sedimentology

.gitattributes

@@ -1,35 +1,6 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+optimizer.pt filter=lfs diff=lfs merge=lfs -text
+pytorch_model.bin filter=lfs diff=lfs merge=lfs -text
+ice_text.model filter=lfs diff=lfs merge=lfs -text
+rng_state.pth filter=lfs diff=lfs merge=lfs -text
+scheduler.pt filter=lfs diff=lfs merge=lfs -text
+training_args.bin filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,55 @@
+# GeoLLM
+**Large Language Model for Geology**
+
+Large language models are used to organize geology-related knowledge (geology, geophysics, geophysical logging, etc.). This version uses the [ChatGLM-6B](https://github.com/THUDM/ChatGLM-6B) base model and fine-tunes it using P-tuning.
+
 ---
-license: gpl-3.0
+
+### Sedimentology
+
+Sedimentology, the study of sedimentary rocks and the processes by which they are formed, includes and is related to a large number of phenomena. Sedimentology includes the five fundamental processes defined by the term sediaentation --weathering, erosion, transportation, deposition and diagenesis.
+
+**Datasets：**《沉积岩石学（第四版）》 朱筱敏
+
+**Model：** ChatGLM-6B
+
+**Fine-tuning：** P-Tuning v2
+
+**Before fine-tuning**
+
+```
+response, history = model.chat(tokenizer, "什么是沉积岩石学？", history=[])
+response
+
+沉积岩石学是一门研究沉积岩的形成、结构和成分的学科，主要关注地球表面上不同条件下的沉积过程和岩石形成机制，包括岩浆沉积、冰川沉积、洪水沉积、海洋沉积等。沉积岩石学在地质学、地球物理学、地球化学、材料科学等领域都有广泛应用，因为沉积岩是许多自然和人工地质工程的基础，如地质勘探、矿产资源开发、土木工程、环境科学等。沉积岩石学的研究对象包括沉积岩的地质特征、成分和构造特征，以及沉积岩与地壳、岩浆和变质岩的关系。研究方法包括沉积岩分析、岩相学分析、岩浆动力学分析等。
+```
+
+**After fine-tuning**
+
+```
+response, history = model.chat(tokenizer, "什么是沉积岩石学？", history=[])
+response
+
+沉积岩石学是研究沉积岩的物质成分、结构构造、岩石类型、沉积物沉积作用和沉积物质形成环境以及沉积岩分布规律的一门科学。
+```
+
+**Error Analysis:** We meticulously refined the model by approximately 500 entries from academic textbooks, subsequently applying P-Tuning v2 for optimization. Detailed control of parameters was not conducted for the time being. Given the scarcity of the training data and the fine-tuning parameters, the outcomes might exhibit some irregularities.
+
+**Results Analysis:** It is evident that the fine-tuned model shows enhanced reliability(more precise and concise) when providing answers within specialized knowledge domains. Moving forward, we will persist in enriching our training data and optimizing our fine-tuning methodologies in order to yield superior results.
+
 ---
+
+### TODO
+
+1. Geophysical Exploration
+
+2. Geophysical logging
+
+3. Petroleum Geology
+
+etc...
+
+---
+
+### Related Resources
+1. [ChatGLM-6B](https://github.com/THUDM/ChatGLM-6B): ChatGLM-6B is an open bilingual language model based on General Language Model (GLM) framework, with 6.2 billion parameters.

config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "/content/drive/MyDrive/ChatGLM/chatglm-6b-slim",
+  "architectures": [
+    "ChatGLMForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_chatglm.ChatGLMConfig",
+    "AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration",
+    "AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration"
+  },
+  "bos_token_id": 130004,
+  "eos_token_id": 130005,
+  "hidden_size": 4096,
+  "inner_hidden_size": 16384,
+  "layernorm_epsilon": 1e-05,
+  "max_sequence_length": 2048,
+  "model_type": "chatglm",
+  "num_attention_heads": 32,
+  "num_layers": 28,
+  "pad_token_id": 3,
+  "position_encoding_2d": true,
+  "pre_seq_len": 128,
+  "prefix_projection": false,
+  "quantization_bit": 4,
+  "torch_dtype": "float16",
+  "transformers_version": "4.27.1",
+  "use_cache": true,
+  "vocab_size": 130528
+}

configuration_chatglm.py

@@ -0,0 +1,100 @@
+""" ChatGLM model configuration """
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class ChatGLMConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`~ChatGLMModel`].
+    It is used to instantiate an ChatGLM model according to the specified arguments, defining the model
+    architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
+    the ChatGLM-6B [THUDM/ChatGLM-6B](https://huggingface.co/THUDM/chatglm-6b) architecture.
+    We remove 20K image tokens on top of ChatGLM-6B to save memories.
+
+    Configuration objects inherit from  [`PretrainedConfig`] and can be used
+    to control the model outputs. Read the documentation from  [`PretrainedConfig`]
+    for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 150528):
+            Vocabulary size of the ChatGLM-6B model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`~ChatGLMModel`] or
+            [`~TFChatGLMModel`].
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the encoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 28):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        inner_hidden_size (`int`, *optional*, defaults to 16384):
+            Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        max_sequence_length (`int`, *optional*, defaults to 512):
+            The maximum sequence length that this model might ever be used with.
+            Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
+        layernorm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon used by the layer normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether the model should return the last key/values attentions (not used by all models).
+        Example:
+
+    ```python
+    >>> from configuration_chatglm import ChatGLMConfig
+    >>> from modeling_chatglm import ChatGLMModel
+
+    >>> # Initializing a ChatGLM-6B THUDM/ChatGLM-6B style configuration
+    >>> configuration = ChatGLMConfig()
+
+    >>> # Initializing a model from the THUDM/ChatGLM-6B style configuration
+    >>> model = ChatGLMModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```
+"""
+    model_type = "chatglm"
+
+    def __init__(
+            self,
+            vocab_size=130528,
+            hidden_size=4096,
+            num_layers=28,
+            num_attention_heads=32,
+            layernorm_epsilon=1e-5,
+            use_cache=False,
+            bos_token_id=130004,
+            eos_token_id=130005,
+            pad_token_id=0,
+            max_sequence_length=2048,
+            inner_hidden_size=16384,
+            position_encoding_2d=True,
+            quantization_bit=0,
+            pre_seq_len=None,
+            prefix_projection=False,
+            **kwargs
+    ):
+        self.num_layers = num_layers
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_attention_heads = num_attention_heads
+        self.max_sequence_length = max_sequence_length
+        self.layernorm_epsilon = layernorm_epsilon
+        self.inner_hidden_size = inner_hidden_size
+        self.use_cache = use_cache
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.position_encoding_2d = position_encoding_2d
+        self.quantization_bit = quantization_bit
+        self.pre_seq_len = pre_seq_len
+        self.prefix_projection = prefix_projection
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            **kwargs
+        )

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 130004,
+  "eos_token_id": 130005,
+  "pad_token_id": 3,
+  "transformers_version": "4.27.1"
+}

ice_text.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99871e0c85db81ad7af1028854fd091cd5778c8414ae9d94bbbc10d02c831c21
+size 2699926

modeling_chatglm.py

@@ -0,0 +1,1406 @@
+""" PyTorch ChatGLM model. """
+
+import math
+import copy
+import os
+import warnings
+import re
+import sys
+
+import torch
+import torch.utils.checkpoint
+import torch.nn.functional as F
+from torch import nn
+from torch.nn import CrossEntropyLoss, LayerNorm
+from torch.nn.utils import skip_init
+from typing import Optional, Tuple, Union, List, Callable, Dict, Any
+
+from transformers.utils import (
+    add_code_sample_docstrings,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+)
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+    BaseModelOutputWithPastAndCrossAttentions,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+from transformers.generation.logits_process import LogitsProcessor
+from transformers.generation.utils import LogitsProcessorList, StoppingCriteriaList, GenerationConfig, ModelOutput
+
+from .configuration_chatglm import ChatGLMConfig
+
+# flags required to enable jit fusion kernels
+
+if sys.platform != 'darwin':
+    torch._C._jit_set_profiling_mode(False)
+    torch._C._jit_set_profiling_executor(False)
+    torch._C._jit_override_can_fuse_on_cpu(True)
+    torch._C._jit_override_can_fuse_on_gpu(True)
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "silver/ChatGLM-6B"
+_CONFIG_FOR_DOC = "ChatGLM6BConfig"
+
+CHATGLM_6B_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "silver/chatglm-6b-slim",
+    # See all ChatGLM-6B models at https://huggingface.co/models?filter=chatglm
+    # See the slim model at https://huggingface.co/silver/chatglm-6b-slim
+]
+
+
+class InvalidScoreLogitsProcessor(LogitsProcessor):
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
+        if torch.isnan(scores).any() or torch.isinf(scores).any():
+            scores.zero_()
+            scores[..., 5] = 5e4
+        return scores
+
+
+def load_tf_weights_in_chatglm_6b(model, config, tf_checkpoint_path):
+    """Load tf checkpoints in a pytorch model."""
+    try:
+        import re
+
+        import numpy as np
+        import tensorflow as tf
+    except ImportError:
+        logger.error(
+            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
+            "https://www.tensorflow.org/install/ for installation instructions."
+        )
+        raise
+    tf_path = os.path.abspath(tf_checkpoint_path)
+    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
+    # Load weights from TF model
+    init_vars = tf.train.list_variables(tf_path)
+    names = []
+    arrays = []
+    for name, shape in init_vars:
+        logger.info(f"Loading TF weight {name} with shape {shape}")
+        array = tf.train.load_variable(tf_path, name)
+        names.append(name)
+        arrays.append(array)
+
+    for name, array in zip(names, arrays):
+        name = name.split("/")
+        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
+        # which are not required for using pretrained model
+        if any(
+                n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
+                for n in name
+        ):
+            logger.info(f"Skipping {'/'.join(name)}")
+            continue
+        pointer = model
+        for m_name in name:
+            if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
+                scope_names = re.split(r"_(\d+)", m_name)
+            else:
+                scope_names = [m_name]
+            if scope_names[0] == "kernel" or scope_names[0] == "gamma":
+                pointer = getattr(pointer, "weight")
+            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
+                pointer = getattr(pointer, "bias")
+            elif scope_names[0] == "output_weights":
+                pointer = getattr(pointer, "weight")
+            elif scope_names[0] == "squad":
+                pointer = getattr(pointer, "classifier")
+            else:
+                try:
+                    pointer = getattr(pointer, scope_names[0])
+                except AttributeError:
+                    logger.info(f"Skipping {'/'.join(name)}")
+                    continue
+            if len(scope_names) >= 2:
+                num = int(scope_names[1])
+                pointer = pointer[num]
+        if m_name[-11:] == "_embeddings":
+            pointer = getattr(pointer, "weight")
+        elif m_name == "kernel":
+            array = np.transpose(array)
+        try:
+            assert (
+                    pointer.shape == array.shape
+            ), f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
+        except AssertionError as e:
+            e.args += (pointer.shape, array.shape)
+            raise
+        logger.info(f"Initialize PyTorch weight {name}")
+        pointer.data = torch.from_numpy(array)
+    return model
+
+
+class PrefixEncoder(torch.nn.Module):
+    """
+    The torch.nn model to encode the prefix
+    Input shape: (batch-size, prefix-length)
+    Output shape: (batch-size, prefix-length, 2*layers*hidden)
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.prefix_projection = config.prefix_projection
+        if self.prefix_projection:
+            # Use a two-layer MLP to encode the prefix
+            self.embedding = torch.nn.Embedding(config.pre_seq_len, config.hidden_size)
+            self.trans = torch.nn.Sequential(
+                torch.nn.Linear(config.hidden_size, config.hidden_size),
+                torch.nn.Tanh(),
+                torch.nn.Linear(config.hidden_size, config.num_layers * config.hidden_size * 2)
+            )
+        else:
+            self.embedding = torch.nn.Embedding(config.pre_seq_len, config.num_layers * config.hidden_size * 2)
+
+    def forward(self, prefix: torch.Tensor):
+        if self.prefix_projection:
+            prefix_tokens = self.embedding(prefix)
+            past_key_values = self.trans(prefix_tokens)
+        else:
+            past_key_values = self.embedding(prefix)
+        return past_key_values
+
+
+@torch.jit.script
+def gelu_impl(x):
+    """OpenAI's gelu implementation."""
+    return 0.5 * x * (1.0 + torch.tanh(0.7978845608028654 * x *
+                                       (1.0 + 0.044715 * x * x)))
+
+
+def gelu(x):
+    return gelu_impl(x)
+
+
+class RotaryEmbedding(torch.nn.Module):
+    def __init__(self, dim, base=10000, precision=torch.half, learnable=False):
+        super().__init__()
+        inv_freq = 1. / (base ** (torch.arange(0, dim, 2).float() / dim))
+        inv_freq = inv_freq.half()
+        self.learnable = learnable
+        if learnable:
+            self.inv_freq = torch.nn.Parameter(inv_freq)
+            self.max_seq_len_cached = None
+        else:
+            self.register_buffer('inv_freq', inv_freq)
+            self.max_seq_len_cached = None
+            self.cos_cached = None
+            self.sin_cached = None
+        self.precision = precision
+
+    def _load_from_state_dict(self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys,
+                              error_msgs):
+        pass
+
+    def forward(self, x, seq_dim=1, seq_len=None):
+        if seq_len is None:
+            seq_len = x.shape[seq_dim]
+        if self.max_seq_len_cached is None or (seq_len > self.max_seq_len_cached):
+            self.max_seq_len_cached = None if self.learnable else seq_len
+            t = torch.arange(seq_len, device=x.device, dtype=self.inv_freq.dtype)
+            freqs = torch.einsum('i,j->ij', t, self.inv_freq)
+            # Different from paper, but it uses a different permutation in order to obtain the same calculation
+            emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
+            if self.precision == torch.bfloat16:
+                emb = emb.float()
+
+            # [sx, 1 (b * np), hn]
+            cos_cached = emb.cos()[:, None, :]
+            sin_cached = emb.sin()[:, None, :]
+            if self.precision == torch.bfloat16:
+                cos_cached = cos_cached.bfloat16()
+                sin_cached = sin_cached.bfloat16()
+            if self.learnable:
+                return cos_cached, sin_cached
+            self.cos_cached, self.sin_cached = cos_cached, sin_cached
+        return self.cos_cached[:seq_len, ...], self.sin_cached[:seq_len, ...]
