1

app.py

@@ -0,0 +1,121 @@
+import os
+import platform
+import signal
+from transformers import AutoTokenizer, AutoModel, AutoConfig, AutoModelForCausalLM
+import readline
+import torch
+from accelerate import infer_auto_device_map, init_empty_weights, load_checkpoint_and_dispatch
+import gradio as gr
+import time
+
+model_path = "THUDM/chatglm-6b-int4"
+# 载入Tokenizer
+tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
+# Fine-tuning 后的表现测试
+config = AutoConfig.from_pretrained(model_path, trust_remote_code=True, pre_seq_len=128)
+model = AutoModel.from_pretrained(model_path, config=config, trust_remote_code=True)
+# 此处使用你的 ptuning 工作目录
+prefix_state_dict = torch.load(os.path.join("./xiaowo", "pytorch_model.bin"))
+new_prefix_state_dict = {}
+for k, v in prefix_state_dict.items():
+    new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
+model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
+
+model = model.float()
+model.transformer.prefix_encoder.float()
+model = model.eval()
+
+#剩下的直接抄web_demo.py了家人们
+"""Override Chatbot.postprocess"""
+
+
+def postprocess(self, y):
+    if y is None:
+        return []
+    for i, (message, response) in enumerate(y):
+        y[i] = (
+            None if message is None else mdtex2html.convert((message)),
+            None if response is None else mdtex2html.convert(response),
+        )
+    return y
+
+
+gr.Chatbot.postprocess = postprocess
+
+
+def parse_text(text):
+    """copy from https://github.com/GaiZhenbiao/ChuanhuChatGPT/"""
+    lines = text.split("\n")
+    lines = [line for line in lines if line != ""]
+    count = 0
+    for i, line in enumerate(lines):
+        if "```" in line:
+            count += 1
+            items = line.split('`')
+            if count % 2 == 1:
+                lines[i] = f'<pre><code class="language-{items[-1]}">'
+            else:
+                lines[i] = f'<br></code></pre>'
+        else:
+            if i > 0:
+                if count % 2 == 1:
+                    line = line.replace("`", "\`")
+                    line = line.replace("<", "&lt;")
+                    line = line.replace(">", "&gt;")
+                    line = line.replace(" ", "&nbsp;")
+                    line = line.replace("*", "&ast;")
+                    line = line.replace("_", "&lowbar;")
+                    line = line.replace("-", "&#45;")
+                    line = line.replace(".", "&#46;")
+                    line = line.replace("!", "&#33;")
+                    line = line.replace("(", "&#40;")
+                    line = line.replace(")", "&#41;")
+                    line = line.replace("$", "&#36;")
+                lines[i] = "<br>"+line
+    text = "".join(lines)
+    return text
+
+
+def predict(input, chatbot, max_length, top_p, temperature, history):
+    chatbot.append((parse_text(input), ""))
+    for response, history in model.stream_chat(tokenizer, input, history, max_length=max_length, top_p=top_p,
+                                               temperature=temperature):
+        chatbot[-1] = (parse_text(input), parse_text(response))
+
+        yield chatbot, history
+
+
+def reset_user_input():
+    return gr.update(value='')
+
+
+def reset_state():
+    return [], []
+
+
+with gr.Blocks() as demo:
+    gr.HTML("""<h1 align="center">ChatGLM</h1>""")
+
+    chatbot = gr.Chatbot()
+    with gr.Row():
+        with gr.Column(scale=4):
+            with gr.Column(scale=12):
+                user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=10).style(
+                    container=False)
+            with gr.Column(min_width=32, scale=1):
+                submitBtn = gr.Button("Submit", variant="primary")
+        with gr.Column(scale=1):
+            emptyBtn = gr.Button("Clear History")
+            max_length = gr.Slider(0, 4096, value=2048, step=1.0, label="Maximum length", interactive=True)
+            top_p = gr.Slider(0, 1, value=0.7, step=0.01, label="Top P", interactive=True)
+            temperature = gr.Slider(0, 1, value=0.95, step=0.01, label="Temperature", interactive=True)
+
+    history = gr.State([])
+
+    submitBtn.click(predict, [user_input, chatbot, max_length, top_p, temperature, history], [chatbot, history],
+                    show_progress=True)
+    submitBtn.click(reset_user_input, [], [user_input])
+
+    emptyBtn.click(reset_state, outputs=[chatbot, history], show_progress=True)
+
+demo.queue().launch(share=False, inbrowser=True)

requirements.txt

@@ -0,0 +1,8 @@
+protobuf
+transformers==4.27.1
+cpm_kernels
+torch>=1.10
+gradio
+mdtex2html
+sentencepiece
+accelerate

