequencenumber() const override;
+
+   protected:
+    at::Tensor tensor_;
+    std::unique_ptr<::gloo::transport::UnboundBuffer> buffer_;
+    int srcRank_;
+    const uint64_t seq_;
+  };
+
+  struct TORCH_API Options : public Backend::Options {
+    explicit Options(
+        std::chrono::milliseconds timeout = kBackendDefaultTimeout);
+
+    // return intrusive_ptr of the object
+    static c10::intrusive_ptr<Options> create(
+        std::chrono::milliseconds timeout = kBackendDefaultTimeout) {
+      return c10::make_intrusive<Options>(timeout);
+    }
+
+    std::vector<std::shared_ptr<::gloo::transport::Device>> devices;
+    int threads;
+  };
+
+  const std::string getBackendName() const override {
+    return std::string(GLOO_BACKEND_NAME);
+  }
+
+  // Helper functions to create a new device object.
+  // They are static functions on this class to keep them logically
+  // separate from the rest of the code base (e.g. torch/csrc/distributed).
+
+  // Create new device instance for specific interface.
+  static std::shared_ptr<::gloo::transport::Device> createDeviceForInterface(
+      const std::string& interface);
+
+  // Create new device instance for specific hostname or address.
+  static std::shared_ptr<::gloo::transport::Device> createDeviceForHostname(
+      const std::string& hostname);
+
+  // Create new device instance.
+  // It tries to resolve this machine's hostname and bind to that address.
+  // If that fails (i.e. the hostname doesn't resolve to an address), it
+  // falls back to binding to the loopback address.
+  static std::shared_ptr<::gloo::transport::Device> createDefaultDevice();
+
+  // Create ProcessGroupGloo instance.
+  static c10::intrusive_ptr<ProcessGroupGloo> createProcessGroupGloo(
+      const c10::intrusive_ptr<Store>& store,
+      int rank,
+      int size,
+      std::chrono::milliseconds timeout);
+
+  explicit ProcessGroupGloo(
+      const c10::intrusive_ptr<Store>& store,
+      int rank,
+      int size,
+      c10::intrusive_ptr<Options> options = Options::create());
+
+  ~ProcessGroupGloo() override;
+
+  c10::intrusive_ptr<Options> getOptions() {
+    return options_;
+  }
+
+  c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& tensors,
+      const BroadcastOptions& opts = BroadcastOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce_sparse(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceCoalescedOptions& opts =
+          AllreduceCoalescedOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& tensors,
+      const ReduceOptions& opts = ReduceOptions()) override;
+
+  c10::intrusive_ptr<Work> _reduce_scatter_base(
+      at::Tensor& outputTensor,
+      at::Tensor& inputTensor,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& output_tensor,
+      at::Tensor& input_tensor,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& output_lists,
+      std::vector<at::Tensor>& input_list,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather_into_tensor_coalesced(
+      std::vector<at::Tensor>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const GatherOptions& opts = GatherOptions()) override;
+
+  c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& outputs,
+      std::vector<std::vector<at::Tensor>>& inputs,
+      const ScatterOptions& opts = ScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& outputs,
+      std::vector<std::vector<at::Tensor>>& inputs,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce_scatter_tensor_coalesced(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& outputTensor,
+      at::Tensor& inputTensor,
+      std::vector<int64_t>& outputCounts,
+      std::vector<int64_t>& inputCounts,
+      const AllToAllOptions& opts = AllToAllOptions()) override;
+
+  c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& tensors,
+      int dstRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& tensors,
+      int srcRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& tensors,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& opts = BarrierOptions()) override;
+
+  void enableCollectivesTiming() override;
+
+  const std::unique_ptr<::gloo::rendezvous::Store>& _getStore() const {
+    return store_;
+  }
+
+  // Similar to barrier(), but blocks rank 0 until all other ranks have
+  // acknowledged that they are alive (through send/recv from rank 0). Rank 0
+  // is able to report all failed ranks if waitAllRanks = true, otherwise
+  // reports the first rank it detected as failed.
+  void monitoredBarrier(
+      const BarrierOptions& opts = BarrierOptions(),
+      bool waitAllRanks = false) override;
+
+  // Agrees on an initial sequence number for the whole group by having rank 0
+  // create it and broadcast it to other ranks using the store.
+  void setSequenceNumberForGroup() override;
+
+  // Retrieves the current sequence number for the whole group, which should be
+  // in sync. If the returned number is not consistent across the group, it
+  // may indicate that there is some sort of collective desynchronization.
+  uint64_t getSequenceNumberForGroup() override;
+
+  int getNumThreads() {
+    return options_->threads;
+  }
+
+ protected:
+  std::unique_ptr<::gloo::rendezvous::Store> store_;
+  const c10::intrusive_ptr<Options> options_;
+
+  // Every Gloo context represents a set of connections to its peers.
+  // In order to use more than one device (or allow for parallelism on
+  // a single device), you need multiple contexts.
+  std::vector<std::shared_ptr<::gloo::Context>> contexts_;
+  std::vector<std::thread> threads_;
+  bool stop_;
+
+  // Incremented for every collective we kick off.
+  // The value is used as tag for collective operations. Collectives are kicked
+  // off in identical order across processes. Therefore the tag can be used
+  // to match up operations during concurrent execution.
+  uint32_t collectiveCounter_;
+
+  // Returns next collective tag to use (uses collectiveCounter_).
+  uint32_t nextTag();
+
+  // Returns the context to use for the specified tag.
+  // With `nextTag` returning an increasing number, this should lead
+  // to contexts being used in a round-robin fashion.
+  std::shared_ptr<::gloo::Context> getContext(uint32_t tag);
+
+  // Entrypoint for worker threads.
+  void runLoop(int workerIndex);
+
+  // Queue work to run on worker thread.
+  void enqueue(c10::intrusive_ptr<AsyncWork> work);
+
+  // Keep both a queue of pending work, and a vector with in progress work.
+  // Both of these can only be mutated when holding the queue lock.
+  // We keep both around instead of just the queue, so we can grab a weak_ptr
+  // to all in progress and pending work when executing a barrier.
+  // When executing a barrier, we need to ensure that all prior work
+  // has completed before completing itself.
+  std::deque<c10::intrusive_ptr<AsyncWork>> workQueue_;
+  std::vector<c10::intrusive_ptr<AsyncWork>> workInProgress_;
+  std::mutex workMutex_;
+  std::condition_variable workProduceCV_;
+  std::condition_variable workConsumeCV_;
+  uint64_t seq_{0};
+};
+
+} // namespace c10d
+
+#endif // USE_C10D_GLOO