+
+    def _apply(self, fn):
+        if self.cos_cached is not None:
+            self.cos_cached = fn(self.cos_cached)
+        if self.sin_cached is not None:
+            self.sin_cached = fn(self.sin_cached)
+        return super()._apply(fn)
+
+
+def rotate_half(x):
+    x1, x2 = x[..., :x.shape[-1] // 2], x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=x1.ndim - 1)  # dim=-1 triggers a bug in earlier torch versions
+
+
+@torch.jit.script
+def apply_rotary_pos_emb_index(q, k, cos, sin, position_id):
+    # position_id: [sq, b], q, k: [sq, b, np, hn], cos: [sq, 1, hn] -> [sq, b, 1, hn]
+    cos, sin = F.embedding(position_id, cos.squeeze(1)).unsqueeze(2), \
+        F.embedding(position_id, sin.squeeze(1)).unsqueeze(2)
+    q, k = (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
+    return q, k
+
+
+def attention_fn(
+        self,
+        query_layer,
+        key_layer,
+        value_layer,
+        attention_mask,
+        hidden_size_per_partition,
+        layer_id,
+        layer_past=None,
+        scaling_attention_score=True,
+        use_cache=False,
+):
+    if layer_past is not None:
+        past_key, past_value = layer_past[0], layer_past[1]
+        key_layer = torch.cat((past_key, key_layer), dim=0)
+        value_layer = torch.cat((past_value, value_layer), dim=0)
+
+    # seqlen, batch, num_attention_heads, hidden_size_per_attention_head
+    seq_len, b, nh, hidden_size = key_layer.shape
+
+    if use_cache:
+        present = (key_layer, value_layer)
+    else:
+        present = None
+
+    query_key_layer_scaling_coeff = float(layer_id + 1)
+    if scaling_attention_score:
+        query_layer = query_layer / (math.sqrt(hidden_size) * query_key_layer_scaling_coeff)
+
+    # ===================================
+    # Raw attention scores. [b, np, s, s]
+    # ===================================
+
+    # [b, np, sq, sk]
+    output_size = (query_layer.size(1), query_layer.size(2), query_layer.size(0), key_layer.size(0))
+
+    # [sq, b, np, hn] -> [sq, b * np, hn]
+    query_layer = query_layer.view(output_size[2], output_size[0] * output_size[1], -1)
+    # [sk, b, np, hn] -> [sk, b * np, hn]
+    key_layer = key_layer.view(output_size[3], output_size[0] * output_size[1], -1)
+
+    matmul_result = torch.empty(
+        output_size[0] * output_size[1],
+        output_size[2],
+        output_size[3],
+        dtype=query_layer.dtype,
+        device=query_layer.device,
+    )
+
+    matmul_result = torch.baddbmm(
+        matmul_result,
+        query_layer.transpose(0, 1),  # [b * np, sq, hn]
+        key_layer.transpose(0, 1).transpose(1, 2),  # [b * np, hn, sk]
+        beta=0.0,
+        alpha=1.0,
+    )
+
+    # change view to [b, np, sq, sk]
+    attention_scores = matmul_result.view(*output_size)
+
+    if self.scale_mask_softmax:
+        self.scale_mask_softmax.scale = query_key_layer_scaling_coeff
+        attention_probs = self.scale_mask_softmax(attention_scores, attention_mask.contiguous())
+    else:
+        if not (attention_mask == 0).all():
+            # if auto-regressive, skip
+            attention_scores.masked_fill_(attention_mask, -10000.0)
+        dtype = attention_scores.dtype
+        attention_scores = attention_scores.float()
+        attention_scores = attention_scores * query_key_layer_scaling_coeff
+
+        attention_probs = F.softmax(attention_scores, dim=-1)
+
+        attention_probs = attention_probs.type(dtype)
+
+    # =========================
+    # Context layer. [sq, b, hp]
+    # =========================
+
+    # value_layer -> context layer.
+    # [sk, b, np, hn] --> [b, np, sq, hn]
+
+    # context layer shape: [b, np, sq, hn]
+    output_size = (value_layer.size(1), value_layer.size(2), query_layer.size(0), value_layer.size(3))
+
+    # change view [sk, b * np, hn]
+    value_layer = value_layer.view(value_layer.size(0), output_size[0] * output_size[1], -1)
+
+    # change view [b * np, sq, sk]
+    attention_probs = attention_probs.view(output_size[0] * output_size[1], output_size[2], -1)
+
+    # matmul: [b * np, sq, hn]
+    context_layer = torch.bmm(attention_probs, value_layer.transpose(0, 1))
+
+    # change view [b, np, sq, hn]
+    context_layer = context_layer.view(*output_size)
+
+    # [b, np, sq, hn] --> [sq, b, np, hn]
+    context_layer = context_layer.permute(2, 0, 1, 3).contiguous()
+
+    # [sq, b, np, hn] --> [sq, b, hp]
+    new_context_layer_shape = context_layer.size()[:-2] + (hidden_size_per_partition,)
+    context_layer = context_layer.view(*new_context_layer_shape)
+
+    outputs = (context_layer, present, attention_probs)
+
+    return outputs
+
+
+class SelfAttention(torch.nn.Module):
+    def __init__(self, hidden_size, num_attention_heads,
+                 layer_id, hidden_size_per_attention_head=None, bias=True,
+                 params_dtype=torch.float, position_encoding_2d=True):
+        super(SelfAttention, self).__init__()
+
+        self.layer_id = layer_id
+        self.hidden_size = hidden_size
+        self.hidden_size_per_partition = hidden_size
+        self.num_attention_heads = num_attention_heads
+        self.num_attention_heads_per_partition = num_attention_heads
+        self.position_encoding_2d = position_encoding_2d
+        self.rotary_emb = RotaryEmbedding(
+            self.hidden_size // (self.num_attention_heads * 2)
+            if position_encoding_2d
+            else self.hidden_size // self.num_attention_heads,
+            base=10000,
+            precision=torch.half,
+            learnable=False,
+        )
+
+        self.scale_mask_softmax = None
+
+        if hidden_size_per_attention_head is None:
+            self.hidden_size_per_attention_head = hidden_size // num_attention_heads
+        else:
+            self.hidden_size_per_attention_head = hidden_size_per_attention_head
+
+        self.inner_hidden_size = num_attention_heads * self.hidden_size_per_attention_head
+
+        # Strided linear layer.
+        self.query_key_value = skip_init(
+            torch.nn.Linear,
+            hidden_size,
+            3 * self.inner_hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+        self.dense = skip_init(
+            torch.nn.Linear,
+            self.inner_hidden_size,
+            hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+    @staticmethod
+    def attention_mask_func(attention_scores, attention_mask):
+        attention_scores.masked_fill_(attention_mask, -10000.0)
+        return attention_scores
+
+    def split_tensor_along_last_dim(self, tensor, num_partitions,
+                                    contiguous_split_chunks=False):
+        """Split a tensor along its last dimension.
+        Arguments:
+            tensor: input tensor.
+            num_partitions: number of partitions to split the tensor
+            contiguous_split_chunks: If True, make each chunk contiguous
+                                    in memory.
+        """
+        # Get the size and dimension.
+        last_dim = tensor.dim() - 1
+        last_dim_size = tensor.size()[last_dim] // num_partitions
+        # Split.
+        tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
+        # Note: torch.split does not create contiguous tensors by default.
+        if contiguous_split_chunks:
+            return tuple(chunk.contiguous() for chunk in tensor_list)
+
+        return tensor_list
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            position_ids,
+            attention_mask: torch.Tensor,
+            layer_id,
+            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+            use_cache: bool = False,
+            output_attentions: bool = False,
+    ):
+        """
+        hidden_states: [seq_len, batch, hidden_size]
+        attention_mask: [(1, 1), seq_len, seq_len]
+        """
+
+        # [seq_len, batch, 3 * hidden_size]
+        mixed_raw_layer = self.query_key_value(hidden_states)
+
+        # [seq_len, batch, 3 * hidden_size] --> [seq_len, batch, num_attention_heads, 3 * hidden_size_per_attention_head]
+        new_tensor_shape = mixed_raw_layer.size()[:-1] + (
+            self.num_attention_heads_per_partition,
+            3 * self.hidden_size_per_attention_head,
+        )
+        mixed_raw_layer = mixed_raw_layer.view(*new_tensor_shape)
+
+        # [seq_len, batch, num_attention_heads, hidden_size_per_attention_head]
+        (query_layer, key_layer, value_layer) = self.split_tensor_along_last_dim(mixed_raw_layer, 3)
+
+        if self.position_encoding_2d:
+            q1, q2 = query_layer.chunk(2, dim=(query_layer.ndim - 1))
+            k1, k2 = key_layer.chunk(2, dim=(key_layer.ndim - 1))
+            cos, sin = self.rotary_emb(q1, seq_len=position_ids.max() + 1)
+            position_ids, block_position_ids = position_ids[:, 0, :].transpose(0, 1).contiguous(), \
+                position_ids[:, 1, :].transpose(0, 1).contiguous()
+            q1, k1 = apply_rotary_pos_emb_index(q1, k1, cos, sin, position_ids)
+            q2, k2 = apply_rotary_pos_emb_index(q2, k2, cos, sin, block_position_ids)
+            query_layer = torch.concat([q1, q2], dim=(q1.ndim - 1))
+            key_layer = torch.concat([k1, k2], dim=(k1.ndim - 1))
+        else:
+            position_ids = position_ids.transpose(0, 1)
+            cos, sin = self.rotary_emb(value_layer, seq_len=position_ids.max() + 1)
+            # [seq_len, batch, num_attention_heads, hidden_size_per_attention_head]
+            query_layer, key_layer = apply_rotary_pos_emb_index(query_layer, key_layer, cos, sin, position_ids)
+
+        # [seq_len, batch, hidden_size]
+        context_layer, present, attention_probs = attention_fn(
+            self=self,
+            query_layer=query_layer,
+            key_layer=key_layer,
+            value_layer=value_layer,
+            attention_mask=attention_mask,
+            hidden_size_per_partition=self.hidden_size_per_partition,
+            layer_id=layer_id,
+            layer_past=layer_past,
+            use_cache=use_cache
+        )
+
+        output = self.dense(context_layer)
+
+        outputs = (output, present)
+
+        if output_attentions:
+            outputs += (attention_probs,)
+
+        return outputs  # output, present, attention_probs
+
+
+class GEGLU(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.activation_fn = F.gelu
+
+    def forward(self, x):
+        # dim=-1 breaks in jit for pt<1.10
+        x1, x2 = x.chunk(2, dim=(x.ndim - 1))
+        return x1 * self.activation_fn(x2)
+
+
+class GLU(torch.nn.Module):
+    def __init__(self, hidden_size, inner_hidden_size=None,
+                 layer_id=None, bias=True, activation_func=gelu, params_dtype=torch.float):
+        super(GLU, self).__init__()
+        self.layer_id = layer_id
+        self.activation_func = activation_func
+
+        # Project to 4h.
+        self.hidden_size = hidden_size
+        if inner_hidden_size is None:
+            inner_hidden_size = 4 * hidden_size
+        self.inner_hidden_size = inner_hidden_size
+        self.dense_h_to_4h = skip_init(
+            torch.nn.Linear,
+            self.hidden_size,
+            self.inner_hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+        # Project back to h.
+        self.dense_4h_to_h = skip_init(
+            torch.nn.Linear,
+            self.inner_hidden_size,
+            self.hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+    def forward(self, hidden_states):
+        """
+        hidden_states: [seq_len, batch, hidden_size]
+        """
+
+        # [seq_len, batch, inner_hidden_size]
+        intermediate_parallel = self.dense_h_to_4h(hidden_states)
+
+        intermediate_parallel = self.activation_func(intermediate_parallel)
+
+        output = self.dense_4h_to_h(intermediate_parallel)
+
+        return output
+
+
+class GLMBlock(torch.nn.Module):
+    def __init__(
+            self,
+            hidden_size,
+            num_attention_heads,
+            layernorm_epsilon,
+            layer_id,
+            inner_hidden_size=None,
+            hidden_size_per_attention_head=None,
+            layernorm=LayerNorm,
+            use_bias=True,
+            params_dtype=torch.float,
+            num_layers=28,
+            position_encoding_2d=True
+    ):
+        super(GLMBlock, self).__init__()
+        # Set output layer initialization if not provided.
+
+        self.layer_id = layer_id
+
+        # Layernorm on the input data.
+        self.input_layernorm = layernorm(hidden_size, eps=layernorm_epsilon)
+
+        self.position_encoding_2d = position_encoding_2d
+
+        # Self attention.
+        self.attention = SelfAttention(
+            hidden_size,
+            num_attention_heads,
+            layer_id,
+            hidden_size_per_attention_head=hidden_size_per_attention_head,
+            bias=use_bias,
+            params_dtype=params_dtype,
+            position_encoding_2d=self.position_encoding_2d
+        )
+
+        # Layernorm on the input data.
+        self.post_attention_layernorm = layernorm(hidden_size, eps=layernorm_epsilon)
+
+        self.num_layers = num_layers
+
+        # GLU
+        self.mlp = GLU(
+            hidden_size,
+            inner_hidden_size=inner_hidden_size,
+            bias=use_bias,
+            layer_id=layer_id,
+            params_dtype=params_dtype,
+        )
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            position_ids,
+            attention_mask: torch.Tensor,
+            layer_id,
+            layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+            use_cache: bool = False,
+            output_attentions: bool = False,
+    ):
+        """
+        hidden_states: [seq_len, batch, hidden_size]
+        attention_mask: [(1, 1), seq_len, seq_len]
+        """
+
+        # Layer norm at the begining of the transformer layer.
+        # [seq_len, batch, hidden_size]
+        attention_input = self.input_layernorm(hidden_states)
+
+        # Self attention.
+        attention_outputs = self.attention(
+            attention_input,
+            position_ids,
+            attention_mask=attention_mask,
+            layer_id=layer_id,
+            layer_past=layer_past,
+            use_cache=use_cache,
+            output_attentions=output_attentions
+        )
+
+        attention_output = attention_outputs[0]
+
+        outputs = attention_outputs[1:]
+
+        # Residual connection.
+        alpha = (2 * self.num_layers) ** 0.5
+        hidden_states = attention_input * alpha + attention_output
+
+        mlp_input = self.post_attention_layernorm(hidden_states)
+
+        # MLP.
+        mlp_output = self.mlp(mlp_input)
+
+        # Second residual connection.
+        output = mlp_input * alpha + mlp_output
+
+        if use_cache:
+            outputs = (output,) + outputs
+        else:
+            outputs = (output,) + outputs[1:]
+
+        return outputs  # hidden_states, present, attentions
+
+
+class ChatGLMPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and
+    a simple interface for downloading and loading pretrained models.
+    """
+
+    is_parallelizable = False
+    supports_gradient_checkpointing = True
+    config_class = ChatGLMConfig
+    base_model_prefix = "transformer"
+    _no_split_modules = ["GLMBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module: nn.Module):
+        """Initialize the weights."""