utils.py

@@ -0,0 +1,54 @@
+import os
+from typing import Dict, Tuple, Union, Optional
+
+from torch.nn import Module
+from transformers import AutoModel
+
+
+def auto_configure_device_map(num_gpus: int) -> Dict[str, int]:
+    # transformer.word_embeddings 占用1层
+    # transformer.final_layernorm 和 lm_head 占用1层
+    # transformer.layers 占用 28 层
+    # 总共30层分配到num_gpus张卡上
+    num_trans_layers = 28
+    per_gpu_layers = 30 / num_gpus
+
+    # bugfix: 在linux中调用torch.embedding传入的weight,input不在同一device上,导致RuntimeError
+    # windows下 model.device 会被设置成 transformer.word_embeddings.device
+    # linux下 model.device 会被设置成 lm_head.device
+    # 在调用chat或者stream_chat时,input_ids会被放到model.device上
+    # 如果transformer.word_embeddings.device和model.device不同,则会导致RuntimeError
+    # 因此这里将transformer.word_embeddings,transformer.final_layernorm,lm_head都放到第一张卡上
+    device_map = {'transformer.word_embeddings': 0,
+                  'transformer.final_layernorm': 0, 'lm_head': 0}
+
+    used = 2
+    gpu_target = 0
+    for i in range(num_trans_layers):
+        if used >= per_gpu_layers:
+            gpu_target += 1
+            used = 0
+        assert gpu_target < num_gpus
+        device_map[f'transformer.layers.{i}'] = gpu_target
+        used += 1
+
+    return device_map
+
+
+def load_model_on_gpus(checkpoint_path: Union[str, os.PathLike], num_gpus: int = 2,
+                       device_map: Optional[Dict[str, int]] = None, **kwargs) -> Module:
+    if num_gpus < 2 and device_map is None:
+        model = AutoModel.from_pretrained(checkpoint_path, trust_remote_code=True, **kwargs).half().cuda()
+    else:
+        from accelerate import dispatch_model
+
+        model = AutoModel.from_pretrained(checkpoint_path, trust_remote_code=True, **kwargs).half()
+
+        if device_map is None:
+            device_map = auto_configure_device_map(num_gpus)
+
+        model = dispatch_model(model, device_map=device_map)
+
+    return model
+
+

xiaowo/config.json

@@ -0,0 +1,32 @@
+{
+  "_name_or_path": "THUDM/chatglm-6b-int4",
+  "architectures": [
+    "ChatGLMForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_chatglm.ChatGLMConfig",
+    "AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration",
+    "AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration"
+  },
+  "bos_token_id": 130004,
+  "eos_token_id": 130005,
+  "gmask_token_id": 130001,
+  "hidden_size": 4096,
+  "inner_hidden_size": 16384,
+  "layernorm_epsilon": 1e-05,
+  "mask_token_id": 130000,
+  "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,
+  "quantization_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.27.1",
+  "use_cache": true,
+  "vocab_size": 130528
+}

xiaowo/configuration_chatglm.py

@@ -0,0 +1,105 @@
+""" 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.
+
+    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=150528,
+            hidden_size=4096,
+            num_layers=28,
+            num_attention_heads=32,
+            layernorm_epsilon=1e-5,
+            use_cache=False,
+            bos_token_id=150004,
+            eos_token_id=150005,
+            mask_token_id=150000,
+            gmask_token_id=150001,
+            pad_token_id=0,
+            max_sequence_length=2048,
+            inner_hidden_size=16384,
+            position_encoding_2d=True,
+            quantization_bit=0,
+            quantization_embeddings=False,
+            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.mask_token_id = mask_token_id
+        self.gmask_token_id = gmask_token_id
+        self.position_encoding_2d = position_encoding_2d
+        self.quantization_bit = quantization_bit
+        self.quantization_embeddings = quantization_embeddings
+        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
+        )

xiaowo/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"
+}

xiaowo/ice_text.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e974d9a69c242ce014c88c2b26089270f6198f3c0b700a887666cd3e816f17e
+size 2706249

