docker/bloom13b/Model-References/MLPERF3.1/Training/benchmarks/gpt3/megatron/initialize.py

# coding=utf-8
# Copyright (c) 2023 Habana Labs, Ltd. an Intel Company.
# Copyright (c) 2020, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Megatron initialization."""

import random
import os
import time
import shutil

import numpy as np
import torch

from megatron import fused_kernels
from megatron import get_adlr_autoresume
from megatron import get_args
from megatron import get_tensorboard_writer
from megatron import mpu
from megatron.global_vars import set_global_variables
from megatron.mpu import (set_tensor_model_parallel_rank,
                          set_tensor_model_parallel_world_size,
                          _set_global_memory_buffer)

import deepspeed
import deepspeed.utils.groups as groups

from mlperf_common.logging import MLLoggerWrapper
from mlperf_common.frameworks.pyt import PyTCommunicationHandler
from mlperf_logging import mllog

def initialize_megatron(extra_args_provider=None, args_defaults={},
                        ignore_unknown_args=False, allow_no_cuda=False):
    """Set global variables, initialize distributed, and
    set autoresume and random seeds.
    `allow_no_cuda` should not be set unless using megatron for cpu only
    data processing. In general this arg should not be set unless you know
    what you are doing.
    Returns a function to finalize distributed env initialization
    (optionally, only when args.lazy_mpu_init == True)
    """

    # Parse args, build tokenizer, and set adlr-autoresume,
    # tensorboard-writer, and timers.
    set_global_variables(extra_args_provider=extra_args_provider,
                         args_defaults=args_defaults,
                         ignore_unknown_args=ignore_unknown_args)

    args = get_args()
    if args.rank == 0:
        mllog.config(filename=args.mllog_output_path)
    mllogger = MLLoggerWrapper(PyTCommunicationHandler())
    mllogger.start(key=mllogger.constants.INIT_START, sync=False)

    if os.getenv("P2P_DUMMY_MODE_PHASE") != "2":
        if args.world_size != 1 and os.getenv("HLS_MODULE_ID") is None:
            if args.local_rank == None:
                print("Non-existent HLS_MODULE_ID provided: Setting env var HLS_MODULE_ID=0")
                os.environ["HLS_MODULE_ID"] = "0"
            else:
                print("Non-existent HLS_MODULE_ID provided: Setting env var HLS_MODULE_ID=", args.local_rank)
                os.environ["HLS_MODULE_ID"] = str(args.local_rank)

    # profiler config, must be done before hpu initialization
    if args.profile is not None:
        os.environ['HABANA_PROFILE'] = 'profile_api_with_nics'
        shutil.rmtree('.graph_dumps', ignore_errors=True)


    # torch.distributed initialization
    def finish_mpu_init():
        args = get_args()
        # Pytorch distributed.
        _initialize_distributed()

        # Random seeds for reproducibility.
        if args.rank == 0:
            print('> setting random seeds to {} ...'.format(args.seed))
        _set_random_seed(args.seed)

    if  args.lazy_mpu_init:
        args.use_cpu_initialization=True
        # delayed initialization of DDP-related stuff
        # We only set basic DDP globals
        set_tensor_model_parallel_world_size(args.tensor_model_parallel_size)
        # and return function for external DDP manager
        # to call when it has DDP initialized
        set_tensor_model_parallel_rank(args.rank)
        return finish_mpu_init, mllogger
    else:
        # Megatron's MPU is the master. Complete initialization right away.
        finish_mpu_init()

        # Initialize memory buffers.
        _initialize_mem_buffs()

        # Initialize global memory buffer (used with sequence parallel)
        _set_global_memory_buffer()

        # Autoresume.
        _init_autoresume()

        # Compile dependencies.
        _compile_dependencies()

        # No continuation function
        return None, mllogger


def _compile_dependencies():

    args = get_args()

    # =========================
    # Compile dataset C++ code.
    # =========================
    # TODO: move this to ninja
    if args.local_rank == 0:
        start_time = time.time()
        print('> compiling dataset index builder ...')
        from megatron.data.dataset_utils import compile_helper
        compile_helper()
        print('>>> done with dataset index builder. Compilation time: {:.3f} '
              'seconds'.format(time.time() - start_time), flush=True)