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp

@@ -0,0 +1,1097 @@
+#pragma once
+
+#if defined(__linux__)
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+#endif
+
+#ifdef USE_C10D_NCCL
+
+#include <atomic>
+#include <chrono>
+#include <future>
+#include <iostream>
+#include <list>
+#include <mutex>
+#include <thread>
+#include <unordered_map>
+
+#include <torch/csrc/distributed/c10d/Backend.hpp>
+#include <torch/csrc/distributed/c10d/NCCLUtils.hpp>
+#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
+#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <torch/csrc/distributed/c10d/intra_node_comm.hpp>
+
+#include <ATen/DynamicLibrary.h>
+#include <ATen/cuda/CUDAContext.h>
+#include <ATen/cuda/CUDAEvent.h>
+#include <c10/core/Stream.h>
+#include <c10/core/StreamGuard.h>
+#include <c10/cuda/CUDACachingAllocator.h>
+#include <c10/cuda/CUDAGuard.h>
+#include <c10/cuda/CUDAStream.h>
+
+#include <torch/custom_class.h>
+
+namespace c10d {
+
+// Control whether or not wait() is blocking or non-blocking.
+static std::vector<std::string> TORCH_NCCL_BLOCKING_WAIT = {
+    "TORCH_NCCL_BLOCKING_WAIT",
+    "NCCL_BLOCKING_WAIT"};
+
+// Control whether or not we perform Async Error Handling with NCCL.
+static std::vector<std::string> TORCH_NCCL_ASYNC_ERROR_HANDLING = {
+    "TORCH_NCCL_ASYNC_ERROR_HANDLING",
+    "NCCL_ASYNC_ERROR_HANDLING"};
+
+// Control whether dumping debug info on watchdog
+// timeout is enabled. This variable must be set together with
+// TORCH_NCCL_ENABLE_MONITORING=1 and TORCH_NCCL_TRACE_BUFFER_SIZE > 0.
+static std::vector<std::string> TORCH_NCCL_DUMP_ON_TIMEOUT = {
+    "TORCH_NCCL_DUMP_ON_TIMEOUT"};
+
+// Control whether Desync Debug is enabled. This variable must be set
+// together with TORCH_NCCL_ASYNC_ERROR_HANDLING.
+static std::vector<std::string> TORCH_NCCL_DESYNC_DEBUG = {
+    "TORCH_NCCL_DESYNC_DEBUG",
+    "NCCL_DESYNC_DEBUG"};
+
+// Enable recording start-events for all ProcessGroupNCCL collectives, and
+// compute accurate collective timing per-collective. (Note: end-events are
+// recorded by default. Turn on this flag can increase chances of a watchdog
+// hang due to performing a CUDA event query which eventually calls
+// cudaEventElapsedTime() API.
+static std::vector<std::string> TORCH_NCCL_ENABLE_TIMING = {
+    "TORCH_NCCL_ENABLE_TIMING",
+    "NCCL_ENABLE_TIMING"};
+
+// Enable monitoring thread which aborts the process when the ProcessGroupNCCL
+// Watchdog thread gets stuck and no heartbeat is detected after
+// TORCH_NCCL_HEARTBEAT_TIMEOUT_SEC. This can happen due to calling CUDA/NCCL
+// APIs that may hang. It is Useful to prevent jobs being stuck for a prolonged
+// time than necessary tying up cluster resources.
+static std::vector<std::string> TORCH_NCCL_ENABLE_MONITORING = {
+    "TORCH_NCCL_ENABLE_MONITORING"};
+
+// Control the watchdog heartbeat timeout period after which the monitoring
+// thread will abort the process.
+static std::vector<std::string> TORCH_NCCL_HEARTBEAT_TIMEOUT_SEC = {
+    "TORCH_NCCL_HEARTBEAT_TIMEOUT_SEC"};
+
+// The maximum number of events we store in the flight recorder's ring buffer.
+// (One event could be the start or end of a collective, for example).
+static std::vector<std::string> TORCH_NCCL_TRACE_BUFFER_SIZE = {
+    "TORCH_NCCL_TRACE_BUFFER_SIZE"};
+
+// Control how much extra time we will wait for dumping the debugging info
+// before we exit and throws timeout exception.
+static std::vector<std::string> TORCH_NCCL_WAIT_TIMEOUT_DUMP_MILSEC = {
+    "TORCH_NCCL_WAIT_TIMEOUT_DUMP_MILSEC"};
+
+// Control the interval inside the watchdog thread to check the coordinated
+// signal from other ranks, e.g. to dump the debugging information.
+static std::vector<std::string> TORCH_NCCL_COORD_CHECK_MILSEC = {
+    "TORCH_NCCL_COORD_CHECK_MILSEC"};
+
+// Whether to abort the communicators when users call destroy_process_group().
+// If yes, communicators will be aborted when destroy_process_group is called,
+// but not in destructor.
+static std::vector<std::string> TORCH_NCCL_ABORT_IN_DESTROY_PG = {
+    "TORCH_NCCL_ABORT_IN_DESTROY_PG"};
+
+constexpr const char* NCCL_BACKEND_NAME = "nccl";
+
+constexpr const char* TIMEOUT_DUMP = "timeout_dump";
+
+constexpr const int kWorkStatusUpdatePeriodMs = 10 * 1000; // 10 seconds
+
+constexpr auto kProcessGroupNCCLDefaultTimeout =
+    std::chrono::milliseconds(10 * 60 * 1000);
+
+// NoHandling: do not handle asynchronous NCCL errors
+// TearDown: tear down process upon error, see `WorkNCCL::handleException`
+// CleanUpOnly: just clean up collectives and abort communicators without
+// tearing down process SkipCleanUp: (this is a temporary option and can be
+// removed in future) tear down process without cleaning up NCCL communicators.
+// This should be used as a last resort in case `ncclCommAbort` itself is
+// hanging
+enum ErrorHandlingMode {
+  NoHandling = 0,
+  TearDown = 1,
+  CleanUpOnly = 2,
+  SkipCleanUp = 3
+};
+
+#define SHOULD_CLEAN_UP(a) (a != NoHandling && a != SkipCleanUp)
+
+#define SHOULD_TEAR_DOWN(a) (a != NoHandling && a != CleanUpOnly)
+
+#define PRINT_COLLECTIVE_HASH_SIGNATURE(phase, opType, numel, hashValue)      \
+  LOG(WARNING) << logPrefix() << "Hash of " << phase << " to NCCL " << opType \
+               << " with size " << numel << " is " << hashValue;
+
+// If set, ProcessGroupNCCL doesn't use recordStream calls to ensure
+// caching allocator safety for tensors used on both user-facing and
+// internal comm streams.
+// Instead, it stashes live references to those tensors until after
+// user-facing streams are synced with comm streams.
+// See stashed_for_allocator_safety_ below.
+static std::vector<std::string> TORCH_NCCL_AVOID_RECORD_STREAMS = {
+    "TORCH_NCCL_AVOID_RECORD_STREAMS"};
+
+// If set, ProcessGroupNCCL registers postAlloc and preFree hooks to cuda cache
+// allocator so that whenever a tensor is allocated or freed, ProcessGroupNCCL
+// can register/deregister the tensor on all available NCCL communicators.
+static std::vector<std::string> TORCH_NCCL_USE_TENSOR_REGISTER_ALLOCATOR_HOOK =
+    {"TORCH_NCCL_USE_TENSOR_REGISTER_ALLOCATOR_HOOK",
+     "NCCL_USE_TENSOR_REGISTER_ALLOCATOR_HOOK"};
+
+#if defined(__linux__)
+struct DumpPipe {
+  DumpPipe(int rank) {
+    std::string fileStem =
+        getCvarString({"TORCH_NCCL_DEBUG_INFO_PIPE_FILE"}, "");
+    if (fileStem.empty() ||
+        getCvarInt({"TORCH_NCCL_TRACE_BUFFER_SIZE"}, 0) <= 0) {
+      return;
+    }
+    TORCH_CHECK(!fileStem.empty(), "TORCH_NCCL_DEBUG_INFO_TEMP_FILE is empty");
+    std::string filename = c10::str(fileStem, rank, ".pipe");
+    TORCH_CHECK(
+        unlink(filename.c_str()) != -1 || errno == ENOENT,
+        "Error removing existing named pipe ",
+        filename);
+    TORCH_CHECK(
+        mkfifo(filename.c_str(), 0666) != -1,
+        "Error creating named pipe ",
+        filename);
+    fd_ = open(filename.c_str(), O_RDONLY | O_NONBLOCK);
+    LOG(INFO) << "Pipe file " << filename
+              << " has been opened, write to it to trigger NCCL Debug Dump.";
+    TORCH_CHECK(fd_ != -1, "Error opening named pipe ", filename);
+  }
+  bool shouldDump() {
+    if (fd_ == -1) {
+      return false;
+    }
+    char buf[128];
+    // non-blocking from O_NONBLOCK above.
+    // Ignore EINTR because we already will poll this
+    // again later.
+    ssize_t bytesRead = read(fd_, &buf, 128);
+    return bytesRead > 0;
+  }
+  ~DumpPipe() {
+    if (fd_ != -1) {
+      close(fd_);
+    }
+  }
+
+ private:
+  int fd_ = -1;
+};
+#else
+struct DumpPipe {
+  DumpPipe(int rank) {}
+  bool shouldDump() {
+    return false;
+  }
+};
+#endif
+
+// ProcessGroupNCCL implements NCCL bindings for c10d.
+//
+// All functions of the class are expected to be called in the same order
+// across all processes in the process group.  This is the only way that we
+// can guarantee to match up the same calls among all processes.
+//
+// All NCCL functions provided by this class are asynchronous functions. More
+// specifically, each NCCL call is scheduled on a separate CUDA stream that is
+// different from the current CUDA stream. This is for the purpose of
+// achieving potentially concurrency and better performance. As a result,
+// it is the callers' responsibility to make sure that the CUDA stream their
+// code works on needs to wait for the NCCL operation from
+// this class.
+//
+// This can be done by calling:
+//
+// either WorkNCCL::wait() or WorkNCCL::synchronize(), both achieves the same
+// functionality and are synonyms.
+//
+// Also note that WorkNCCL::finishedGPUExecution() is a helper function only
+// provided by ProcessGroupNCCL to check if the NCCL operation of WorkNCCL has
+// finished execution on the GPU (not just scheduled).
+//
+// Example on using the NCCL process group
+//
+//   ProcessGroupNCCL pg(store, rank, size);
+//   std::shared_ptr<WorkNCCL> work = pg.allreduce(tensors);
+//
+//   // At this point, NCCL kernel has already by queued successfully
+//   // Now, let current stream wait for the NCCL to finish, this function is
+//   // async operation as well
+//
+//   work->wait()
+//
+//   // Now continue on other work in the current stream.
+class TORCH_API ProcessGroupNCCL : public Backend {
+ public:
+  class WorkNCCL : public Work, public std::enable_shared_from_this<WorkNCCL> {
+   public:
+    friend struct WorkInfo;
+
+    // Constructor takes a list of CUDA devices
+    WorkNCCL(
+        at::Device& device,
+        int rank,
+        OpType opType,
+        uint64_t seq,
+        const char* profilingTitle = nullptr,
+        const c10::optional<std::vector<at::Tensor>>& inputs = c10::nullopt,
+        bool desyncDebug = false,
+        bool enableTiming = false,
+        DebugLevel distDebugLevel = DebugLevel::Off);
+    // Copy constructor doing partial copy without outputs_. Cleanup thread
+    // monitors and removes finished works. However it will deadlock when
+    // destructs outputs_ tensors who are view tensors in autograd graph.
+    WorkNCCL(const WorkNCCL& w);
+
+    ~WorkNCCL() override;
+
+    // Checks if the NCCL kernel has started to execute.
+    bool isStarted();
+
+    // Checks if request has completed. In this specific case of NCCL, it checks
+    // if the NCCL operation has completed on the GPU in its own NCCL stream.
+    // Non-blocking operation.
+    bool isCompleted() override;
+
+    bool isSuccess() const override;
+
+    // Same as calling synchronize() for NCCL work.
+    bool wait(std::chrono::milliseconds timeout = kNoTimeout) override;
+
+    void abort() override;
+
+    // Let current stream wait on the completing of the NCCL work
+    // Throws on exceptions. Blocking operation, which will wait for work
+    // completion.
+    void synchronize() override;
+
+    // Synchronize streams by blocking each on the NCCL stream
+    void synchronizeStream();
+
+    // Helper function to handle exception (throw if needed).
+    void handleException(ErrorHandlingMode asyncErrorHandling);
+
+    // Helper function that checks if the NCCL kernels have finished
+    // execution on the GPUs
+    bool finishedGPUExecution();
+
+    // Get a Future object that will be marked as completed internally.
+    c10::intrusive_ptr<c10::ivalue::Future> getFuture() override;
+
+    float getDuration() const override;
+
+    uint64_t getSequencenumber() const override;
+
+    const std::string& logPrefix() const;
+
+    // Helper function that sets an exception_ptr on the WorkNCCL object.
+    void setException(std::exception_ptr exception_ptr);
+
+    // Helper function that returns True if the WorkNCCL object has timed out
+    // and False otherwise.
+    // In case of timeout, set exception on the WorkNCCL object.
+    bool checkTimeout(
+        c10::optional<std::chrono::milliseconds> timeout = c10::nullopt);
+
+    std::vector<at::Tensor> result() override;
+
+   protected:
+    // The cached list of CUDA devices to operate on
+    at::Device device_;
+
+    // The start CUDA event of NCCL operator tracking this work item. These
+    // start CUDA events are needed by desync debugging if enabled.
+    std::shared_ptr<at::cuda::CUDAEvent> ncclStartEvent_;
+
+    // The end CUDA event of NCCL operator tracking this work item.
+    std::shared_ptr<at::cuda::CUDAEvent> ncclEndEvent_;
+
+    // The NCCL communicator used for this work item.
+    std::shared_ptr<NCCLComm> ncclComm_;
+
+    // Tensors used for barrier op
+    at::Tensor barrierTensor_;
+
+    // Clone of blockingWait_ from ProcessGroupNCCL.
+    bool blockingWait_ = false;
+
+    // Clone of avoidRecordStreams_ from ProcessGroupNCCL.
+    bool avoidRecordStreams_ = false;
+
+    // Clone of opTimeout_ from ProcessGroupNCCL.
+    std::chrono::milliseconds opTimeout_;
+
+    // Time point representing when the work started.
+    std::chrono::time_point<std::chrono::steady_clock> workStartTime_;
+
+    // Record the collective sequential number.
+    uint64_t seq_;
+
+    // Indicates if the nccl start event has been updated to the store trace.
+    // This will be used by desync debug.
+    bool startTraceUpdated_{false};
+
+    // Record collective sizes for debug. We only record the size on the first
+    // device as multi-device per process is deprecated
+    size_t numelIn_ = -1;
+    size_t numelOut_ = -1;
+
+    // Wrapper method for the static checkForNCCLErrors which can be overridden
+    // for tests.
+    virtual std::exception_ptr checkForNCCLErrors();
+
+    friend std::ostream& operator<<(
+        std::ostream& output,
+        const WorkNCCL& workNCCL);
+
+   private:
+    // Helper function for synchronize
+    void synchronizeInternal(std::chrono::milliseconds timeout);
+
+    // Checks for NCCL errors and sets an appropriate exception_ptr.
+    void checkAndSetException();
+
+    // Just checks whether GPU execution has started, without modifying
+    // exception_ptr.
+    bool startedGPUExecutionInternal() const;
+
+    // Just checks whether GPU execution has completed, without modifying
+    // exception_ptr.
+    bool finishedGPUExecutionInternal() const;
+
+    // Reference to the store so that we can write aborted communicators
+    // to the store.
+    c10::intrusive_ptr<Store> store_;
+
+    // Store a reference to NCCL collective's outputs, used by result and to
+    // give a more descriptive message when representing the Work as a string.
+    std::shared_ptr<std::vector<at::Tensor>> outputs_;
+
+    // TORCH_NCCL_AVOID_RECORD_STREAMS implementation helper.
+    // Stores references to participating non-output tensors (ie inputs,
+    // flattened intermediates).
+    // We'll clear this list in synchronizeStream, just after user-facing
+    // stream(s) are synced with the nccl work stream(s).
+    // By keeping these refs (as well as outputs_) alive until after the
+    // collective's work rejoins the user-facing streams, we achieve
+    // caching allocator safety without any recordStream calls.
+    // For in-place collectives, some refs stashed here may alias outputs_,
+    // but that doesn't do any harm.
+    std::shared_ptr<std::vector<at::Tensor>> stashed_for_allocator_safety_;
+
+    // The future returned by getFuture.
+    c10::intrusive_ptr<at::ivalue::Future> future_;
+
+    bool timingEnabled_;
+    // unique id used to tell the trace buffer that this
+    // work has completed
+    c10::optional<uint64_t> trace_id_;
+    DebugLevel distDebugLevel_;
+    friend class ProcessGroupNCCL;
+  };
+
+  struct Options : Backend::Options {
+    // NOTE: timeout in ProcessGroupNCCL::Options denote the timeout for
+    // operations. This is only used when blockingWait_ is enabled.
+    explicit Options(bool is_high_priority_stream = false);
+
+    // return intrusive_ptr of the object
+    static c10::intrusive_ptr<Options> create(
+        bool is_high_priority_stream = false) {
+      return c10::make_intrusive<Options>(is_high_priority_stream);
+    }
+
+    // Schedule NCCL operations on high priority CUDA streams
+    bool is_high_priority_stream;
+
+#ifdef NCCL_HAS_COMM_NONBLOCKING
+    // Configure ranks
+    ncclConfig_t config = NCCL_CONFIG_INITIALIZER;
+#endif
+
+    // Optional "parent" backend and color to create communicators from
+    // via `ncclCommSplit`
+    std::shared_ptr<ProcessGroupNCCL> split_from;
+    int64_t split_color{0};
+    std::vector<uint64_t> global_ranks_in_group;
+  };
+
+  // If you wish to create multiple process groups, each with a potentially
+  // different rank and size, you can do so by passing a new store instance
+  // to each one. If you have only a single store object, you can
+  // use the `c10d::PrefixStore` to derive scoped instances.
+  // This is also what the Python API in torch.distributed does.
+  //
+  // The process group instance keeps a reference to the store because
+  // it may be used long after the constructor runs. In fact, the constructor
+  // doesn't create any NCCL communicators. A single NCCL communicator can
+  // only be used on a specific set of devices, and are therefore created
+  // on-demand when a collective runs. If another collective is executed later,
+  // against a different set of devices, the process group creates another NCCL
+  // communicator. These NCCL communicators are cached and reused if possible.
+  //
+  ProcessGroupNCCL(
+      const c10::intrusive_ptr<Store>& store,
+      int rank,
+      int size,
+      c10::intrusive_ptr<Options> options = Options::create());
+
+  // This constructor includes the deprecated `groupName` argument.
+  // If you have existing code that uses the `groupName`, you can replace
+  // it by specifying a `c10d::PrefixStore(groupName, store)` for store.
+  C10_DEPRECATED ProcessGroupNCCL(
+      const c10::intrusive_ptr<Store>& store,
+      int rank,
+      int size,
+      const std::string& groupName,
+      c10::intrusive_ptr<Options> options = Options::create())
+      : ProcessGroupNCCL(store, rank, size, options) {}
+
+  ~ProcessGroupNCCL() override;
+
+  uint64_t getUid() {
+    return static_cast<uint64_t>(uid_);
+  }
+
+  c10::intrusive_ptr<Options> getOptions() {
+    return options_;
+  }
+
+  const std::string getBackendName() const override {
+    return std::string(NCCL_BACKEND_NAME);
+  }
+
+  bool supportsSplitting() const override {
+    return true;
+  }
+
+  void startCoalescing() override;
+
+  c10::intrusive_ptr<Work> endCoalescing() override;
+
+  // For specifying a composite optype, such as ALLGATHER and REDUCE_SCATTER
+  c10::intrusive_ptr<Work> endCoalescing(OpType optype);
+
+  c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& tensors,
+      const BroadcastOptions& opts = BroadcastOptions()) override;
+
+  c10::intrusive_ptr<Work> _broadcast_oop(
+      at::Tensor& outputTensors,
+      at::Tensor& inputTensors,
+      const BroadcastOptions& opts = BroadcastOptions());
+
+  c10::intrusive_ptr<Work> allreduce_sparse(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceCoalescedOptions& opts =
+          AllreduceCoalescedOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& tensors,
+      const ReduceOptions& opts = ReduceOptions()) override;
+
+  c10::intrusive_ptr<Work> _reduce_oop(
+      at::Tensor& outputTensors,
+      at::Tensor& inputTensors,
+      const ReduceOptions& opts = ReduceOptions());
+
+  c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& outputbuffer,
+      at::Tensor& inputbuffer,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& outputTensorLists,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather_into_tensor_coalesced(
+      std::vector<at::Tensor>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> _reduce_scatter_base(
+      at::Tensor& outputTensor,
+      at::Tensor& inputTensor,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce_scatter_tensor_coalesced(
+      std::vector<at::Tensor>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& opts = BarrierOptions()) override;
+
+  c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& outputTensor,
+      at::Tensor& inputTensor,
+      std::vector<int64_t>& outputSplitSizes,
+      std::vector<int64_t>& inputSplitSizes,
+      const AllToAllOptions& opts = AllToAllOptions()) override;
+
+  c10::intrusive_ptr<Work> alltoall(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllToAllOptions& opts = AllToAllOptions()) override;
+
+  c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& tensors,
+      int dstRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& tensors,
+      int srcRank,
+      int tag) override;
+
+  void groupStart();
+
+  void groupEnd();
+
+  void groupEndNonblocking(std::shared_ptr<NCCLComm> comm);
+
+  c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const GatherOptions& opts = GatherOptions()) override;
+
+  c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ScatterOptions& opts = ScatterOptions()) override;
+
+  // Unsupported Ops
+  c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& tensors,
+      int tag) override;
+
+  // Agrees on an initial sequence number for the whole group by having rank 0
+  // create it and broadcast it to other ranks using the store.
+  void setSequenceNumberForGroup() override;
+
+  // Retrieves the current sequence number for the whole group, which should be
+  // in sync. If the returned number is not consistent across the group, it
+  // may indicate that there is some sort of collective desynchronization.
+  uint64_t getSequenceNumberForGroup() override;
+
+  // Return the total number of splits the communicators held by this process
+  // group have performed.
+  uint64_t getCommSplitCounter() const;
+
+  void registerOnCompletionHook(
+      std::function<void(std::shared_ptr<WorkInfo>)>&& hook) override;
+  void waitForPendingWorks() override;
+
+  void enableCollectivesTiming() override;
+
+  // Helper function for iteratively aborting communicators in the provided map
+  void abortCommsFromMap(
+      std::unordered_map<std::string, std::shared_ptr<NCCLComm>>& ncclCommsMap,
+      c10::optional<std::string> abortReason);
+
+  c10::intrusive_ptr<intra_node_comm::IntraNodeComm> initIntraNodeComm();
+
+  // Provides an API to abort the ProcessGroup (similar to ncclCommAbort)
+  // instead of relying on ProcessGroupNCCL destructor.
+  // return true if abort is successful, otherwise false
+  bool abort(c10::optional<std::string> abortReason = c10::nullopt);
+
+  void shutdown(c10::optional<std::string> reason = c10::nullopt);
+
+  void eagerConnectSingleDevice(at::Device device) override;
+
+  void performNocolorSplit(at::Device device);
+
+ protected:
+  // Helper that broadcasts nccl unique ID to all ranks through the store
+  void broadcastUniqueNCCLID(
+      ncclUniqueId* ncclID,
+      bool isSingleP2POp,
+      const std::string& devicesKey,
+      int p2pRank);
+
+  // Helper that either looks up the cached NCCL communicators or creates
+  // a new set of NCCL communicators as a cache entry
+  std::shared_ptr<NCCLComm> getNCCLComm(
+      const std::string& deviceKey,
+      at::Device& device,
+      OpType opType,
+      int p2pRank = 0,
+      bool isSendRecvSelf = false);
+
+  // Wrapper method which can be overridden for tests.
+  virtual std::exception_ptr checkForNCCLErrors(
+      std::shared_ptr<NCCLComm>& ncclComm);
+
+  // Ensure thaht if record is True, the work obj will be enqueued via
+  // workEnqueue
+  virtual c10::intrusive_ptr<ProcessGroupNCCL::WorkNCCL> initWork(
+      at::Device& device,
+      int rank,
+      OpType opType,
+      const char* profilingTitle = nullptr,
+      const std::vector<at::Tensor>& inputs = {},
+      const std::vector<at::Tensor>& outputs = {},
+      bool record = false);
+
+  // In the timeout case and we will dump debug info such as the NCCL flight
+  // recorder to storage. Down the road, if we have more complicated or blocking
+  // operations, we might need to use a side thread to do it.
+  bool dumpDebuggingInfo();
+
+ private:
+  int globalRankStart;
+  int globalRankStride;
+
+  // Helper that encapsulates work shared across all collective communication
+  // primitives.  The callbacks have the following signatures:
+  //
+  //    ncclResult_t fn(at::Tensor& input, at::Tensor& output,
+  //                    ncclComm_t, at::cuda::CUDAStream&);
+  //    void {pre,post}(std::vector<at::cuda::CUDAStream&>);
+  template <typename Fn>
+  c10::intrusive_ptr<Work> collective(
+      at::Tensor& input,
+      at::Tensor& output,
+      Fn fn,
+      OpType opType,
+      const char* profilingTitle = nullptr,
+      bool avoidRecordStreams = false);
+
+  template <typename Fn, typename PreProcess, typename PostProcess>
+  c10::intrusive_ptr<Work> collective(
+      at::Tensor& input,
+      at::Tensor& output,
+      Fn fn,
+      PreProcess pre,
+      PostProcess post,
+      OpType opType,
+      const char* profilingTitle = nullptr,
+      bool avoidRecordStreams = false);
+
+  template <typename Fn>
+  c10::intrusive_ptr<Work> collectiveCoalesced(
+      std::vector<at::Tensor>& input,
+      std::vector<at::Tensor>& output,
+      Fn fn,
+      OpType opType,
+      const char* profilingTitle = nullptr,
+      bool avoidRecordStreams = false);
+
+  // Helper that encapsulates work shared across point-to-point communication
+  // primitives. It is the same structure as the helper used for collective
+  // communication primitives.
+  template <typename Fn>
+  c10::intrusive_ptr<Work> pointToPoint(
+      at::Tensor& tensor,
+      Fn fn,
+      int peer,
+      OpType opType,
+      const char* profilingTitle = nullptr);
+
+  template <typename Fn, typename PreProcess, typename PostProcess>
+  c10::intrusive_ptr<Work> pointToPoint(
+      at::Tensor& tensor,
+      Fn fn,
+      int peer,
+      OpType opType,
+      PreProcess pre,
+      PostProcess post,
+      const char* profilingTitle);
+
+  c10::intrusive_ptr<Work> allreduce_impl(
+      at::Tensor& tensor,
+      const AllreduceOptions& opts = AllreduceOptions());
+
+  // Checks for NCCL errors on each of the communicators and returns an
+  // appropriate exception_ptr (nullptr if no errors).
+  static std::exception_ptr checkForNCCLErrorsInternal(
+      std::shared_ptr<NCCLComm>& ncclComm);
+
+  // Function that runs as part of a separate thread and checks for errors on
+  // NCCL communicators. We need a separate thread to check for NCCL errors
+  // since we can't rely on the user calling certain methods like wait(),
+  // isCompleted() etc. to detect and remediate errors. In addition to this, we
+  // need a mechanism to safely abort and remove NCCL communicators from our
+  // cache. This can be done cleanly by having a thread for the ProcessGroupNCCL
+  // class. Attempting to modify the communicator cache from the WorkNCCL class
+  // might run into issues with object lifetime since the ProcessGroupNCCL
+  // object might get destroyed before the WorkNCCL object.
+  void ncclCommWatchdog();
+
+  // Return the CUDA device most likely associated with this backend.
+  // If we aren't bound to a specific device, there is no strict
+  // guarantee that this heuristic is the correct assignment of ranks
+  // to GPUs that Python layers use, but in practice it tends to be.
+  // Fortunately we don't rely on this for correctness of any tensor
+  // operations, just for ancillary uses like barriers.
+  at::Device guessDeviceForRank() const;
+
+  // Destroys initialized NCCL communicators in devNCCLComMap_ given by input
+  // key. Throws if there are no communicators to destroy. Also removes
+  // communicators from the cache and clears used device indices.
+  void destroyNCCLComms(const std::string& devNCCLCommMapKey);
+
+  // Watchdog's inside loop.
+  // Takes care of cleaning up completed work, and aborting upon failure or
+  // timeout.
+  void watchdogHandler();
+
+  void runHookLoop();
+
+  // Desync debug helper
+  void logWorkStart(WorkNCCL& work);
+
+  // Desync debug helper
+  void logWorkEnd(WorkNCCL& work);
+
+  // Generates a prefix that is unique to this process group and rank, for
+  // disambiguating logs
+  std::string createLogPrefix() const;
+
+  // Returns the unique prefix created in createLogPrefix
+  const std::string& logPrefix() const;
+
+  // Returns the global rank of the device. This function assumes that users
+  // always create a default global process group(PG) which includes all
+  // devices. It is called in the constructor of ProcessGroupNCCL, so it always
+  // return the rank_ of the the very first PG created, aka, default global PG.
+  const int& globalRank() const;
+
+  // Returns the global ranks of a PG.
+  const std::vector<uint64_t>& groupRanks() const;
+
+ protected:
+  // Function that runs as part of a separate thread aside from watchdog
+  // thread because we need to check the heartbeat from watchdog thread
+  // so that when we get stuck in some NCCL/CUDA calls,
+  // we can dump the debugging information and abort the process.
+  virtual void heartbeatMonitor();
+
+  // Function that directly trigger std::abort so that the whole process
+  // gets terminated.
+  virtual void terminateProcess(std::string errMsg);
+
+  // A helper function to wait for a future to complete or timeout.
+  void waitForFutureOrTimeout(
+      std::future<bool>& fut,
+      const std::chrono::milliseconds& timeOutMilSec,
+      const std::string& futDescription,
+      bool throwException = false);
+
+  // When watchdog timeout, this function will be called and return debug info
+  // for users. For now we only get information from retrieveDesyncReport.
+  // We are working on enabling more useful debug information for watchdog
+  // timeout.
+  virtual std::string getNCCLWatchdogDebugInfo();
+
+  static const int64_t kWatchdogThreadSleepMillis;
+
+  // The store is used to broadcast the NCCL unique ID of rank 0. This store
+  // comes with prefix and it is different across ProcessGroup NCCL instances
+  // (aka, different ProcessGroups).
+  c10::intrusive_ptr<Store> store_;
+
+  // Reference to the store without prefix so that keys are same across all
+  // ProcessGroup NCCL instances and (key, value) pairs written to the store are
+  // global.
+  c10::intrusive_ptr<Store> globalStore_;
+
+  bool storeError_{false};
+
+  const c10::intrusive_ptr<Options> options_;
+
+  // The number of NCCL communicators that have been created during
+  // the lifetime of this process group. This sequence number is
+  // used to scope keys used in the store.
+  uint64_t ncclCommCounter_{0};
+
+  // The store keys to trace the last NCCL collective kernel CUDA events - start
+  // event and end event respectively. These are used to do desync root cause
+  // analysis.
+  const std::string traceKeyStart_;
+  const std::string traceKeyEnd_;
+
+  // The NCCL communicator that the process group has cached.
+  //
+  // For collective operations:
+  // The key is a list of GPU devices that an operation is operating on
+  // The GPU devices are stored in a device sequence and the cache NCCL
+  // communicator is associated with this GPU device sequence
+  //
+  // e.g. If the process group op only uses device 0, then the value of
+  // the used device string stored (value of the hashmap) would be "0".
+  //
+  //      If the process group op uses device 0 - 7 and the each tensor of the
+  //      input tensor list is on device, 0, 1, 2, 3, 4, 5, 6, 7 separately,
+  //      then the value of the used device string (key) stored would be
+  //      "0,1,2,3,4,5,6,7"
+  //
+  //      If the process group op uses device 0 - 7 and the each tensor of the
+  //      input tensor list is on device, 0, 4, 5, 6, 7, 1, 2, 3 separately,
+  //      then the value of the used device string stored would be
+  //      "0,4,5,6,7,1,2,3"
+  //
+  //      Note that the order of the device for the tensor list matters.
+  //
+  // For point-to-point operations:
+  // The key is a string of my current rank and the peer process rank.
+  // e.g. If process 1 and process 2 are involved in a point-to-point
+  // communication, the key will be "1:2" on both processes. Note: this is for
+  // the scenario where there is only 1 GPU per process. When it comes to
+  // multiple GPUs per process, this part may need to redesigned.
+  std::unordered_map<std::string, std::shared_ptr<NCCLComm>> devNCCLCommMap_;
+
+  // The NCCL communicators currently in process of being initialized.
+  std::unordered_map<std::string, std::shared_ptr<NCCLComm>>
+      inInitializationCommMap_;
+
+  // Map from ncclUniqueId to appropriate communicator.
+  std::unordered_map<std::string, std::shared_ptr<NCCLComm>> ncclIdToCommMap_;
+
+  // Mutex to guard maps like devNCCLCommMap_ and ncclIdToCommMap_.
+  std::mutex mutex_;
+
+  // Heartbeat of watchdog thread.
+  std::atomic_uint64_t heartbeat_;
+
+  // The time interval used for deciding whether there is no watchdog heartbeat.
+  int heartbeatTimeoutInSec_;
+
+  // timeout for the dump to finish.
+  int waitTimeoutDumpInMilSec_;
+
+  // Interval of check coordinated signals in ProcessGroupNCCL from other ranks
+  // e.g., trigger the dump of the debugging info for timeout when notified.
+  int coordCheckIntervalMilSec_;
+
+  // Size of ring buffer where we store NCCL Traces for debugging.
+  int ncclTraceBufferSize_;
+
+  // We gate the heartbeat monitor thread so that we can roll it out gradually.
+  std::atomic<bool> monitorThreadEnabled_;
+
+  // Monitor thread which checks the heartbeat of Watchdog thread.
+  // If the monitor thread finds there is no heartbeat, it will dump debug info
+  // and then kill the watchdog thread to avoid hang.
+  std::thread ncclHeartbeatMonitorThread_;
+
+  // Watchdog thread which looks for errors on the cached NCCL communicators.
+  std::thread ncclCommWatchdogThread_;
+
+  std::thread onCompletionHookThread_;
+
+  // Whether or not we should terminate the watchdog and workCleanup threads.
+  std::atomic<bool> terminateProcessGroup_;
+
+  // Whether or not we should terminate the heartbeat monitoring threads.
+  std::atomic<bool> terminateHeartbeatMonitorThread_;
+
+  // Whether we are in the shutdown mode when we are trying to get debug info,
+  // such as desync report.
+  std::atomic<bool> collectiveDebugInfoMode_;
+
+  // Whether there are hooks pending to be fired
+  std::atomic<bool> hasPendingHooks_;
+
+  // This is the signal from watchdog threads to indicate whether the monitor
+  // thread should dump. Making it static so that it is accessiable from all the
+  // PGs. With this flag, monitor thread would dump debug info under any one of
+  // the 3 conditions: 1: this flag is set to true by the watchdog thread when
+  // it detects a timeout. 2: timeout signal is received from
+  // other ranks through tcpstore 3: no heartbeat of watchdog Note that only the
+  // monitor thread from PG0 should dump the debug info and only once
+  static std::atomic<bool> shouldDump_;
+
+  // Mutex to Guard workMetaList_
+  std::mutex workMetaListMutex_;
+
+  // Mutex to Guard monitorWakeUpCV_
+  std::mutex monitorMutex_;
+
+  bool writeDebugInfo_ = false;
+
+  // Condition Variable for watchdog thread sleep
+  std::condition_variable workMetaListCV_;
+
+  // Condition Variable for monitor thread to wake up early
+  std::condition_variable monitorWakeUpCV_;
+
+  // Vector to Store WorkNCCL pointers
+  std::list<ProcessGroupNCCL::WorkNCCL> workMetaList_;
+
+  std::chrono::time_point<std::chrono::steady_clock> lastWorkListUpdateTime_;
+
+  // Mutex to Guard workMetaList_
+  std::mutex completedWorkListMutex_;
+
+  // Condition Variable for watchdog thread sleep
+  std::condition_variable completedWorkListCV_;
+
+  std::list<ProcessGroupNCCL::WorkNCCL> completedWorkList_;
+
+  // Add Work Pointer to workVector
+  void workEnqueue(c10::intrusive_ptr<ProcessGroupNCCL::WorkNCCL>);
+
+  // The CUDA streams used by NCCL kernels
+  std::unordered_map<std::string, at::cuda::CUDAStream> ncclStreams_;
+
+  // The CUDA events used to sync NCCL streams
+  std::unordered_map<std::string, at::cuda::CUDAEvent> ncclEvents_;
+
+  // Device Indexes used for all collectives in this group
+  std::set<int> usedDeviceIdxs_;
+
+  // Flag to denote if a coalescing groupStart/groupEnd block is active
+  int coalescing_state_ = 0;
+
+  // Stores device indexes for all collectives run inside a coalescing block
+  std::vector<at::Device> coalescedDevices_;
+
+  // Stores communicators for all collectives run inside a coalescing block
+  std::vector<std::shared_ptr<NCCLComm>> coalescedComms_;
+
+  // map from the key: "group name + pg counter (ID)" to the
+  // unique NCCL ID count. This needs to be group and pg specific
+  //
+  // For each process group, we need a uniform unique NCCL ID counter to ensure
+  // that NCCL operation in this process group can be completed successfully.
+  // Since each process group ID belongs to a group name, the key to this map
+  // is a combination of group name and ProcessGroupNCCL ID.
+  static std::unordered_map<std::string, ssize_t> pgUniqueNCCLIDCnt_;
+
+  // map from group name to the pg counter (ID) within that group
+  //
+  // For each group with the "group name" (which is the key), we need to
+  // keep track of a unique process group ID when creating a new
+  // ProcessGroupNCCL for this "group name". Therefore, the value of this
+  // map keeps the unique ProcessGroupNCCL's ID for a specific group with
+  // the "group name". The reason we need a per-group process group ID counter
+  // is that different group can have different ranks and we need ensure that
+  // each group has its own uniform process group ID for all its ranks.
+  static std::unordered_map<std::string, ssize_t> processGroupCounterMap_;
+
+  // Whether or not wait() and synchronize() are blocking operations that wait
+  // for the operation to complete.
+  bool blockingWait_ = false;
+
+  // Whether to abort the communicators when users call destroy_process_group().
+  // If yes, communicators will be aborted when destroy_process_group is called,
+  // but not in destructor.
+  bool abortInDestroyProcessGroup_ = false;
+
+  // Whether or not to hook the cache allocator to register all allocated
+  // tensors
+  bool useTensorRegisterAllocatorHook_ = false;
+
+  // Whether or not the workCleanupThread is used to perform async error
+  // handling.
+  ErrorHandlingMode asyncErrorHandling_ = NoHandling;
+
+  // Whether or not to enable timeout root cause analysis.
+  bool desyncDebug_;
+
+  // Whether or not to dump debug info on timeout
+  bool dumpOnTimeout_;
+
+  // Whether or not to create start CUDAEvent and enable timing for start
+  // and end events. Note that enableTiming_ is always true if desyncDebug_
+  // is set to true.
+  std::atomic<bool> enableTiming_;
+
+  // Flag to enable the print of hash value of input/output of collectives for
+  // verification.
+  std::atomic<bool> enableCollecticeHashDebug_;
+
+  // Whether or not TORCH_NCCL_AVOID_RECORD_STREAMS was set
+  bool avoidRecordStreams_ = false;
+
+  // Set of communicators that this process group has aborted and their
+  // ncclUniqueId has been written to the store. We don't need a lock
+  // for this map since only the watchdog thread accesses this set. The
+  // set contains the string representation of ncclUniqueId.
+  std::unordered_set<std::string> abortedComms_;
+
+  // The number of active ncclGroupStart() calls. This counter will be increased
+  // by 1 when ncclGroupStart() is called and decreased by 1 when ncclGroupEnd()
+  // is called.
+  static thread_local uint64_t ncclActiveGroupCounter_;
+
+  // Counting for the sequential number of NCCL collective call.
+  // (specifically, how many actual kernels we launched, which differs from
+  // op_id_ when coalescing is enabled)
+  uint64_t seq_{0};
+
+  // Incrementing counter for logical operations (collective or p2p) issued on
+  // the ProcessGroup
+  uint64_t op_id_{0};
+
+  // the sequential number of the last colletive enqueued into workMetaList_
+  // This is useful for indentifying a rank that has not join a collective
+  uint64_t lastEnqueuedSeq_;
+
+  // the sequential number of the last colletive completed marked by
+  // the watchdog thread
+  uint64_t lastCompletedSeq_;
+
+  std::exception_ptr watchDogException_ = nullptr;
+
+  size_t uid_;
+
+  std::string logPrefix_;
+
+  c10::intrusive_ptr<intra_node_comm::IntraNodeComm> intraNodeComm_;
+};
+
+TORCH_API std::string dump_nccl_trace();
+
+// Gets a mutable reference to a global optional function.  Heartbeat Monitor
+// will query this function and if available, call it to dump traces. Inside
+// fbcode, we store a function here that uses an internal tool for process
+// tracing
+TORCH_API c10::optional<std::function<std::string()>>& get_cpp_trace_dumper();
+
+// Similar to get_cpp_trace_dumper, this stores a function defined in
+// torch-python layer that lets us check whether the GIL can be acquired,
+// helpful for instrumenting in cases where a hang was observed.
+typedef bool (*gil_checker_t)();
+
+TORCH_API gil_checker_t& get_gil_checker();
+} // namespace c10d
+
+#endif // USE_C10D_NCCL