xiaowo/modeling_chatglm.py

@@ -0,0 +1,1471 @@
+""" 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 = "THUDM/ChatGLM-6B"
+_CONFIG_FOR_DOC = "ChatGLM6BConfig"
+
+CHATGLM_6B_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "THUDM/chatglm-6b",
+    # See all ChatGLM-6B models at https://huggingface.co/models?filter=chatglm
+]
+
+
+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.zeros(
+        1, 1, 1,
+        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
+
+
+def default_init(cls, *args, **kwargs):
+    return cls(*args, **kwargs)
+
+
+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, empty_init=True):
+        if empty_init:
+            init_method = skip_init
+        else:
+            init_method = default_init
+        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 = init_method(
+            torch.nn.Linear,
+            hidden_size,
+            3 * self.inner_hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+
+        self.dense = init_method(
+            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, empty_init=True):
+        super(GLU, self).__init__()
+        if empty_init:
+            init_method = skip_init
+        else:
+            init_method = default_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 = init_method(
+            torch.nn.Linear,
+            self.hidden_size,
+            self.inner_hidden_size,
+            bias=bias,
+            dtype=params_dtype,
+        )
+        # Project back to h.
+        self.dense_4h_to_h = init_method(
+            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,
+            empty_init=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,
+            empty_init=empty_init
+        )
+
+        # 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,
+            empty_init=empty_init
+        )
+
+    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, use_gmasks=None):
+        batch_size, seq_length = input_ids.shape
+        if use_gmasks is None:
+            use_gmasks = [False] * batch_size
+        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)
+            for i, context_length in enumerate(context_lengths):
+                if not use_gmasks[i]:
+                    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, empty_init=True):
+        super().__init__(config)
+        if empty_init:
+            init_method = skip_init
+        else:
+            init_method = default_init
+        # 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 = init_method(
+            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,
+                empty_init=empty_init
+            )
+
+        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 = self.config.mask_token_id, self.config.gmask_token_id
+                seqs = input_ids.tolist()
+
+                mask_positions, use_gmasks = [], []
+                for seq in seqs:
+                    mask_token = gMASK if gMASK in seq else MASK
+                    use_gmask = mask_token == gMASK
+                    mask_positions.append(seq.index(mask_token))
+                    use_gmasks.append(use_gmask)
+
+                position_ids = self.get_position_ids(
+                    input_ids,
+                    mask_positions=mask_positions,
+                    device=input_ids.device,
+                    use_gmasks=use_gmasks
+                )
+
+        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(hidden_states.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, empty_init=True):
+        super().__init__(config)
+        if empty_init:
+            init_method = skip_init
+        else:
+            init_method = default_init
+
+        # 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, empty_init=empty_init)
+
+        self.lm_head = init_method(
+            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, self.config.quantization_embeddings, use_quantization_cache=True, 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 = self.config.mask_token_id, self.config.gmask_token_id
+        seqs = input_ids.tolist()
+        mask_positions, use_gmasks = [], []
+        for seq in seqs:
+            mask_token = gMASK if gMASK in seq else MASK
+            use_gmask = mask_token == gMASK
+            mask_positions.append(seq.index(mask_token))
+            use_gmasks.append(use_gmask)
+
+        # 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,
+                    use_gmasks=use_gmasks
+                )
+
+            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, quantize_embeddings=False, use_quantization_cache=False, empty_init=False, **kwargs):
+        if bits == 0:
+            return
+
+        from .quantization import quantize, QuantizedEmbedding, QuantizedLinear, load_cpu_kernel
+
+        if self.quantized:
+            if self.device == torch.device("cpu"):
+                logger.info("Already quantized, reloading cpu kernel.")
+                load_cpu_kernel(**kwargs)
+            else:
+                logger.info("Already quantized.")
+            return self
+
+        self.quantized = True
+
+        self.config.quantization_bit = bits
+        self.config.quantization_embeddings = quantize_embeddings
+
+        self.transformer = quantize(self.transformer, bits, use_quantization_cache=use_quantization_cache, empty_init=empty_init, **kwargs)
+
+        if self.device == torch.device("cpu"):
+            dtype = torch.float32
+        else:
+            dtype = torch.half
+
+        if quantize_embeddings:
+            logger.info("Applying quantization to embeddings")
+            self.transformer.word_embeddings = QuantizedEmbedding(
+                weight_bit_width=bits,
+                weight_tensor=self.transformer.word_embeddings.weight.to(self.device),
+                num_embeddings=self.transformer.word_embeddings.num_embeddings,
+                embedding_dim=self.transformer.word_embeddings.embedding_dim,
+                dtype=dtype,
+                empty_init=empty_init,
+                device=self.transformer.word_embeddings.weight.device,
+            )
+            self.lm_head = QuantizedLinear(
+                weight_bit_width=bits,
+                weight_tensor=self.lm_head.weight.to(self.device),
+                bias_tensor=None,
+                in_features=self.lm_head.in_features,
+                out_features=self.lm_head.out_features,
+                bias=False,
+                quantized_weight=self.transformer.word_embeddings.weight,
+                quantized_weight_scale=self.transformer.word_embeddings.weight_scale,
+                dtype=dtype,
+                empty_init=empty_init,
+                device=self.lm_head.weight.device,
+            )
+
+        return self

xiaowo/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7736d0582427d706ac424dcfe385990b4ba35f6481b446ae1fcaf041cc5e662
+size 234882351

xiaowo/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c64795fba60fdadcfd8791a3b5dd9fd877febb3c560b99a25aab42b8118421d5
+size 117441341