    # ==================
    # Load fused kernels
    # ==================

    # Custom kernel constraints check.
    seq_len = args.seq_length
    attn_batch_size = \
        (args.num_attention_heads / args.tensor_model_parallel_size) * \
        args.micro_batch_size
    # Constraints on sequence length and attn_batch_size to enable warp based
    # optimization and upper triangular optimization (for causal mask)
    custom_kernel_constraint = seq_len > 16 and seq_len <=2048 and \
        seq_len % 4 == 0 and attn_batch_size % 4 == 0
    # Print a warning.
    if not ((args.fp16 or args.bf16) and
            custom_kernel_constraint and
            args.masked_softmax_fusion):
        if args.rank == 0:
            print('WARNING: constraints for invoking optimized'
                  ' fused softmax kernel are not met. We default'
                  ' back to unfused kernel invocations.', flush=True)

    # Always build on rank zero first.
    if _is_rank_0():
        start_time = time.time()
        print('> compiling and loading fused kernels ...', flush=True)
        if args.device.type == 'cuda':
            fused_kernels.load(args)
        torch.distributed.barrier()
    else:
        torch.distributed.barrier()
        if args.device.type == 'cuda':
            fused_kernels.load(args)
    # Simple barrier to make sure all ranks have passed the
    # compilation phase successfully before moving on to the
    # rest of the program. We think this might ensure that
    # the lock is released.
    torch.distributed.barrier()
    if _is_rank_0():
        print('>>> done with compiling and loading fused kernels. '
              'Compilation time: {:.3f} seconds'.format(
                  time.time() - start_time), flush=True)


def setup_deepspeed_random_and_activation_checkpointing(args):
    '''Optional DeepSpeed Activation Checkpointing features.
    Gives access to partition activations, contiguous memory optimizations
    and cpu checkpointing.
    Activation checkpoint requires keep track of the random states
    and setting the random seed for each MP process. Megatron uses
    mpu.get_cuda_rng_tracker and mpu.model_parallel_cuda_manual_seed
    for keeping track of the random states and setting the random seeds.
    Since they are used in places outside of activation checkpointing,
    we overwrite them to maintain consistency.
    This must be called before all the calls to mpu.model_parallel_cuda_manual_seed
    '''
    num_layers = args.num_layers // args.checkpoint_num_layers
    num_layers = num_layers if args.num_layers % args.checkpoint_num_layers == 0 else num_layers + 1
    if args.split_transformers:
        num_layers *= 2

    deepspeed.checkpointing.configure(
        mpu,
        partition_activations=args.partition_activations,
        contiguous_checkpointing=args.contigious_checkpointing,
        num_checkpoints=num_layers,
        checkpoint_in_cpu=args.checkpoint_in_cpu,
        synchronize=args.synchronize_each_layer,
        profile=args.profile_backward)

    mpu.checkpoint = deepspeed.checkpointing.checkpoint
    mpu.get_cuda_rng_tracker = deepspeed.checkpointing.get_cuda_rng_tracker
    mpu.model_parallel_cuda_manual_seed = deepspeed.checkpointing.model_parallel_cuda_manual_seed

def update_wa_env_var(key, value):
    if key not in os.environ.keys():
        os.environ[key] = value

def _initialize_distributed():
    """Initialize torch.distributed and mpu."""
    args = get_args()
    if torch.distributed.is_initialized():

        if args.rank == 0:
            print('torch distributed is already initialized, '
                  'skipping initialization ...', flush=True)
        args.rank = torch.distributed.get_rank()
        args.world_size = torch.distributed.get_world_size()

    else:
        print("_initialize_distributed: Initializing with below params:")
        print("args.local_rank:", args.local_rank)
        print("args.world_size:", args.world_size)
        print("args.rank:", args.rank)
        # TODO SW-65249 need to align behavior between device types
        device_count = None
        print("args.distributed_backend:", args.distributed_backend)
        if args.distributed_backend == 'hccl':
            import habana_frameworks.torch as htcore
            device_count = htcore.hpu.device_count()
            if args.hpu_deterministic:
                assert args.use_hpu, f"--hpu-deterministic supported only with --use-hpu flag"
                htcore.hpu.setDeterministic(True)
            print("hccl device_count: ", device_count)
        elif args.distributed_backend == 'nccl':
            device_count = torch.cuda.device_count()
        elif args.distributed_backend == 'gloo':
            # no limit of devices when working on CPU, setting 8.
            device_count = int(os.getenv('GPUS_PER_NODE', '8'))
        else:
            assert False, f"Unsupported backend {args.distributed_backend}"