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/ProcessGroupRoundRobin.hpp

@@ -0,0 +1,113 @@
+#pragma once
+
+#include <vector>
+
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+
+namespace c10d {
+
+constexpr const char* ROUND_ROBIN_BACKEND_NAME = "round_robin";
+
+// ProcessGroupRoundRobin implements simple load balancing.
+//
+// It is constructed with multiple processes groups. Each call is dispatched to
+// one of the specified process groups in a round robin fashion. Each process
+// group instance must have the same rank and size.
+//
+// All functions of the class are expected to be called in the same order
+// across all processes in the process group. This is the only way that we
+// can guarantee to match up the same calls among all processes.
+//
+class TORCH_API ProcessGroupRoundRobin final : public ProcessGroup {
+ public:
+  explicit ProcessGroupRoundRobin(
+      int rank,
+      int size,
+      std::vector<c10::intrusive_ptr<ProcessGroup>> processGroups);
+
+  ~ProcessGroupRoundRobin() override;
+
+  const std::string getBackendName() const override {
+    return std::string(ROUND_ROBIN_BACKEND_NAME);
+  }
+
+  c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& tensors,
+      const BroadcastOptions& opts = BroadcastOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceCoalescedOptions& opts =
+          AllreduceCoalescedOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& tensors,
+      const ReduceOptions& opts = ReduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& outputTensorLists,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& outputs,
+      std::vector<at::Tensor>& inputs,
+      const GatherOptions& opts = GatherOptions()) override;
+
+  c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& outputs,
+      std::vector<std::vector<at::Tensor>>& inputs,
+      const ScatterOptions& opts = ScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& outputs,
+      std::vector<std::vector<at::Tensor>>& inputs,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& outputTensor,
+      at::Tensor& inputTensor,
+      std::vector<int64_t>& outputSplitSizes,
+      std::vector<int64_t>& inputSplitSizes,
+      const AllToAllOptions& opts = AllToAllOptions()) override;
+
+  c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& tensors,
+      int dstRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& tensors,
+      int srcRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& tensors,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& opts = BarrierOptions()) override;
+
+ private:
+  std::vector<c10::intrusive_ptr<ProcessGroup>> processGroups_;
+  std::vector<c10::intrusive_ptr<ProcessGroup>>::const_iterator iterator_;
+
+  // Returns the next ProcessGroup to use.
+  const c10::intrusive_ptr<ProcessGroup>& next();
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/ProcessGroupWrapper.hpp

@@ -0,0 +1,140 @@
+#pragma once
+
+#ifdef USE_C10D_GLOO
+
+#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <torch/csrc/distributed/c10d/Types.hpp>
+#include <torch/csrc/distributed/c10d/Utils.hpp>
+
+namespace c10d {
+
+class TORCH_API ProcessGroupWrapper : public Backend {
+ public:
+  explicit ProcessGroupWrapper(
+      c10::intrusive_ptr<Backend> backend,
+      c10::intrusive_ptr<Backend> glooBackend);
+
+  const std::string getBackendName() const override;
+
+  c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& data,
+      const BroadcastOptions& opts = BroadcastOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& data,
+      const AllreduceOptions& opts = AllreduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceCoalescedOptions& opts =
+          AllreduceCoalescedOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce(
+      std::vector<at::Tensor>& tensors,
+      const ReduceOptions& opts = ReduceOptions()) override;
+
+  c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> _allgather_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  // This function is deprecated and will be moved out of ProcessGroup to comms:
+  // * do not add dependencies on this function,
+  // * do not implement it in your ProcessGroup, implement _allgather_base
+  //   instead.
+  c10::intrusive_ptr<Work> allgather_coalesced(
+      std::vector<std::vector<at::Tensor>>& outputTensorLists,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override;
+
+  c10::intrusive_ptr<Work> gather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const GatherOptions& opts = GatherOptions()) override;
+
+  c10::intrusive_ptr<Work> scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ScatterOptions& opts = ScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override;
+
+  c10::intrusive_ptr<Work> alltoall_base(
+      at::Tensor& outputTensor,
+      at::Tensor& inputTensor,
+      std::vector<int64_t>& outputSplitSizes,
+      std::vector<int64_t>& inputSplitSizes,
+      const AllToAllOptions& opts = AllToAllOptions()) override;
+
+  c10::intrusive_ptr<Work> alltoall(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllToAllOptions& opts = AllToAllOptions()) override;
+
+  void monitoredBarrier(const BarrierOptions& opts, bool waitAllRanks = false)
+      override;
+
+  // Agrees on an initial sequence number for the whole group by having rank 0
+  // create it and broadcast it to other ranks using the store. Only implemented
+  // for GLOO and NCCL backends currently.
+  // dont implement this
+  void setSequenceNumberForGroup() override;
+
+  // Retrieves the current sequence number for the whole group, which should be
+  // in sync. If the returned number is not consistent across the group, it
+  // may indicate that there is some sort of collective desynchronization.
+  uint64_t getSequenceNumberForGroup() override; // just call underlying
+
+  c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& tensors,
+      int dstRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& tensors,
+      int srcRank,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> recvAnysource(
+      std::vector<at::Tensor>& tensors,
+      int tag) override;
+
+  c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& opts = BarrierOptions()) override;
+
+  c10::intrusive_ptr<Work> _reduce_scatter_base(
+      at::Tensor& outputBuffer,
+      at::Tensor& inputBuffer,
+      const ReduceScatterOptions& opts) override;
+
+  void startCoalescing() override;
+
+  c10::intrusive_ptr<Work> endCoalescing() override;
+
+  c10::intrusive_ptr<Backend> getWrappedPg() const;
+
+ private:
+  // Underlying process group that actual application collectives will be
+  // dispatched to
+  c10::intrusive_ptr<Backend> backend_;
+  // Gloo process group responsible for internal coordination such as monitored
+  // barrier, sequence number checking, collective fingerprint collecting.
+  c10::intrusive_ptr<Backend> glooBackend_;
+  // Conducts several checks to ensure that the underlying collective is well
+  // formed with the goal of notifying the user about incorrect collective use
+  // in the application.
+  void runCollectiveChecks(
+      OpType op_type,
+      const std::vector<at::Tensor>& tensors);
+};
+} // namespace c10d
+
+#endif // USE_C10D_GLOO

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/PyProcessGroup.hpp

@@ -0,0 +1,219 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <torch/csrc/jit/python/pybind_utils.h>
+#include <torch/csrc/utils/pybind.h>
+
+namespace c10d {
+
+// PyProcessGroup is a pybind11 trampoline class to allow a Python
+// class to inherit from torch.distributed.ProcessGroup
+class PyProcessGroup : public ProcessGroup {
+ public:
+  // PyWork is a pybind11 trampoline class to allow a Python
+  // class to inherit from torch.distributed.Work
+  class PyWork : public Work {
+   public:
+    PyWork() = default;
+
+    bool wait(std::chrono::milliseconds timeout = kNoTimeout) override {
+      PYBIND11_OVERRIDE(
+          bool, /* Return type */
+          Work, /* Parent class */
+          wait, /* Name of function in C++ */
+          timeout);
+    }
+
+    c10::intrusive_ptr<c10::ivalue::Future> getFuture() override {
+      // We cannot use PYBIND11_OVERRIDE because:
+      // 1. We have to >MANUALLY< unwrap the PyFutureWrapper and
+      // 2. The python name is get_future
+      pybind11::gil_scoped_acquire gil;
+      auto override =
+          pybind11::get_override(static_cast<const Work*>(this), "get_future");
+
+      if (override) {
+        py::object o = override();
+        auto futWrapper =
+            o.cast<std::shared_ptr<torch::jit::PythonFutureWrapper>>();
+        return futWrapper->fut;
+      }
+
+      return Work::getFuture();
+    }
+  };
+
+  using ProcessGroup::ProcessGroup;
+
+  const std::string getBackendName() const override {
+    PYBIND11_OVERRIDE_PURE(
+        std::string, /* Return type */
+        ProcessGroup, /* Parent class */
+        getBackendName, /* Name of function in C++ */
+    );
+  }
+
+  c10::intrusive_ptr<Work> allgather(
+      std::vector<std::vector<at::Tensor>>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        allgather, /* Name of function in C++ */
+        outputTensors,
+        inputTensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> allgather_into_tensor_coalesced(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const AllgatherOptions& opts = AllgatherOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        allgather_into_tensor_coalesced, /* Name of function in C++ */
+        outputTensors,
+        inputTensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> allreduce(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceOptions& opts = AllreduceOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        allreduce, /* Name of function in C++ */
+        tensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> allreduce_coalesced(
+      std::vector<at::Tensor>& tensors,
+      const AllreduceCoalescedOptions& opts =
+          AllreduceCoalescedOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        allreduce_coalesced, /* Name of function in C++ */
+        tensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> barrier(
+      const BarrierOptions& opts = BarrierOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        barrier, /* Name of function in C++ */
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> broadcast(
+      std::vector<at::Tensor>& tensors,
+      const BroadcastOptions& opts = BroadcastOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        broadcast, /* Name of function in C++ */
+        tensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> reduce_scatter(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<std::vector<at::Tensor>>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        reduce_scatter, /* Name of function in C++ */
+        outputTensors,
+        inputTensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> reduce_scatter_tensor_coalesced(
+      std::vector<at::Tensor>& outputTensors,
+      std::vector<at::Tensor>& inputTensors,
+      const ReduceScatterOptions& opts = ReduceScatterOptions()) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        reduce_scatter_tensor_coalesced, /* Name of function in C++ */
+        outputTensors,
+        inputTensors,
+        opts);
+  }
+
+  c10::intrusive_ptr<Work> send(
+      std::vector<at::Tensor>& tensors,
+      int dstRank,
+      int tag) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        send, /* Name of function in C++ */
+        tensors,
+        dstRank,
+        tag);
+  }
+
+  c10::intrusive_ptr<Work> recv(
+      std::vector<at::Tensor>& tensors,
+      int srcRank,
+      int tag) override {
+    PYBIND11_OVERRIDE(
+        c10::intrusive_ptr<Work>, /* Return type */
+        ProcessGroup, /* Parent class */
+        recv, /* Name of function in C++ */
+        tensors,
+        srcRank,
+        tag);
+  }
+};
+
+class TORCH_PYTHON_API PythonOnCompletionHook {
+ public:
+  // Wraps a py::object hook and acquires Python GIL in dtor before
+  // destructing the hook object.
+  PythonOnCompletionHook(py::object hook) : hook_(std::move(hook)) {}
+
+  ~PythonOnCompletionHook() {
+    py::gil_scoped_acquire ag;
+    hook_.dec_ref();
+    // Explicitly set hook_ to nullptr to prevent py::object's dtor
+    // to decref on the PyObject again.
+    // See Note [Destructing py::object] in python_ivalue.h
+    hook_.ptr() = nullptr;
+  }
+
+  void operator()(std::shared_ptr<WorkInfo> workInfo) const {
+    std::exception_ptr eptr;
+    {
+      py::gil_scoped_acquire acquire;
+      try {
+        hook_(workInfo);
+      } catch (py::error_already_set& e) {
+        // py::error_already_set requires GIL to destruct, take
+        // special care.
+        eptr = std::make_exception_ptr(std::runtime_error(e.what()));
+        e.restore();
+        PyErr_Clear();
+      } catch (std::exception& e) {
+        eptr = std::current_exception();
+      }
+    }
+    // No more Python-related stuff at this point, i.e., this
+    // exception can be captured and handled by PG backend.
+    if (eptr)
+      std::rethrow_exception(eptr);
+  }
+
+ private:
+  py::object hook_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/RankLocal.hpp

@@ -0,0 +1,73 @@
+
+#pragma once
+
+#include <shared_mutex>
+
+#include <torch/csrc/autograd/function.h>
+
+namespace c10d {
+
+// `RankLocal` maintains a unique instance of T for each non-autograd thread.
+// For non-autograd threads, `RankLocal<T>::get()` functions similar to
+// thread_local. For autograd threads, `RankLocal<T>::get()` returns the
+// instance of T corresponding to the enqueuing non-autograd thread. The
+// mechanism allows for rank-specific context shared between forward and
+// backward. It works for both the one-rank-per-process and one-rank-per-thread
+// scenarios.
+//
+// NOTE: RankLocal doesn't make the underlying objects thread-safe.
+template <typename T>
+class RankLocal {
+ public:
+  RankLocal(const RankLocal&) = delete;
+  RankLocal& operator=(const RankLocal&) = delete;
+
+  static T& get() {
+    // Fast path: non-autograd threads can simply return
+    // the object reference cached in TLS.
+    if (cached_ != nullptr) {
+      return *cached_;
+    }
+    const auto node = torch::autograd::get_current_node();
+    auto fwd_thread_id = node == nullptr ? at::RecordFunction::currentThreadId()
+                                         : node->thread_id();
+    // Optimistically acquire the read lock first, since most likely we are in
+    // an autograd thread and the object has already been constructed.
+    {
+      std::shared_lock read_lock(lock_);
+      auto it = thread_id_to_rank_local_.find(fwd_thread_id);
+      if (it != thread_id_to_rank_local_.end()) {
+        // Cache for non-autograd threads
+        if (node == nullptr) {
+          cached_ = &it->second;
+        }
+        return it->second;
+      }
+    }
+
+    std::unique_lock write_lock(lock_);
+    auto [it, _] = thread_id_to_rank_local_.try_emplace(fwd_thread_id);
+    // Cache for non-autograd threads
+    if (node == nullptr) {
+      cached_ = &it->second;
+    }
+    return it->second;
+  }
+
+ private:
+  RankLocal(){};
+  thread_local static T* cached_;
+  static std::unordered_map<uint64_t, T> thread_id_to_rank_local_;
+  static std::shared_mutex lock_;
+};
+
+template <typename T>
+thread_local T* RankLocal<T>::cached_ = nullptr;
+
+template <typename T>
+std::unordered_map<uint64_t, T> RankLocal<T>::thread_id_to_rank_local_;
+
+template <typename T>
+std::shared_mutex RankLocal<T>::lock_;
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/Store.hpp

@@ -0,0 +1,101 @@
+#pragma once
+
+#include <chrono>
+#include <cstdint>
+#include <stdexcept>
+#include <string>
+#include <vector>
+
+#include <c10/macros/Macros.h>
+#include <torch/custom_class.h>
+
+namespace c10d {
+
+// callback function will be given arguments (optional<string> oldValue,
+// optional<string> newValue)
+using WatchKeyCallback =
+    std::function<void(c10::optional<std::string>, c10::optional<std::string>)>;
+
+class TORCH_API Store : public torch::CustomClassHolder {
+ public:
+  static constexpr std::chrono::milliseconds kDefaultTimeout =
+      std::chrono::seconds(300);
+  static constexpr std::chrono::milliseconds kNoTimeout =
+      std::chrono::milliseconds::zero();
+
+  Store() : timeout_(kDefaultTimeout) {}
+
+  explicit Store(const std::chrono::milliseconds& timeout)
+      : timeout_(timeout) {}
+
+  Store(const Store&) = default;
+  Store(Store&&) noexcept = default;
+
+  ~Store() override = default;
+
+  void set(const std::string& key, const std::string& value);
+
+  virtual void set(
+      const std::string& key,
+      const std::vector<uint8_t>& value) = 0;
+
+  std::string compareSet(
+      const std::string& key,
+      const std::string& currentValue,
+      const std::string& newValue);
+
+  virtual std::vector<uint8_t> compareSet(
+      const std::string& key,
+      const std::vector<uint8_t>& currentValue,
+      const std::vector<uint8_t>& newValue) {
+    TORCH_INTERNAL_ASSERT(false, "Not implemented.");
+  }
+
+  std::string get_to_str(const std::string& key);
+
+  virtual std::vector<uint8_t> get(const std::string& key) = 0;
+
+  virtual int64_t add(const std::string& key, int64_t value) = 0;
+
+  virtual bool deleteKey(const std::string& key) = 0;
+
+  virtual bool check(const std::vector<std::string>& keys) = 0;
+
+  virtual int64_t getNumKeys() = 0;
+
+  virtual void wait(const std::vector<std::string>& keys) = 0;
+
+  virtual void wait(
+      const std::vector<std::string>& keys,
+      const std::chrono::milliseconds& timeout) = 0;
+
+  virtual const std::chrono::milliseconds& getTimeout() const noexcept;
+
+  virtual void setTimeout(const std::chrono::milliseconds& timeout);
+
+  // watchKey() is deprecated and no longer supported.
+  virtual void watchKey(
+      const std::string& /* unused */,
+      WatchKeyCallback /* unused */) {
+    TORCH_CHECK(false, "watchKey is deprecated, no implementation support it.");
+  }
+
+  virtual void append(
+      const std::string& key,
+      const std::vector<uint8_t>& value);
+
+  virtual std::vector<std::vector<uint8_t>> multiGet(
+      const std::vector<std::string>& keys);
+
+  virtual void multiSet(
+      const std::vector<std::string>& keys,
+      const std::vector<std::vector<uint8_t>>& values);
+
+  // Returns true if this store support append, multiGet and multiSet
+  virtual bool hasExtendedApi() const;
+
+ protected:
+  std::chrono::milliseconds timeout_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/TCPStore.hpp

@@ -0,0 +1,164 @@
+#pragma once
+
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+
+#include <torch/csrc/distributed/c10d/Store.hpp>
+
+namespace c10d {
+namespace detail {
+
+class TCPServer;
+
+class TCPClient;
+
+struct SocketAddress {
+  std::string host{};
+  std::uint16_t port{};
+};
+
+class Counter {
+ public:
+  void update(double val);
+  std::unordered_map<std::string, double> observe() const;
+
+  double mean() const noexcept {
+    return mean_;
+  }
+  int64_t count() const noexcept {
+    return count_;
+  }
+  double variance() const noexcept {
+    return m2_ / count_;
+  }
+  double sample_variance() const noexcept {
+    return m2_ / (count_ - 1);
+  }
+
+ private:
+  int64_t count_ = 0;
+  double mean_ = 0;
+  double m2_ = 0;
+};
+
+} // namespace detail
+
+struct TCPStoreOptions {
+  static constexpr std::uint16_t kDefaultPort = 29500;
+
+  std::uint16_t port = kDefaultPort;
+  bool isServer = false;
+  c10::optional<std::size_t> numWorkers = c10::nullopt;
+  bool waitWorkers = true;
+  std::chrono::milliseconds timeout = Store::kDefaultTimeout;
+
+  // A boolean value indicating whether multiple store instances can be
+  // initialized with the same host:port pair.
+  bool multiTenant = false;
+
+  // If specified, and if isServer is true, the underlying TCPServer will take
+  // over the bound socket associated to this fd. This option is useful to avoid
+  // port assignment races in certain scenarios.
+  c10::optional<int> masterListenFd = c10::nullopt;
+
+  // A boolean value indicating whether to use the experimental libUV backend.
+  bool useLibUV = false;
+};
+
+class TORCH_API TCPStore : public Store {
+ public:
+  explicit TCPStore(std::string host, const TCPStoreOptions& opts = {});
+
+  [[deprecated("Use TCPStore(host, opts) instead.")]] explicit TCPStore(
+      const std::string& masterAddr,
+      std::uint16_t masterPort,
+      c10::optional<int> numWorkers = c10::nullopt,
+      bool isServer = false,
+      const std::chrono::milliseconds& timeout = kDefaultTimeout,
+      bool waitWorkers = true);
+
+  ~TCPStore() override;
+
+  void set(const std::string& key, const std::vector<uint8_t>& value) override;
+
+  std::vector<uint8_t> compareSet(
+      const std::string& key,
+      const std::vector<uint8_t>& expectedValue,
+      const std::vector<uint8_t>& desiredValue) override;
+
+  std::vector<uint8_t> get(const std::string& key) override;
+
+  int64_t add(const std::string& key, int64_t value) override;
+
+  bool deleteKey(const std::string& key) override;
+
+  bool check(const std::vector<std::string>& keys) override;
+
+  int64_t getNumKeys() override;
+
+  void wait(const std::vector<std::string>& keys) override;
+
+  void wait(
+      const std::vector<std::string>& keys,
+      const std::chrono::milliseconds& timeout) override;
+
+  void append(const std::string& key, const std::vector<uint8_t>& value)
+      override;
+
+  std::vector<std::vector<uint8_t>> multiGet(
+      const std::vector<std::string>& keys) override;
+
+  void multiSet(
+      const std::vector<std::string>& keys,
+      const std::vector<std::vector<uint8_t>>& values) override;
+
+  bool hasExtendedApi() const override;
+
+  // Waits for all workers to join.
+  void waitForWorkers();
+
+  // Returns the hostname used by the TCPStore.
+  const std::string& getHost() const noexcept {
+    return addr_.host;
+  }
+
+  // Returns the port used by the TCPStore.
+  std::uint16_t getPort() const noexcept {
+    return addr_.port;
+  }
+
+  std::unordered_map<std::string, std::unordered_map<std::string, double>>
+  collectClientCounters() const noexcept;
+
+  bool isLibUvBackend() const noexcept {
+    return usingLibUv_;
+  }
+
+  // note(xilunwu): this function is only for internal testing
+  void _splitSet(const std::string& key, const std::vector<uint8_t>& data);
+
+ private:
+  int64_t incrementValueBy(const std::string& key, int64_t delta);
+
+  void validate(void);
+
+  std::vector<uint8_t> doGet(const std::string& key);
+
+  void doWait(
+      c10::ArrayRef<std::string> keys,
+      std::chrono::milliseconds timeout);
+
+  detail::SocketAddress addr_;
+  std::shared_ptr<detail::TCPServer> server_;
+  std::unique_ptr<detail::TCPClient> client_;
+  c10::optional<std::size_t> numWorkers_;
+
+  const std::string initKey_ = "init/";
+  const std::string keyPrefix_ = "/";
+  std::mutex activeOpLock_;
+  std::unordered_map<std::string, detail::Counter> clientCounters_;
+  bool usingLibUv_ = false;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/TraceUtils.h