xiaowo/quantization.py

@@ -0,0 +1,533 @@
+from torch.nn import Linear, Embedding
+from torch.nn.parameter import Parameter
+import torch.nn.functional as F
+
+import os
+import bz2
+import torch
+import base64
+import ctypes
+import sys
+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:", 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
+
+
+class W8A16LinearCPU(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, inp: torch.Tensor, quant_w: torch.Tensor, scale_w: torch.Tensor, weight_bit_width,
+                quantization_cache=None):
+        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_float(quant_w, scale_w, weight_bit_width, quantization_cache=quantization_cache)
+        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_float(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
+
+
+default_cpu_kernel_code_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "quantization_kernels.c")
+default_cpu_kernel_code = "QlpoOTFBWSZTWXLbSoQAAgzbgERwQXxmTwAAr/ff3kABt0Q2oRVT0hpo9RtEAAAAyBEiSQ9EGjQGQAAAwANGhowjJoNGmgMEUplMTNSMJ5TQaDJpsoMyRMj8P4mZzFSVVwqSXG8GG7MlVwiToYEQwVD7noBxMhNfkeZYtYFtbgOBUSIGtIQjhNHCEnPJsadhb3yBmRIOD3TeAtNLSaU5GgvKUBWSNuuOIHmVt0YhW6rsmDMDUjeUJGJ64R1Jm5lrh0Aa0tKjhFwPdWcGogxLDSXPWQUWTM8Sd3Qz1HMYNxx3HMeiNqNo4jeRDEfZ3gUSHIcU/heomq0vEzL1Msz5KKGxH8FrNOYw3KaxdqaEmNHYMxJFgQbR0DyRknL2L4kwUSxKRdhjRpEtUqilVfggFL1klaMS3PPRDfNqbBOPWO7m4JTVGhS9QTBDDJaEbLbrUQNB+IpJSKQbG5SZZ5gkwJEhJ3aYKJipZ/i7kinChIOW2lQg"
+default_cpu_parallel_kernel_code_path = os.path.join(os.path.dirname(os.path.abspath(__file__)),
+                                                     "quantization_kernels_parallel.c")
+default_cpu_parallel_kernel_code = "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"
+
+cpu_kernels = None
+
+
+class CPUKernel:
+    def __init__(self, kernel_file="", source_code=default_cpu_kernel_code_path, compile_parallel_kernel=None,
+                 parallel_num=None):
+        self.load = False
+        self.int8WeightExtractionFloat = None
+        self.int4WeightExtractionFloat = None
+        self.int4WeightCompression = None
+        self.SetNumThreads = lambda x: x
+
+        try:
+            if not os.path.exists(default_cpu_kernel_code_path):
+                with open(default_cpu_kernel_code_path, "w", encoding="utf-8") as file:
+                    code = default_cpu_kernel_code
+                    cpu_quantization_code = bz2.decompress(base64.b64decode(code)).decode()
+                    file.write(cpu_quantization_code)
+
+            if not os.path.exists(default_cpu_parallel_kernel_code_path):
+                with open(default_cpu_parallel_kernel_code_path, "w", encoding="utf-8") as file:
+                    code = default_cpu_parallel_kernel_code
+                    cpu_quantization_code = bz2.decompress(base64.b64decode(code)).decode()
+                    file.write(cpu_quantization_code)
+
+        except Exception as ex:
+            print("Error when generating default cpu kernel code(can be ignored when using custom kernels).")
+
+        if compile_parallel_kernel is None:
+            compile_parallel_kernel = bool(int(os.cpu_count()) >= 4)
+
+        if compile_parallel_kernel and source_code == default_cpu_kernel_code_path:
+            source_code = default_cpu_parallel_kernel_code_path
+
+        kernels = None
+
+        if (not kernel_file) or (not os.path.exists(kernel_file)):
+            print("No compiled kernel found.")
+            try:
+                if os.path.exists(source_code):
+                    print("Compiling kernels :", source_code)
+                    kernel_file = source_code[:-2] + ".so"
+
+                    if compile_parallel_kernel:
+                        if sys.platform != 'darwin':
+                            compile_command = "gcc -O3 -fPIC -pthread -fopenmp -std=c99 {} -shared -o {}".format(
+                                source_code, kernel_file)
+                        else:
+                            compile_command = "clang -O3 -fPIC -pthread -Xclang -fopenmp -lomp -std=c99 {} -shared -o {}".format(
+                                source_code, kernel_file)
+                        print("Compiling", compile_command)
+                        exit_state = os.system(compile_command)
+                        if not exit_state:
+                            try:
+                                kernels = ctypes.cdll.LoadLibrary(kernel_file)
+                                print("Load kernel :", kernel_file)
+                            except:
+                                kernels = None
+                                print("Load parallel cpu kernel failed, using default cpu kernel code:")
+                                import traceback
+                                exception = traceback.format_exc()
+                                print(exception)
+                        else:
+                            print("Compile default cpu kernel failed, using default cpu kernel code.")
+
+                        if kernels is None:  # adjust config, use default cpu kernel
+                            compile_parallel_kernel = False
+                            source_code = default_cpu_kernel_code_path
+                            kernel_file = source_code[:-2] + ".so"
+
+                    if kernels is None:
+                        compile_command = "gcc -O3 -fPIC -std=c99 {} -shared -o {}".format(source_code, kernel_file)
+                        print("Compiling", compile_command)