        # Manually set the device ids.
        if device_count > 0:
            device = args.rank % device_count
            if args.local_rank is not None:
                assert args.local_rank == device, \
                    'expected local-rank to be the same as rank % device-count.'
            else:
                args.local_rank = device
        else:
            assert False, "Error: device_count is not positive"

        if args.distributed_backend == 'hccl':
            device = torch.device('hpu')
        elif args.distributed_backend == 'nccl':
            torch.cuda.set_device(device)
            device = torch.device('cuda')
        elif args.distributed_backend == 'gloo':
            device = torch.device('cpu')
        else:
            assert False, f"Unsupported backend {args.distributed_backend}"

        args.device = device

        if args.rank == 0:
            print('> initializing torch distributed ...', flush=True)

        # Call the init process
        init_method = 'tcp://'
        master_ip = os.getenv('MASTER_ADDR', 'localhost')
        master_port = os.getenv('MASTER_PORT', '6000')
        init_method += master_ip + ':' + master_port

        if args.distributed_backend == "hccl":
            import habana_frameworks.torch.core
        if args.deepspeed or args.ds_inference:
            deepspeed.init_distributed(dist_backend=args.distributed_backend)
        else:
            torch.distributed.init_process_group(
                backend=args.distributed_backend,
                world_size=args.world_size, rank=args.rank,
                init_method=init_method)
    # Set the tensor model-parallel, pipeline model-parallel, and
    # data-parallel communicators.
    if device_count > 0:
        if mpu.model_parallel_is_initialized():
            print('model parallel is already initialized')
        else:
            mpu.initialize_model_parallel(args.tensor_model_parallel_size,
                                          args.pipeline_model_parallel_size,
                                          args.virtual_pipeline_model_parallel_size)

    if args.deepspeed and args.deepspeed_activation_checkpointing:
        setup_deepspeed_random_and_activation_checkpointing(args)


def _init_autoresume():
    """Set autoresume start time."""
    autoresume = get_adlr_autoresume()
    if autoresume:
        torch.distributed.barrier()
        autoresume.init()
        torch.distributed.barrier()


def _set_random_seed(seed_):
    """Set random seed for reproducability."""
    if seed_ is not None and seed_ > 0:
        # Ensure that different pipeline MP stages get different seeds.
        seed = seed_ + (100 * mpu.get_pipeline_model_parallel_rank())
        random.seed(seed)
        np.random.seed(seed)
        torch.manual_seed(seed)
        if ((get_args().device.type == "cuda" and torch.cuda.device_count() > 0) or
            (get_args().device.type == "hpu" and torch.hpu.device_count() > 0)):
                mpu.model_parallel_cuda_manual_seed(seed)
    else:
        raise ValueError('Seed ({}) should be a positive integer.'.format(seed))


def write_args_to_tensorboard():
    """Write arguments to tensorboard."""
    args = get_args()
    writer = get_tensorboard_writer()
    if writer:
        for arg in vars(args):
            writer.add_text(arg, str(getattr(args, arg)),
                            global_step=args.iteration)


def _initialize_mem_buffs():
    """Initialize manually allocated static memory."""
    args = get_args()

    # Initialize memory for checkpointed activations.
    if args.distribute_checkpointed_activations:
        mpu.init_checkpointed_activations_memory_buffer()

def _is_rank_0():
    """Check whether it is rank 0. For AML, check if it is rank 0 of a node"""
    if torch.distributed.is_initialized():
        if torch.distributed.get_rank() == 0 or (
            'AZUREML_EXPERIMENT_ID' in os.environ and torch.distributed.get_rank() % torch.cuda.device_count() == 0
            ):
            return True
        else:
            return False
    else:
        return True