+        return
+
+    def get_masks(self, input_ids, device):
+        batch_size, seq_length = input_ids.shape
+        context_lengths = [seq.tolist().index(self.config.bos_token_id) for seq in input_ids]
+        attention_mask = torch.ones((batch_size, seq_length, seq_length), device=device)
+        attention_mask.tril_()
+        for i, context_length in enumerate(context_lengths):
+            attention_mask[i, :, :context_length] = 1
+        attention_mask.unsqueeze_(1)
+        attention_mask = (attention_mask < 0.5).bool()
+
+        return attention_mask
+
+    def get_position_ids(self, input_ids, mask_positions, device, gmask=False):
+        batch_size, seq_length = input_ids.shape
+        context_lengths = [seq.tolist().index(self.config.bos_token_id) for seq in input_ids]
+        if self.position_encoding_2d:
+            position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
+            for i, context_length in enumerate(context_lengths):
+                position_ids[i, context_length:] = mask_positions[i]
+            block_position_ids = [torch.cat((
+                torch.zeros(context_length, dtype=torch.long, device=device),
+                torch.arange(seq_length - context_length, dtype=torch.long, device=device) + 1
+            )) for context_length in context_lengths]
+            block_position_ids = torch.stack(block_position_ids, dim=0)
+            position_ids = torch.stack((position_ids, block_position_ids), dim=1)
+        else:
+            position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
+            if not gmask:
+                for i, context_length in enumerate(context_lengths):
+                    position_ids[context_length:] = mask_positions[i]
+
+        return position_ids
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, ChatGLMModel):
+            module.gradient_checkpointing = value
+
+
+CHATGLM_6B_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general
+    usage and behavior.
+
+    Parameters:
+        config ([`~ChatGLM6BConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the configuration.
+            Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+CHATGLM_6B_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`ChatGLM6BTokenizer`].
+            See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings.
+            Selected in the range `[0, config.max_position_embeddings - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
+            This is useful if you want more control over how to convert *input_ids* indices into associated vectors
+            than the model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare ChatGLM-6B Model transformer outputting raw hidden-states without any specific head on top.",
+    CHATGLM_6B_START_DOCSTRING,
+)
+class ChatGLMModel(ChatGLMPreTrainedModel):
+    """
+
+    The model can behave as an encoder (with only self-attention) as well
+    as a decoder, in which case a layer of cross-attention is added between
+    the self-attention layers, following the architecture described in [Attention is
+    all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani,
+    Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
+
+    To behave as an decoder the model needs to be initialized with the
+    `is_decoder` argument of the configuration set to `True`.
+    To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder`
+    argument and `add_cross_attention` set to `True`; an
+    `encoder_hidden_states` is then expected as an input to the forward pass.
+    """
+
+    def __init__(self, config: ChatGLMConfig):
+        super().__init__(config)
+
+        # recording parameters
+        self.max_sequence_length = config.max_sequence_length
+        self.hidden_size = config.hidden_size
+        self.params_dtype = torch.half
+        self.num_attention_heads = config.num_attention_heads
+        self.vocab_size = config.vocab_size
+        self.num_layers = config.num_layers
+        self.layernorm_epsilon = config.layernorm_epsilon
+        self.inner_hidden_size = config.inner_hidden_size
+        self.hidden_size_per_attention_head = self.hidden_size // self.num_attention_heads
+        self.position_encoding_2d = config.position_encoding_2d
+        self.pre_seq_len = config.pre_seq_len
+        self.prefix_projection = config.prefix_projection
+
+        self.word_embeddings = skip_init(
+            torch.nn.Embedding,
+            num_embeddings=self.vocab_size, embedding_dim=self.hidden_size,
+            dtype=self.params_dtype
+        )
+        self.gradient_checkpointing = False
+
+        def get_layer(layer_id):
+            return GLMBlock(
+                self.hidden_size,
+                self.num_attention_heads,
+                self.layernorm_epsilon,
+                layer_id,
+                inner_hidden_size=self.inner_hidden_size,
+                hidden_size_per_attention_head=self.hidden_size_per_attention_head,
+                layernorm=LayerNorm,
+                use_bias=True,
+                params_dtype=self.params_dtype,
+                position_encoding_2d=self.position_encoding_2d,
+            )
+
+        self.layers = torch.nn.ModuleList(
+            [get_layer(layer_id) for layer_id in range(self.num_layers)]
+        )
+
+        # Final layer norm before output.
+        self.final_layernorm = LayerNorm(self.hidden_size, eps=self.layernorm_epsilon)
+
+        if self.pre_seq_len is not None:
+            for param in self.parameters():
+                param.requires_grad = False
+            self.prefix_tokens = torch.arange(self.pre_seq_len).long()
+            self.prefix_encoder = PrefixEncoder(config)
+            self.dropout = torch.nn.Dropout(0.1)
+
+            # total_params = sum(p.numel() for p in self.parameters())
+            # trainable_params = sum(p.numel() for p in self.parameters() if p.requires_grad)
+            # print("Using p-tuning v2: # trainable_params = {} / {}".format(trainable_params, total_params))
+
+    def get_input_embeddings(self):
+        return self.word_embeddings
+
+    def set_input_embeddings(self, new_embeddings: torch.Tensor):
+        self.word_embeddings = new_embeddings
+
+    def get_prompt(self, batch_size, device, dtype=torch.half):
+        prefix_tokens = self.prefix_tokens.unsqueeze(0).expand(batch_size, -1).to(device)
+        past_key_values = self.prefix_encoder(prefix_tokens).type(dtype)
+        past_key_values = past_key_values.view(
+            batch_size,
+            self.pre_seq_len,
+            self.num_layers * 2,
+            self.num_attention_heads,
+            self.hidden_size // self.num_attention_heads
+        )
+        # seq_len, b, nh, hidden_size
+        past_key_values = self.dropout(past_key_values)
+        past_key_values = past_key_values.permute([2, 1, 0, 3, 4]).split(2)
+        # past_key_values = [(v[0], v[1]) for v in past_key_values]
+        return past_key_values
+
+    @add_start_docstrings_to_model_forward(CHATGLM_6B_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPastAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            inputs_embeds: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPast]:
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape[:2]
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape[:2]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.word_embeddings(input_ids)
+
+        if past_key_values is None:
+            if self.pre_seq_len is not None:
+                past_key_values = self.get_prompt(batch_size=input_ids.shape[0], device=input_ids.device,
+                                                  dtype=inputs_embeds.dtype)
+            else:
+                past_key_values = tuple([None] * len(self.layers))
+
+            if attention_mask is None:
+                attention_mask = self.get_masks(
+                    input_ids,
+                    device=input_ids.device
+                )
+
+
+            if position_ids is None:
+                MASK, gMASK = 130000, 130001
+                mask_token = MASK if MASK in input_ids else gMASK
+                use_gmask = False if MASK in input_ids else gMASK
+
+                mask_positions = [seq.tolist().index(mask_token) for seq in input_ids]
+                position_ids = self.get_position_ids(
+                    input_ids,
+                    mask_positions=mask_positions,
+                    device=input_ids.device,
+                    gmask=use_gmask
+                )
+
+        if self.pre_seq_len is not None and attention_mask is not None:
+            prefix_attention_mask = torch.ones(batch_size, 1, input_ids.size(-1), self.pre_seq_len).to(
+                attention_mask.device)
+            prefix_attention_mask = (prefix_attention_mask < 0.5).bool()
+            attention_mask = torch.cat((prefix_attention_mask, attention_mask), dim=3)
+
+        # [seq_len, batch, hidden_size]
+        hidden_states = inputs_embeds.transpose(0, 1)
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+
+        if attention_mask is None:
+            attention_mask = torch.zeros(1, 1, device=input_ids.device).bool()
+
+        else:
+            attention_mask = attention_mask.to(input_ids.device)
+
+        for i, layer in enumerate(self.layers):
+
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+            layer_past = past_key_values[i]
+
+            if self.gradient_checkpointing and self.training:
+                layer_ret = torch.utils.checkpoint.checkpoint(
+                    layer,
+                    hidden_states,
+                    position_ids,
+                    attention_mask,
+                    torch.tensor(i),
+                    layer_past,
+                    use_cache,
+                    output_attentions
+                )
+            else:
+                layer_ret = layer(
+                    hidden_states,
+                    position_ids=position_ids,
+                    attention_mask=attention_mask,
+                    layer_id=torch.tensor(i),
+                    layer_past=layer_past,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions
+                )
+
+            hidden_states = layer_ret[0]
+
+            if use_cache:
+                presents = presents + (layer_ret[1],)
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (layer_ret[2 if use_cache else 1],)
+
+        # Final layer norm.
+        hidden_states = self.final_layernorm(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+class ChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):
+    def __init__(self, config: ChatGLMConfig):
+        super().__init__(config)
+
+        # self.hidden_size = config.hidden_size
+        # self.params_dtype = torch.half
+        # self.vocab_size = config.vocab_size
+        self.max_sequence_length = config.max_sequence_length
+
+        self.position_encoding_2d = config.position_encoding_2d
+
+        self.transformer = ChatGLMModel(config)
+
+        self.lm_head = skip_init(
+            nn.Linear,
+            config.hidden_size,
+            config.vocab_size,
+            bias=False,
+            dtype=torch.half
+        )
+
+        self.config = config
+
+        self.quantized = False
+
+        if self.config.quantization_bit:
+            self.quantize(self.config.quantization_bit, empty_init=True)
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def _update_model_kwargs_for_generation(
+        self,
+        outputs: ModelOutput,
+        model_kwargs: Dict[str, Any],
+        is_encoder_decoder: bool = False,
+        standardize_cache_format: bool = False,
+    ) -> Dict[str, Any]:
+        # update past_key_values
+        model_kwargs["past_key_values"] = self._extract_past_from_model_output(
+            outputs, standardize_cache_format=standardize_cache_format
+        )
+
+        # update attention mask
+        if "attention_mask" in model_kwargs:
+            attention_mask = model_kwargs["attention_mask"]
+            if attention_mask is not None and attention_mask.dtype == torch.bool:
+                attention_mask = torch.cat(
+                    [attention_mask, attention_mask.new_ones((*attention_mask.shape[:3], 1))], dim=3)
+                new_attention_mask = attention_mask[:, :, -1:].clone()
+                new_attention_mask[..., -1] = False
+                model_kwargs["attention_mask"] = torch.cat(
+                    [attention_mask, new_attention_mask], dim=2
+                )
+
+        # update position ids
+        if "position_ids" in model_kwargs:
+            position_ids = model_kwargs["position_ids"]
+            new_position_id = position_ids[..., -1:].clone()
+            new_position_id[:, 1, :] += 1
+            model_kwargs["position_ids"] = torch.cat(
+                [position_ids, new_position_id], dim=-1
+            )
+
+        return model_kwargs
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids: torch.LongTensor,
+            past: Optional[torch.Tensor] = None,
+            past_key_values: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.Tensor] = None,
+            **kwargs
+    ) -> dict:
+        batch_size, seq_length = input_ids.shape
+        MASK, gMASK = 130000, 130001
+        mask_token = MASK if MASK in input_ids else gMASK
+        use_gmask = False if MASK in input_ids else gMASK
+        seqs = input_ids.tolist()
+        mask_positions = [seq.index(mask_token) for seq in seqs]
+
+        # only last token for input_ids if past is not None
+        if past is not None or past_key_values is not None:
+            last_token = input_ids[:, -1].unsqueeze(-1)
+            if attention_mask is not None and attention_mask.dtype == torch.bool:
+                attention_mask = attention_mask[:, :, -1:]
+            else:
+                attention_mask = None
+            if position_ids is not None:
+                position_ids = position_ids[..., -1:]
+            else:
+                context_lengths = [seq.index(self.config.bos_token_id) for seq in seqs]
+                if self.position_encoding_2d:
+                    position_ids = torch.tensor(
+                        [[mask_position, seq_length - context_length] for mask_position, context_length in
+                         zip(mask_positions, context_lengths)], dtype=torch.long, device=input_ids.device).unsqueeze(-1)
+                else:
+                    position_ids = torch.tensor([mask_position for mask_position in mask_positions], dtype=torch.long,
+                                                device=input_ids.device).unsqueeze(-1)
+
+            if past is None:
+                past = past_key_values
+            return {
+                "input_ids": last_token,
+                "past_key_values": past,
+                "position_ids": position_ids,
+                "attention_mask": attention_mask
+            }
+        else:
+            if attention_mask is not None and attention_mask.dtype != torch.bool:
+                logger.warning_once(f"The dtype of attention mask ({attention_mask.dtype}) is not bool")
+                attention_mask = None
+            if attention_mask is None:
+                attention_mask = self.get_masks(
+                    input_ids,
+                    device=input_ids.device
+                )
+            if position_ids is None:
+                position_ids = self.get_position_ids(
+                    input_ids,
+                    device=input_ids.device,
+                    mask_positions=mask_positions,
+                    gmask=use_gmask
+                )
+
+            return {
+                "input_ids": input_ids,
+                "past_key_values": past,
+                "position_ids": position_ids,
+                "attention_mask": attention_mask
+            }
+
+    def forward(
+            self,
+            input_ids: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            labels: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ):
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids=input_ids,
+            position_ids=position_ids,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = transformer_outputs[0]
+
+        lm_logits = self.lm_head(hidden_states).permute(1, 0, 2).contiguous()
+
+        loss = None
+        if labels is not None:
+            lm_logits = lm_logits.to(torch.float32)
+
+            # Shift so that tokens < n predict n
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss(ignore_index=-100)
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            lm_logits = lm_logits.to(hidden_states.dtype)
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+            past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
+    ) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
+        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+
+        Output shares the same memory storage as `past`.