+                        exit_state = os.system(compile_command)
+                        if not exit_state:
+                            try:
+                                kernels = ctypes.cdll.LoadLibrary(kernel_file)
+                                print("Load kernel :", kernel_file)
+                            except:
+                                kernels = None
+                                print("Load default cpu kernel failed:")
+                                import traceback
+                                exception = traceback.format_exc()
+                                print(exception)
+                        else:
+                            print("Compile default cpu kernel failed.")
+                else:
+                    print("Kernel source code not found.")
+                    return
+            except:
+                print("Failed to build cpu kernel:")
+                import traceback
+                exception = traceback.format_exc()
+                print(exception)
+                return
+        else:
+            try:
+                kernels = ctypes.cdll.LoadLibrary(kernel_file)
+                print("Load kernel :", kernel_file)
+            except:
+                kernels = None
+                print("Load custom cpu kernel failed:")
+                import traceback
+                exception = traceback.format_exc()
+                print(exception)
+
+        if kernels is not None:
+            self.int8WeightExtractionFloat = kernels.extract_int8_weight_to_float
+            self.int4WeightExtractionFloat = kernels.extract_int4_weight_to_float
+            self.int4WeightCompression = kernels.compress_int4_weight
+            if compile_parallel_kernel:
+                try:
+                    self.SetNumThreads = kernels.set_num_threads
+                except:
+                    print("No set_num_threads() found in kernel.")
+            self.load = True
+        else:
+            print("Failed to load kernel.")
+            return
+
+        if compile_parallel_kernel:
+            if parallel_num is None:
+                parallel_num = max(os.cpu_count() // 2, 1)
+            print("Setting CPU quantization kernel threads to", parallel_num)
+            if parallel_num < 4:
+                print("Parallel kernel is not recommended when parallel num < 4.")
+            self.SetNumThreads(parallel_num)
+
+        self.parallel_num = parallel_num
+
+
+def compress_int4_weight(weight: torch.Tensor):  # (n, m)
+    """compress weight on cpu or cuda to int4"""
+    if weight.device == torch.device("cpu"):
+        assert isinstance(cpu_kernels, CPUKernel)
+        n, m = weight.size(0), weight.size(1)
+        assert m % 2 == 0
+        m = m // 2
+        out = torch.empty(n, m, dtype=torch.int8, device="cpu")
+        cpu_kernels.int4WeightCompression(
+            ctypes.c_void_p(weight.data_ptr()),
+            ctypes.c_void_p(out.data_ptr()),
+            ctypes.c_int32(n),
+            ctypes.c_int32(m)
+        )
+        return out
+    else:
+        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
+
+
+def extract_weight_to_float(weight: torch.Tensor, scale_list: torch.Tensor, source_bit_width: int,
+                            quantization_cache=None):
+    """extract weight on cpu to float32"""
+    if source_bit_width == 8:
+        func = cpu_kernels.int8WeightExtractionFloat
+    elif source_bit_width == 4:
+        func = cpu_kernels.int4WeightExtractionFloat
+    else:
+        assert False, "Unsupported bit-width"
+
+    n, m = weight.size(0), weight.size(1)
+
+    if quantization_cache is not None:
+        out = quantization_cache
+        func(
+            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.tensor
+    else:
+        out = torch.empty(n, m * (8 // source_bit_width), dtype=torch.float, device="cpu")
+        func(
+            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 CacheTensor():
+    def __init__(self, *args, **kwargs):
+        self.tensor = torch.empty(*args, **kwargs)
+
+    def to(self, *args, **kwargs):
+        self.tensor = self.tensor.to(*args, **kwargs)
+
+    def data_ptr(self):
+        return self.tensor.data_ptr()
+
+
+class QuantizedLinear(Linear):
+    def __init__(self, weight_bit_width: int, weight_tensor=None, bias_tensor=None, quantized_weight=None,
+                 quantized_weight_scale=None, quantization_cache=None, empty_init=False, *args, **kwargs):
+        super(QuantizedLinear, self).__init__(*args, **kwargs)
+        self.weight_bit_width = weight_bit_width
+        self.quantization_cache = quantization_cache
+
+        if (quantized_weight is not None) and (quantized_weight_scale is not None):
+            del self.weight
+            self.weight = Parameter(quantized_weight.to(kwargs["device"]), requires_grad=False)
+            self.weight_scale = Parameter(quantized_weight_scale.to(kwargs["device"]), requires_grad=False)
+        else:
+            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)).to(
+                    kwargs["dtype"])
+                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 reset_parameters(self):
+        """To accelerate initialization"""
+        pass
+
+    def forward(self, input):
+        if self.weight.device == torch.device("cpu"):
+            output = W8A16LinearCPU.apply(input, self.weight, self.weight_scale, self.weight_bit_width,
+                                          self.quantization_cache)