@@ -0,0 +1,723 @@
+#pragma once
+
+#include <c10/util/ApproximateClock.h>
+#include <c10/util/irange.h>
+#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <torch/csrc/distributed/c10d/Types.hpp>
+#include <torch/csrc/jit/serialization/pickler.h>
+#include <torch/csrc/profiler/combined_traceback.h>
+
+#include <sys/types.h>
+
+#include <cstdlib>
+#include <string>
+#include <system_error>
+#include <vector>
+namespace c10d {
+
+/* Trace Utils Related to TORCH_NCCL_DESYNC_DEBUG */
+
+inline std::string getTraceStartKey(const std::string& pgName, int rank) {
+  return pgName + "_" + std::to_string(rank) + "_trace_start";
+}
+
+inline std::string getTraceEndKey(const std::string& pgName, int rank) {
+  return pgName + "_" + std::to_string(rank) + "_trace_end";
+}
+
+inline bool traceUpdate(
+    c10::intrusive_ptr<Store>& store,
+    const std::string& key,
+    uint64_t seq,
+    const std::string& col) {
+  std::vector<uint8_t> value(col.size() + sizeof(seq) + 1);
+  memcpy(value.data(), &seq, sizeof(seq));
+  memcpy(value.data() + sizeof(seq), col.data(), col.size());
+  try {
+    store->set(key, value);
+    return true;
+  } catch (...) {
+    LOG(ERROR) << "Store is down while updating #" << seq << " with key "
+               << key;
+    return false;
+  }
+  return true;
+}
+
+enum TraceDebugEvent {
+  kEventStart,
+  kEventEnd,
+};
+// <seq, <rank, <col, start/end>>>
+using TraceMap =
+    std::map<uint64_t, std::map<int, std::pair<std::string, TraceDebugEvent>>>;
+
+inline std::string ranksToString(const std::vector<int>& ranks) {
+  std::string str;
+  for (int rank : ranks) {
+    if (str.empty()) {
+      str = std::to_string(rank);
+    } else {
+      str += ", " + std::to_string(rank);
+    }
+  }
+  return str;
+}
+
+inline std::string ranksFromTrace(
+    const std::vector<std::pair<int, std::string>>& items) {
+  std::string ranks;
+  for (auto& p : items) {
+    if (ranks.empty()) {
+      ranks = std::to_string(p.first);
+    } else {
+      ranks += ", " + std::to_string(p.first);
+    }
+  }
+  return ranks;
+}
+
+inline std::string analyzeMissingRanks(const std::vector<int>& missingRanks) {
+  return c10::str(
+      "\n\t - To our best knowledge, ranks [",
+      ranksToString(missingRanks),
+      "] are the lagging ranks that caused this timeout. "
+      "They never joined any collectives");
+}
+
+inline std::string analyzeLaggingRanks(const TraceMap& traceMap) {
+  uint64_t lagSeq = traceMap.begin()->first;
+  std::vector<int> startRanks;
+  std::vector<int> endRanks;
+  for (auto& p : traceMap.begin()->second) {
+    if (p.second.second == kEventStart) {
+      startRanks.push_back(p.first);
+    } else {
+      endRanks.push_back(p.first);
+    }
+  }
+  std::string report =
+      "\n\t - To our best knowledge, the lagging/dead/mismatched ranks "
+      "that caused the desync are:";
+  if (startRanks.size()) {
+    report += c10::str(
+        "\n\t   - [",
+        ranksToString(startRanks),
+        "] joined but didn't finish collective #",
+        lagSeq,
+        " (count from 1)");
+  }
+  if (endRanks.size()) {
+    report += c10::str(
+        "\n\t     [",
+        ranksToString(endRanks),
+        "] finished collective #",
+        lagSeq,
+        ", but didn't join collective #",
+        lagSeq + 1,
+        " (count from 1)");
+  }
+  return report;
+}
+
+inline std::string dumpSnapshot(TraceMap& traceMap) {
+  std::string report = "\n\t - Snapshot of ranks' latest states:";
+  for (auto& tracePair : traceMap) {
+    uint64_t seq = tracePair.first;
+    std::map<int, std::pair<std::string, TraceDebugEvent>>& subMap =
+        tracePair.second;
+
+    std::unordered_map<std::string, std::vector<int>> collectivesStart;
+    std::unordered_map<std::string, std::vector<int>> collectivesEnd;
+    for (auto& p : subMap) {
+      int rank = p.first;
+      const std::string& col = p.second.first;
+      if (p.second.second == kEventStart) {
+        collectivesStart[col].push_back(rank);
+      } else {
+        collectivesEnd[col].push_back(rank);
+      }
+    }
+
+    if (collectivesStart.size()) {
+      report += c10::str("\n\t   #", seq, " started ranks:");
+      for (auto& mapPair : collectivesStart) {
+        report += c10::str(
+            "\n\t     [",
+            ranksToString(mapPair.second),
+            "] started ",
+            mapPair.first);
+      }
+    }
+    if (collectivesEnd.size()) {
+      report += c10::str("\n\t   #", seq, " finished ranks:");
+      for (auto& mapPair : collectivesEnd) {
+        report += c10::str(
+            "\n\t     [",
+            ranksToString(mapPair.second),
+            "] finished ",
+            mapPair.first);
+      }
+    }
+  }
+  return report;
+}
+
+inline bool parseTraceValue(
+    c10::intrusive_ptr<Store>& store,
+    const std::string& key,
+    uint64_t& seq,
+    std::string& col) {
+  try {
+    std::vector<uint8_t> traceValue = store->get(key);
+    memcpy(&seq, traceValue.data(), sizeof(seq));
+    std::string colName((char*)traceValue.data() + sizeof(seq));
+    col = colName;
+    return true;
+  } catch (...) {
+    LOG(ERROR) << "Store is down while getting key " << key;
+    return false;
+  }
+  return true;
+}
+
+inline std::string retrieveDesyncReport(
+    c10::intrusive_ptr<Store>& store,
+    const std::string& pgName,
+    int myRank,
+    int worldSize) {
+  std::string report;
+
+  uint64_t thisSeq;
+  std::string thisCol;
+
+  std::vector<int> missingRanks;
+  TraceMap traceMap;
+
+  for (const auto rank : c10::irange(worldSize)) {
+    // Build traceMapStart.
+    uint64_t seqStart;
+    {
+      std::string traceKeyStart = getTraceStartKey(pgName, rank);
+      if (!store->check({traceKeyStart})) {
+        missingRanks.push_back(rank);
+        continue;
+      }
+      std::string col;
+      if (!parseTraceValue(store, traceKeyStart, seqStart, col)) {
+        return report;
+      }
+      traceMap[seqStart].emplace(rank, std::make_pair(col, kEventStart));
+      if (rank == myRank) {
+        thisSeq = seqStart;
+        thisCol = std::move(col);
+      }
+    }
+
+    // Build traceMapEnd.
+    {
+      std::string traceKeyEnd = getTraceEndKey(pgName, rank);
+      if (!store->check({traceKeyEnd})) {
+        continue;
+      }
+      uint64_t seq;
+      std::string col;
+      if (!parseTraceValue(store, traceKeyEnd, seq, col)) {
+        return report;
+      }
+      if (seq == seqStart) {
+        traceMap[seq][rank].second = kEventEnd;
+      }
+    }
+  }
+
+  TORCH_INTERNAL_ASSERT(
+      !missingRanks.empty() || !traceMap.empty(),
+      "Trace shouldn't be empty while enabled GLOO_ASYNC_TIMEOUT_DEBUG");
+  TORCH_INTERNAL_ASSERT(
+      !thisCol.empty(),
+      "Timeout rank [",
+      myRank,
+      "] must have collective tracking iteam in c10::Store trace");
+  TORCH_INTERNAL_ASSERT(
+      traceMap[thisSeq][myRank].second == kEventStart,
+      "Timeout rank [",
+      myRank,
+      "] last trace item must be kEventStart. thisSeq = ",
+      thisSeq,
+      ", col = ",
+      thisCol);
+
+  report += c10::str(
+      "\n\t - [", myRank, "] Timeout at collective: ", thisCol, ", #", thisSeq);
+
+  if (!missingRanks.empty()) {
+    report += analyzeMissingRanks(missingRanks);
+  } else {
+    report += analyzeLaggingRanks(traceMap);
+    report += dumpSnapshot(traceMap);
+  }
+
+  return report;
+}
+
+/* Trace Utils Related to Flight Recorder */
+
+/* Note: this is only used by PGNCCL (could be generalized in an ideal world but
+ * wasn't done that way, so isn't expected to be fully general at the moment) */
+
+#ifdef USE_C10D_NCCL
+
+/* Helper used by work::getDuration() and nccl flight recorder */
+float getDurationFromEvent(
+    at::cuda::CUDAEvent& ncclStartEvent,
+    at::cuda::CUDAEvent& ncclEndEvent) {
+  TORCH_CHECK(
+      ncclEndEvent.query(),
+      "getDuration can only be called after work is succeeded.")
+  return ncclStartEvent.elapsed_time(ncclEndEvent);
+}
+
+DebugInfoWriter::~DebugInfoWriter() = default;
+
+void DebugInfoWriter::write(const std::string& ncclTrace) {
+  // Open a file for writing. The ios::binary flag is used to write data as
+  // binary.
+  std::ofstream file(filename_, std::ios::binary);
+
+  // Check if the file was opened successfully.
+  if (!file.is_open()) {
+    LOG(ERROR) << "Error opening file for writing NCCLPG debug info: "
+               << filename_;
+    return;
+  }
+
+  file.write(ncclTrace.data(), ncclTrace.size());
+  LOG(INFO) << "Finished writing NCCLPG debug info to " << filename_;
+}
+
+DebugInfoWriter& DebugInfoWriter::getWriter(int rank) {
+  if (writer_ == nullptr) {
+    std::string fileNamePrefix = getCvarString(
+        {"TORCH_NCCL_DEBUG_INFO_TEMP_FILE"}, "/tmp/nccl_trace_rank_");
+    // Using std::unique_ptr here to auto-delete the writer object
+    // when the pointer itself is destroyed.
+    std::unique_ptr<DebugInfoWriter> writerPtr(
+        new DebugInfoWriter(fileNamePrefix, rank));
+    DebugInfoWriter::registerWriter(std::move(writerPtr));
+  }
+  return *writer_;
+}
+
+void DebugInfoWriter::registerWriter(std::unique_ptr<DebugInfoWriter> writer) {
+  TORCH_CHECK_WITH(
+      DistBackendError,
+      hasWriterRegistered_.load() == false,
+      "debugInfoWriter already registered");
+  hasWriterRegistered_.store(true);
+  writer_ = std::move(writer);
+}
+
+std::unique_ptr<DebugInfoWriter> DebugInfoWriter::writer_ = nullptr;
+std::atomic<bool> DebugInfoWriter::hasWriterRegistered_(false);
+
+inline std::string pickle_str(const c10::IValue& v) {
+  std::vector<char> result;
+  {
+    auto writer = [&](const char* data, size_t size) {
+      result.insert(result.end(), data, data + size);
+    };
+    torch::jit::Pickler pickler(
+        writer, nullptr, nullptr, nullptr, nullptr, false);
+    pickler.protocol();
+    pickler.pushIValue(v);
+    pickler.stop();
+  }
+  return std::string(result.begin(), result.end());
+}
+
+inline std::string get_python_cpp_trace() {
+  // usage:
+  // LOG(INFO) << "stacktrace: "
+  //           << get_python_cpp_trace();
+  // warn: might be slow in getting cpp traces
+  // because of slow/broken addr2line
+  // in different system libs
+  std::shared_ptr<torch::CapturedTraceback> tb =
+      torch::CapturedTraceback::gather(
+          /*python=*/true, /*script=*/true, /*cpp=*/true);
+  torch::SymbolizedTracebacks s_tbs = torch::symbolize({tb.get()});
+  const auto& s_tb = s_tbs.tracebacks.at(0);
+  std::stringstream oss;
+  for (auto idx : c10::irange(s_tb.size())) {
+    auto frame_id = s_tb[idx];
+    const auto& frame = s_tbs.all_frames.at(frame_id);
+    oss << "#" << idx << " " << frame.funcname << " from " << frame.filename
+        << ":" << frame.lineno << std::endl;
+  }
+  return oss.str();
+}
+
+inline c10::Dict<c10::IValue, c10::IValue> new_dict() {
+  return c10::Dict<c10::IValue, c10::IValue>(
+      c10::AnyType::get(), c10::AnyType::get());
+}
+
+inline c10::List<c10::IValue> new_list() {
+  return c10::List<c10::IValue>(c10::AnyType::get());
+}
+
+struct NCCLTraceBuffer {
+  static NCCLTraceBuffer* get() {
+    // intentionally leak on exit
+    // because this will hold python state that may get destructed
+    static NCCLTraceBuffer* instance = new NCCLTraceBuffer();
+    return instance;
+  }
+  NCCLTraceBuffer() {
+    max_entries_ = getCvarInt({"TORCH_NCCL_TRACE_BUFFER_SIZE"}, 0);
+    capture_cpp_stack_ = getCvarBool({"TORCH_NCCL_TRACE_CPP_STACK"}, false);
+    enabled_ = max_entries_ > 0;
+    pg_id_to_ranks_ = {};
+  }
+  using Event = at::cuda::CUDAEvent;
+  struct Entry {
+    size_t id_; // incremented id in the trace buffer
+                // used to figure out where in the circular entries
+                // buffer this entry will be located to
+                // update state information
+    size_t pg_id_;
+
+    // Both seq_id_ and op_id_ are per_pg incrementing counters
+    // seq_id refers to actual kernel launches (e.g. 1 per coalesced group)
+    // op_id refers to logical operations (e.g. one per op inside coalesced
+    // group)
+    size_t seq_id_;
+    size_t op_id_;
+    std::string profiling_name_;
+
+    std::shared_ptr<torch::CapturedTraceback> traceback_;
+    // we borrow pointers to start_ and end_ so we can query the state
+    // on reporting. However, once the event is completed, the call
+    // to `complete` will clear these.
+    Event *start_, *end_;
+
+    // timestamp when the entry was created, likely close to the time the work
+    // was 'enqueued'- not necessarily started
+    c10::time_t time_created_;
+    c10::optional<float> duration_;
+
+    // timestamp when our CPU threads discovered that the kernel started.
+    // will always be _after_ it actually started, and can be very late
+    // if the watchdog thread got stuck on CUDA APIs.
+    c10::optional<c10::time_t> time_discovered_started_;
+
+    // timestamp when our CPU threads discovered that the kernel completed.
+    // will always be _after_ it actually complated, and can be the same time
+    // as the discovery of the start if the watchdog thread is stuck on CUDA
+    // APIs
+    c10::optional<c10::time_t> time_discovered_completed_;
+
+    // size information for input/output tensors
+    c10::SmallVector<int, 4> input_dims_;
+    c10::SmallVector<int, 4> output_dims_;
+    c10::SmallVector<int64_t, 8> sizes_; // flattened from inputs, outputs
+    bool retired_ = false; // is this work entry no longer in the workMetaList_?
+                           // a retired but not completed event has timed out
+  };
+
+  bool enabled_ = false;
+  bool capture_cpp_stack_ = false;
+  std::mutex mutex_;
+  std::vector<Entry> entries_;
+  size_t max_entries_ = 0;
+  size_t next_ = 0;
+  size_t id_ = 0;
+  std::map<size_t, std::vector<uint64_t>> pg_id_to_ranks_;
+
+  c10::optional<size_t> record(
+      size_t pg_id,
+      size_t seq_id,
+      size_t op_id,
+      std::string profiling_name,
+      const std::vector<at::Tensor>& inputs,
+      const std::vector<at::Tensor>& outputs,
+      Event* start,
+      Event* end) {
+    if (!enabled_) {
+      return c10::nullopt;
+    }
+    auto traceback =
+        torch::CapturedTraceback::gather(true, true, capture_cpp_stack_);
+    std::lock_guard<std::mutex> guard(mutex_);
+
+    auto te = Entry{
+        id_,
+        pg_id,
+        seq_id,
+        op_id,
+        std::move(profiling_name),
+        std::move(traceback),
+        std::move(start),
+        std::move(end),
+        c10::getTime()};
+
+    for (const auto& input : inputs) {
+      c10::IntArrayRef sizes = input.sizes();
+      te.input_dims_.push_back(sizes.size());
+      te.sizes_.insert(te.sizes_.end(), sizes.begin(), sizes.end());
+    }
+
+    for (const auto& output : outputs) {
+      c10::IntArrayRef sizes = output.sizes();
+      te.output_dims_.push_back(sizes.size());
+      te.sizes_.insert(te.sizes_.end(), sizes.begin(), sizes.end());
+    }
+
+    if (entries_.size() < max_entries_) {
+      entries_.emplace_back(std::move(te));
+    } else {
+      entries_[next_++] = std::move(te);
+      if (next_ == max_entries_) {
+        next_ = 0;
+      }
+    }
+    return id_++;
+  }
+
+  void record_pg_ranks(size_t pg_id, std::vector<uint64_t> ranks) {
+    if (!enabled_) {
+      return;
+    }
+    std::lock_guard<std::mutex> guard(mutex_);
+    pg_id_to_ranks_[pg_id] = ranks;
+  }
+
+  void update_state(Entry& r) {
+    if (r.start_ != nullptr) {
+      bool started = r.start_->query();
+      if (started && !r.time_discovered_started_) {
+        r.time_discovered_started_ = c10::getTime();
+      }
+    }
+    if (r.end_ != nullptr) {
+      bool completed = r.end_->query();
+      if (completed && !r.time_discovered_completed_) {
+        r.time_discovered_completed_ = c10::getTime();
+      }
+    }
+  }
+
+  std::vector<Entry> dump_entries() {
+    std::lock_guard<std::mutex> guard(mutex_);
+    std::vector<Entry> result;
+    result.reserve(entries_.size());
+    result.insert(result.end(), entries_.begin() + next_, entries_.end());
+    result.insert(result.end(), entries_.begin(), entries_.begin() + next_);
+    // query any remaining events
+    for (auto& r : result) {
+      update_state(r);
+      r.start_ = r.end_ = nullptr;
+    }
+    return result;
+  }
+
+  /*
+  Mark an Event as completed and free its events.
+
+  This is called by the watchdog thread, and is asynchronous from the
+  perspective of the main thread.
+
+  compute_duration defaults to true since retire_id is only called in the
+  watchdog thread, which is currently a place we call cuda APIs which may hang,
+  but care should be taken to avoid computing duration in any function that must
+  never hang. (timing must also be enabled for compute_duration - see
+  TORCH_NCCL_ENABLE_TIMING).
+  */
+  void retire_id(c10::optional<size_t> id, bool compute_duration = true) {
+    if (!enabled_ || !id) {
+      return;
+    }
+
+    bool can_compute_duration = false;
+    Event* startEvent = nullptr;
+    Event* endEvent = nullptr;
+    c10::optional<float> duration = c10::nullopt;
+
+    std::unique_lock<std::mutex> guard(mutex_);
+
+    Entry* entry = &entries_.at(*id % max_entries_);
+    if (entry->id_ == *id) {
+      update_state(*entry);
+
+      if (compute_duration) {
+        can_compute_duration = entry->time_discovered_completed_.has_value() &&
+            entry->start_ && entry->end_;
+        startEvent = entry->start_;
+        endEvent = entry->end_;
+      }
+    }
+
+    if (can_compute_duration) {
+      // Compute duration without without holding the lock, because
+      // cudaEventDuration() can hang, and we need to acquire the lock before we
+      // can dump(), which we never want to block.
+      guard.unlock();
+      duration = getDurationFromEvent(*startEvent, *endEvent);
+      guard.lock();
+
+      // Refresh the entry pointer, see if the entry has been overwritten
+      entry = &entries_.at(*id % max_entries_);
+      if (entry->id_ != *id) {
+        LOG(INFO)
+            << "retire_id abandoned for id " << *id
+            << ", event was overwritten while waiting to compute duration.";
+        return;
+      }
+      if (duration.has_value()) {
+        entry->duration_ = duration.value();
+      }
+    }
+
+    entry->retired_ = true;
+    entry->start_ = entry->end_ = nullptr;
+  }
+
+  std::string dump(
+      const c10::optional<std::unordered_map<
+          std::string,
+          std::unordered_map<std::string, std::string>>>& ncclDumpMap) {
+    auto result = dump_entries();
+    auto entries = new_list();
+    c10::IValue entries_key = "entries";
+    c10::IValue nccl_comm_key = "nccl_comm_state";
+    c10::IValue version_key = "version";
+    // Update whenever changing contents or formatting of the dump
+    // (minor when adding fields, major when changing existing fields)
+    c10::IValue version_val = "1.4";
+    c10::IValue pg_config_key = "pg_config";
+    c10::IValue record_id_key = "record_id";
+    c10::IValue pg_id_key = "pg_id";
+    c10::IValue seq_id_key = "seq_id";
+    c10::IValue op_id_key = "op_id";
+    c10::IValue profiling_name_key = "profiling_name";
+    c10::IValue input_sizes_key = "input_sizes";
+    c10::IValue output_sizes_key = "output_sizes";
+    c10::IValue time_created_key = "time_created_ns";
+    c10::IValue duration_key = "duration_ms";
+
+    c10::IValue frames_key = "frames";
+    c10::IValue state_key = "state";
+    c10::IValue line_key = "line";
+    c10::IValue name_key = "name";
+    c10::IValue filename_key = "filename";
+    c10::IValue retired_key = "retired";
+    c10::IValue time_discovered_started_key = "time_discovered_started_ns";
+    c10::IValue time_discovered_completed_key = "time_discovered_completed_ns";
+
+    std::vector<torch::CapturedTraceback*> tracebacks;
+    for (auto& e : result) {
+      tracebacks.push_back(e.traceback_.get());
+    }
+    torch::SymbolizedTracebacks stracebacks = torch::symbolize(tracebacks);
+    std::vector<c10::IValue> all_frames;
+    for (const auto& f : stracebacks.all_frames) {
+      auto d = new_dict();
+      d.insert(name_key, f.funcname);
+      d.insert(filename_key, f.filename);
+      d.insert(line_key, int64_t(f.lineno));
+      all_frames.emplace_back(std::move(d));
+    }
+
+    for (auto i : c10::irange(result.size())) {
+      auto& e = result.at(i);
+      auto& tb = stracebacks.tracebacks.at(i);
+      auto dict = new_dict();
+      dict.insert(record_id_key, int64_t(e.id_));
+      dict.insert(pg_id_key, int64_t(e.pg_id_));
+      dict.insert(seq_id_key, int64_t(e.seq_id_));
+      dict.insert(op_id_key, int64_t(e.op_id_));
+      dict.insert(profiling_name_key, e.profiling_name_);
+      dict.insert(time_created_key, int64_t(e.time_created_));
+      if (e.duration_) {
+        dict.insert(duration_key, *e.duration_);
+      }
+
+      auto it = e.sizes_.begin();
+      auto read_sizes = [&](const c10::SmallVector<int, 4>& dims) {
+        auto sizes = new_list();
+        for (auto dim : dims) {
+          auto arg_sizes = new_list();
+          for (auto i : c10::irange(dim)) {
+            (void)i;
+            arg_sizes.push_back(*it++);
+          }
+          sizes.push_back(arg_sizes);
+        }
+        return sizes;
+      };
+
+      dict.insert(input_sizes_key, read_sizes(e.input_dims_));
+      dict.insert(output_sizes_key, read_sizes(e.output_dims_));
+      if (e.time_discovered_completed_.has_value()) {
+        dict.insert(state_key, "completed");
+      } else if (e.time_discovered_started_.has_value()) {
+        dict.insert(state_key, "started");
+      } else {
+        dict.insert(state_key, "scheduled");
+      }
+
+      dict.insert(
+          time_discovered_started_key,
+          e.time_discovered_started_.has_value()
+              ? int64_t(*e.time_discovered_started_)
+              : c10::IValue());
+      dict.insert(
+          time_discovered_completed_key,
+          e.time_discovered_completed_.has_value()
+              ? int64_t(*e.time_discovered_completed_)
+              : c10::IValue());
+      dict.insert(retired_key, e.retired_);
+
+      auto frames = new_list();
+      for (int64_t frame : tb) {
+        frames.push_back(all_frames.at(frame));
+      }
+      dict.insert(frames_key, frames);
+      entries.push_back(dict);
+    }
+    auto pg_config = new_dict();
+    for (const auto& [pg_id, ranks] : pg_id_to_ranks_) {
+      auto pg_ranks = new_list();
+      for (const auto& rank : ranks) {
+        pg_ranks.push_back(static_cast<int>(rank));
+      }
+      pg_config.insert(static_cast<int>(pg_id), pg_ranks);
+    }
+
+    // convert ncclDumpMap into a dictionary
+    auto per_comm_dict = new_dict();
+    if (ncclDumpMap.has_value()) {
+      for (const auto& [ncclId, ncclDump] : ncclDumpMap.value()) {
+        auto inner_dict = new_dict();
+        for (const auto& [key, value] : ncclDump) {
+          inner_dict.insert(key, value);
+        }
+        per_comm_dict.insert(ncclId, inner_dict);
+      }
+    }
+
+    auto dict = new_dict();
+    dict.insert(entries_key, entries);
+    dict.insert(version_key, version_val);
+    if (per_comm_dict.size() > 0) {
+      dict.insert(nccl_comm_key, per_comm_dict);
+    }
+    dict.insert(pg_config_key, pg_config);
+
+    return pickle_str(dict);
+  }
+};
+
+#endif
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/UCCTracing.hpp