+        """
+        return tuple(
+            (
+                layer_past[0].index_select(1, beam_idx.to(layer_past[0].device)),
+                layer_past[1].index_select(1, beam_idx.to(layer_past[1].device)),
+            )
+            for layer_past in past
+        )
+
+    def process_response(self, response):
+        response = response.strip()
+        response = response.replace("[[训练时间]]", "2023年")
+        punkts = [
+            [",", "，"],
+            ["!", "！"],
+            [":", "："],
+            [";", "；"],
+            ["\?", "？"],
+        ]
+        for item in punkts:
+            response = re.sub(r"([\u4e00-\u9fff])%s" % item[0], r"\1%s" % item[1], response)
+            response = re.sub(r"%s([\u4e00-\u9fff])" % item[0], r"%s\1" % item[1], response)
+        return response
+
+    @torch.no_grad()
+    def chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, max_length: int = 2048, num_beams=1,
+             do_sample=True, top_p=0.7, temperature=0.95, logits_processor=None, **kwargs):
+        if history is None:
+            history = []
+        if logits_processor is None:
+            logits_processor = LogitsProcessorList()
+        logits_processor.append(InvalidScoreLogitsProcessor())
+        gen_kwargs = {"max_length": max_length, "num_beams": num_beams, "do_sample": do_sample, "top_p": top_p,
+                      "temperature": temperature, "logits_processor": logits_processor, **kwargs}
+        if not history:
+            prompt = query
+        else:
+            prompt = ""
+            for i, (old_query, response) in enumerate(history):
+                prompt += "[Round {}]\n问：{}\n答：{}\n".format(i, old_query, response)
+            prompt += "[Round {}]\n问：{}\n答：".format(len(history), query)
+        inputs = tokenizer([prompt], return_tensors="pt")
+        inputs = inputs.to(self.device)
+        outputs = self.generate(**inputs, **gen_kwargs)
+        outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):]
+        response = tokenizer.decode(outputs)
+        response = self.process_response(response)
+        history = history + [(query, response)]
+        return response, history
+
+    @torch.no_grad()
+    def stream_chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, max_length: int = 2048,
+                    do_sample=True, top_p=0.7, temperature=0.95, logits_processor=None, **kwargs):
+        if history is None:
+            history = []
+        if logits_processor is None:
+            logits_processor = LogitsProcessorList()
+        logits_processor.append(InvalidScoreLogitsProcessor())
+        gen_kwargs = {"max_length": max_length, "do_sample": do_sample, "top_p": top_p,
+                      "temperature": temperature, "logits_processor": logits_processor, **kwargs}
+        if not history:
+            prompt = query
+        else:
+            prompt = ""
+            for i, (old_query, response) in enumerate(history):
+                prompt += "[Round {}]\n问：{}\n答：{}\n".format(i, old_query, response)
+            prompt += "[Round {}]\n问：{}\n答：".format(len(history), query)
+        inputs = tokenizer([prompt], return_tensors="pt")
+        inputs = inputs.to(self.device)
+        for outputs in self.stream_generate(**inputs, **gen_kwargs):
+            outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):]
+            response = tokenizer.decode(outputs)
+            response = self.process_response(response)
+            new_history = history + [(query, response)]
+            yield response, new_history
+
+    @torch.no_grad()
+    def stream_generate(
+            self,
+            input_ids,
+            generation_config: Optional[GenerationConfig] = None,
+            logits_processor: Optional[LogitsProcessorList] = None,
+            stopping_criteria: Optional[StoppingCriteriaList] = None,
+            prefix_allowed_tokens_fn: Optional[Callable[[int, torch.Tensor], List[int]]] = None,
+            **kwargs,
+    ):
+        batch_size, input_ids_seq_length = input_ids.shape[0], input_ids.shape[-1]
+
+        if generation_config is None:
+            generation_config = self.generation_config
+        generation_config = copy.deepcopy(generation_config)
+        model_kwargs = generation_config.update(**kwargs)
+        bos_token_id, eos_token_id = generation_config.bos_token_id, generation_config.eos_token_id
+
+        if isinstance(eos_token_id, int):
+            eos_token_id = [eos_token_id]
+
+        has_default_max_length = kwargs.get("max_length") is None and generation_config.max_length is not None
+        if has_default_max_length and generation_config.max_new_tokens is None:
+            warnings.warn(
+                f"Using `max_length`'s default ({generation_config.max_length}) to control the generation length. "
+                "This behaviour is deprecated and will be removed from the config in v5 of Transformers -- we"
+                " recommend using `max_new_tokens` to control the maximum length of the generation.",
+                UserWarning,
+            )
+        elif generation_config.max_new_tokens is not None:
+            generation_config.max_length = generation_config.max_new_tokens + input_ids_seq_length
+            if not has_default_max_length:
+                logger.warn(
+                    f"Both `max_new_tokens` (={generation_config.max_new_tokens}) and `max_length`(="
+                    f"{generation_config.max_length}) seem to have been set. `max_new_tokens` will take precedence. "
+                    "Please refer to the documentation for more information. "
+                    "(https://huggingface.co/docs/transformers/main/en/main_classes/text_generation)",
+                    UserWarning,
+                )
+
+        if input_ids_seq_length >= generation_config.max_length:
+            input_ids_string = "decoder_input_ids" if self.config.is_encoder_decoder else "input_ids"
+            logger.warning(
+                f"Input length of {input_ids_string} is {input_ids_seq_length}, but `max_length` is set to"
+                f" {generation_config.max_length}. This can lead to unexpected behavior. You should consider"
+                " increasing `max_new_tokens`."
+            )
+
+        # 2. Set generation parameters if not already defined
+        logits_processor = logits_processor if logits_processor is not None else LogitsProcessorList()
+        stopping_criteria = stopping_criteria if stopping_criteria is not None else StoppingCriteriaList()
+
+        logits_processor = self._get_logits_processor(
+            generation_config=generation_config,
+            input_ids_seq_length=input_ids_seq_length,
+            encoder_input_ids=input_ids,
+            prefix_allowed_tokens_fn=prefix_allowed_tokens_fn,
+            logits_processor=logits_processor,
+        )
+
+        stopping_criteria = self._get_stopping_criteria(
+            generation_config=generation_config, stopping_criteria=stopping_criteria
+        )
+        logits_warper = self._get_logits_warper(generation_config)
+
+        unfinished_sequences = input_ids.new(input_ids.shape[0]).fill_(1)
+        scores = None
+        while True:
+            model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
+            # forward pass to get next token
+            outputs = self(
+                **model_inputs,
+                return_dict=True,
+                output_attentions=False,
+                output_hidden_states=False,
+            )
+
+            next_token_logits = outputs.logits[:, -1, :]
+
+            # pre-process distribution
+            next_token_scores = logits_processor(input_ids, next_token_logits)
+            next_token_scores = logits_warper(input_ids, next_token_scores)
+
+            # sample
+            probs = nn.functional.softmax(next_token_scores, dim=-1)
+            if generation_config.do_sample:
+                next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
+            else:
+                next_tokens = torch.argmax(probs, dim=-1)
+
+            # update generated ids, model inputs, and length for next step
+            input_ids = torch.cat([input_ids, next_tokens[:, None]], dim=-1)
+            model_kwargs = self._update_model_kwargs_for_generation(
+                outputs, model_kwargs, is_encoder_decoder=self.config.is_encoder_decoder
+            )
+            unfinished_sequences = unfinished_sequences.mul((sum(next_tokens != i for i in eos_token_id)).long())
+
+            # stop when each sentence is finished, or if we exceed the maximum length
+            if unfinished_sequences.max() == 0 or stopping_criteria(input_ids, scores):
+                break
+            yield input_ids
+
+    def quantize(self, bits: int, empty_init=False, **kwargs):
+        if bits == 0:
+            return
+
+        from .quantization import quantize
+
+        if self.quantized:
+            logger.info("Already quantized.")
+            return self
+
+        self.quantized = True
+
+        self.config.quantization_bit = bits
+
+        self.transformer = quantize(self.transformer, bits, empty_init=empty_init, **kwargs)
+        return self

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec688fe5ecf20587f9e8ad2c1af552fd19872c1f077073cf428dd49d74d6c1ad
+size 134

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d043ef7609416a4b52624ca09449dc4925ad776fbe1e40db2c5ca482fb55655
+size 134

quantization.py

@@ -0,0 +1,201 @@
+from torch.nn import Linear
+from torch.nn.parameter import Parameter
+
+import bz2
+import torch
+import base64
+import ctypes
+from transformers.utils import logging
+
+from typing import List
+from functools import partial
+
+logger = logging.get_logger(__name__)
+
+try:
+    from cpm_kernels.kernels.base import LazyKernelCModule, KernelFunction, round_up
+
+    class Kernel:
+        def __init__(self, code: bytes, function_names: List[str]):
+            self.code = code
+            self._function_names = function_names
+            self._cmodule = LazyKernelCModule(self.code)
+
+            for name in self._function_names:
+                setattr(self, name, KernelFunction(self._cmodule, name))
+
+    quantization_code = "$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"
+
+    kernels = Kernel(
+        bz2.decompress(base64.b64decode(quantization_code)),
+        [
+            "int4WeightCompression",
+            "int4WeightExtractionFloat",
+            "int4WeightExtractionHalf",
+            "int8WeightExtractionFloat",
+            "int8WeightExtractionHalf",
+        ],
+    )
+except Exception as exception:
+    kernels = None
+    logger.warning("Failed to load cpm_kernels:" + str(exception))
+
+
+class W8A16Linear(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, inp: torch.Tensor, quant_w: torch.Tensor, scale_w: torch.Tensor, weight_bit_width):
+        ctx.inp_shape = inp.size()
+        ctx.weight_bit_width = weight_bit_width
+        out_features = quant_w.size(0)
+        inp = inp.contiguous().view(-1, inp.size(-1))
+        weight = extract_weight_to_half(quant_w, scale_w, weight_bit_width)
+        ctx.weight_shape = weight.size()
+        output = inp.mm(weight.t())
+        ctx.save_for_backward(inp, quant_w, scale_w)
+        return output.view(*(ctx.inp_shape[:-1] + (out_features,)))
+
+    @staticmethod
+    def backward(ctx, grad_output: torch.Tensor):
+        inp, quant_w, scale_w = ctx.saved_tensors
+        weight = extract_weight_to_half(quant_w, scale_w, ctx.weight_bit_width)
+        grad_output = grad_output.contiguous().view(-1, weight.size(0))
+        grad_input = grad_output.mm(weight)
+        grad_weight = grad_output.t().mm(inp)
+        return grad_input.view(ctx.inp_shape), grad_weight.view(ctx.weight_shape), None, None
+
+
+def compress_int4_weight(weight: torch.Tensor):  # (n, m)
+    with torch.cuda.device(weight.device):
+        n, m = weight.size(0), weight.size(1)
+        assert m % 2 == 0
+        m = m // 2
+        out = torch.empty(n, m, dtype=torch.int8, device="cuda")
+        stream = torch.cuda.current_stream()
+
+        gridDim = (n, 1, 1)
+        blockDim = (min(round_up(m, 32), 1024), 1, 1)
+
+        kernels.int4WeightCompression(
+            gridDim,
+            blockDim,
+            0,
+            stream,
+            [ctypes.c_void_p(weight.data_ptr()), ctypes.c_void_p(out.data_ptr()), ctypes.c_int32(n), ctypes.c_int32(m)],
+        )
+        return out
+
+
+def extract_weight_to_half(weight: torch.Tensor, scale_list: torch.Tensor, source_bit_width: int):
+    if source_bit_width == 8:
+        func = kernels.int8WeightExtractionHalf
+    elif source_bit_width == 4:
+        func = kernels.int4WeightExtractionHalf
+    else:
+        assert False, "Unsupported bit-width"
+
+    with torch.cuda.device(weight.device):
+        n, m = weight.size(0), weight.size(1)
+        out = torch.empty(n, m * (8 // source_bit_width), dtype=torch.half, device="cuda")
+        stream = torch.cuda.current_stream()
+
+        gridDim = (n, 1, 1)
+        blockDim = (min(round_up(m, 32), 1024), 1, 1)
+
+        func(
+            gridDim,
+            blockDim,
+            0,
+            stream,
+            [
+                ctypes.c_void_p(weight.data_ptr()),
+                ctypes.c_void_p(scale_list.data_ptr()),
+                ctypes.c_void_p(out.data_ptr()),
+                ctypes.c_int32(n),
+                ctypes.c_int32(m),
+            ],
+        )
+        return out
+
+
+class QuantizedLinear(Linear):
+    def __init__(self, weight_bit_width: int, weight_tensor=None, bias_tensor=None, empty_init=False, *args, **kwargs):
+        super(QuantizedLinear, self).__init__(*args, **kwargs)
+        self.weight_bit_width = weight_bit_width
+
+        shape = self.weight.shape
+        del self.weight
+
+        if weight_tensor is None or empty_init:
+            self.weight = torch.empty(
+                shape[0], shape[1] * weight_bit_width // 8, dtype=torch.int8, device=kwargs["device"]
+            )
+            self.weight_scale = torch.empty(shape[0], dtype=kwargs["dtype"], device=kwargs["device"])
+        else:
+            self.weight_scale = (weight_tensor.abs().max(dim=-1).values / ((2 ** (weight_bit_width - 1)) - 1)).half()
+            self.weight = torch.round(weight_tensor / self.weight_scale[:, None]).to(torch.int8)
+            if weight_bit_width == 4:
+                self.weight = compress_int4_weight(self.weight)
+
+        self.weight = Parameter(self.weight.to(kwargs["device"]), requires_grad=False)
+        self.weight_scale = Parameter(self.weight_scale.to(kwargs["device"]), requires_grad=False)
+        if bias_tensor is not None:
+            self.bias = Parameter(bias_tensor.to(kwargs["device"]), requires_grad=False)
+        else:
+            self.bias = None
+
+    def forward(self, input):
+        output = W8A16Linear.apply(input, self.weight, self.weight_scale, self.weight_bit_width)
+        if self.bias is not None:
+            output = output + self.bias
+        return output
+
+
+def quantize(model, weight_bit_width, empty_init=False, **kwargs):
+    """Replace fp16 linear with quantized linear"""
+
+    for layer in model.layers:
+        layer.attention.query_key_value = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.attention.query_key_value.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.attention.query_key_value.bias,
+            in_features=layer.attention.query_key_value.in_features,
+            out_features=layer.attention.query_key_value.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.attention.query_key_value.weight.device,
+            empty_init=empty_init
+        )
+        layer.attention.dense = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.attention.dense.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.attention.dense.bias,
+            in_features=layer.attention.dense.in_features,
+            out_features=layer.attention.dense.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.attention.dense.weight.device,
+            empty_init=empty_init
+        )
+        layer.mlp.dense_h_to_4h = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.mlp.dense_h_to_4h.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.mlp.dense_h_to_4h.bias,
+            in_features=layer.mlp.dense_h_to_4h.in_features,
+            out_features=layer.mlp.dense_h_to_4h.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.mlp.dense_h_to_4h.weight.device,
+            empty_init=empty_init
+        )
+        layer.mlp.dense_4h_to_h = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight_tensor=layer.mlp.dense_4h_to_h.weight.to(torch.cuda.current_device()),
+            bias_tensor=layer.mlp.dense_4h_to_h.bias,
+            in_features=layer.mlp.dense_4h_to_h.in_features,
+            out_features=layer.mlp.dense_4h_to_h.out_features,
+            bias=True,
+            dtype=torch.half,
+            device=layer.mlp.dense_4h_to_h.weight.device,
+            empty_init=empty_init
+        )
+    return model

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e37dc82f73d28016090cbbda0b59b3662b6e32a61e6714fbc9bf56073442baba
+size 14575

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:789a1605f05260342a605be992efe3dec445d76b75c62bdef618345fc0105cc0
+size 627

special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "bos_token": "<sop>",
+  "eos_token": "</s>",
+  "mask_token": "[MASK]",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

tokenization_chatglm.py

@@ -0,0 +1,429 @@
+"""Tokenization classes for ChatGLM."""