+        else:
+            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 _apply(self, fn):
+        self_obj = super()._apply(fn)
+        if self.quantization_cache is not None:
+            self.quantization_cache.to(self_obj.weight.device)
+            self.quantization_cache.to(self_obj.weight_scale.dtype)
+        return self_obj
+
+
+class QuantizedEmbedding(Embedding):  # TODO: backward, check empty_init
+    def __init__(self, weight_bit_width: int, weight_tensor=None, quantized_weight=None, quantized_weight_scale=None,
+                 empty_init=False, *args, **kwargs):
+        super(QuantizedEmbedding, self).__init__(*args, **kwargs)
+        self.weight_bit_width = weight_bit_width
+
+        if (quantized_weight is not None) and (quantized_weight_scale is not None):
+            del self.weight
+            self.weight = Parameter(quantized_weight.to(kwargs["device"]), requires_grad=False)
+            self.weight_scale = Parameter(quantized_weight_scale.to(kwargs["device"]), requires_grad=False)
+        else:
+            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)).to(
+                    kwargs["dtype"])
+                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)
+
+    def forward(self, input):
+        if self.weight.device == torch.device("cpu"):
+            original_weight = extract_weight_to_float(weight=self.weight, scale_list=self.weight_scale,
+                                                      source_bit_width=self.weight_bit_width)
+        else:
+            original_weight = extract_weight_to_half(weight=self.weight, scale_list=self.weight_scale,
+                                                     source_bit_width=self.weight_bit_width)
+        output = F.embedding(
+            input, original_weight, self.padding_idx, self.max_norm,
+            self.norm_type, self.scale_grad_by_freq, self.sparse
+        )
+        return output
+
+
+def load_cpu_kernel(**kwargs):
+    global cpu_kernels
+    cpu_kernels = CPUKernel(**kwargs)
+
+
+def quantize(model, weight_bit_width, use_quantization_cache=False, empty_init=False, **kwargs):
+    """Replace fp16 linear with quantized linear"""
+
+    query_key_value_quantization_cache = None
+    dense_quantization_cache = None
+    dense_h_to_4h_quantization_cache = None
+    dense_4h_to_h_quantization_cache = None
+
+    load_cpu_kernel(**kwargs)
+    if not cpu_kernels.load:
+        if kernels is None:  # CUDA kernels failed
+            print("Cannot load cpu or cuda kernel, quantization failed:")
+            assert kernels is not None
+        print("Cannot load cpu kernel, don't use quantized model on cpu.")
+
+    current_device = model.device
+
+    if model.device == torch.device("cpu"):
+        dtype = torch.float32
+    else:
+        dtype = torch.half
+
+    QuantizedLinearWithPara = partial(
+        QuantizedLinear,
+        weight_bit_width=weight_bit_width,
+        bias=True,
+        dtype=dtype,
+        empty_init=empty_init
+    )
+
+    if use_quantization_cache:
+        print("Using quantization cache")
+        layer = model.layers[0]
+        weight = layer.attention.query_key_value.weight
+        n, m = weight.size(0), weight.size(1)
+        query_key_value_quantization_cache = CacheTensor(n, m, dtype=dtype, device=current_device, requires_grad=False)
+        weight = layer.attention.dense.weight
+        n, m = weight.size(0), weight.size(1)
+        dense_quantization_cache = CacheTensor(n, m, dtype=dtype, device=current_device, requires_grad=False)
+        weight = layer.mlp.dense_h_to_4h.weight
+        n, m = weight.size(0), weight.size(1)
+        dense_h_to_4h_quantization_cache = CacheTensor(n, m, dtype=dtype, device=current_device, requires_grad=False)
+        weight = layer.mlp.dense_4h_to_h.weight
+        n, m = weight.size(0), weight.size(1)
+        dense_4h_to_h_quantization_cache = CacheTensor(n, m, dtype=dtype, device=current_device, requires_grad=False)
+
+    print("Applying quantization to glm layers")
+
+    for layer in model.layers:
+        layer.attention.query_key_value = QuantizedLinearWithPara(
+            weight_tensor=layer.attention.query_key_value.weight.to(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,
+            device=layer.attention.query_key_value.weight.device,
+            quantization_cache=query_key_value_quantization_cache
+        )
+        layer.attention.dense = QuantizedLinearWithPara(
+            weight_tensor=layer.attention.dense.weight.to(current_device),
+            bias_tensor=layer.attention.dense.bias,
+            in_features=layer.attention.dense.in_features,
+            out_features=layer.attention.dense.out_features,
+            device=layer.attention.dense.weight.device,
+            quantization_cache=dense_quantization_cache
+        )
+        layer.mlp.dense_h_to_4h = QuantizedLinearWithPara(
+            weight_tensor=layer.mlp.dense_h_to_4h.weight.to(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,
+            device=layer.mlp.dense_h_to_4h.weight.device,
+            quantization_cache=dense_h_to_4h_quantization_cache
+        )
+        layer.mlp.dense_4h_to_h = QuantizedLinearWithPara(
+            weight_tensor=layer.mlp.dense_4h_to_h.weight.to(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,
+            device=layer.mlp.dense_4h_to_h.weight.device,
+            quantization_cache=dense_4h_to_h_quantization_cache
+        )
+    return model