@@ -0,0 +1,58 @@
+#pragma once
+
+#ifdef USE_C10D_UCC
+
+#include <torch/csrc/distributed/c10d/UCCUtils.hpp>
+
+namespace c10d {
+
+#define RECORD_COMMS_TRACE(                                                    \
+    _comms_tracer, _work, _opType, _rank, _comm_size, _inTensors, _outTensors) \
+  do {                                                                         \
+    if (torch_ucc_config.enable_comms_logger) {                                \
+      _comms_tracer->recordComms(                                              \
+          opTypeToString(_opType),                                             \
+          (uintptr_t)_work.get(),                                              \
+          _rank,                                                               \
+          _comm_size,                                                          \
+          _inTensors,                                                          \
+          _outTensors);                                                        \
+    }                                                                          \
+  } while (0)
+
+// interfaces to collect communication traces
+class TORCH_API CommTraceLogger : public torch::CustomClassHolder {
+ private:
+  std::vector<std::string> comms_trace_;
+  std::vector<std::string> curBlocks_; /* unused */
+  std::vector<int64_t> curOutSplitSizes_;
+  std::vector<int64_t> curInSplitSizes_;
+  int curRoot_ = -1;
+  unsigned long seqnum = 0;
+
+ public:
+  void setCurBlock(const std::string& name); /* unused */
+  void popBlock(); /* unused */
+  // record root info if applicable, e.g., broadcast, gather, scatter
+  void recordOptionalInfo(int root = -1);
+  // record input/output splits of Alltoallv
+  void recordOptionalInfo(
+      const std::vector<int64_t>& outputSplitSizes = {},
+      const std::vector<int64_t>& inputSplitSizes = {});
+  // record essential comms information
+  void recordComms(
+      const std::string& collName,
+      const uintptr_t workReq = 0,
+      const int rank = -1,
+      const int world_size = -1,
+      const std::vector<at::Tensor>& inputTensors = {},
+      const std::vector<at::Tensor>& outputTensor = {});
+  // return collected comms traces
+  std::vector<std::string>& getCommsTrace() {
+    return comms_trace_;
+  }
+};
+
+} // namespace c10d
+
+#endif // USE_C10D_UCC

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/UCCUtils.hpp

@@ -0,0 +1,187 @@
+#pragma once
+
+#ifdef USE_C10D_UCC
+
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <ucc/api/ucc.h>
+
+namespace c10d {
+
+// Macro to generate the error message on a non-successful UCC return value.
+#define TORCH_UCC_GET_ERROR_MSG(_err, _error_msg, _result) \
+  do {                                                     \
+    _err = c10::str(                                       \
+        "[",                                               \
+        std::string(__FILE__),                             \
+        ":",                                               \
+        std::to_string(__LINE__),                          \
+        "] ",                                              \
+        logger->getLogPrefix(),                            \
+        _error_msg,                                        \
+        ", error code ",                                   \
+        _result,                                           \
+        ": ",                                              \
+        ucc_status_string(_result),                        \
+        ", system error code ",                            \
+        errno);                                            \
+  } while (0)
+
+// Macro to throw on a non-successful UCC return value.
+#define TORCH_UCC_CHECK(_cmd, _error_msg)               \
+  do {                                                  \
+    ucc_status_t result = _cmd;                         \
+    if (result != UCC_OK) {                             \
+      std::string err;                                  \
+      TORCH_UCC_GET_ERROR_MSG(err, _error_msg, result); \
+      TORCH_CHECK(false, err);                          \
+    }                                                   \
+  } while (0)
+
+// Macro and throw on a non-successful UCC return value and free its request.
+#define TORCH_UCC_CHECK_REQUEST(_request, _cmd, _error_msg) \
+  do {                                                      \
+    ucc_status_t result = _cmd;                             \
+    if (result != UCC_OK) {                                 \
+      std::string err;                                      \
+      TORCH_UCC_GET_ERROR_MSG(err, _error_msg, result);     \
+      if (_request != nullptr) {                            \
+        ucc_collective_finalize(_request);                  \
+      }                                                     \
+      TORCH_CHECK(false, err);                              \
+    }                                                       \
+  } while (0)
+
+// Macros to print logs with unified format
+#define TORCH_UCC_LOG_ERROR(_phase, _msg) \
+  LOG(ERROR) << logger->getLogPrefix(_phase) << "[ERROR] " << _msg;
+#define TORCH_UCC_LOG_INFO(_phase, _msg) \
+  LOG(INFO) << logger->getLogPrefix(_phase) << "[INFO] " << _msg;
+#define TORCH_UCC_LOG_DEBUG(_phase, _msg) \
+  VLOG(1) << logger->getLogPrefix(_phase) << "[DEBUG] " << _msg;
+
+enum torch_ucc_phase_t {
+  TORCH_UCC_UNKNOWN = -1,
+  TORCH_UCC_INIT,
+  TORCH_UCC_HEALTH_CHECK,
+  TORCH_UCC_READY,
+  TORCH_UCC_COLL_POST,
+  TORCH_UCC_COLL_PROGRESS,
+  TORCH_UCC_FINALIZE,
+};
+
+const std::map<torch_ucc_phase_t, std::string> ucc_phase_map = {
+    {TORCH_UCC_UNKNOWN, "UNKNOWN"},
+    {TORCH_UCC_INIT, "INIT"},
+    {TORCH_UCC_HEALTH_CHECK, "HEALTH_CHECK"},
+    {TORCH_UCC_READY, "READY"},
+    {TORCH_UCC_COLL_POST, "COLL_POST"},
+    {TORCH_UCC_COLL_PROGRESS, "COLL_PROGRESS"},
+    {TORCH_UCC_FINALIZE, "FINALIZE"},
+};
+
+class CommTraceLogger;
+
+class TORCH_API ProcessGroupUCCLogger : public torch::CustomClassHolder {
+ public:
+  ProcessGroupUCCLogger();
+  ProcessGroupUCCLogger(std::string log_prefix, torch_ucc_phase_t phase);
+
+  std::string getLogPrefix(torch_ucc_phase_t phase = TORCH_UCC_UNKNOWN);
+  void setLogPrefix(std::string log_prefix);
+  inline void setPhase(torch_ucc_phase_t phase) {
+    local_phase = phase;
+  }
+
+  void initCommsTracer();
+  void flushComms(int rank, int world_size);
+  std::shared_ptr<CommTraceLogger> trace_generator = nullptr;
+
+ protected:
+  std::string log_prefix;
+  torch_ucc_phase_t local_phase = TORCH_UCC_UNKNOWN;
+  bool initialized_CommTraceLogger = false;
+};
+
+struct torch_ucc_oob_coll_info_t {
+  c10::intrusive_ptr<Store> store;
+  uint32_t comm_id;
+  int rank;
+  int size;
+  void* rbuf;
+  size_t msglen;
+  std::string getKey(std::string key) {
+    return std::to_string(comm_id) + key;
+  }
+};
+
+class CommBase {
+ public:
+  CommBase(const c10::intrusive_ptr<ProcessGroupUCCLogger>& logger_)
+      : logger(logger_) {}
+  virtual void progress() = 0;
+  virtual void free_request(ucc_coll_req_h request) = 0;
+  virtual ~CommBase() {}
+  c10::intrusive_ptr<ProcessGroupUCCLogger> logger;
+};
+class CommUCC : public CommBase {
+ public:
+  ucc_lib_h lib{nullptr};
+  ucc_context_h context{nullptr};
+
+ public:
+  void progress() override;
+  CommUCC(
+      std::shared_ptr<torch_ucc_oob_coll_info_t> oob,
+      const c10::intrusive_ptr<ProcessGroupUCCLogger>& logger);
+  void free_request(ucc_coll_req_h request) override;
+  ~CommUCC();
+};
+
+ucc_status_t oob_allgather(
+    void* sbuf,
+    void* rbuf,
+    size_t msglen,
+    void* coll_info,
+    void** req);
+
+ucc_status_t oob_allgather_test(void* req);
+
+ucc_status_t oob_allgather_free(void* req);
+
+// trim: remove spaces before and after the string view
+// implementation borrowed from https://stackoverflow.com/a/17976541
+inline c10::string_view trim(c10::string_view s) {
+  auto wsfront = std::find_if_not(
+      s.begin(), s.end(), [](int c) { return std::isspace(c); });
+  auto wsback = std::find_if_not(s.rbegin(), s.rend(), [](int c) {
+                  return std::isspace(c);
+                }).base();
+  return (
+      wsback <= wsfront ? "" : s.substr(wsfront - s.begin(), wsback - wsfront));
+}
+
+inline std::string tolower(c10::string_view s) {
+  std::string result;
+  result.reserve(s.size());
+  for (auto c : s) {
+    result.push_back(std::tolower(c));
+  }
+  return result;
+}
+
+inline std::vector<std::string> parse_list(std::string list) {
+  std::vector<std::string> result;
+  list = tolower(trim(list));
+  while (!list.empty()) {
+    const auto end_pos = list.find_first_of(',');
+    const auto token = trim(list.substr(0, end_pos));
+    result.push_back(std::string(token));
+    list = (end_pos != c10::string_view::npos) ? list.substr(end_pos + 1) : "";
+  }
+  return result;
+}
+
+} // namespace c10d
+
+#endif // USE_C10D_UCC

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/UnixSockUtils.hpp

@@ -0,0 +1,27 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/Utils.hpp>
+
+namespace c10d {
+namespace tcputil {
+
+#define CONNECT_SOCKET_OFFSET 2
+
+inline int poll(struct pollfd* fds, unsigned long nfds, int timeout) {
+  return ::poll(fds, nfds, timeout);
+}
+
+inline void addPollfd(
+    std::vector<struct pollfd>& fds,
+    int socket,
+    short events) {
+  fds.push_back({.fd = socket, .events = events});
+}
+
+inline struct ::pollfd getPollfd(int socket, short events) {
+  struct ::pollfd res = {.fd = socket, .events = events};
+  return res;
+}
+
+} // namespace tcputil
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/Utils.hpp