+from typing import List, Optional, Union
+import os
+
+from transformers.tokenization_utils import PreTrainedTokenizer
+from icetk.text_tokenizer import TextTokenizer
+import icetk.sentencepiece_model_pb2 as sp_model
+from transformers.utils import logging, PaddingStrategy
+from transformers.tokenization_utils_base import EncodedInput, BatchEncoding
+from typing import Dict
+import numpy as np
+
+logger = logging.get_logger(__name__)
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "silver/chatglm-6b-slim": 2048,
+}
+
+
+class SPTokenizer:
+    def __init__(
+        self,
+        vocab_file,
+        max_blank_length=80,
+        byte_fallback=True,
+    ):
+        assert vocab_file is not None
+        self.vocab_file = vocab_file
+        self.special_tokens = ["[MASK]", "[gMASK]", "[sMASK]", "<unused_0>", "<sop>", "<eop>", "<ENC>", "<dBLOCK>"]
+        self.max_blank_length = max_blank_length
+        self.byte_fallback = byte_fallback
+        self.text_tokenizer = self._build_text_tokenizer(encode_special_tokens=False)
+        self.special_text_tokenizer = self._build_text_tokenizer(encode_special_tokens=True)
+
+    @staticmethod
+    def _configure_tokenizer(
+        text_tokenizer: TextTokenizer,
+        special_tokens: List[str],
+        max_blank_length: int,
+        byte_fallback: bool,
+        encode_special_tokens=False,
+    ):
+        # special token
+        special_token_type = 4 if encode_special_tokens else 3  # 3 - CONTROL, 4 - USER_DEFINE
+        for token in special_tokens:
+            text_tokenizer.proto.pieces.append(
+                sp_model.ModelProto.SentencePiece(piece=token, score=0.0, type=special_token_type)
+            )
+        # whitespaces
+        for token in [SPTokenizer.get_tab_token()] + [
+            SPTokenizer.get_blank_token(i) for i in range(2, max_blank_length + 1)
+        ]:
+            text_tokenizer.proto.pieces.append(sp_model.ModelProto.SentencePiece(piece=token, score=0.0, type=4))
+        # byte fallback
+        if byte_fallback:
+            text_tokenizer.proto.trainer_spec.byte_fallback = True
+            for i in range(256):
+                text_tokenizer.proto.pieces.append(
+                    sp_model.ModelProto.SentencePiece(piece="<0x{:02X}>".format(i), score=0.0, type=6)
+                )
+        text_tokenizer.refresh()
+
+    def _build_text_tokenizer(self, encode_special_tokens=False):
+        tokenizer = TextTokenizer(self.vocab_file)
+        self._configure_tokenizer(
+            tokenizer, self.special_tokens, self.max_blank_length, self.byte_fallback, encode_special_tokens
+        )
+        return tokenizer
+
+    def _get_text_tokenizer(self, encode_special_tokens=False):
+        if encode_special_tokens:
+            return self.special_text_tokenizer
+        else:
+            return self.text_tokenizer
+
+    @staticmethod
+    def get_blank_token(length: int):
+        assert length >= 2
+        return f"<|blank_{length}|>"
+
+    @staticmethod
+    def get_tab_token():
+        return f"<|tab|>"
+
+    @property
+    def num_text_tokens(self):
+        return self.text_tokenizer.num_tokens
+
+    @property
+    def num_tokens(self):
+        return self.num_text_tokens
+
+    @staticmethod
+    def _encode_whitespaces(text: str, max_len: int = 80):
+        text = text.replace("\t", SPTokenizer.get_tab_token())
+        for i in range(max_len, 1, -1):
+            text = text.replace(" " * i, SPTokenizer.get_blank_token(i))
+        return text
+
+    def _preprocess(self, text: str, linebreak=True, whitespaces=True):
+        if linebreak:
+            text = text.replace("\n", "<n>")
+        if whitespaces:
+            text = self._encode_whitespaces(text, max_len=self.max_blank_length)
+        return text
+
+    def encode(
+        self, text: str, linebreak=True, whitespaces=True, special_tokens=False, add_dummy_prefix=True
+    ) -> List[int]:
+        """
+        @param text: Text to encode.
+        @param linebreak: Whether to encode newline (\n) in text.
+        @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
+        @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
+        @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
+        """
+        text = self._preprocess(text, linebreak, whitespaces)
+        if not add_dummy_prefix:
+            text = "<n>" + text
+        tmp = self._get_text_tokenizer(encode_special_tokens=special_tokens).encode(text)
+        tokens = [x for x in tmp]
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def decode(self, text_ids: List[int], special_tokens=False) -> str:
+        ids = [int(_id) for _id in text_ids]
+        ids = [_id for _id in ids if _id >= 0]
+        text = self._get_text_tokenizer(encode_special_tokens=special_tokens).decode(ids)
+        text = text.replace("<n>", "\n")
+        text = text.replace(SPTokenizer.get_tab_token(), "\t")
+        for i in range(2, self.max_blank_length + 1):
+            text = text.replace(self.get_blank_token(i), " " * i)
+        return text
+
+    def tokenize(
+        self, text: str, linebreak=True, whitespaces=True, special_tokens=False, add_dummy_prefix=True
+    ) -> List[str]:
+        """
+        @param text: Text to encode.
+        @param linebreak: Whether to encode newline (\n) in text.
+        @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
+        @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
+        @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
+        """
+        text = self._preprocess(text, linebreak, whitespaces)
+        if not add_dummy_prefix:
+            text = "<n>" + text
+        tokens = self._get_text_tokenizer(encode_special_tokens=special_tokens).tokenize(text)
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def __getitem__(self, x: Union[int, str]):
+        if isinstance(x, int):
+            return self.text_tokenizer.convert_id_to_token(x)
+        elif isinstance(x, str):
+            return self.text_tokenizer.convert_token_to_id(x)
+        else:
+            raise ValueError("The key should be str or int.")
+
+
+class ChatGLMTokenizer(PreTrainedTokenizer):
+    """
+    Construct a ChatGLM tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = {"vocab_file": "ice_text.model"}
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["input_ids", "attention_mask", "position_ids"]
+
+    def __init__(
+            self,
+            vocab_file,
+            do_lower_case=False,
+            remove_space=False,
+            bos_token='sop',
+            eos_token='eos',
+            eop_token='eop',
+            mask_token='[MASK]',
+            gmask_token='[gMASK]',
+            padding_side="left",
+            **kwargs
+    ) -> None:
+        super().__init__(
+            do_lower_case=do_lower_case,
+            remove_space=remove_space,
+            padding_side=padding_side,
+            **kwargs
+        )
+
+        self.do_lower_case = do_lower_case
+        self.remove_space = remove_space
+        self.vocab_file = vocab_file
+
+        self.bos_token = bos_token
+        self.eos_token = eos_token
+        self.eop_token = eop_token
+        self.mask_token = mask_token
+        self.gmask_token = gmask_token
+
+        self.sp_tokenizer = SPTokenizer(vocab_file)
+
+        """ Initialisation """
+
+    @property
+    def eop_token_id(self) -> Optional[int]:
+        """
+        `Optional[int]`: Id of the end of sentence token in the vocabulary. Returns `None` if the token has not been
+        set.
+        """
+        if self.eop_token is None:
+            return None
+        return self.convert_tokens_to_ids(self.eop_token)
+
+    @property
+    def vocab_size(self):
+        """ Returns vocab size """
+        return self.sp_tokenizer.num_tokens
+
+    def get_vocab(self):
+        """ Returns vocab as a dict """
+        vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def preprocess_text(self, inputs):
+        if self.remove_space:
+            outputs = " ".join(inputs.strip().split())
+        else:
+            outputs = inputs
+
+        if self.do_lower_case:
+            outputs = outputs.lower()
+
+        return outputs
+
+    def _tokenize(self, text, **kwargs):
+        """ Returns a tokenized string. """
+        text = self.preprocess_text(text)
+
+        seq = self.sp_tokenizer.tokenize(text)
+
+        return seq
+
+    def decode(
+            self,
+            token_ids: Union[List[int], List[List[int]]],
+            skip_special_tokens: bool = False,
+            clean_up_tokenization_spaces: bool = True,
+            spaces_between_special_tokens: bool = True,
+            **kwargs
+    ) -> str:
+        if isinstance(token_ids[0], list):
+            tokens = []
+            for single_token_ids in token_ids:
+                if self.pad_token_id in single_token_ids:  # remove pad
+                    single_token_ids = list(filter((self.pad_token_id).__ne__, single_token_ids))
+                tokens.append(self.sp_tokenizer.decode(single_token_ids))
+            return (tokens)
+        else:
+            if self.pad_token_id in token_ids:  # remove pad
+                token_ids = list(filter((self.pad_token_id).__ne__, token_ids))
+            return self.sp_tokenizer.decode(token_ids)
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+        return self.sp_tokenizer[token]
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.sp_tokenizer[index]
+
+    def save_vocabulary(self, save_directory, filename_prefix=None):
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+            filename_prefix (`str`, *optional*):
+                An optional prefix to add to the named of the saved files.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if os.path.isdir(save_directory):
+            vocab_file = os.path.join(
+                save_directory, self.vocab_files_names["vocab_file"]
+            )
+        else:
+            vocab_file = save_directory
+
+        with open(self.vocab_file, 'rb') as fin:
+            proto_str = fin.read()
+
+        with open(vocab_file, "wb") as writer:
+            writer.write(proto_str)
+
+        return (vocab_file,)
+
+    def build_inputs_with_special_tokens(
+            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
+        adding special tokens. A BERT sequence has the following format:
+
+        - single sequence: `[CLS] X [SEP]`
+        - pair of sequences: `[CLS] A [SEP] B [SEP]`
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs to which the special tokens will be added.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
+        """
+        mask_ids = self.sp_tokenizer[self.mask_token]
+        gmask_ids = self.sp_tokenizer[self.gmask_token]
+        eop_id = self.sp_tokenizer[self.eop_token]
+        if mask_ids not in token_ids_0 and gmask_ids not in token_ids_0:
+            token_ids_0 += [gmask_ids]
+
+        if token_ids_0[-1] != mask_ids and token_ids_0[-1] != gmask_ids:
+            token_ids_0 += [self.sp_tokenizer[self.eos_token]]
+
+        token_ids_0 += [self.sp_tokenizer[self.bos_token]]
+
+        if token_ids_1 is not None:
+            if not token_ids_1 or token_ids_1[-1] != eop_id:
+                token_ids_1 += [eop_id]
+            token_ids_0 += token_ids_1
+
+        return token_ids_0
+
+    def _pad(
+        self,
+        encoded_inputs: Union[Dict[str, EncodedInput], BatchEncoding],
+        max_length: Optional[int] = None,
+        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
+        pad_to_multiple_of: Optional[int] = None,
+        return_attention_mask: Optional[bool] = None,
+    ) -> dict:
+        """
+        Pad encoded inputs (on left/right and up to predefined length or max length in the batch)
+
+        Args:
+            encoded_inputs:
+                Dictionary of tokenized inputs (`List[int]`) or batch of tokenized inputs (`List[List[int]]`).
+            max_length: maximum length of the returned list and optionally padding length (see below).
+                Will truncate by taking into account the special tokens.
+            padding_strategy: PaddingStrategy to use for padding.
+
+                - PaddingStrategy.LONGEST Pad to the longest sequence in the batch
+                - PaddingStrategy.MAX_LENGTH: Pad to the max length (default)
+                - PaddingStrategy.DO_NOT_PAD: Do not pad
+                The tokenizer padding sides are defined in self.padding_side:
+
+                    - 'left': pads on the left of the sequences
+                    - 'right': pads on the right of the sequences
+            pad_to_multiple_of: (optional) Integer if set will pad the sequence to a multiple of the provided value.
+                This is especially useful to enable the use of Tensor Core on NVIDIA hardware with compute capability
+                `>= 7.5` (Volta).
+            return_attention_mask:
+                (optional) Set to False to avoid returning attention mask (default: set to model specifics)
+        """
+        # Load from model defaults
+        bos_token_id = self.sp_tokenizer[self.bos_token]
+        mask_token_id = self.sp_tokenizer[self.mask_token]
+        gmask_token_id = self.sp_tokenizer[self.gmask_token]
+        assert self.padding_side == "left"
+
+        required_input = encoded_inputs[self.model_input_names[0]]
+        seq_length = len(required_input)
+
+        if padding_strategy == PaddingStrategy.LONGEST:
+            max_length = len(required_input)
+
+        if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
+            max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of
+
+        needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) != max_length
+
+        # Initialize attention mask if not present.