xiaowo/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7bb7f4ea82efa2762df9b1d92c3cd635e2f206648536bff15c82e5349882c08b
+size 14575

xiaowo/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76865769b4b6b37c95369d34996cca04197a697394c214078eea0941cf10ccb9
+size 627

xiaowo/special_tokens_map.json

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

xiaowo/tokenization_chatglm.py

@@ -0,0 +1,443 @@
+"""Tokenization classes for ChatGLM."""
+from typing import List, Optional, Union
+import os
+
+from transformers.tokenization_utils import PreTrainedTokenizer
+from transformers.utils import logging, PaddingStrategy
+from transformers.tokenization_utils_base import EncodedInput, BatchEncoding
+from typing import Dict
+import sentencepiece as spm
+import numpy as np
+
+logger = logging.get_logger(__name__)
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "THUDM/chatglm-6b": 2048,
+}
+
+
+class TextTokenizer:
+    def __init__(self, model_path):
+        self.sp = spm.SentencePieceProcessor()
+        self.sp.Load(model_path)
+        self.num_tokens = self.sp.vocab_size()
+
+    def encode(self, text):
+        return self.sp.EncodeAsIds(text)
+
+    def decode(self, ids: List[int]):
+        return self.sp.DecodeIds(ids)
+
+    def tokenize(self, text):
+        return self.sp.EncodeAsPieces(text)
+
+    def convert_tokens_to_string(self, tokens):
+        return self.sp.DecodePieces(tokens)
+
+    def convert_tokens_to_ids(self, tokens):
+        return [self.sp.PieceToId(token) for token in tokens]
+
+    def convert_token_to_id(self, token):
+        return self.sp.PieceToId(token)
+
+    def convert_id_to_token(self, idx):
+        return self.sp.IdToPiece(idx)
+
+    def __len__(self):
+        return self.num_tokens
+
+
+class SPTokenizer:
+    def __init__(
+            self,
+            vocab_file,
+            num_image_tokens=20000,
+            max_blank_length=80,
+            byte_fallback=True,
+    ):
+        assert vocab_file is not None
+        self.vocab_file = vocab_file
+        self.num_image_tokens = num_image_tokens
+        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 = TextTokenizer(vocab_file)
+
+    def _get_text_tokenizer(self):
+        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_image_tokens + 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, 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(text)
+        tokens = [x + self.num_image_tokens for x in tmp]
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def postprocess(self, text):
+        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 decode(self, text_ids: List[int]) -> str:
+        ids = [int(_id) - self.num_image_tokens for _id in text_ids]
+        ids = [_id for _id in ids if _id >= 0]
+        text = self._get_text_tokenizer().decode(ids)
+        text = self.postprocess(text)
+        return text
+
+    def decode_tokens(self, tokens: List[str]) -> str:
+        text = self._get_text_tokenizer().convert_tokens_to_string(tokens)
+        text = self.postprocess(text)
+        return text
+
+    def tokenize(
+            self, text: str, linebreak=True, whitespaces=True, 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().tokenize(text)
+        return tokens if add_dummy_prefix else tokens[2:]
+
+    def __getitem__(self, x: Union[int, str]):
+        if isinstance(x, int):
+            if x < self.num_image_tokens:
+                return "<image_{}>".format(x)
+            else:
+                return self.text_tokenizer.convert_id_to_token(x - self.num_image_tokens)
+        elif isinstance(x, str):
+            if x.startswith("<image_") and x.endswith(">") and x[7:-1].isdigit():
+                return int(x[7:-1])
+            else:
+                return self.text_tokenizer.convert_token_to_id(x) + self.num_image_tokens
+        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='<eop>',
+            end_token='</s>',
+            mask_token='[MASK]',
+            gmask_token='[gMASK]',
+            padding_side="left",
+            pad_token="<pad>",
+            unk_token="<unk>",
+            num_image_tokens=20000,
+            **kwargs
+    ) -> None:
+        super().__init__(
+            do_lower_case=do_lower_case,
+            remove_space=remove_space,
+            padding_side=padding_side,
+            bos_token=bos_token,
+            eos_token=eos_token,
+            end_token=end_token,
+            mask_token=mask_token,
+            gmask_token=gmask_token,
+            pad_token=pad_token,
+            unk_token=unk_token,
+            num_image_tokens=num_image_tokens,
+            **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.end_token = end_token
+        self.mask_token = mask_token
+        self.gmask_token = gmask_token
+
+        self.sp_tokenizer = SPTokenizer(vocab_file, num_image_tokens=num_image_tokens)
+
+        """ Initialisation """
+
+    @property
+    def gmask_token_id(self) -> Optional[int]:
+        if self.gmask_token is None:
+            return None
+        return self.convert_tokens_to_ids(self.gmask_token)
+
+    @property
+    def end_token_id(self) -> Optional[int]:
+        """
+        `Optional[int]`: Id of the end of context token in the vocabulary. Returns `None` if the token has not been
+        set.
+        """
+        if self.end_token is None:
+            return None
+        return self.convert_tokens_to_ids(self.end_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 convert_tokens_to_string(self, tokens: List[str]) -> str:
+        return self.sp_tokenizer.decode_tokens(tokens)
+
+    def _decode(
+            self,
+            token_ids: Union[int, List[int]],
+            **kwargs
+    ) -> str:
+        if isinstance(token_ids, int):
+            token_ids = [token_ids]
+        if len(token_ids) == 0:
+            return ""
+        if self.pad_token_id in token_ids:  # remove pad
+            token_ids = list(filter((self.pad_token_id).__ne__, token_ids))
+        return super()._decode(token_ids, **kwargs)
+
+    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.
+        """
+        gmask_id = self.sp_tokenizer[self.gmask_token]
+        eos_id = self.sp_tokenizer[self.eos_token]
+        token_ids_0 = token_ids_0 + [gmask_id, self.sp_tokenizer[self.bos_token]]
+        if token_ids_1 is not None:
+            token_ids_0 = token_ids_0 + token_ids_1 + [eos_id]
+        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:
+                if bos_token_id in required_input:
+                    context_length = required_input.index(bos_token_id)
+                else:
+                    context_length = seq_length
+                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

xiaowo/tokenizer_config.json

@@ -0,0 +1,22 @@
+{
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_chatglm.ChatGLMTokenizer",
+      null
+    ]
+  },
+  "bos_token": "<sop>",
+  "do_lower_case": false,
+  "end_token": "</s>",
+  "eos_token": "<eop>",
+  "gmask_token": "[gMASK]",
+  "mask_token": "[MASK]",
+  "model_max_length": 1000000000000000019884624838656,
+  "num_image_tokens": 0,
+  "pad_token": "<pad>",
+  "padding_side": "left",
+  "remove_space": false,
+  "special_tokens_map_file": null,
+  "tokenizer_class": "ChatGLMTokenizer",
+  "unk_token": "<unk>"
+}