@@ -0,0 +1,731 @@
+#pragma once
+
+#include <ATen/ATen.h>
+#include <c10/util/Exception.h>
+#include <c10/util/accumulate.h>
+#include <c10/util/irange.h>
+#include <torch/csrc/distributed/c10d/Types.hpp>
+
+#ifdef _WIN32
+#include <winsock2.h>
+#include <ws2tcpip.h>
+typedef SSIZE_T ssize_t;
+#pragma comment(lib, "Ws2_32.lib")
+#else
+#include <fcntl.h>
+#include <netdb.h>
+#include <sys/poll.h>
+#include <sys/socket.h>
+#include <unistd.h>
+#endif
+
+#include <sys/types.h>
+
+#include <chrono>
+#include <cstdint>
+#include <cstdlib>
+#include <functional>
+#include <limits>
+#include <string>
+#include <system_error>
+#include <tuple>
+#include <vector>
+
+namespace c10d {
+
+TORCH_API size_t getTensorsNumel(const std::vector<at::Tensor>& tensors);
+
+// Retrieve tensor shapes from a given tensor.
+TORCH_API std::vector<at::Tensor> getTensorShapes(
+    const std::vector<at::Tensor>& tensors);
+
+// Use -2 to represent unset state of env vars
+#define C10D_ENV_NOT_SET -2
+
+// Turns at::IntArrayRef into "(1, 2, 3, 4)".
+inline std::string toString(at::IntArrayRef l) {
+  std::stringstream ss;
+  ss << "(";
+  for (const auto i : c10::irange(l.size())) {
+    if (i > 0) {
+      ss << ", ";
+    }
+    ss << l[i];
+  }
+  ss << ")";
+  return ss.str();
+}
+
+inline std::string toString(const c10::Layout& layout) {
+  std::stringstream ss;
+  ss << layout;
+  return ss.str();
+}
+
+inline void assertSameType(
+    const at::DeprecatedTypeProperties& type,
+    const std::vector<at::Tensor>& tensors) {
+  for (const auto i : c10::irange(tensors.size())) {
+    if (!tensors[i].options().type_equal(type.options())) {
+      const std::string expected = type.toString();
+      const std::string actual = tensors[i].toString();
+      throw std::invalid_argument(
+          "mixed types (" + expected + " and " + actual + ")");
+    }
+  }
+}
+
+inline std::vector<std::string> split(
+    char separator,
+    const std::string& string) {
+  std::vector<std::string> pieces;
+  std::stringstream ss(string);
+  std::string item;
+  while (std::getline(ss, item, separator)) {
+    pieces.push_back(std::move(item));
+  }
+  return pieces;
+}
+
+inline std::string getCvarString(
+    const std::vector<std::string>& env,
+    const char* def) {
+  const char* ret = def;
+
+  if (env.empty()) {
+    TORCH_CHECK(false, "No environment variables passed");
+    return ret;
+  }
+
+  /* parse environment variable in reverse order, so the early
+   * versions of a variable get higher priority than the latter
+   * versions of the same variable */
+  for (int i = env.size() - 1; i >= 0; i--) {
+    const char* val = std::getenv(env[i].c_str());
+    if (val == nullptr) {
+      continue;
+    } else if (i) {
+      TORCH_WARN(
+          "Environment variable " + env[i] + " is deprecated; use " + env[0] +
+          " instead");
+    }
+
+    ret = val;
+  }
+
+  return ret;
+}
+
+inline int getCvarInt(const std::vector<std::string>& env, int def) {
+  int ret = def;
+
+  if (env.empty()) {
+    TORCH_CHECK(false, "No environment variables passed");
+    return ret;
+  }
+
+  /* parse environment variable in reverse order, so the early
+   * versions of a variable get higher priority than the latter
+   * versions of the same variable */
+  for (int i = env.size() - 1; i >= 0; i--) {
+    char* val = std::getenv(env[i].c_str());
+    if (val == nullptr) {
+      continue;
+    } else if (i) {
+      TORCH_WARN(
+          "Environment variable " + env[i] + " is deprecated; use " + env[0] +
+          " instead");
+    }
+
+    try {
+      ret = std::stoi(val);
+    } catch (std::exception& e) {
+      TORCH_CHECK(false, "Invalid value for environment variable: " + env[i]);
+    }
+  }
+
+  return ret;
+}
+
+inline bool getCvarBool(const std::vector<std::string>& env, bool def) {
+  bool ret = def;
+
+  if (env.empty()) {
+    TORCH_CHECK(false, "No environment variables passed");
+    return ret;
+  }
+
+  /* parse environment variable in reverse order, so the early
+   * versions of a variable get higher priority than the latter
+   * versions of the same variable */
+  for (int i = env.size() - 1; i >= 0; i--) {
+    char* val_ = std::getenv(env[i].c_str());
+    if (val_ == nullptr) {
+      continue;
+    } else if (i) {
+      TORCH_WARN(
+          "Environment variable " + env[i] + " is deprecated; use " + env[0] +
+          " instead");
+    }
+
+    std::string val = std::string(val_);
+    for (auto& x : val) {
+      x = std::tolower(x);
+    }
+
+    if (val == "y" || val == "yes" || val == "1" || val == "t" ||
+        val == "true") {
+      ret = true;
+    } else if (
+        val == "n" || val == "no" || val == "0" || val == "f" ||
+        val == "false") {
+      ret = false;
+    } else {
+      TORCH_CHECK(false, "Invalid value for environment variable: " + env[i]);
+      return ret;
+    }
+  }
+
+  return ret;
+}
+
+inline void assertSameSizes(
+    const at::IntArrayRef& sizes,
+    const std::vector<at::Tensor>& tensors) {
+  for (const auto i : c10::irange(tensors.size())) {
+    if (!tensors[i].sizes().equals(sizes)) {
+      const auto expected = toString(sizes);
+      const auto actual = toString(tensors[i].sizes());
+      throw std::invalid_argument(
+          "mixed sizes (" + expected + " and " + actual + ")");
+    }
+  }
+}
+
+inline void assertSameSizeAndType(const std::vector<at::Tensor>& tensors) {
+  // Ensure we have at least one tensor
+  if (tensors.empty()) {
+    throw std::invalid_argument("argument is empty");
+  }
+
+  // Ensure all tensors have identical type and shape
+  auto options = tensors[0].options();
+  auto sizes = tensors[0].sizes();
+  for (const auto i : c10::irange(1, tensors.size())) {
+    if (!tensors[i].options().type_equal(options)) {
+      const auto expected = toString(options);
+      const auto actual = toString(tensors[i].options());
+      throw std::invalid_argument(
+          "argument contains mixed types (" + expected + " and " + actual +
+          ")");
+    }
+    if (!tensors[i].sizes().equals(sizes)) {
+      const auto expected = toString(sizes);
+      const auto actual = toString(tensors[i].sizes());
+      throw std::invalid_argument(
+          "argument contains mixed sizes (" + expected + " and " + actual +
+          ")");
+    }
+  }
+}
+
+inline void assertTypeMatch(
+    std::function<void(const std::string&)> fn,
+    const at::DeprecatedTypeProperties& type,
+    const at::ArrayRef<at::Tensor> tensors,
+    size_t index) {
+  if (!tensors[index].options().type_equal(type.options())) {
+    fn("invalid tensor type at index " + std::to_string(index) + " (expected " +
+       type.toString() + ", got " + tensors[index].toString() + ")");
+  }
+}
+
+inline void assertTypeMatch(
+    std::function<void(const std::string&)> fn,
+    const at::TensorOptions& options,
+    const at::ArrayRef<at::Tensor> tensors,
+    size_t index) {
+  if (!tensors[index].options().type_equal(options)) {
+    fn("invalid tensor type at index " + std::to_string(index) + " (expected " +
+       toString(options) + ", got " + toString(tensors[index].options()) + ")");
+  }
+}
+
+inline void assertSizesMatch(
+    std::function<void(const std::string&)> fn,
+    const at::IntArrayRef& sizes,
+    const at::ArrayRef<at::Tensor> tensors,
+    size_t index) {
+  if (tensors[index].sizes() != sizes) {
+    fn("invalid tensor size at index " + std::to_string(index) + " (expected " +
+       toString(sizes) + ", got " + toString(tensors[index].sizes()) + ")");
+  }
+}
+
+inline void assertLayoutMatch(
+    std::function<void(const std::string&)> fn,
+    const c10::Layout& expected,
+    const at::ArrayRef<at::Tensor> tensors,
+    size_t index) {
+  const auto& actual = tensors[index].layout();
+  if (actual != expected) {
+    fn("invalid tensor layout at index " + std::to_string(index) +
+       " (expected " + toString(expected) + ", got " + toString(actual) + ")");
+  }
+}
+
+inline void assertLayoutMatch(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  const auto& layout = tensors[0].layout();
+  for (const auto i : c10::irange(1, tensors.size())) {
+    assertLayoutMatch(fn, layout, tensors, i);
+  }
+}
+
+inline void assertNonEmpty(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  if (tensors.empty()) {
+    fn("requires non-empty tensor list");
+  }
+}
+
+inline void assertSingleElement(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  if (tensors.size() != 1) {
+    fn("requires a single-element tensor list");
+  }
+}
+
+inline void assertSingleElementInput(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  if (tensors.size() != 1) {
+    fn("requires a single-element input tensor list");
+  }
+}
+
+inline void assertSingleElementOutput(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  if (tensors.size() != 1) {
+    fn("requires a single-element output tensor list");
+  }
+}
+
+inline void assertRootRank(
+    std::function<void(const std::string&)> fn,
+    int rank,
+    int size) {
+  if (rank < 0 || rank >= size) {
+    fn("invalid root rank: " + std::to_string(rank));
+  }
+}
+
+inline void assertRootTensor(
+    std::function<void(const std::string&)> fn,
+    int rank,
+    int size) {
+  if (rank < 0 || rank >= size) {
+    fn("invalid root tensor: " + std::to_string(rank));
+  }
+}
+
+inline void assertDense(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  const auto& layout = tensors[0].layout();
+  if (layout != at::kStrided) {
+    fn("only supports dense tensors");
+  }
+}
+
+inline void assertCPU(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  const auto& device = tensors[0].device();
+  if (device.type() != at::kCPU) {
+    fn("only supports CPU tensors");
+  }
+}
+
+inline void assertSameDevice(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  if (tensors.size() < 2) {
+    return;
+  }
+  const auto& device = tensors[0].device();
+  for (const auto i : c10::irange(1, tensors.size())) {
+    if (tensors[i].device() != device) {
+      fn("tensors should be on the same device");
+    }
+  }
+}
+
+inline void assertTypeAndSizesMatch(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors,
+    const at::DeprecatedTypeProperties& type,
+    const at::IntArrayRef& sizes) {
+  for (const auto i : c10::irange(tensors.size())) {
+    assertTypeMatch(fn, type, tensors, i);
+    assertSizesMatch(fn, sizes, tensors, i);
+  }
+}
+
+inline void assertTypeAndSizesMatch(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors,
+    const at::TensorOptions& options,
+    const at::IntArrayRef& sizes) {
+  for (const auto i : c10::irange(tensors.size())) {
+    assertTypeMatch(fn, options, tensors, i);
+    assertSizesMatch(fn, sizes, tensors, i);
+  }
+}
+
+inline void assertTypeAndSizesMatch(
+    std::function<void(const std::string&)> fn,
+    const at::ArrayRef<at::Tensor> tensors) {
+  const auto& options = tensors[0].options();
+  const auto sizes = tensors[0].sizes();
+  assertTypeAndSizesMatch(fn, tensors.slice(1), options, sizes);
+}
+
+// Copied from ATen/core/functional.h.
+template <typename F, typename T>
+inline auto fmap(T& inputs, const F& fn)
+    -> std::vector<decltype(fn(*inputs.begin()))> {
+  std::vector<decltype(fn(*inputs.begin()))> r;
+  r.reserve(inputs.size());
+  for (auto& input : inputs) {
+    r.push_back(fn(input));
+  }
+  return r;
+}
+
+// Copied from torch/csrc/utils/tensor_flatten.h.
+inline at::Tensor flattenDenseTensors(at::TensorList tensors) {
+  static const auto flatten = [](const at::Tensor& t) {
+    return t.contiguous().view({-1});
+  };
+  if (tensors.size() == 1) {
+    return flatten(tensors[0]);
+  }
+  return at::cat(::c10d::fmap(tensors, flatten));
+}
+
+inline at::Tensor newLikeFlat(
+    std::vector<std::vector<at::Tensor>>& tensors,
+    size_t deviceIdx) {
+  if (tensors.empty() || tensors[0].empty()) {
+    TORCH_CHECK(false, "Received an empty list");
+  }
+  if (deviceIdx >= tensors.size()) {
+    TORCH_CHECK(false, "Invalid device index");
+  }
+  auto& t = tensors[deviceIdx][0];
+  auto device = t.device();
+  for (const auto i : c10::irange(1, tensors[deviceIdx].size())) {
+    if (tensors[deviceIdx][i].device() != device) {
+      TORCH_CHECK(false, "Expecting all tensors on the same device");
+    }
+  }
+  at::DeviceGuard gpuGuard(device);
+  std::vector<int64_t> sizes{static_cast<int64_t>(tensors[deviceIdx].size())};
+  std::vector<int64_t> strides{static_cast<int64_t>(t.numel())};
+  sizes.insert(sizes.end(), t.sizes().begin(), t.sizes().end());
+  strides.insert(strides.end(), t.strides().begin(), t.strides().end());
+  return at::empty_strided(
+      sizes, strides, t.options().memory_format(c10::nullopt));
+}
+
+inline at::Tensor newLikeFlat(std::vector<at::Tensor>& tensors) {
+  if (tensors.empty()) {
+    TORCH_CHECK(false, "Received an empty list");
+  }
+  auto& t = tensors[0];
+  at::DeviceGuard gpuGuard(t.device());
+  std::vector<int64_t> sizes{static_cast<int64_t>(tensors.size())};
+  sizes.insert(sizes.end(), t.sizes().begin(), t.sizes().end());
+  return at::empty(sizes, t.options());
+}
+
+inline std::vector<std::vector<int64_t>> getSizes(
+    const std::vector<at::Tensor>& tensors) {
+  std::vector<std::vector<int64_t>> sizes(tensors.size());
+  for (const auto i : c10::irange(tensors.size())) {
+    sizes[i] = tensors[i].sizes().vec();
+  }
+  return sizes;
+}
+
+inline std::vector<int> getDevices(const std::vector<at::Tensor>& tensors) {
+  std::vector<int> devices(tensors.size(), -1);
+  if (tensors[0].device().is_cuda()) {
+    for (const auto i : c10::irange(tensors.size())) {
+      devices[i] = tensors[i].storage().device().index();
+    }
+  }
+  return devices;
+}
+
+template <typename T>
+inline T* getDataPointer(const at::Tensor& tensor) {
+  // This method is only used in ProcessGroupGloo for now. Call sites must make
+  // sure that the input tensor is contiguous. It is OK if the tensor does not
+  // start from the beginning of the storage. For example, it could come from
+  // chunk(..., dim=0)[1]. Hence, we need to use data_ptr() instead of
+  // tensor.storage().data()
+  // NB: not using tensor.data<T>() because tensor is not aware of gloo::TYPE
+  return static_cast<T*>(tensor.data_ptr());
+}
+
+template <typename T>
+std::vector<T*> getDataPointers(const std::vector<at::Tensor>& tensors) {
+  std::vector<T*> ptrs(tensors.size());
+  for (const auto i : c10::irange(tensors.size())) {
+    ptrs[i] = getDataPointer<T>(tensors[i]);
+  }
+  return ptrs;
+}
+
+// For alltoall split size sanity check
+inline void checkSplitSizes(
+    const std::vector<int64_t>& split_sizes,
+    const at::Tensor& tensor,
+    int group_size) {
+  if (split_sizes.empty()) {
+    TORCH_CHECK(
+        tensor.size(0) % group_size == 0,
+        "Tensor's dim 0 does not divide equally across group size");
+  } else {
+    TORCH_CHECK(
+        split_sizes.size() == static_cast<size_t>(group_size),
+        "Number of tensor splits not equal to group size");
+    const auto sum = c10::sum_integers(split_sizes);
+    TORCH_CHECK(
+        sum == tensor.size(0), "Split sizes doesn't match total dim 0 size");
+  }
+}
+
+// Compute alltoall lengths and offsets, handling multi-dimension tensors
+template <typename T>
+size_t computeLengthsAndOffsets(
+    const std::vector<int64_t>& split_sizes,
+    const at::Tensor& tensor,
+    std::vector<T>* lengths,
+    std::vector<T>* offsets) {
+  size_t group_size = lengths->size();
+  bool equal_splits = false;
+  size_t dim0_size = tensor.size(0);
+  size_t row_size = (dim0_size ? tensor.numel() / dim0_size : 1);
+  size_t split_size = 0;
+  size_t offset = 0;
+
+  if (split_sizes.empty()) {
+    equal_splits = true;
+    split_size = tensor.size(0) / group_size;
+  }
+  for (const auto i : c10::irange(group_size)) {
+    size_t length = row_size * (equal_splits ? split_size : split_sizes[i]);
+    (*lengths)[i] = length;
+    (*offsets)[i] = offset;
+    // TODO: see if we should add overflow protection for offset
+    offset += length;
+  }
+  return offset;
+}
+
+template <typename T>
+size_t computeLengthsAndOffsets(
+    const std::vector<at::Tensor>& tensors,
+    std::vector<T>* lengths,
+    std::vector<T>* offsets) {
+  size_t group_size = lengths->size();
+  size_t offset = 0;
+  for (const auto i : c10::irange(group_size)) {
+    size_t length = tensors[i].numel();
+    (*lengths)[i] = length;
+    (*offsets)[i] = offset;
+    offset += length;
+  }
+  return offset;
+}
+
+using RankType = uint32_t;
+using SizeType = uint64_t;
+
+// `errno` is only meaningful when it fails. E.g., a  successful `fork()` sets
+// `errno` to `EINVAL` in child process on some macos
+// (https://stackoverflow.com/a/20295079), and thus `errno` should really only
+// be inspected if an error occurred.
+//
+// `success_cond` is an expression used to check if an error has happend. So for
+// `fork()`, we can use `SYSCHECK(pid = fork(), pid != -1)`. The function output
+// is stored in variable `__output` and may be used in `success_cond`.
+#ifdef _WIN32
+#define SYSCHECK(expr, success_cond)                                      \
+  while (true) {                                                          \
+    auto __output = (expr);                                               \
+    auto errno_local = WSAGetLastError();                                 \
+    (void)__output;                                                       \
+    if (!(success_cond)) {                                                \
+      if (errno == EINTR) {                                               \
+        continue;                                                         \
+      } else if (                                                         \
+          errno_local == WSAETIMEDOUT || errno_local == WSAEWOULDBLOCK) { \
+        C10_THROW_ERROR(DistNetworkError, "Socket Timeout");              \
+      } else {                                                            \
+        C10_THROW_ERROR(DistNetworkError, std::strerror(errno_local));    \
+      }                                                                   \
+    } else {                                                              \
+      break;                                                              \
+    }                                                                     \
+  }
+#else
+#define SYSCHECK(expr, success_cond)                             \
+  while (true) {                                                 \
+    auto __output = (expr);                                      \
+    (void)__output;                                              \
+    if (!(success_cond)) {                                       \
+      if (errno == EINTR) {                                      \
+        continue;                                                \
+      } else if (errno == EAGAIN || errno == EWOULDBLOCK) {      \
+        C10_THROW_ERROR(DistNetworkError, "Socket Timeout");     \
+      } else {                                                   \
+        C10_THROW_ERROR(DistNetworkError, std::strerror(errno)); \
+      }                                                          \
+    } else {                                                     \
+      break;                                                     \
+    }                                                            \
+  }
+#endif
+
+// Most functions indicate error by returning `-1`. This is a helper macro for
+// this common case with `SYSCHECK`.
+// Since SOCKET_ERROR = -1 in MSVC, so also leverage SYSCHECK_ERR_RETURN_NEG1
+#define SYSCHECK_ERR_RETURN_NEG1(expr) SYSCHECK(expr, __output != -1)
+
+namespace tcputil {
+
+// Send and receive
+template <typename T>
+void sendBytes(
+    int socket,
+    const T* buffer,
+    size_t length,
+    bool moreData = false) {
+  size_t bytesToSend = sizeof(T) * length;
+  if (bytesToSend == 0) {
+    return;
+  }
+
+  auto bytes = reinterpret_cast<const uint8_t*>(buffer);
+  uint8_t* currentBytes = const_cast<uint8_t*>(bytes);
+
+  int flags = 0;
+
+#ifdef MSG_MORE
+  if (moreData) { // there is more data to send
+    flags |= MSG_MORE;
+  }
+#endif
+
+// Ignore SIGPIPE as the send() return value is always checked for error
+#ifdef MSG_NOSIGNAL
+  flags |= MSG_NOSIGNAL;
+#endif
+
+  while (bytesToSend > 0) {
+    ssize_t bytesSent;
+    SYSCHECK_ERR_RETURN_NEG1(
+        bytesSent =
+            ::send(socket, (const char*)currentBytes, bytesToSend, flags))
+    if (bytesSent == 0) {
+      C10_THROW_ERROR(DistNetworkError, std::strerror(ECONNRESET));
+    }
+
+    bytesToSend -= bytesSent;
+    currentBytes += bytesSent;
+  }
+}
+
+template <typename T>
+void recvBytes(int socket, T* buffer, size_t length) {
+  size_t bytesToReceive = sizeof(T) * length;
+  if (bytesToReceive == 0) {
+    return;
+  }
+
+  auto bytes = reinterpret_cast<uint8_t*>(buffer);
+  uint8_t* currentBytes = bytes;
+
+  while (bytesToReceive > 0) {
+    ssize_t bytesReceived;
+    SYSCHECK_ERR_RETURN_NEG1(
+        bytesReceived = recv(socket, (char*)currentBytes, bytesToReceive, 0))
+    if (bytesReceived == 0) {
+      C10_THROW_ERROR(DistNetworkError, std::strerror(ECONNRESET));
+    }
+
+    bytesToReceive -= bytesReceived;
+    currentBytes += bytesReceived;
+  }
+}
+
+// send a vector's length and data
+template <typename T>
+void sendVector(int socket, const std::vector<T>& vec, bool moreData = false) {
+  SizeType size = vec.size();
+  sendBytes<SizeType>(socket, &size, 1, true);
+  sendBytes<T>(socket, vec.data(), size, moreData);
+}
+
+// receive a vector as sent in sendVector
+template <typename T>
+std::vector<T> recvVector(int socket) {
+  SizeType valueSize;
+  recvBytes<SizeType>(socket, &valueSize, 1);
+  std::vector<T> value(valueSize);
+  recvBytes<T>(socket, value.data(), value.size());
+  return value;
+}
+
+// this is only for convenience when sending rvalues
+template <typename T>
+void sendValue(int socket, const T& value, bool moreData = false) {
+  sendBytes<T>(socket, &value, 1, moreData);
+}
+
+template <typename T>
+T recvValue(int socket) {
+  T value;
+  recvBytes<T>(socket, &value, 1);
+  return value;
+}
+
+// send a string's length and data
+inline void sendString(
+    int socket,
+    const std::string& str,
+    bool moreData = false) {
+  SizeType size = str.size();
+  sendBytes<SizeType>(socket, &size, 1, true);
+  sendBytes<char>(socket, str.data(), size, moreData);
+}
+
+// receive a string as sent in sendString
+inline std::string recvString(int socket) {
+  SizeType valueSize;
+  recvBytes<SizeType>(socket, &valueSize, 1);
+  std::vector<char> value(valueSize);
+  recvBytes<char>(socket, value.data(), value.size());
+  return std::string(value.data(), value.size());
+}
+
+} // namespace tcputil
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/WinSockUtils.hpp

@@ -0,0 +1,27 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/Utils.hpp>
+
+namespace c10d {
+namespace tcputil {
+
+#define CONNECT_SOCKET_OFFSET 1
+
+inline int poll(struct pollfd* fdArray, unsigned long fds, int timeout) {
+  return WSAPoll(fdArray, fds, timeout);
+}
+
+inline void addPollfd(
+    std::vector<struct pollfd>& fds,
+    int socket,
+    short events) {
+  fds.push_back({(SOCKET)socket, events});
+}
+
+inline struct ::pollfd getPollfd(int socket, short events) {
+  struct ::pollfd res = {(SOCKET)socket, events};
+  return res;
+}
+
+} // namespace tcputil
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/Work.hpp

@@ -0,0 +1,164 @@
+#pragma once
+
+#include <ATen/ATen.h>
+#include <stdexcept>
+#include <vector>
+
+constexpr auto kNoTimeout = std::chrono::milliseconds(0);
+
+namespace c10d {
+
+constexpr const char* const kSeqNumStoreKey = "SEQ_NUM_STORE_KEY";
+
+enum class OpType : std::uint8_t {
+  BROADCAST = 0,
+  ALLREDUCE = 1,
+  ALLREDUCE_COALESCED = 2,
+  REDUCE = 3,
+  ALLGATHER = 4,
+  _ALLGATHER_BASE = 5,
+  ALLGATHER_COALESCED = 6,
+  GATHER = 7,
+  SCATTER = 8,
+  REDUCE_SCATTER = 9,
+  ALLTOALL_BASE = 10,
+  ALLTOALL = 11,
+  SEND = 12,
+  RECV = 13,
+  RECVANYSOURCE = 14,
+  BARRIER = 15,
+  _REDUCE_SCATTER_BASE = 16,
+  COALESCED = 17,
+  _ALLREDUCE_SPARSE = 18,
+  UNKNOWN = 100,
+};
+
+// Converts OpType to human readable string.
+TORCH_API std::string opTypeToString(OpType opType);
+
+// Whether or not an OP is an p2p op (SEND, RECV, RECVANYSOURCE)
+TORCH_API bool isP2POp(OpType opType, bool batchP2P = false);
+
+// Please do not use Work API, it is going away, to be
+// replaced by ivalue::Future.
+// Python binding for this class might change, please do not assume
+// this will be bound using pybind.
+class TORCH_API Work : public torch::CustomClassHolder {
+ public:
+  Work(
+      int rank = -1,
+      OpType opType = OpType::UNKNOWN,
+      const char* profilingTitle = nullptr,
+      const c10::optional<std::vector<at::Tensor>>& inputTensors =
+          c10::nullopt);
+
+  ~Work() override;
+
+  // Checks if request has completed. Non-blocking operation.
+  virtual bool isCompleted();
+
+  // Returns if the work completed successfully.
+  // If false, the exception function can be called to get details.
+  virtual bool isSuccess() const;
+
+  // Returns exception if isSuccess() returned false.
+  virtual std::exception_ptr exception() const;
+
+  // Returns source rank if this objects represents a recv-from-any.
+  virtual int sourceRank() const;
+
+  // Returns result tensors, if applicable.
+  // If work is not supposed to have result, we return empty list.
+  virtual std::vector<at::Tensor> result();
+
+  // Ensures that operations on the output tensors that are invoked
+  // after this function returns are correctly sequenced after the
+  // asynchronous completion of this work.
+  //
+  // For CUDA tensors, it inserts stream synchronization such that
+  // the streams of the caller wait for completion of the
+  // asynchronous operations on the destination tensors.
+  //
+  // For CPU tensors, it is currently a nop.
+  //
+  // This function should only be used if the caller polls for
+  // completion through the `isCompleted` function, it has returned
+  // true, and the `isSuccess` function also has returned true.
+  //
+  virtual void synchronize();
+
+  // Waits until request completes. Blocking operation.
+  // Throws if the work completed with an exception.
+  // Returns false if the work is aborted.
+  // Otherwise, it always returns true, indicating the work is completed.
+  //
+  // Functionally equivalent to:
+  //
+  //   while (!isCompleted()) { /* nop */ }
+  //   auto success = isSuccess();
+  //   if (!success) { std::rethrow_exception(exception()); }
+  //   return success;
+  //
+  virtual bool wait(std::chrono::milliseconds timeout = kNoTimeout);
+
+  virtual void abort();
+
+  // Returns a Future object that will be associated with the completion of
+  // work. Only NCCL backend is currently supported.
+  virtual c10::intrusive_ptr<c10::ivalue::Future> getFuture();
+
+  virtual float getDuration() const;
+
+  virtual uint64_t getSequencenumber() const;
+
+  OpType retrieveOpType() const;
+
+  static c10::intrusive_ptr<Work> create_from_future(
+      const c10::intrusive_ptr<c10::ivalue::Future>&);
+
+ protected:
+  // Completes the work object and optionally sets the exception in a
+  // thread-safe manner. Notifies all waiting condition variables as well.
+  void finish(std::exception_ptr exception = nullptr);
+
+  // Similar to finish, but throws an exception if one is already set or
+  // provided by the user.
+  void finishAndThrow(std::exception_ptr exception);
+
+  mutable std::mutex mutex_;
+  std::condition_variable cv_;
+  bool completed_ = false;
+  std::exception_ptr exception_;
+
+  // Current rank of the node.
+  const int rank_;
+
+  // Operation type that this work object refers to.
+  OpType opType_;
+
+  // When profiling, the callback to record end of operation event. This
+  // callback needs to be called when collective operation is complete.
+  std::function<void()> recordFunctionEndCallback_;
+};
+
+struct TORCH_API WorkInfo {
+  WorkInfo(
+      const OpType& opType,
+      const uint64_t seq,
+      const std::chrono::time_point<std::chrono::system_clock>& timeStarted,
+      const std::chrono::time_point<std::chrono::system_clock>& timeFinished,
+      const std::chrono::duration<float>& activeDuration)
+      : opType(opType),
+        seq(seq),
+        timeStarted(timeStarted),
+        timeFinished(timeFinished),
+        activeDuration(activeDuration) {}
+
+  OpType opType;
+  uint64_t seq;
+  std::chrono::time_point<std::chrono::system_clock> timeStarted;
+  std::chrono::time_point<std::chrono::system_clock> timeFinished;
+  std::chrono::duration<float> activeDuration;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/c10d.h

@@ -0,0 +1,13 @@
+#pragma once
+
+#include <torch/csrc/python_headers.h>
+
+namespace torch {
+namespace distributed {
+namespace c10d {
+
+PyMethodDef* python_functions();
+
+} // namespace c10d
+} // namespace distributed
+} // namespace torch

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/comm.hpp

@@ -0,0 +1,140 @@
+#pragma once
+
+#include <ATen/ATen.h>
+#include <ATen/core/ivalue.h>
+#include <torch/csrc/Export.h>
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <utility>
+
+namespace c10d {
+
+// Broadcast many tensors to all processes in the process group.
+TORCH_API void broadcast_coalesced(
+    const c10::intrusive_ptr<c10d::ProcessGroup>& process_group,
+    at::TensorList tensors,
+    size_t buffer_size,
+    int rank = 0);
+
+// This class passes bucket contents tensor to DDP communication hook.
+class TORCH_API GradBucket {
+ public:
+  explicit GradBucket(
+      size_t index,
+      size_t bucket_count,
+      at::Tensor tensor,
+      std::vector<size_t> offsets,
+      std::vector<size_t> lengths,
+      std::vector<c10::IntArrayRef> sizes_vec,
+      std::vector<at::Tensor> parameters,
+      c10::optional<at::Tensor> sparse_grad_indices)
+      : index_(index),
+        bucket_count_(bucket_count),
+        buffer_(std::move(tensor)),
+        offsets_(std::move(offsets)),
+        lengths_(std::move(lengths)),
+        sizes_vec_(std::move(sizes_vec)),
+        parameters_(std::move(parameters)),
+        sparse_grad_indices_(std::move(sparse_grad_indices)) {}
+
+  // Returns the index of the bucket, which is unique across all the buckets.
+  size_t getIndex() const {
+    return index_;
+  }
+
+  const at::Tensor& getBuffer() const {
+    return buffer_;
+  }
+
+  // Returns a mutable buffer compared with the above method.
+  at::Tensor& getBufferRef() {
+    return buffer_;
+  }
+
+  // Overwrites the buffer at a specific index.
+  void setBuffer(at::Tensor& buffer) {
+    buffer_ = buffer;
+  }
+
+  // Each tensor in the list that getGradients corresponds to a
+  // parameter.
+  std::vector<at::Tensor> getGradients() const;
+
+  // Returns model parameters belonging to this bucket. They are returned in the
+  // same order as gradient tensors via getGradients(). For example,
+  // getParameters[i] will have its gradient stored in
+  // getGradients[i]
+  const std::vector<at::Tensor> getParameters() const {
+    return parameters_;
+  }
+
+  // Returns whther this bucket is the last bucket to allreduce in an iteration.
+  bool isLast() const {
+    return index_ == bucket_count_ - 1;
+  }
+
+  c10::optional<at::Tensor>& getSparseGradIndices() {
+    return sparse_grad_indices_;
+  }
+
+ private:
+  size_t index_;
+  size_t bucket_count_;
+  at::Tensor buffer_;
+
+  // Per-variable info in buffer_.
+  std::vector<size_t> offsets_;
+  std::vector<size_t> lengths_;
+  std::vector<c10::IntArrayRef> sizes_vec_;
+
+  // Model parameters for this bucket.
+  const std::vector<at::Tensor> parameters_;
+
+  // Predefined sparse indices for this bucket (only used for sparse tensors).
+  // The gradients will be updated to have indices with these tensor values
+  c10::optional<at::Tensor> sparse_grad_indices_;
+};
+
+// Base class of both `PythonCommHook` and `CppCommHook`.
+// Requires implementing 1) `runHook` method that communicates gradients
+// asynchronously, and 2) `parseHookResult` method that converts the hook
+// result into a tensor.
+class TORCH_API CommHookInterface {
+ public:
+  virtual ~CommHookInterface() = default;
+
+  // Passes the input grad bucket to the registered communication hook.
+  // Once the tensor in the bucket are ready, kicks off the hook asynchronously
+  // and returns a future that holds the communication results.
+  virtual c10::intrusive_ptr<c10::ivalue::Future> runHook(
+      GradBucket& bucket) = 0;
+
+  // Returns the resulting tensor once the communication hook result is
+  // ready. The resulting tensor will then be copied to the grads of
+  // individual parameters.
+  virtual at::Tensor parseHookResult(const c10::IValue& result) = 0;
+};
+
+namespace detail {
+// This helper function is called both by CppCommHookInterface below and inside
+// reducer.
+TORCH_API at::Tensor parseCppCommHookResult(const c10::IValue& result);
+} // namespace detail
+
+// This CppCommHook interface only requires implementing runHook method that
+// potentially uses a state.
+template <typename T>
+class CppCommHookInterface : public CommHookInterface {
+ public:
+  explicit CppCommHookInterface(T state) : state_(std::move(state)) {}
+
+  ~CppCommHookInterface() override = default;
+
+  at::Tensor parseHookResult(const c10::IValue& result) override {
+    return detail::parseCppCommHookResult(result);
+  }
+
+ protected:
+  T state_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/debug.h