+        if max_length is not None:
+            if "attention_mask" not in encoded_inputs:
+                if bos_token_id in required_input:
+                    context_length = required_input.index(bos_token_id)
+                else:
+                    context_length = seq_length
+                attention_mask = np.ones((1, seq_length, seq_length))
+                attention_mask = np.tril(attention_mask)
+                attention_mask[:, :, :context_length] = 1
+                attention_mask = np.bool_(attention_mask < 0.5)
+                encoded_inputs["attention_mask"] = attention_mask
+
+            if "position_ids" not in encoded_inputs:
+                position_ids = np.arange(seq_length, dtype=np.int64)
+                mask_token = mask_token_id if mask_token_id in required_input else gmask_token_id
+                if mask_token in required_input:
+                    mask_position = required_input.index(mask_token)
+                    position_ids[context_length:] = mask_position
+                block_position_ids = np.concatenate(
+                    [np.zeros(context_length, dtype=np.int64),
+                     np.arange(1, seq_length - context_length + 1, dtype=np.int64)])
+                encoded_inputs["position_ids"] = np.stack([position_ids, block_position_ids], axis=0)
+
+        if needs_to_be_padded:
+            difference = max_length - len(required_input)
+
+            if "attention_mask" in encoded_inputs:
+                encoded_inputs["attention_mask"] = np.pad(encoded_inputs["attention_mask"],
+                                                          pad_width=[(0, 0), (difference, 0), (difference, 0)],
+                                                          mode='constant', constant_values=True)
+            if "token_type_ids" in encoded_inputs:
+                encoded_inputs["token_type_ids"] = [self.pad_token_type_id] * difference + encoded_inputs[
+                    "token_type_ids"
+                ]
+            if "special_tokens_mask" in encoded_inputs:
+                encoded_inputs["special_tokens_mask"] = [1] * difference + encoded_inputs["special_tokens_mask"]
+            if "position_ids" in encoded_inputs:
+                encoded_inputs["position_ids"] = np.pad(encoded_inputs["position_ids"],
+                                                        pad_width=[(0, 0), (difference, 0)])
+            encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input
+
+        return encoded_inputs

tokenizer_config.json

@@ -0,0 +1,16 @@
+{
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_chatglm.ChatGLMTokenizer",
+      null
+    ]
+  },
+  "do_lower_case": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "padding_side": "left",
+  "remove_space": false,
+  "special_tokens_map_file": null,
+  "tokenizer_class": "ChatGLMTokenizer",
+  "unk_token": "<unk>"
+}

trainer_state.json

@@ -0,0 +1,1816 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 130.43478260869566,
+  "global_step": 3000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.019933333333333334,
+      "loss": 4.2601,
+      "step": 10
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.019866666666666668,
+      "loss": 3.5698,
+      "step": 20
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0198,
+      "loss": 3.3843,
+      "step": 30
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 0.019733333333333335,
+      "loss": 3.167,
+      "step": 40
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 0.019666666666666666,
+      "loss": 3.0689,
+      "step": 50
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 0.0196,
+      "loss": 2.814,
+      "step": 60
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 0.019533333333333333,
+      "loss": 2.7346,
+      "step": 70
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 0.019466666666666667,
+      "loss": 2.3168,
+      "step": 80
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 0.0194,
+      "loss": 2.3365,
+      "step": 90
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 0.019333333333333334,
+      "loss": 1.9153,
+      "step": 100
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 0.019266666666666668,
+      "loss": 1.8378,
+      "step": 110
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 0.0192,
+      "loss": 1.5231,
+      "step": 120
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 0.019133333333333332,
+      "loss": 1.278,
+      "step": 130
+    },
+    {
+      "epoch": 6.09,
+      "learning_rate": 0.01906666666666667,
+      "loss": 1.1264,
+      "step": 140
+    },
+    {
+      "epoch": 6.52,
+      "learning_rate": 0.019,
+      "loss": 0.8317,
+      "step": 150
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 0.018933333333333333,
+      "loss": 0.7888,
+      "step": 160
+    },
+    {
+      "epoch": 7.39,
+      "learning_rate": 0.018866666666666667,
+      "loss": 0.5529,
+      "step": 170
+    },
+    {
+      "epoch": 7.83,
+      "learning_rate": 0.0188,
+      "loss": 0.4765,
+      "step": 180
+    },
+    {
+      "epoch": 8.26,
+      "learning_rate": 0.018733333333333334,
+      "loss": 0.348,
+      "step": 190
+    },
+    {
+      "epoch": 8.7,
+      "learning_rate": 0.018666666666666668,
+      "loss": 0.2838,
+      "step": 200
+    },
+    {
+      "epoch": 9.13,
+      "learning_rate": 0.018600000000000002,
+      "loss": 0.2262,
+      "step": 210
+    },
+    {
+      "epoch": 9.57,
+      "learning_rate": 0.018533333333333332,
+      "loss": 0.1665,
+      "step": 220
+    },
+    {
+      "epoch": 10.0,
+      "learning_rate": 0.018466666666666666,
+      "loss": 0.1613,
+      "step": 230
+    },
+    {
+      "epoch": 10.43,
+      "learning_rate": 0.0184,
+      "loss": 0.1014,
+      "step": 240
+    },
+    {
+      "epoch": 10.87,
+      "learning_rate": 0.018333333333333333,
+      "loss": 0.104,
+      "step": 250
+    },
+    {
+      "epoch": 11.3,
+      "learning_rate": 0.018266666666666667,
+      "loss": 0.0812,
+      "step": 260
+    },
+    {
+      "epoch": 11.74,
+      "learning_rate": 0.0182,
+      "loss": 0.0777,
+      "step": 270
+    },
+    {
+      "epoch": 12.17,
+      "learning_rate": 0.01813333333333333,
+      "loss": 0.0651,
+      "step": 280
+    },
+    {
+      "epoch": 12.61,
+      "learning_rate": 0.01806666666666667,
+      "loss": 0.054,
+      "step": 290
+    },
+    {
+      "epoch": 13.04,
+      "learning_rate": 0.018000000000000002,
+      "loss": 0.0502,
+      "step": 300
+    },
+    {
+      "epoch": 13.48,
+      "learning_rate": 0.017933333333333332,
+      "loss": 0.0348,
+      "step": 310
+    },
+    {
+      "epoch": 13.91,
+      "learning_rate": 0.017866666666666666,
+      "loss": 0.0388,
+      "step": 320
+    },
+    {
+      "epoch": 14.35,
+      "learning_rate": 0.0178,
+      "loss": 0.0328,
+      "step": 330
+    },
+    {
+      "epoch": 14.78,
+      "learning_rate": 0.017733333333333334,
+      "loss": 0.0346,
+      "step": 340
+    },
+    {
+      "epoch": 15.22,
+      "learning_rate": 0.017666666666666667,
+      "loss": 0.029,
+      "step": 350
+    },
+    {
+      "epoch": 15.65,
+      "learning_rate": 0.0176,
+      "loss": 0.0342,
+      "step": 360
+    },
+    {
+      "epoch": 16.09,
+      "learning_rate": 0.017533333333333335,
+      "loss": 0.0317,
+      "step": 370
+    },
+    {
+      "epoch": 16.52,
+      "learning_rate": 0.017466666666666665,
+      "loss": 0.0253,
+      "step": 380
+    },
+    {
+      "epoch": 16.96,
+      "learning_rate": 0.0174,
+      "loss": 0.0237,
+      "step": 390
+    },
+    {
+      "epoch": 17.39,
+      "learning_rate": 0.017333333333333336,
+      "loss": 0.0216,
+      "step": 400
+    },
+    {
+      "epoch": 17.83,
+      "learning_rate": 0.017266666666666666,
+      "loss": 0.0238,
+      "step": 410
+    },
+    {
+      "epoch": 18.26,
+      "learning_rate": 0.0172,
+      "loss": 0.0203,
+      "step": 420
+    },
+    {
+      "epoch": 18.7,
+      "learning_rate": 0.017133333333333334,
+      "loss": 0.0254,
+      "step": 430
+    },
+    {
+      "epoch": 19.13,
+      "learning_rate": 0.017066666666666667,
+      "loss": 0.0231,
+      "step": 440
+    },
+    {
+      "epoch": 19.57,
+      "learning_rate": 0.017,
+      "loss": 0.0185,
+      "step": 450
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 0.016933333333333335,
+      "loss": 0.0205,
+      "step": 460
+    },
+    {
+      "epoch": 20.43,
+      "learning_rate": 0.01686666666666667,
+      "loss": 0.0183,
+      "step": 470
+    },
+    {
+      "epoch": 20.87,
+      "learning_rate": 0.0168,
+      "loss": 0.0246,
+      "step": 480
+    },
+    {
+      "epoch": 21.3,
+      "learning_rate": 0.016733333333333333,
+      "loss": 0.0181,
+      "step": 490
+    },
+    {
+      "epoch": 21.74,
+      "learning_rate": 0.016666666666666666,
+      "loss": 0.0198,
+      "step": 500
+    },
+    {
+      "epoch": 22.17,
+      "learning_rate": 0.0166,
+      "loss": 0.0282,
+      "step": 510
+    },
+    {
+      "epoch": 22.61,
+      "learning_rate": 0.016533333333333334,
+      "loss": 0.0161,
+      "step": 520
+    },
+    {
+      "epoch": 23.04,
+      "learning_rate": 0.016466666666666668,
+      "loss": 0.0158,
+      "step": 530
+    },
+    {
+      "epoch": 23.48,
+      "learning_rate": 0.016399999999999998,
+      "loss": 0.0138,
+      "step": 540
+    },
+    {
+      "epoch": 23.91,
+      "learning_rate": 0.01633333333333333,
+      "loss": 0.0191,
+      "step": 550
+    },
+    {
+      "epoch": 24.35,
+      "learning_rate": 0.01626666666666667,
+      "loss": 0.0147,
+      "step": 560
+    },
+    {
+      "epoch": 24.78,
+      "learning_rate": 0.016200000000000003,
+      "loss": 0.0146,
+      "step": 570
+    },
+    {
+      "epoch": 25.22,
+      "learning_rate": 0.016133333333333333,
+      "loss": 0.0116,
+      "step": 580
+    },
+    {
+      "epoch": 25.65,
+      "learning_rate": 0.016066666666666667,
+      "loss": 0.0126,
+      "step": 590
+    },
+    {
+      "epoch": 26.09,
+      "learning_rate": 0.016,
+      "loss": 0.0166,
+      "step": 600
+    },
+    {
+      "epoch": 26.52,
+      "learning_rate": 0.015933333333333334,
+      "loss": 0.014,
+      "step": 610
+    },
+    {
+      "epoch": 26.96,
+      "learning_rate": 0.015866666666666668,
+      "loss": 0.0105,
+      "step": 620
+    },
+    {
+      "epoch": 27.39,
+      "learning_rate": 0.0158,
+      "loss": 0.0098,
+      "step": 630
+    },
+    {
+      "epoch": 27.83,
+      "learning_rate": 0.015733333333333332,
+      "loss": 0.0134,
+      "step": 640
+    },
+    {
+      "epoch": 28.26,
+      "learning_rate": 0.015666666666666666,
+      "loss": 0.0138,
+      "step": 650
+    },
+    {
+      "epoch": 28.7,
+      "learning_rate": 0.015600000000000001,
+      "loss": 0.0132,
+      "step": 660
+    },
+    {
+      "epoch": 29.13,
+      "learning_rate": 0.015533333333333333,
+      "loss": 0.0106,
+      "step": 670
+    },
+    {
+      "epoch": 29.57,
+      "learning_rate": 0.015466666666666667,
+      "loss": 0.0101,
+      "step": 680
+    },
+    {
+      "epoch": 30.0,
+      "learning_rate": 0.0154,
+      "loss": 0.0095,
+      "step": 690
+    },
+    {
+      "epoch": 30.43,
+      "learning_rate": 0.015333333333333334,
+      "loss": 0.0079,
+      "step": 700
+    },
+    {
+      "epoch": 30.87,
+      "learning_rate": 0.015266666666666666,
+      "loss": 0.0107,
+      "step": 710
+    },
+    {
+      "epoch": 31.3,
+      "learning_rate": 0.0152,
+      "loss": 0.0089,
+      "step": 720
+    },
+    {
+      "epoch": 31.74,
+      "learning_rate": 0.015133333333333334,
+      "loss": 0.0095,
+      "step": 730
+    },
+    {
+      "epoch": 32.17,
+      "learning_rate": 0.015066666666666666,
+      "loss": 0.0073,
+      "step": 740
+    },
+    {
+      "epoch": 32.61,
+      "learning_rate": 0.015,
+      "loss": 0.0085,
+      "step": 750
+    },
+    {
+      "epoch": 33.04,
+      "learning_rate": 0.014933333333333335,
+      "loss": 0.0088,
+      "step": 760
+    },
+    {
+      "epoch": 33.48,
+      "learning_rate": 0.014866666666666667,
+      "loss": 0.0077,
+      "step": 770
+    },
+    {
+      "epoch": 33.91,
+      "learning_rate": 0.0148,
+      "loss": 0.0081,
+      "step": 780
+    },
+    {
+      "epoch": 34.35,
+      "learning_rate": 0.014733333333333334,
+      "loss": 0.0063,
+      "step": 790
+    },
+    {
+      "epoch": 34.78,
+      "learning_rate": 0.014666666666666666,
+      "loss": 0.0105,
+      "step": 800
+    },
+    {
+      "epoch": 35.22,
+      "learning_rate": 0.0146,
+      "loss": 0.007,
+      "step": 810
+    },
+    {
+      "epoch": 35.65,
+      "learning_rate": 0.014533333333333334,
+      "loss": 0.0075,
+      "step": 820
+    },
+    {
+      "epoch": 36.09,
+      "learning_rate": 0.014466666666666668,
+      "loss": 0.0092,
+      "step": 830
+    },
+    {
+      "epoch": 36.52,
+      "learning_rate": 0.0144,
+      "loss": 0.0074,
+      "step": 840
+    },
+    {
+      "epoch": 36.96,
+      "learning_rate": 0.014333333333333333,
+      "loss": 0.0099,
+      "step": 850
+    },
+    {
+      "epoch": 37.39,
+      "learning_rate": 0.014266666666666667,
+      "loss": 0.0077,