@@ -0,0 +1,23 @@
+// Copyright (c) Meta Platforms, Inc. and its affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
+
+#pragma once
+
+#include <c10/macros/Macros.h>
+
+namespace c10d {
+
+enum class DebugLevel { Off = 0, Info = 1, Detail = 2 };
+
+TORCH_API void setDebugLevel(DebugLevel level);
+
+// Sets the debug level based on the value of the `TORCH_DISTRIBUTED_DEBUG`
+// environment variable.
+TORCH_API void setDebugLevelFromEnvironment();
+
+TORCH_API DebugLevel debug_level() noexcept;
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/default_comm_hooks.hpp

@@ -0,0 +1,52 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <torch/csrc/distributed/c10d/comm.hpp>
+
+namespace c10d {
+
+enum class BuiltinCommHookType {
+  ALLREDUCE = 1,
+  FP16_COMPRESS = 2,
+};
+
+class AllReduceCommHook
+    : public CppCommHookInterface<c10::intrusive_ptr<ProcessGroup>> {
+ public:
+  explicit AllReduceCommHook(const c10::intrusive_ptr<ProcessGroup>& state)
+      : CppCommHookInterface<c10::intrusive_ptr<ProcessGroup>>(state) {}
+
+  ~AllReduceCommHook() override = default;
+
+  c10::intrusive_ptr<c10::ivalue::Future> runHook(GradBucket& bucket) override;
+};
+
+class FP16CompressCommHook
+    : public CppCommHookInterface<c10::intrusive_ptr<ProcessGroup>> {
+ public:
+  explicit FP16CompressCommHook(const c10::intrusive_ptr<ProcessGroup>& state)
+      : CppCommHookInterface<c10::intrusive_ptr<ProcessGroup>>(state) {}
+
+  ~FP16CompressCommHook() override = default;
+
+  c10::intrusive_ptr<c10::ivalue::Future> runHook(GradBucket& bucket) override;
+};
+
+// Almost same as AllReduceCommHook, but without division inside the hook.
+// This enables the optimization of fusing copy and division and saves one scan
+// over all the input parameters, when no communication hook is provided by the
+// user. Only used internally and not released as a public built-in
+// communication hook.
+class _AllReduceBySumCommHook
+    : public CppCommHookInterface<c10::intrusive_ptr<ProcessGroup>> {
+ public:
+  explicit _AllReduceBySumCommHook(
+      const c10::intrusive_ptr<ProcessGroup>& state)
+      : CppCommHookInterface<c10::intrusive_ptr<ProcessGroup>>(state) {}
+
+  ~_AllReduceBySumCommHook() override = default;
+
+  c10::intrusive_ptr<c10::ivalue::Future> runHook(GradBucket& bucket) override;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/error.h

@@ -0,0 +1,56 @@
+// Copyright (c) Facebook, Inc. and its affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
+
+#pragma once
+
+#include <cstring>
+#include <system_error>
+
+#include <fmt/format.h>
+
+namespace fmt {
+
+template <>
+struct formatter<std::error_category> {
+  constexpr decltype(auto) parse(format_parse_context& ctx) const {
+    return ctx.begin();
+  }
+
+  template <typename FormatContext>
+  decltype(auto) format(const std::error_category& cat, FormatContext& ctx)
+      const {
+    if (std::strcmp(cat.name(), "generic") == 0) {
+      return fmt::format_to(ctx.out(), "errno");
+    } else {
+      return fmt::format_to(ctx.out(), "{} error", cat.name());
+    }
+  }
+};
+
+template <>
+struct formatter<std::error_code> {
+  constexpr decltype(auto) parse(format_parse_context& ctx) const {
+    return ctx.begin();
+  }
+
+  template <typename FormatContext>
+  decltype(auto) format(const std::error_code& err, FormatContext& ctx) const {
+    return fmt::format_to(
+        ctx.out(), "({}: {} - {})", err.category(), err.value(), err.message());
+  }
+};
+
+} // namespace fmt
+
+namespace c10d {
+namespace detail {
+
+inline std::error_code lastError() noexcept {
+  return std::error_code{errno, std::generic_category()};
+}
+
+} // namespace detail
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/exception.h

@@ -0,0 +1,33 @@
+// Copyright (c) Facebook, Inc. and its affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
+
+#pragma once
+
+#include <stdexcept>
+
+#include <c10/macros/Macros.h>
+#include <c10/util/Exception.h>
+
+// Utility macro similar to C10_THROW_ERROR, the major difference is that this
+// macro handles exception types defined in the c10d namespace, whereas
+// C10_THROW_ERROR requires an exception to be defined in the c10 namespace.
+#define C10D_THROW_ERROR(err_type, msg) \
+  throw ::c10d::err_type(               \
+      {__func__, __FILE__, static_cast<uint32_t>(__LINE__)}, msg)
+
+namespace c10d {
+
+using c10::DistNetworkError;
+
+class TORCH_API SocketError : public DistNetworkError {
+  using DistNetworkError::DistNetworkError;
+};
+
+class TORCH_API TimeoutError : public DistNetworkError {
+  using DistNetworkError::DistNetworkError;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/logger.hpp

@@ -0,0 +1,104 @@
+#include <c10/util/Logging.h>
+#include <torch/csrc/distributed/c10d/reducer.hpp>
+
+#include <mutex>
+
+namespace c10d {
+
+class TORCH_API Logger {
+ public:
+  explicit Logger(std::shared_ptr<c10d::Reducer> reducer);
+  // Set logging data that can be got during DistributedDataParallel
+  // construction time.
+  void set_construction_data_and_log(
+      const std::string& module_name,
+      const std::vector<int>& device_ids,
+      int output_device,
+      bool broadcast_buffers,
+      bool has_sync_bn,
+      bool static_graph);
+
+  void set_static_graph();
+
+  // An interface for users to get DDPLoggingData and log them
+  // in the applications. Explanation of logging fields are in
+  // "struct DDPLoggingData" of "torch/c10/util/Logging.h".
+  at::DDPLoggingData get_ddp_logging_data();
+
+  // Stream insertion operator for logging data to stream under
+  // TORCH_DISTRIBUTED_DEBUG.
+  friend std::ostream& operator<<(std::ostream& output, const Logger& logger);
+
+  ~Logger() noexcept(false) {
+    // Log if DDP graph is static in Logger dtor instead of Reducer dtor since
+    // Logger is deleted before Reducer.
+    log_if_graph_static(reducer_->ddp_graph_static());
+  }
+
+  // Set environment variables.
+  void set_env_variables();
+  // Set parameters stats.
+  void set_parameter_stats();
+  // Get size of each bucket (Bytes).
+  std::vector<int64_t> get_bucket_sizes();
+  // Get variable indices for each bucket.
+  std::vector<std::vector<size_t>> get_per_bucket_variable_indices();
+  // Set comm. hook, if used
+  void set_comm_hook(const std::string& hook);
+  // Set running with uneven input detection (model.join() context manager)
+  void set_uneven_input_join();
+
+  // Reset performance stats at current iteration
+  void reset_performance_stats();
+
+  // Calculate avg stats using cpu timer and gpu timer
+  // that has been recorded in reducer.
+  void calculate_avg_time(
+      int64_t& avg_time,
+      int64_t& time_duration,
+      Timer& timer,
+      Timer::Event start_event,
+      Timer::Event end_event);
+
+  // Set the absolute time of the event that has been recorded in reducer.
+  void set_event_time(int64_t& event_time, Timer& timer, Timer::Event event);
+  // Set stats that can be collected only during
+  // training loop. It is called at the beginning of forward call
+  // to record the run time stats of sampled iterations that previously ran.
+  // GPU performance stats are collected only for single process
+  // single device program and single device module right now.
+  // TODO to support single process multiple devices and multi device modules,
+  // events need to be created and recorded on multiple devices.
+  void set_runtime_stats_and_log();
+
+  // Called when DDP/reducer is failing with an error. The
+  // logging data structure will have two fields filled: "has_error" indicating
+  // that this iteration encountered an error and other fields are not valid,
+  // and "error", a string which contains the error message that DDP failed
+  // with.
+  template <typename... Args>
+  void set_error_and_log(const std::string& ddp_error, const Args&... args) {
+    ddp_logging_data_->ints_map["has_error"] = 1;
+    auto err = c10::str(ddp_error, args...);
+    ddp_logging_data_->strs_map["error"] = err;
+    // Report the iteration we are erroring at so user knows how many examples
+    // successfully processed before this error was hit.
+    ddp_logging_data_->ints_map["iteration"] = reducer_->num_iterations_;
+    at::LogPyTorchDDPUsage(*ddp_logging_data_);
+  }
+
+  // When running without static graph, called when reducer is destroyed to log
+  // if graph was actually static and is a candidate for static graph
+  // optimization.
+  void log_if_graph_static(bool is_static);
+
+ private:
+  // ddp_logging_data_ is used to hold all the ddp related logging
+  // data fields.
+  std::unique_ptr<at::DDPLoggingData> ddp_logging_data_;
+  std::shared_ptr<c10d::Reducer> reducer_;
+  // track the number of iterations when runtime stats are collected so far.
+  long num_iterations_stats_recorded_ = 0;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/logging.h

@@ -0,0 +1,51 @@
+// Copyright (c) Meta Platforms, Inc. and its affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
+
+#pragma once
+
+#include <string>
+
+#include <c10/macros/Macros.h>
+#include <c10/util/Logging.h>
+#include <fmt/format.h>
+
+namespace c10d {
+namespace detail {
+
+enum class LogLevel { Trace, Debug, Info, Warning, Error };
+
+TORCH_API bool isLogLevelEnabled(LogLevel level) noexcept;
+
+template <typename... T>
+std::string formatLogMessage(fmt::string_view fmt, T&&... args) {
+  return fmt::vformat(fmt, fmt::make_format_args(args...));
+}
+
+} // namespace detail
+} // namespace c10d
+
+#define C10D_ERROR(...)                                                      \
+  LOG_IF(                                                                    \
+      ERROR, c10d::detail::isLogLevelEnabled(c10d::detail::LogLevel::Error)) \
+      << "[c10d] " << c10d::detail::formatLogMessage(__VA_ARGS__)
+
+#define C10D_WARNING(...)                                               \
+  LOG_IF(                                                               \
+      WARNING,                                                          \
+      c10d::detail::isLogLevelEnabled(c10d::detail::LogLevel::Warning)) \
+      << "[c10d] " << c10d::detail::formatLogMessage(__VA_ARGS__)
+
+#define C10D_INFO(...)                                                        \
+  LOG_IF(INFO, c10d::detail::isLogLevelEnabled(c10d::detail::LogLevel::Info)) \
+      << "[c10d] " << c10d::detail::formatLogMessage(__VA_ARGS__)
+
+#define C10D_DEBUG(...)                                                        \
+  LOG_IF(INFO, c10d::detail::isLogLevelEnabled(c10d::detail::LogLevel::Debug)) \
+      << "[c10d - debug] " << c10d::detail::formatLogMessage(__VA_ARGS__)
+
+#define C10D_TRACE(...)                                                        \
+  LOG_IF(INFO, c10d::detail::isLogLevelEnabled(c10d::detail::LogLevel::Trace)) \
+      << "[c10d - trace] " << c10d::detail::formatLogMessage(__VA_ARGS__)