+      "step": 860
+    },
+    {
+      "epoch": 37.83,
+      "learning_rate": 0.014199999999999999,
+      "loss": 0.0156,
+      "step": 870
+    },
+    {
+      "epoch": 38.26,
+      "learning_rate": 0.014133333333333333,
+      "loss": 0.0103,
+      "step": 880
+    },
+    {
+      "epoch": 38.7,
+      "learning_rate": 0.014066666666666668,
+      "loss": 0.0118,
+      "step": 890
+    },
+    {
+      "epoch": 39.13,
+      "learning_rate": 0.013999999999999999,
+      "loss": 0.0109,
+      "step": 900
+    },
+    {
+      "epoch": 39.57,
+      "learning_rate": 0.013933333333333334,
+      "loss": 0.0104,
+      "step": 910
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 0.013866666666666668,
+      "loss": 0.0127,
+      "step": 920
+    },
+    {
+      "epoch": 40.43,
+      "learning_rate": 0.0138,
+      "loss": 0.0098,
+      "step": 930
+    },
+    {
+      "epoch": 40.87,
+      "learning_rate": 0.013733333333333334,
+      "loss": 0.0135,
+      "step": 940
+    },
+    {
+      "epoch": 41.3,
+      "learning_rate": 0.013666666666666667,
+      "loss": 0.0104,
+      "step": 950
+    },
+    {
+      "epoch": 41.74,
+      "learning_rate": 0.013600000000000001,
+      "loss": 0.0113,
+      "step": 960
+    },
+    {
+      "epoch": 42.17,
+      "learning_rate": 0.013533333333333333,
+      "loss": 0.0105,
+      "step": 970
+    },
+    {
+      "epoch": 42.61,
+      "learning_rate": 0.013466666666666667,
+      "loss": 0.0113,
+      "step": 980
+    },
+    {
+      "epoch": 43.04,
+      "learning_rate": 0.0134,
+      "loss": 0.0144,
+      "step": 990
+    },
+    {
+      "epoch": 43.48,
+      "learning_rate": 0.013333333333333332,
+      "loss": 0.0137,
+      "step": 1000
+    },
+    {
+      "epoch": 43.91,
+      "learning_rate": 0.013266666666666666,
+      "loss": 0.0181,
+      "step": 1010
+    },
+    {
+      "epoch": 44.35,
+      "learning_rate": 0.013200000000000002,
+      "loss": 0.0157,
+      "step": 1020
+    },
+    {
+      "epoch": 44.78,
+      "learning_rate": 0.013133333333333332,
+      "loss": 0.0178,
+      "step": 1030
+    },
+    {
+      "epoch": 45.22,
+      "learning_rate": 0.013066666666666667,
+      "loss": 0.0214,
+      "step": 1040
+    },
+    {
+      "epoch": 45.65,
+      "learning_rate": 0.013000000000000001,
+      "loss": 0.021,
+      "step": 1050
+    },
+    {
+      "epoch": 46.09,
+      "learning_rate": 0.012933333333333333,
+      "loss": 0.0282,
+      "step": 1060
+    },
+    {
+      "epoch": 46.52,
+      "learning_rate": 0.012866666666666667,
+      "loss": 0.023,
+      "step": 1070
+    },
+    {
+      "epoch": 46.96,
+      "learning_rate": 0.0128,
+      "loss": 0.0257,
+      "step": 1080
+    },
+    {
+      "epoch": 47.39,
+      "learning_rate": 0.012733333333333334,
+      "loss": 0.0199,
+      "step": 1090
+    },
+    {
+      "epoch": 47.83,
+      "learning_rate": 0.012666666666666666,
+      "loss": 0.0148,
+      "step": 1100
+    },
+    {
+      "epoch": 48.26,
+      "learning_rate": 0.0126,
+      "loss": 0.0167,
+      "step": 1110
+    },
+    {
+      "epoch": 48.7,
+      "learning_rate": 0.012533333333333334,
+      "loss": 0.0125,
+      "step": 1120
+    },
+    {
+      "epoch": 49.13,
+      "learning_rate": 0.012466666666666666,
+      "loss": 0.0134,
+      "step": 1130
+    },
+    {
+      "epoch": 49.57,
+      "learning_rate": 0.0124,
+      "loss": 0.0105,
+      "step": 1140
+    },
+    {
+      "epoch": 50.0,
+      "learning_rate": 0.012333333333333335,
+      "loss": 0.0125,
+      "step": 1150
+    },
+    {
+      "epoch": 50.43,
+      "learning_rate": 0.012266666666666665,
+      "loss": 0.0105,
+      "step": 1160
+    },
+    {
+      "epoch": 50.87,
+      "learning_rate": 0.0122,
+      "loss": 0.0119,
+      "step": 1170
+    },
+    {
+      "epoch": 51.3,
+      "learning_rate": 0.012133333333333335,
+      "loss": 0.0094,
+      "step": 1180
+    },
+    {
+      "epoch": 51.74,
+      "learning_rate": 0.012066666666666668,
+      "loss": 0.0089,
+      "step": 1190
+    },
+    {
+      "epoch": 52.17,
+      "learning_rate": 0.012,
+      "loss": 0.0085,
+      "step": 1200
+    },
+    {
+      "epoch": 52.61,
+      "learning_rate": 0.011933333333333334,
+      "loss": 0.007,
+      "step": 1210
+    },
+    {
+      "epoch": 53.04,
+      "learning_rate": 0.011866666666666668,
+      "loss": 0.0071,
+      "step": 1220
+    },
+    {
+      "epoch": 53.48,
+      "learning_rate": 0.0118,
+      "loss": 0.0089,
+      "step": 1230
+    },
+    {
+      "epoch": 53.91,
+      "learning_rate": 0.011733333333333333,
+      "loss": 0.0082,
+      "step": 1240
+    },
+    {
+      "epoch": 54.35,
+      "learning_rate": 0.011666666666666667,
+      "loss": 0.007,
+      "step": 1250
+    },
+    {
+      "epoch": 54.78,
+      "learning_rate": 0.0116,
+      "loss": 0.0071,
+      "step": 1260
+    },
+    {
+      "epoch": 55.22,
+      "learning_rate": 0.011533333333333333,
+      "loss": 0.0086,
+      "step": 1270
+    },
+    {
+      "epoch": 55.65,
+      "learning_rate": 0.011466666666666667,
+      "loss": 0.0114,
+      "step": 1280
+    },
+    {
+      "epoch": 56.09,
+      "learning_rate": 0.011399999999999999,
+      "loss": 0.0145,
+      "step": 1290
+    },
+    {
+      "epoch": 56.52,
+      "learning_rate": 0.011333333333333332,
+      "loss": 0.0109,
+      "step": 1300
+    },
+    {
+      "epoch": 56.96,
+      "learning_rate": 0.011266666666666668,
+      "loss": 0.0221,
+      "step": 1310
+    },
+    {
+      "epoch": 57.39,
+      "learning_rate": 0.011200000000000002,
+      "loss": 0.0102,
+      "step": 1320
+    },
+    {
+      "epoch": 57.83,
+      "learning_rate": 0.011133333333333334,
+      "loss": 0.009,
+      "step": 1330
+    },
+    {
+      "epoch": 58.26,
+      "learning_rate": 0.011066666666666667,
+      "loss": 0.0097,
+      "step": 1340
+    },
+    {
+      "epoch": 58.7,
+      "learning_rate": 0.011000000000000001,
+      "loss": 0.0094,
+      "step": 1350
+    },
+    {
+      "epoch": 59.13,
+      "learning_rate": 0.010933333333333333,
+      "loss": 0.0064,
+      "step": 1360
+    },
+    {
+      "epoch": 59.57,
+      "learning_rate": 0.010866666666666667,
+      "loss": 0.0061,
+      "step": 1370
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 0.0108,
+      "loss": 0.0078,
+      "step": 1380
+    },
+    {
+      "epoch": 60.43,
+      "learning_rate": 0.010733333333333333,
+      "loss": 0.005,
+      "step": 1390
+    },
+    {
+      "epoch": 60.87,
+      "learning_rate": 0.010666666666666666,
+      "loss": 0.0068,
+      "step": 1400
+    },
+    {
+      "epoch": 61.3,
+      "learning_rate": 0.0106,
+      "loss": 0.0051,
+      "step": 1410
+    },
+    {
+      "epoch": 61.74,
+      "learning_rate": 0.010533333333333332,
+      "loss": 0.0053,
+      "step": 1420
+    },
+    {
+      "epoch": 62.17,
+      "learning_rate": 0.010466666666666666,
+      "loss": 0.0061,
+      "step": 1430
+    },
+    {
+      "epoch": 62.61,
+      "learning_rate": 0.010400000000000001,
+      "loss": 0.0055,
+      "step": 1440
+    },
+    {
+      "epoch": 63.04,
+      "learning_rate": 0.010333333333333335,
+      "loss": 0.0063,
+      "step": 1450
+    },
+    {
+      "epoch": 63.48,
+      "learning_rate": 0.010266666666666667,
+      "loss": 0.0049,
+      "step": 1460
+    },
+    {
+      "epoch": 63.91,
+      "learning_rate": 0.0102,
+      "loss": 0.0071,
+      "step": 1470
+    },
+    {
+      "epoch": 64.35,
+      "learning_rate": 0.010133333333333334,
+      "loss": 0.0044,
+      "step": 1480
+    },
+    {
+      "epoch": 64.78,
+      "learning_rate": 0.010066666666666666,
+      "loss": 0.0056,
+      "step": 1490
+    },
+    {
+      "epoch": 65.22,
+      "learning_rate": 0.01,
+      "loss": 0.0044,
+      "step": 1500
+    },
+    {
+      "epoch": 65.65,
+      "learning_rate": 0.009933333333333334,
+      "loss": 0.0058,
+      "step": 1510
+    },
+    {
+      "epoch": 66.09,
+      "learning_rate": 0.009866666666666668,
+      "loss": 0.0056,
+      "step": 1520
+    },
+    {
+      "epoch": 66.52,
+      "learning_rate": 0.0098,
+      "loss": 0.0051,
+      "step": 1530
+    },
+    {
+      "epoch": 66.96,
+      "learning_rate": 0.009733333333333333,
+      "loss": 0.0065,
+      "step": 1540
+    },
+    {
+      "epoch": 67.39,
+      "learning_rate": 0.009666666666666667,
+      "loss": 0.0037,
+      "step": 1550
+    },
+    {
+      "epoch": 67.83,
+      "learning_rate": 0.0096,
+      "loss": 0.0056,
+      "step": 1560
+    },
+    {
+      "epoch": 68.26,
+      "learning_rate": 0.009533333333333335,
+      "loss": 0.0045,
+      "step": 1570
+    },
+    {
+      "epoch": 68.7,
+      "learning_rate": 0.009466666666666667,
+      "loss": 0.0059,
+      "step": 1580
+    },
+    {
+      "epoch": 69.13,
+      "learning_rate": 0.0094,
+      "loss": 0.0047,
+      "step": 1590
+    },
+    {
+      "epoch": 69.57,
+      "learning_rate": 0.009333333333333334,
+      "loss": 0.0051,
+      "step": 1600
+    },
+    {
+      "epoch": 70.0,
+      "learning_rate": 0.009266666666666666,
+      "loss": 0.0052,
+      "step": 1610
+    },
+    {
+      "epoch": 70.43,
+      "learning_rate": 0.0092,
+      "loss": 0.0041,
+      "step": 1620
+    },
+    {
+      "epoch": 70.87,
+      "learning_rate": 0.009133333333333334,
+      "loss": 0.0056,
+      "step": 1630
+    },
+    {
+      "epoch": 71.3,
+      "learning_rate": 0.009066666666666666,
+      "loss": 0.005,
+      "step": 1640
+    },
+    {
+      "epoch": 71.74,
+      "learning_rate": 0.009000000000000001,
+      "loss": 0.0051,
+      "step": 1650
+    },
+    {
+      "epoch": 72.17,
+      "learning_rate": 0.008933333333333333,
+      "loss": 0.0049,
+      "step": 1660
+    },
+    {
+      "epoch": 72.61,
+      "learning_rate": 0.008866666666666667,
+      "loss": 0.0046,
+      "step": 1670
+    },
+    {
+      "epoch": 73.04,
+      "learning_rate": 0.0088,
+      "loss": 0.0056,
+      "step": 1680
+    },
+    {
+      "epoch": 73.48,
+      "learning_rate": 0.008733333333333333,
+      "loss": 0.004,
+      "step": 1690
+    },
+    {
+      "epoch": 73.91,
+      "learning_rate": 0.008666666666666668,
+      "loss": 0.0043,
+      "step": 1700
+    },
+    {
+      "epoch": 74.35,
+      "learning_rate": 0.0086,
+      "loss": 0.0046,
+      "step": 1710
+    },
+    {
+      "epoch": 74.78,
+      "learning_rate": 0.008533333333333334,
+      "loss": 0.0058,
+      "step": 1720
+    },
+    {
+      "epoch": 75.22,
+      "learning_rate": 0.008466666666666667,
+      "loss": 0.0046,
+      "step": 1730
+    },
+    {
+      "epoch": 75.65,
+      "learning_rate": 0.0084,
+      "loss": 0.0056,
+      "step": 1740
+    },
+    {
+      "epoch": 76.09,
+      "learning_rate": 0.008333333333333333,
+      "loss": 0.0048,
+      "step": 1750
+    },
+    {
+      "epoch": 76.52,
+      "learning_rate": 0.008266666666666667,
+      "loss": 0.0033,
+      "step": 1760
+    },
+    {
+      "epoch": 76.96,
+      "learning_rate": 0.008199999999999999,
+      "loss": 0.0043,
+      "step": 1770
+    },
+    {
+      "epoch": 77.39,
+      "learning_rate": 0.008133333333333334,
+      "loss": 0.0034,
+      "step": 1780
+    },
+    {
+      "epoch": 77.83,
+      "learning_rate": 0.008066666666666666,
+      "loss": 0.0053,
+      "step": 1790
+    },
+    {
+      "epoch": 78.26,
+      "learning_rate": 0.008,
+      "loss": 0.0044,
+      "step": 1800
+    },
+    {
+      "epoch": 78.7,
+      "learning_rate": 0.007933333333333334,
+      "loss": 0.0036,
+      "step": 1810
+    },
+    {
+      "epoch": 79.13,
+      "learning_rate": 0.007866666666666666,
+      "loss": 0.0046,
+      "step": 1820
+    },
+    {
+      "epoch": 79.57,
+      "learning_rate": 0.0078000000000000005,
+      "loss": 0.0039,
+      "step": 1830
+    },
+    {
+      "epoch": 80.0,
+      "learning_rate": 0.007733333333333333,
+      "loss": 0.0042,
+      "step": 1840
+    },
+    {
+      "epoch": 80.43,
+      "learning_rate": 0.007666666666666667,
+      "loss": 0.004,
+      "step": 1850
+    },
+    {
+      "epoch": 80.87,
+      "learning_rate": 0.0076,
+      "loss": 0.0045,
+      "step": 1860
+    },
+    {
+      "epoch": 81.3,
+      "learning_rate": 0.007533333333333333,
+      "loss": 0.0035,
+      "step": 1870
+    },
+    {
+      "epoch": 81.74,
+      "learning_rate": 0.0074666666666666675,
+      "loss": 0.0036,
+      "step": 1880
+    },
+    {
+      "epoch": 82.17,
+      "learning_rate": 0.0074,
+      "loss": 0.0045,
+      "step": 1890
+    },
+    {
+      "epoch": 82.61,
+      "learning_rate": 0.007333333333333333,
+      "loss": 0.0033,
+      "step": 1900
+    },
+    {
+      "epoch": 83.04,
+      "learning_rate": 0.007266666666666667,
+      "loss": 0.0044,
+      "step": 1910
+    },
+    {
+      "epoch": 83.48,
+      "learning_rate": 0.0072,
+      "loss": 0.0037,
+      "step": 1920
+    },
+    {
+      "epoch": 83.91,
+      "learning_rate": 0.0071333333333333335,
+      "loss": 0.004,
+      "step": 1930
+    },
+    {
+      "epoch": 84.35,
+      "learning_rate": 0.007066666666666666,
+      "loss": 0.0043,
+      "step": 1940
+    },
+    {
+      "epoch": 84.78,
+      "learning_rate": 0.006999999999999999,
+      "loss": 0.0039,
+      "step": 1950
+    },
+    {
+      "epoch": 85.22,
+      "learning_rate": 0.006933333333333334,
+      "loss": 0.0044,
+      "step": 1960
+    },
+    {
+      "epoch": 85.65,
+      "learning_rate": 0.006866666666666667,
+      "loss": 0.003,
+      "step": 1970
+    },
+    {
+      "epoch": 86.09,
+      "learning_rate": 0.0068000000000000005,
+      "loss": 0.0054,
+      "step": 1980
+    },
+    {
+      "epoch": 86.52,
+      "learning_rate": 0.006733333333333333,
+      "loss": 0.0041,
+      "step": 1990
+    },
+    {
+      "epoch": 86.96,
+      "learning_rate": 0.006666666666666666,
+      "loss": 0.0055,
+      "step": 2000
+    },
+    {
+      "epoch": 87.39,
+      "learning_rate": 0.006600000000000001,
+      "loss": 0.0028,
+      "step": 2010
+    },
+    {
+      "epoch": 87.83,
+      "learning_rate": 0.006533333333333334,
+      "loss": 0.0045,
+      "step": 2020
+    },
+    {
+      "epoch": 88.26,
+      "learning_rate": 0.006466666666666667,
+      "loss": 0.0035,
+      "step": 2030
+    },
+    {
+      "epoch": 88.7,
+      "learning_rate": 0.0064,
+      "loss": 0.004,
+      "step": 2040
+    },
+    {
+      "epoch": 89.13,
+      "learning_rate": 0.006333333333333333,
+      "loss": 0.0035,
+      "step": 2050
+    },
+    {
+      "epoch": 89.57,
+      "learning_rate": 0.006266666666666667,
+      "loss": 0.004,
+      "step": 2060
+    },
+    {
+      "epoch": 90.0,
+      "learning_rate": 0.0062,
+      "loss": 0.0039,
+      "step": 2070
+    },
+    {
+      "epoch": 90.43,
+      "learning_rate": 0.006133333333333333,
+      "loss": 0.0037,
+      "step": 2080
+    },
+    {
+      "epoch": 90.87,
+      "learning_rate": 0.006066666666666667,
+      "loss": 0.003,
+      "step": 2090
+    },
+    {
+      "epoch": 91.3,
+      "learning_rate": 0.006,
+      "loss": 0.0043,
+      "step": 2100
+    },
+    {
+      "epoch": 91.74,
+      "learning_rate": 0.005933333333333334,
+      "loss": 0.0033,
+      "step": 2110
+    },
+    {
+      "epoch": 92.17,
+      "learning_rate": 0.005866666666666667,
+      "loss": 0.0037,
+      "step": 2120
+    },
+    {
+      "epoch": 92.61,
+      "learning_rate": 0.0058,
+      "loss": 0.0043,
+      "step": 2130
+    },
+    {
+      "epoch": 93.04,
+      "learning_rate": 0.005733333333333333,
+      "loss": 0.0032,
+      "step": 2140
+    },
+    {
+      "epoch": 93.48,
+      "learning_rate": 0.005666666666666666,
+      "loss": 0.003,
+      "step": 2150
+    },
+    {
+      "epoch": 93.91,
+      "learning_rate": 0.005600000000000001,
+      "loss": 0.0043,
+      "step": 2160
+    },
+    {
+      "epoch": 94.35,
+      "learning_rate": 0.005533333333333334,
+      "loss": 0.0044,
+      "step": 2170
+    },
+    {
+      "epoch": 94.78,
+      "learning_rate": 0.0054666666666666665,
+      "loss": 0.0037,
+      "step": 2180
+    },
+    {
+      "epoch": 95.22,
+      "learning_rate": 0.0054,
+      "loss": 0.0062,
+      "step": 2190
+    },
+    {
+      "epoch": 95.65,
+      "learning_rate": 0.005333333333333333,
+      "loss": 0.0031,
+      "step": 2200
+    },
+    {
+      "epoch": 96.09,
+      "learning_rate": 0.005266666666666666,
+      "loss": 0.0046,
+      "step": 2210
+    },
+    {
+      "epoch": 96.52,
+      "learning_rate": 0.005200000000000001,
+      "loss": 0.0036,
+      "step": 2220
+    },
+    {
+      "epoch": 96.96,
+      "learning_rate": 0.0051333333333333335,
+      "loss": 0.0026,
+      "step": 2230
+    },
+    {
+      "epoch": 97.39,
+      "learning_rate": 0.005066666666666667,
+      "loss": 0.0039,
+      "step": 2240
+    },
+    {
+      "epoch": 97.83,
+      "learning_rate": 0.005,
+      "loss": 0.0045,
+      "step": 2250
+    },
+    {
+      "epoch": 98.26,
+      "learning_rate": 0.004933333333333334,
+      "loss": 0.004,
+      "step": 2260
+    },
+    {
+      "epoch": 98.7,
+      "learning_rate": 0.004866666666666667,
+      "loss": 0.0045,
+      "step": 2270
+    },
+    {
+      "epoch": 99.13,
+      "learning_rate": 0.0048,
+      "loss": 0.0028,
+      "step": 2280
+    },
+    {
+      "epoch": 99.57,
+      "learning_rate": 0.004733333333333333,
+      "loss": 0.0039,
+      "step": 2290
+    },
+    {
+      "epoch": 100.0,
+      "learning_rate": 0.004666666666666667,
+      "loss": 0.004,
+      "step": 2300
+    },
+    {
+      "epoch": 100.43,
+      "learning_rate": 0.0046,
+      "loss": 0.003,
+      "step": 2310
+    },
+    {
+      "epoch": 100.87,
+      "learning_rate": 0.004533333333333333,
+      "loss": 0.0043,
+      "step": 2320
+    },
+    {
+      "epoch": 101.3,
+      "learning_rate": 0.0044666666666666665,
+      "loss": 0.0037,
+      "step": 2330
+    },
+    {
+      "epoch": 101.74,
+      "learning_rate": 0.0044,
+      "loss": 0.003,
+      "step": 2340
+    },
+    {
+      "epoch": 102.17,
+      "learning_rate": 0.004333333333333334,
+      "loss": 0.0037,
+      "step": 2350
+    },
+    {
+      "epoch": 102.61,
+      "learning_rate": 0.004266666666666667,
+      "loss": 0.0032,
+      "step": 2360
+    },
+    {
+      "epoch": 103.04,
+      "learning_rate": 0.0042,
+      "loss": 0.0037,
+      "step": 2370
+    },
+    {
+      "epoch": 103.48,
+      "learning_rate": 0.0041333333333333335,
+      "loss": 0.0041,
+      "step": 2380
+    },
+    {
+      "epoch": 103.91,
+      "learning_rate": 0.004066666666666667,
+      "loss": 0.0038,
+      "step": 2390
+    },
+    {
+      "epoch": 104.35,
+      "learning_rate": 0.004,
+      "loss": 0.0022,
+      "step": 2400
+    },
+    {
+      "epoch": 104.78,
+      "learning_rate": 0.003933333333333333,
+      "loss": 0.0033,
+      "step": 2410
+    },
+    {
+      "epoch": 105.22,
+      "learning_rate": 0.0038666666666666667,
+      "loss": 0.0032,
+      "step": 2420
+    },
+    {
+      "epoch": 105.65,
+      "learning_rate": 0.0038,
+      "loss": 0.0039,
+      "step": 2430
+    },
+    {
+      "epoch": 106.09,
+      "learning_rate": 0.0037333333333333337,
+      "loss": 0.0036,
+      "step": 2440
+    },
+    {
+      "epoch": 106.52,
+      "learning_rate": 0.0036666666666666666,
+      "loss": 0.0031,
+      "step": 2450
+    },
+    {
+      "epoch": 106.96,
+      "learning_rate": 0.0036,
+      "loss": 0.0036,
+      "step": 2460
+    },
+    {
+      "epoch": 107.39,
+      "learning_rate": 0.003533333333333333,
+      "loss": 0.0039,
+      "step": 2470
+    },
+    {
+      "epoch": 107.83,
+      "learning_rate": 0.003466666666666667,
+      "loss": 0.0034,
+      "step": 2480
+    },
+    {
+      "epoch": 108.26,
+      "learning_rate": 0.0034000000000000002,
+      "loss": 0.0033,
+      "step": 2490
+    },
+    {
+      "epoch": 108.7,
+      "learning_rate": 0.003333333333333333,
+      "loss": 0.0034,
+      "step": 2500
+    },
+    {
+      "epoch": 109.13,
+      "learning_rate": 0.003266666666666667,
+      "loss": 0.003,
+      "step": 2510
+    },
+    {
+      "epoch": 109.57,
+      "learning_rate": 0.0032,
+      "loss": 0.0033,
+      "step": 2520
+    },
+    {
+      "epoch": 110.0,
+      "learning_rate": 0.0031333333333333335,
+      "loss": 0.0038,
+      "step": 2530
+    },
+    {
+      "epoch": 110.43,
+      "learning_rate": 0.0030666666666666663,
+      "loss": 0.003,
+      "step": 2540
+    },
+    {
+      "epoch": 110.87,
+      "learning_rate": 0.003,
+      "loss": 0.0032,
+      "step": 2550
+    },
+    {
+      "epoch": 111.3,
+      "learning_rate": 0.0029333333333333334,
+      "loss": 0.004,
+      "step": 2560
+    },
+    {
+      "epoch": 111.74,
+      "learning_rate": 0.0028666666666666667,
+      "loss": 0.0029,
+      "step": 2570
+    },
+    {
+      "epoch": 112.17,
+      "learning_rate": 0.0028000000000000004,
+      "loss": 0.0028,
+      "step": 2580
+    },
+    {
+      "epoch": 112.61,
+      "learning_rate": 0.0027333333333333333,
+      "loss": 0.0034,
+      "step": 2590
+    },
+    {
+      "epoch": 113.04,
+      "learning_rate": 0.0026666666666666666,
+      "loss": 0.003,
+      "step": 2600
+    },
+    {
+      "epoch": 113.48,
+      "learning_rate": 0.0026000000000000003,
+      "loss": 0.0027,
+      "step": 2610
+    },
+    {
+      "epoch": 113.91,
+      "learning_rate": 0.0025333333333333336,
+      "loss": 0.0037,
+      "step": 2620
+    },
+    {
+      "epoch": 114.35,
+      "learning_rate": 0.002466666666666667,
+      "loss": 0.003,
+      "step": 2630
+    },
+    {
+      "epoch": 114.78,
+      "learning_rate": 0.0024,
+      "loss": 0.0032,
+      "step": 2640
+    },
+    {
+      "epoch": 115.22,
+      "learning_rate": 0.0023333333333333335,
+      "loss": 0.0029,
+      "step": 2650
+    },
+    {
+      "epoch": 115.65,
+      "learning_rate": 0.0022666666666666664,
+      "loss": 0.0027,
+      "step": 2660
+    },
+    {
+      "epoch": 116.09,
+      "learning_rate": 0.0022,
+      "loss": 0.004,
+      "step": 2670
+    },
+    {
+      "epoch": 116.52,
+      "learning_rate": 0.0021333333333333334,
+      "loss": 0.0032,
+      "step": 2680
+    },
+    {
+      "epoch": 116.96,
+      "learning_rate": 0.0020666666666666667,
+      "loss": 0.0032,
+      "step": 2690
+    },
+    {
+      "epoch": 117.39,
+      "learning_rate": 0.002,
+      "loss": 0.0032,
+      "step": 2700
+    },
+    {
+      "epoch": 117.83,
+      "learning_rate": 0.0019333333333333333,
+      "loss": 0.0034,
+      "step": 2710
+    },
+    {
+      "epoch": 118.26,
+      "learning_rate": 0.0018666666666666669,
+      "loss": 0.0038,
+      "step": 2720
+    },
+    {
+      "epoch": 118.7,
+      "learning_rate": 0.0018,
+      "loss": 0.0028,
+      "step": 2730
+    },
+    {
+      "epoch": 119.13,
+      "learning_rate": 0.0017333333333333335,
+      "loss": 0.0027,
+      "step": 2740
+    },
+    {
+      "epoch": 119.57,
+      "learning_rate": 0.0016666666666666666,
+      "loss": 0.0032,
+      "step": 2750
+    },
+    {
+      "epoch": 120.0,
+      "learning_rate": 0.0016,
+      "loss": 0.0029,
+      "step": 2760
+    },
+    {
+      "epoch": 120.43,
+      "learning_rate": 0.0015333333333333332,
+      "loss": 0.0033,
+      "step": 2770
+    },
+    {
+      "epoch": 120.87,
+      "learning_rate": 0.0014666666666666667,
+      "loss": 0.0036,
+      "step": 2780
+    },
+    {
+      "epoch": 121.3,
+      "learning_rate": 0.0014000000000000002,
+      "loss": 0.0026,
+      "step": 2790
+    },
+    {
+      "epoch": 121.74,
+      "learning_rate": 0.0013333333333333333,
+      "loss": 0.0028,
+      "step": 2800
+    },
+    {
+      "epoch": 122.17,
+      "learning_rate": 0.0012666666666666668,
+      "loss": 0.0026,
+      "step": 2810
+    },
+    {
+      "epoch": 122.61,
+      "learning_rate": 0.0012,
+      "loss": 0.0029,
+      "step": 2820
+    },
+    {
+      "epoch": 123.04,
+      "learning_rate": 0.0011333333333333332,
+      "loss": 0.0036,
+      "step": 2830
+    },
+    {
+      "epoch": 123.48,
+      "learning_rate": 0.0010666666666666667,
+      "loss": 0.0032,
+      "step": 2840
+    },
+    {
+      "epoch": 123.91,
+      "learning_rate": 0.001,
+      "loss": 0.0032,
+      "step": 2850
+    },
+    {
+      "epoch": 124.35,
+      "learning_rate": 0.0009333333333333334,
+      "loss": 0.0027,
+      "step": 2860
+    },
+    {
+      "epoch": 124.78,
+      "learning_rate": 0.0008666666666666667,
+      "loss": 0.0028,
+      "step": 2870
+    },
+    {
+      "epoch": 125.22,
+      "learning_rate": 0.0008,
+      "loss": 0.0027,
+      "step": 2880
+    },
+    {
+      "epoch": 125.65,
+      "learning_rate": 0.0007333333333333333,
+      "loss": 0.0033,
+      "step": 2890
+    },
+    {
+      "epoch": 126.09,
+      "learning_rate": 0.0006666666666666666,
+      "loss": 0.0027,
+      "step": 2900
+    },
+    {
+      "epoch": 126.52,
+      "learning_rate": 0.0006,
+      "loss": 0.0029,
+      "step": 2910
+    },
+    {
+      "epoch": 126.96,
+      "learning_rate": 0.0005333333333333334,
+      "loss": 0.003,
+      "step": 2920
+    },
+    {
+      "epoch": 127.39,
+      "learning_rate": 0.0004666666666666667,
+      "loss": 0.0031,
+      "step": 2930
+    },
+    {
+      "epoch": 127.83,
+      "learning_rate": 0.0004,
+      "loss": 0.0027,
+      "step": 2940
+    },
+    {
+      "epoch": 128.26,
+      "learning_rate": 0.0003333333333333333,
+      "loss": 0.0032,
+      "step": 2950
+    },
+    {
+      "epoch": 128.7,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.0027,
+      "step": 2960
+    },
+    {
+      "epoch": 129.13,
+      "learning_rate": 0.0002,
+      "loss": 0.0027,
+      "step": 2970
+    },
+    {
+      "epoch": 129.57,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.003,
+      "step": 2980
+    },
+    {
+      "epoch": 130.0,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 0.0033,
+      "step": 2990
+    },
+    {
+      "epoch": 130.43,
+      "learning_rate": 0.0,
+      "loss": 0.0028,
+      "step": 3000
+    }
+  ],
+  "max_steps": 3000,
+  "num_train_epochs": 131,
+  "total_flos": 1.0315218674515968e+17,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af29f237716b9a6b665fa2af1ffaca0c4e272abf51a26e1032103f104f27e507
+size 3707