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/reducer.hpp

@@ -0,0 +1,589 @@
+#pragma once
+
+#include <c10/core/ScalarType.h>
+#include <atomic>
+#include <memory>
+#include <mutex>
+#include <tuple>
+#include <unordered_map>
+#include <vector>
+
+#include <ATen/core/ivalue_inl.h>
+#include <c10/macros/Macros.h>
+#include <c10/util/intrusive_ptr.h>
+#include <torch/csrc/autograd/function.h>
+#include <torch/csrc/autograd/profiler.h>
+#include <torch/csrc/autograd/variable.h>
+#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <torch/csrc/distributed/c10d/comm.hpp>
+#include <torch/csrc/distributed/c10d/debug.h>
+#include <torch/csrc/distributed/c10d/default_comm_hooks.hpp>
+#include <torch/csrc/distributed/c10d/reducer_timer.hpp>
+#ifndef _WIN32
+#include <torch/csrc/distributed/autograd/context/context.h>
+#endif
+
+namespace c10d {
+
+constexpr int kDefaultFirstBucketBytes = int(1024 * 1024);
+constexpr int kDefaultBucketBytesCap = int(25 * 1024 * 1024);
+// Collect runtime stats once for every kDDPRuntimeLoggingSampleRate iterations.
+constexpr int kDDPRuntimeLoggingSampleRate = 100;
+
+// Forward declaration
+class Logger;
+
+// Local accumulator type for a single bucket.
+struct BucketAccumulator {
+  std::vector<size_t> indices;
+  size_t size = 0;
+  size_t size_limit = 0;
+};
+
+class TORCH_API Reducer {
+ public:
+  // The constructor takes a list of variables (i.e. parameters) for this
+  // process's single model replica (as DDP assumes single-process
+  // single-device). The bucket assignment for this reducer, `bucket_indices`,
+  // is specified as a list of buckets, each of which is specified as a list of
+  // indices into the bucket's `variables` list.
+  explicit Reducer(
+      std::vector<at::Tensor> params,
+      std::vector<std::vector<size_t>> bucket_indices,
+      std::vector<size_t> per_bucket_size_limits,
+      c10::intrusive_ptr<c10d::ProcessGroup> process_group,
+      std::vector<bool> expect_sparse_gradients,
+      int64_t bucket_bytes_cap,
+      bool find_unused_parameters,
+      bool gradient_as_bucket_view,
+      std::unordered_map<size_t, std::string> param_names,
+      int64_t first_bucket_bytes_cap);
+
+  ~Reducer() noexcept(false);
+
+  // To (re-)initialize bucket assignment, pass a list of buckets, each of
+  // which is specified by a list of indices in the bucket's `variables` list.
+  // This function performs validation that the variables within a bucket
+  // all live on the same device and have the same dimensionality.
+  void initialize_buckets(std::vector<std::vector<size_t>> bucket_indices);
+
+  void autograd_hook(size_t index);
+
+  // This function is called when the forward function has produced an output,
+  // and the user wishes to reduce gradients in the backwards pass.
+  // If they don't, and wish to accumulate gradients before reducing them,
+  // a call to this function can simply be omitted.
+  void prepare_for_backward(const std::vector<at::Tensor>& outputs);
+
+  // Called at the beginning of forward() inside DistributedDataParallel,
+  // right now it captures the starting time of forward in each iteration.
+  void prepare_for_forward();
+
+  // Returns the relative time in nanoseconds when gradients were ready,
+  // with respect to the time `prepare_for_backward` was called. The
+  // vector is for parameters for a single model replica.
+  std::vector<int64_t> get_backward_stats() const {
+    return backward_stats_;
+  }
+
+  // Registers a hook to the reducer. The hook is `CommHookInterface`
+  // type to allow both Python and CPP hooks. This function can only
+  // be called once before calling backward.
+  // Cannot combine with the call of `register_builtin_comm_hook`.
+  void register_comm_hook(std::unique_ptr<CommHookInterface> iface);
+
+  // Registers a built-in C++ comm hook to the reducer. This function can only
+  // be called once before calling backward.
+  // Cannot combine with the call of `register_comm_hook`.
+  void register_builtin_comm_hook(c10d::BuiltinCommHookType comm_hook_type);
+
+  // Informs reducer that optimizer is running in backward, so gradients
+  // don't need to be copied from buckets as the optimizer would've already
+  // been applied.
+  void set_optimizer_in_backward() {
+    optim_in_backward_ = true;
+  };
+
+  // Runs allreduce or installed communication hook given GradBucket instance.
+  c10::intrusive_ptr<c10::ivalue::Future> run_comm_hook(
+      GradBucket& grad_bucket);
+
+  // Runs default allreduce hook.
+  c10::intrusive_ptr<c10::ivalue::Future> run_allreduce_hook(
+      GradBucket& grad_bucket);
+
+  // Returns gradient buckets in sequential order of buckets_. This is the order
+  // in which buckets are reduced across processes. If return_zero_tensors=true,
+  // will return zero tensors of the same shape instead of the true tensors.
+  std::vector<c10d::GradBucket> get_grad_buckets(
+      bool return_zero_tensors = true) const;
+
+  // Rebuild buckets based on rebuilt_params_ and rebuilt_param_indices_
+  // according to when tensors received grads in the backward pass.
+  // TODO this function makes broadcast communication call and
+  // could be overlapped with next forward() call, thus
+  // it could be async. Will make it async when rebuilding buckets for
+  // find_unused_parameters = true case, as we could rebuild buckets more than
+  // once for find_unused_parameters = true case, where subgraphs are trained
+  // and parameter indices order may change more frequently.
+  // For find_unused_parameters = false case, buckets are only rebuilt once,
+  // the performance cost is negligible. Returns true if the buckets were
+  // rebuilt.
+  bool rebuild_buckets();
+
+  void setSparseMetadata(std::map<std::string, at::Tensor>& metadata);
+
+  // Install futures that should be awaited at end of backwards. Currently these
+  // are only used by user-defined custom buffer reduction hooks, but can be
+  // generalized to any user-originating futures that need to be awaited.
+  void install_futures(c10::List<c10::intrusive_ptr<c10::ivalue::Future>> futs);
+
+  // Returns true if we should rebuild buckets, else false. We only rebuild
+  // buckets once after the first iteration and never rebuild them if
+  // find_unused_parameters_.
+  inline bool should_rebuild_buckets() const {
+    return (static_graph_ || !find_unused_parameters_) && !has_rebuilt_bucket_;
+  }
+
+  // Pushes all parameters to be rebuilt.
+  void push_rebuilt_params_for_all_indices();
+
+  // Creates and sets ForwardPassWorkHandle given a Work and the
+  // corresponding tensor being reduced.
+  void set_forward_pass_work_handle(
+      c10::intrusive_ptr<c10d::Work> forwardPassWorkHandle,
+      bool useStaticWorldSize);
+
+  // Retrieve on-device tensors used to track locally unused parameters. It is
+  // a tensor where index i = 1 if the Variable with that index has been used.
+  at::Tensor get_local_used_map_on_device() const;
+
+  // An function for users to set sample_rate of collecting
+  // runtime stats. The time stats will be recorded for the
+  // first 10 iterations, after 10 iterations time stats will be
+  // recorded once every "sample_rate" training iterations.
+  void set_ddp_runtime_logging_sample_rate(int sample_rate);
+
+  // Specify the training graph is static.
+  void set_static_graph();
+
+  // Delay all reduce to be after all gradients' calculation is complete.
+  void delay_all_reduce();
+
+  void set_mixed_precision_param_dtype(c10::ScalarType dtype);
+
+  // Weak reference to associated DDP logger. The reference is weak to avoid
+  // refcycle between reducer and logger.
+  void set_logger(std::weak_ptr<c10d::Logger> logger);
+
+  // When graph is not explicitly set by user as static and has unused
+  // parameters, this will return whether the graph has been static until the
+  // current iteration, which means unused params set has not changed.
+  bool ddp_graph_static();
+
+  // Removes autograd hooks registered by the Reducer on the model parameters.
+  void remove_autograd_hooks();
+
+  // Checks whether or not the reducer has finalized the current backward
+  // iteration.
+  void check_finalized();
+
+  // Updates the underlying process group used by DDP with the new process
+  // group.
+  void update_process_group(
+      c10::intrusive_ptr<c10d::ProcessGroup> new_process_group);
+
+  // Resets reducer state.
+  void reset_state();
+
+ protected:
+  // Forward declaration.
+  struct Bucket;
+
+  void push_rebuilt_params(const size_t& index);
+
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  mutable std::mutex mutex_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  const std::vector<at::Tensor> params_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  c10::intrusive_ptr<::c10d::ProcessGroup> process_group_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  std::vector<bool> expect_sparse_gradients_;
+
+  std::vector<std::shared_ptr<torch::autograd::Node>>
+      grad_accumulators_; // NOLINT(cppcoreguidelines-non-private-member-variables-in-classes)
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  std::unordered_map<torch::autograd::Node*, size_t> gradAccToVariableMap_;
+  std::vector<std::pair<uintptr_t, std::shared_ptr<torch::autograd::Node>>>
+      hooks_; // NOLINT(cppcoreguidelines-non-private-member-variables-in-classes)
+
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  bool expect_autograd_hooks_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  bool require_finalize_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  size_t next_bucket_;
+
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  bool has_marked_unused_parameters_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  const bool find_unused_parameters_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  const bool gradient_as_bucket_view_;
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  std::vector<size_t> unused_parameters_;
+  // Previous iteration's unused params, used for checking if unused parameters
+  // change between iterations. Only filled during the first backwards call.
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  std::vector<size_t> prev_iteration_unused_parameters_;
+  // Whether graph is static or not. When user does not explicitly set static
+  // graph, the only possible dynamism is set of unused parameters changing
+  // between iterations which is tracked by this flag.
+  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
+  bool ddp_graph_static_{true};
+  // Locally used parameter maps indicating if parameters are used locally
+  // during the current iteration or no_sync session if no_sync is on.
+  // Each map is a one-dim int32 tensor of number of parameters. These tensors
+  // are marked in autograd_hook to indicate the corresponding param has been
+  // used, and get allreduced in the end of backward step of current iteration
+  // or no_sync session for figuring out the globally unused parameters.
+  //
+  // local_used_map_:     CPU tensor for bookkeeping locally used params
+  // local_used_map_dev_: dev tensor for reducing globally unused params
+  at::Tensor local_used_map_;
+  at::Tensor local_used_map_dev_;
+  // Indicate that reduction is done and D2H copy is done as well.
+  bool local_used_map_reduced_;
+
+  // Weak pointer to associated DDP logger.
+  std::weak_ptr<c10d::Logger> logger_;
+  // List of futures installed by Reducer::install_futures that should be
+  // awaited at the end of backwards pass.
+  c10::optional<c10::List<c10::intrusive_ptr<c10::ivalue::Future>>>
+      installed_futures_{c10::nullopt};
+  // Mixed precision parameter dtype for bucket type checking.
+  c10::optional<c10::ScalarType> mixed_precision_param_dtype_{c10::nullopt};
+
+  // Work handle for allreduce on local_used_map_
+  c10::intrusive_ptr<c10d::Work> local_used_work_;
+
+  void mark_variable_ready_dense(size_t variable_index);
+
+  void mark_variable_ready_sparse(size_t variable_index);
+
+  void mark_variable_ready(size_t variable_index);
+
+  void mark_bucket_ready(size_t bucket_index);
+
+  void finalize_bucket_dense(Bucket& bucket);
+
+  void finalize_backward();
+
+  // Returns list of model parameters corresponding to the given bucket.
+  // bucket_index is a key to cache after buckets are rebuilt, after which this
+  // mapping never changes.
+  std::vector<at::Tensor> get_variables_for_bucket(
+      size_t bucket_index,
+      const Bucket& bucket) const;
+
+  // Asserts that the reduction for the previous iteration has finished before
+  // rebuilding buckets or kicking off the next one.
+  void ensure_prior_reduction_finished();
+
+  // Broadcast rebuilt buckets from rank 0 to other ranks before initializing
+  // the buckets
+  void sync_bucket_indices(std::vector<std::vector<size_t>>& bucket_indices);
+
+  // We'd like to use DistAutogradContext::GradCallback here but dist autograd
+  // doesn't exist under Windows. So we just directly use the concrete type but
+  // to preserve and enforce our original intent we do a static assert when dist
+  // autograd is available.
+  using GradCallback = std::function<bool(at::Tensor&)>;
+#ifndef _WIN32
+  static_assert(
+      std::is_same<
+          GradCallback,
+          torch::distributed::autograd::DistAutogradContext::GradCallback>::
+          value,
+      "");
+#endif
+  void runGradCallbackForVariable(at::Tensor& variable, GradCallback&& cb);
+
+  // This function is called inside `initialize_buckets()`. It initializes both
+  // `bucket_views_in` and `bucket_views_out` with views for each variable's
+  // gradient into the bucket's flattened `gradients` tensor. Views serve as
+  // entry points to `copy_()` each grad's data in/out of the flattened
+  // `gradients` tensor.
+  void initialize_bucket_views(Bucket& bucket);
+
+  // This function is called inside `finalize_backward`, it happens only if
+  // DDP communication hook was registered to recreate just bucket_views_out
+  // with the result of `future_work`.
+  void populate_bucket_views_out(Bucket& bucket, at::Tensor& tensor);
+
+  // If gradient_as_bucket_view_ is false, after allreduce buckets,
+  // copy bucket results back to grads.
+  void copy_bucket_to_grad(
+      at::Tensor& variable,
+      Reducer::Bucket& bucket,
+      size_t intra_bucket_index,
+      bool global_unused);
+  // Check layout of grad and bucket_view before copying the grad to bucket.
+  void check_grad_layout(const at::Tensor& grad, const at::Tensor& bucket_view);
+
+  // A bucket contains [1..N] gradients to be reduced, where the gradients
+  // have the same dtype and device.
+  // Coalescing gradients together before reducing can result in lower overhead
+  // and/or faster time to completion. Coalescing requires the constituent
+  // gradients to have the same dtype and device, and the resulting flattened
+  // tensor uses that common dtype and device. The flattened tensor is filled
+  // as the corresponding gradients are computed (triggered by autograd hooks),
+  // and the buckets are reduced in a predetermined order consistent across
+  // processes.
+  struct Bucket {
+    // Gradients of the bucket flattened into a 1-dimensional tensor
+    at::Tensor gradients;
+
+    // Views into the `gradients` tensor for each individual gradient
+    // Each view is created with layout (size and stride) matching the
+    // gradient's expected layout (see the "Gradient Layout Contract" in
+    // torch/csrc/autograd/functions/accumulate_grad.h).
+    // `bucket_views_in[i].copy_(grad)` and `grad.copy_(bucket_views_out[i])`
+    // provide convenient ways to copy gradient data in/out of `gradients`,
+    // respectively.
+    // We keep both `bucket_views_in` and `bucket_views_out` because
+    // registering a DDP communication hook may re-initialize
+    // `bucket_views_out` with the value of the hook's `future_work` but we
+    // still need separate views into the bucket's original flattened gradient
+    // to copy in gradient data.
+    std::vector<at::Tensor> bucket_views_in;
+    std::vector<at::Tensor> bucket_views_out;
+
+    // Variables whose gradients are held in this bucket
+    // We use refcounted tensors here so that we can easily unflatten the
+    // bucket's flattened `gradients` tensor into the participating variables
+    // after reduction has completed.
+    std::vector<at::Tensor> variables;
+
+    // Per-variable offset/length into the flattened `gradients` tensor and
+    // the corresponding `GradBucket` instance for communication hooks
+    std::vector<size_t> offsets;
+    std::vector<size_t> lengths;
+
+    // Per-variable sizes slicing into the bucket's `gradients` tensor
+    std::vector<c10::IntArrayRef> sizes_vec;
+
+    // Number of gradients left to be computed before the bucket is ready to
+    // be reduced
+    size_t pending;
+
+    // Global indices of participating variables in the bucket
+    std::vector<size_t> variable_indices;
+
+    // Future work handle for DDP communication hook
+    // If no hook is registered, a temporary vanilla allreduce hook is used.
+    c10::intrusive_ptr<at::ivalue::Future> future_work;
+
+    // If this bucket should expect a single sparse gradient
+    // If `true`, then this implies that `bucket.variables.size() == 1`.
+    bool expect_sparse_gradient = false;
+
+    // Sparse indices tensor
+    c10::optional<at::Tensor> sparse_tensor_indices = c10::nullopt;
+
+    // TODO(@pietern)
+    // Memory copies from gradient tensors into the bucket are potentially
+    // done on different CUDA streams. We record an event for every copy
+    // so that we can synchronize with them prior to kicking off the reduction.
+    // std::vector<at::cuda::CUDAEvent> events;
+  };
+
+  std::vector<Bucket> buckets_;
+
+  // A variable locator locates a particular variable in the reducer's buckets
+  struct VariableLocator {
+    // Index of the bucket containing the variable in the `buckets_` vector
+    size_t bucket_index;
+    // Index of the variable in the bucket, which may be used consistently
+    // across `bucket_views_in`, `bucket_views_out`, `variables`, `offsets`,
+    // `lengths`, `sizes_vec`, and `variable_indices` in `Bucket`
+    size_t intra_bucket_index;
+
+    VariableLocator() = default;
+
+    VariableLocator(size_t bucket_index_, size_t intra_bucket_index_)
+        : bucket_index(bucket_index_),
+          intra_bucket_index(intra_bucket_index_) {}
+  };
+
+  // Map the index of a variable to its location in the bucket structure.
+  std::vector<VariableLocator> variable_locators_;
+
+  // track the number of iterations to synchronize grads in training so far.
+  long num_iterations_;
+  // track distinct iteration of backward call. This is distinct from
+  // num_iterations_, for example in the case of multiple forward before
+  // backward.
+  long num_bwd_calls_;
+  // whether the first autograd hook for a distinct backward pass has been
+  // called.
+  bool first_autograd_hook_called_;
+  // track the number of buckets that have been ready for
+  // communication calls like allReduce or communication hooks.
+  int num_buckets_ready_;
+
+  // Timing information.
+  int64_t backward_compute_start_time_ = -1;
+  std::unique_ptr<Timer> timer_;
+
+  // We collect the relative timestamp of every gradient being ready
+  // when executing autograd. This can be used to derive a timeline of
+  // the point in time buckets were ready, or ideal bucket assignment/ordering.
+  std::vector<int64_t> backward_stats_;
+
+  bool should_collect_runtime_stats();
+  void record_forward_compute_start_time();
+  void record_backward_compute_start_time();
+  void record_backward_compute_end_time();
+  void record_backward_comm_start_time();
+  void record_backward_comm_end_time();
+
+  int get_ddp_runtime_logging_sample_rate();
+  int ddp_runtime_logging_sample_rate_ = kDDPRuntimeLoggingSampleRate;
+
+  bool is_multi_device_module_ = false;
+
+  // Following variables are to help build dynamic bucket order
+  bool has_rebuilt_bucket_;
+  std::vector<at::Tensor> rebuilt_params_;
+  std::vector<int64_t> rebuilt_param_indices_;
+  const int64_t bucket_bytes_cap_;
+
+#ifndef _WIN32
+  struct RpcContext {
+    using ContextPtr = torch::distributed::autograd::ContextPtr;
+    // The shared_ptr is to hold the context instance.
+    ContextPtr context_ptr_holder;
+    std::atomic<ContextPtr::element_type*> context_ptr{nullptr};
+
+    void set(ContextPtr&& new_context_ptr);
+  };
+  RpcContext rpc_context_;
+#endif
+
+  // A struct containing work handle and tensor for allreduce scheduled in
+  // forward pass, if applicable.
+  struct ForwardPassAllreduceWork {
+    c10::intrusive_ptr<c10d::Work> workHandle;
+    at::Tensor resultTensor;
+    // whether we should divide by the initial world_size or the no. of
+    // remaining DDP ranks.
+    bool useStaticWorldSize;
+  };
+
+  // Handle for the currently scheduled allreduce in the forward pass, if
+  // applicable.
+  ForwardPassAllreduceWork forwardPassWorkHandle_;
+
+  // Division factor for reduction of gradients.
+  // Equal to the process group size, with an exception of handling uneven
+  // input.
+  int div_factor_;
+
+  bool static_graph_;
+
+  // Key: size_t (index), Value: the number of times that a variable's
+  // autograd_hook() should be triggered before marking this variable's grad as
+  // ready for communication. Map will not change after 1st iteration.
+  std::unordered_map<size_t, int> numGradHooksTriggeredMap_;
+  // Key: size_t (index), Value: the number of times that a variable's
+  // autograd_hook() are left to be triggered before marking this variable's
+  // grad as ready for communication. Map will change after 1st iteration to
+  // track a grad is ready for communication or not.
+  std::unordered_map<size_t, int> numGradHooksTriggeredMapPerIteration_;
+
+ private:
+  // reset counting for buckets before backward starts
+  void reset_bucket_counting();
+  // search unused parameters beore backward starts
+  void search_unused_parameters(
+      const std::vector<torch::autograd::Variable>& outputs);
+  void set_divide_factor();
+  // kick off all reduce for the ready bucket
+  void all_reduce_bucket(Bucket& bucket);
+  // kick off all reduce to local used map, it can help find global unused
+  // parameters
+  void all_reduce_local_used_map();
+  // initialize locally used parameter maps
+  void initialize_local_used_map();
+  // get current cuda stream
+  const c10::Stream get_current_stream();
+  bool dynamic_graph_find_unused();
+  bool static_graph_first_iteration();
+  bool static_graph_after_first_iteration();
+
+  // comm_hook_ is used to access the DDP communication hook if registered.
+  std::unique_ptr<CommHookInterface> comm_hook_;
+
+  // Sparse metadata contains the indices that will be used
+  // when calling into sparse allreduce.
+  // This is only used in the sparse allreduce collective calls
+  std::unique_ptr<std::map<std::string, at::Tensor>> sparse_metadata_;
+
+  // Debug level setting. It is parsed once when Reducer is constructed, and
+  // remains the same across a single invocation of DDP training.
+  DebugLevel ddp_debug_level_;
+  // Mapping of variable index to fully qualified name of model to notify users
+  // about errors when certain parameters do not get gradient.
+  std::unordered_map<size_t, std::string> param_names_;
+  // Variable indices stored sequentially in order of when the gradient is ready
+  // for the current backwards pass.
+  std::vector<int> grad_ready_order_indices_;
+  // Bytes capacity of first bucket, can be configured by user
+  int64_t first_bucket_bytes_cap_;
+  // Per iteration set of parameter indices that have been marked ready.
+  std::unordered_set<size_t> perIterationReadyParams_;
+  // Retrieves parameter names that have not been marked as ready as part of
+  // previous iteration.
+  std::vector<std::string> getUnmarkedParamsForIteration();
+  // Retrieves parameter indices that have not been marked as ready as part of
+  // previous iteration.
+  std::vector<size_t> getUnmarkedParamIndicesForIteration();
+  // Raises appropriate error if mark_variable_ready is called on the same
+  // variable twice, which is unexpected.
+  void checkAndRaiseMarkedTwiceError(size_t curVariableIndex);
+  // Retrieves parameter corresponding to the given VariableIndex.
+  at::Tensor& get_param_from_index(size_t index);
+
+  // Cached bucket index to model parameter mapping. Populated after buckets
+  // are rebuilt after which this mapping is static.
+  mutable std::unordered_map<size_t, std::vector<at::Tensor>>
+      cached_variables_for_bucket_;
+
+  bool optim_in_backward_{false};
+  friend class Logger;
+};
+
+// This is equivalent to take_tensors but returns indices into the
+// tensor list argument for bucket assignment. Also, it is aware
+// of device placement and will not allow buckets to span devices.
+// The index of tensors[i] assigned to bucket is tensor_indices[i],
+// when tensor_indices is empty, the index of tensors[i] assigned to
+// bucket is i.
+TORCH_API std::tuple<std::vector<std::vector<size_t>>, std::vector<size_t>>
+compute_bucket_assignment_by_size(
+    const std::vector<at::Tensor>& tensors,
+    const std::vector<size_t>& bucket_size,
+    const std::vector<bool>& expect_sparse_gradient = {},
+    const std::vector<int64_t>& tensor_indices = {},
+    const c10::optional<std::weak_ptr<c10d::Logger>>& logger = {});
+
+// Verify models across all processes are the same as model on rank 0 with
+// respect to no. of params and matching dtype/size/layout.
+TORCH_API void verify_params_across_processes(
+    const c10::intrusive_ptr<c10d::ProcessGroup>& process_group,
+    const std::vector<at::Tensor>& params,
+    const c10::optional<std::weak_ptr<c10d::Logger>>& logger);
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/reducer_timer.hpp

@@ -0,0 +1,81 @@
+#pragma once
+#include <c10/util/ApproximateClock.h>
+#include <torch/csrc/autograd/profiler.h>
+
+namespace c10d {
+constexpr int kUnsetTime = -1;
+
+inline int64_t current_time_in_nanos() {
+  return c10::getTime();
+}
+
+class TORCH_API Timer {
+ private:
+  // The timestamp of forward call start time in each iteration.
+  int64_t forward_start_time = kUnsetTime;
+  // The timestamp of backward computation start and end time in each
+  // iteration.
+  int64_t backward_compute_start_time = kUnsetTime;
+  int64_t backward_compute_end_time = kUnsetTime;
+  // The timestamp of first communication call start time in each iteration.
+  int64_t backward_comm_start_time = kUnsetTime;
+  // The timestamp of last communication call end time in each iteration.
+  int64_t backward_comm_end_time = kUnsetTime;
+
+ public:
+  enum class Event {
+    kForwardStart,
+    kBackwardComputeStart,
+    kBackwardComputeEnd,
+    kBackwardCommStart,
+    kBackwardCommEnd,
+  };
+
+  // Record the current event, i.e., mark it as having occurred now. Default
+  // CPU implementation.
+  virtual void record(Event event) {
+    getTimeRef(event) = current_time_in_nanos();
+  }
+
+  // Return the difference between when two events occurred, in nanoseconds.
+  // Or nullopt if one of them hasn't been recorded.
+  virtual c10::optional<int64_t> measureDifference(Event start, Event end) = 0;
+
+  virtual ~Timer() = default;
+
+  // Return host-side timestamp, or nullopt if it has not yet been recorded.
+  c10::optional<int64_t> getTimestamp(Event event) {
+    auto time = getTimeRef(event);
+    if (time == kUnsetTime) {
+      return c10::nullopt;
+    } else {
+      return time;
+    }
+  }
+
+  // Return host-side time member variable corresponding to the given event.
+  int64_t& getTimeRef(Event event) {
+    switch (event) {
+      case Event::kForwardStart:
+        return forward_start_time;
+      case Event::kBackwardComputeStart:
+        return backward_compute_start_time;
+      case Event::kBackwardComputeEnd:
+        return backward_compute_end_time;
+      case Event::kBackwardCommStart:
+        return backward_comm_start_time;
+      case Event::kBackwardCommEnd:
+        return backward_comm_end_time;
+      default:
+        TORCH_INTERNAL_ASSERT(false);
+    }
+  }
+};
+
+TORCH_DECLARE_TYPED_REGISTRY(
+    TimerRegistry,
+    c10::DeviceType,
+    Timer,
+    std::unique_ptr,
+    c10::Device);
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/sequence_num.hpp

@@ -0,0 +1,65 @@
+#pragma once
+
+#include <c10/macros/Macros.h>
+#include <c10/util/Optional.h>
+#include <c10/util/irange.h>
+#include <vector>
+
+namespace c10d {
+const int kUnsetSeqNum = 0;
+
+namespace {
+constexpr int kByteOffset = 8;
+}
+
+// Converts from int to char vec to write in store
+template <typename T>
+inline std::vector<T> toVec(uint64_t num, int numBytes) {
+  std::vector<T> values;
+  // Read off bytes from right to left, pushing them into
+  // char array.
+  for (const auto i : c10::irange(numBytes)) {
+    uint8_t x = (num >> (kByteOffset * i)) & 0xff;
+    values.push_back(static_cast<T>(x));
+  }
+  return values;
+}
+
+// Converts from char vec (such as from store read) to int.
+template <typename T>
+inline uint64_t fromVec(const std::vector<T>& values) {
+  uint64_t num = 0;
+  // Set each byte at the correct location on num
+  for (const auto i : c10::irange(values.size())) {
+    uint8_t x = static_cast<uint8_t>(values[i]);
+    num |= (static_cast<int64_t>(x) << (kByteOffset * i));
+  }
+  return num;
+}
+
+class TORCH_API SequenceNum {
+ public:
+  SequenceNum();
+  explicit SequenceNum(const uint64_t num);
+  // Retrieve num_. Will throw if not set.
+  uint64_t get() const;
+  // Increment num_. Will throw if not set.
+  void increment();
+  // Increment num_ and return the old value. Will throw if not set.
+  uint64_t getAndIncrement();
+  // Sets num_
+  void set(const uint64_t num);
+  // Returns true if this SequenceNum is properly initialized with a value, else
+  // false.
+  bool isSet() const;
+
+  SequenceNum& operator=(const SequenceNum& other);
+
+  SequenceNum(const SequenceNum& other);
+
+ private:
+  c10::optional<uint64_t> num_;
+  mutable std::mutex lock_;
+};
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/c10d/socket.h

@@ -0,0 +1,93 @@
+// Copyright (c) Meta Platforms, Inc. and its affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the BSD-style license found in the
+// LICENSE file in the root directory of this source tree.
+
+#pragma once
+
+#include <chrono>
+#include <cstdint>
+#include <memory>
+#include <string>
+
+#include <c10/macros/Macros.h>
+#include <c10/util/Exception.h>
+#include <torch/csrc/distributed/c10d/exception.h>
+
+namespace c10d {
+namespace detail {
+
+class SocketOptions {
+ public:
+  SocketOptions& prefer_ipv6(bool value) noexcept {
+    prefer_ipv6_ = value;
+
+    return *this;
+  }
+
+  bool prefer_ipv6() const noexcept {
+    return prefer_ipv6_;
+  }
+
+  SocketOptions& connect_timeout(std::chrono::seconds value) noexcept {
+    connect_timeout_ = value;
+
+    return *this;
+  }
+
+  std::chrono::seconds connect_timeout() const noexcept {
+    return connect_timeout_;
+  }
+
+ private:
+  bool prefer_ipv6_ = true;
+  std::chrono::seconds connect_timeout_{30};
+};
+
+class SocketImpl;
+
+class Socket {
+ public:
+  // This function initializes the underlying socket library and must be called
+  // before any other socket function.
+  static void initialize();
+
+  static Socket listen(std::uint16_t port, const SocketOptions& opts = {});
+
+  static Socket listenFromFd(int fd, std::uint16_t expected_port);
+
+  static Socket connect(
+      const std::string& host,
+      std::uint16_t port,
+      const SocketOptions& opts = {});
+
+  Socket() noexcept = default;
+
+  Socket(const Socket& other) = delete;
+
+  Socket& operator=(const Socket& other) = delete;
+
+  Socket(Socket&& other) noexcept;
+
+  Socket& operator=(Socket&& other) noexcept;
+
+  ~Socket();
+
+  Socket accept() const;
+
+  int handle() const noexcept;
+
+  std::uint16_t port() const;
+
+  bool waitForInput(std::chrono::milliseconds timeout);
+
+ private:
+  explicit Socket(std::unique_ptr<SocketImpl>&& impl) noexcept;
+
+  std::unique_ptr<SocketImpl> impl_;
+};
+
+} // namespace detail
+
+} // namespace c10d

venv/lib/python3.10/site-packages/torch/include/torch/csrc/distributed/rpc/agent_utils.h

@@ -0,0 +1,46 @@
+#pragma once
+
+#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
+#include <torch/csrc/distributed/rpc/utils.h>
+
+namespace torch {
+namespace distributed {
+namespace rpc {
+
+// All RPC peers should call into this function at the same time. Each peer
+// provides its own id and name, and this function uses the given Store to
+// gather global name-to-id mapping on all peers.
+TORCH_API std::unordered_map<std::string, worker_id_t> collectNames(
+    ::c10d::PrefixStore store,
+    const worker_id_t selfId,
+    const std::string& selfName,
+    const int worldSize);
+
+// Ranks in dynamic RPC groups will initially call into this to establish the
+// name-to-id mapping for the current peers in the group. The current rank will
+// put its own worker info in the store and discover all the ranks that came
+// before it. NOTE: This needs to be called with the Dynamic RPC group
+// membership management token held.
+TORCH_API std::unordered_map<std::string, worker_id_t> collectCurrentNames(
+    ::c10d::PrefixStore store,
+    const worker_id_t selfId,
+    const std::string& selfName);
+
+// Remove name frmo Store, used in dynamic RPC groups.
+// NOTE: This needs to be called with the Dynamic RPC group
+// membership management token held.
+TORCH_API void removeCurrentName(
+    ::c10d::PrefixStore store,
+    const worker_id_t selfId,
+    const std::string& selfName);
+
+// This performs a synchronization of all call counts by using store.
+// All RPC peers wait for others to join to exit at the same time.
+TORCH_API int syncCallCount(
+    ::c10d::PrefixStore store,
+    const int worldSize,
+    int activeCalls = 0);
+
+} // namespace rpc
+} // namespace distributed
+} // namespace torch