xiaowo/trainer_state.json

@@ -0,0 +1,3016 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.305008235222351,
+  "global_step": 5000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0024984848484848484,
+      "loss": 8.3168,
+      "step": 10
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.002496969696969697,
+      "loss": 7.4992,
+      "step": 20
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0024954545454545455,
+      "loss": 7.1773,
+      "step": 30
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.002493939393939394,
+      "loss": 6.7129,
+      "step": 40
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0024924242424242426,
+      "loss": 5.6141,
+      "step": 50
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.002490909090909091,
+      "loss": 5.4168,
+      "step": 60
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.0024893939393939393,
+      "loss": 5.2236,
+      "step": 70
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 0.002487878787878788,
+      "loss": 5.5969,
+      "step": 80
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024863636363636364,
+      "loss": 5.2027,
+      "step": 90
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024848484848484847,
+      "loss": 4.7475,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024833333333333335,
+      "loss": 4.7779,
+      "step": 110
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.002481818181818182,
+      "loss": 4.4724,
+      "step": 120
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00248030303030303,
+      "loss": 5.1906,
+      "step": 130
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024787878787878785,
+      "loss": 4.949,
+      "step": 140
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024772727272727273,
+      "loss": 4.3598,
+      "step": 150
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.002475757575757576,
+      "loss": 4.6328,
+      "step": 160
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024742424242424244,
+      "loss": 4.6934,
+      "step": 170
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024727272727272727,
+      "loss": 4.8184,
+      "step": 180
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024712121212121215,
+      "loss": 4.6488,
+      "step": 190
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00246969696969697,
+      "loss": 4.7639,
+      "step": 200
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.002468181818181818,
+      "loss": 4.6092,
+      "step": 210
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.002466666666666667,
+      "loss": 4.8875,
+      "step": 220
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024651515151515153,
+      "loss": 4.7232,
+      "step": 230
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.0024636363636363636,
+      "loss": 5.1139,
+      "step": 240
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024621212121212124,
+      "loss": 4.3293,
+      "step": 250
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024606060606060607,
+      "loss": 4.6564,
+      "step": 260
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.002459090909090909,
+      "loss": 4.7779,
+      "step": 270
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.002457575757575758,
+      "loss": 4.6312,
+      "step": 280
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.002456060606060606,
+      "loss": 4.4924,
+      "step": 290
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024545454545454545,
+      "loss": 4.418,
+      "step": 300
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024530303030303032,
+      "loss": 4.2307,
+      "step": 310
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024515151515151516,
+      "loss": 4.3598,
+      "step": 320
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00245,
+      "loss": 4.6166,
+      "step": 330
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024484848484848487,
+      "loss": 4.7531,
+      "step": 340
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.002446969696969697,
+      "loss": 4.5492,
+      "step": 350
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024454545454545454,
+      "loss": 4.5928,
+      "step": 360
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024439393939393937,
+      "loss": 4.576,
+      "step": 370
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.0024424242424242425,
+      "loss": 4.5797,
+      "step": 380
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.002440909090909091,
+      "loss": 4.3904,
+      "step": 390
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.002439393939393939,
+      "loss": 4.6285,
+      "step": 400
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002437878787878788,
+      "loss": 4.6023,
+      "step": 410
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024363636363636362,
+      "loss": 4.3693,
+      "step": 420
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002434848484848485,
+      "loss": 4.585,
+      "step": 430
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024333333333333334,
+      "loss": 4.7988,
+      "step": 440
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002431818181818182,
+      "loss": 4.2064,
+      "step": 450
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024303030303030305,
+      "loss": 4.5627,
+      "step": 460
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002428787878787879,
+      "loss": 4.0707,
+      "step": 470
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024272727272727276,
+      "loss": 4.5716,
+      "step": 480
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002425757575757576,
+      "loss": 4.3047,
+      "step": 490
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024242424242424242,
+      "loss": 4.5617,
+      "step": 500
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002422727272727273,
+      "loss": 4.4297,
+      "step": 510
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024212121212121213,
+      "loss": 4.8227,
+      "step": 520
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024196969696969697,
+      "loss": 4.4693,
+      "step": 530
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0024181818181818185,
+      "loss": 4.3514,
+      "step": 540
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002416666666666667,
+      "loss": 4.368,
+      "step": 550
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002415151515151515,
+      "loss": 4.049,
+      "step": 560
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.002413636363636364,
+      "loss": 4.5527,
+      "step": 570
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0024121212121212122,
+      "loss": 4.7082,
+      "step": 580
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0024106060606060606,
+      "loss": 4.2236,
+      "step": 590
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.002409090909090909,
+      "loss": 4.5283,
+      "step": 600
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0024075757575757577,
+      "loss": 4.5746,
+      "step": 610
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.002406060606060606,
+      "loss": 4.6463,
+      "step": 620
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0024045454545454543,
+      "loss": 4.4469,
+      "step": 630
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.002403030303030303,
+      "loss": 4.6049,
+      "step": 640
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0024015151515151515,
+      "loss": 4.5684,
+      "step": 650
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0024,
+      "loss": 4.3484,
+      "step": 660
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0023984848484848486,
+      "loss": 4.8201,
+      "step": 670
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.002396969696969697,
+      "loss": 4.2961,
+      "step": 680
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0023954545454545452,
+      "loss": 3.9297,
+      "step": 690
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.002393939393939394,
+      "loss": 4.7969,
+      "step": 700
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0023924242424242423,
+      "loss": 4.5076,
+      "step": 710
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.002390909090909091,
+      "loss": 4.8527,
+      "step": 720
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0023893939393939394,
+      "loss": 4.5695,
+      "step": 730
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002387878787878788,
+      "loss": 4.425,
+      "step": 740
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023863636363636366,
+      "loss": 4.2727,
+      "step": 750
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002384848484848485,
+      "loss": 4.5701,
+      "step": 760
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023833333333333337,
+      "loss": 3.857,
+      "step": 770
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002381818181818182,
+      "loss": 4.6637,
+      "step": 780
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023803030303030303,
+      "loss": 4.6662,
+      "step": 790
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002378787878787879,
+      "loss": 4.6828,
+      "step": 800
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023772727272727274,
+      "loss": 4.4822,
+      "step": 810
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023757575757575758,
+      "loss": 4.5971,
+      "step": 820
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002374242424242424,
+      "loss": 4.2271,
+      "step": 830
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002372727272727273,
+      "loss": 3.6249,
+      "step": 840
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002371212121212121,
+      "loss": 4.3729,
+      "step": 850
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023696969696969696,
+      "loss": 4.3625,
+      "step": 860
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023681818181818183,
+      "loss": 3.9025,
+      "step": 870
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023666666666666667,
+      "loss": 4.1387,
+      "step": 880
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.002365151515151515,
+      "loss": 4.1297,
+      "step": 890
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0023636363636363638,
+      "loss": 4.4295,
+      "step": 900
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002362121212121212,
+      "loss": 4.315,
+      "step": 910
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023606060606060604,
+      "loss": 4.6844,
+      "step": 920
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002359090909090909,
+      "loss": 4.3967,
+      "step": 930
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023575757575757575,
+      "loss": 4.3477,
+      "step": 940
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002356060606060606,
+      "loss": 4.3279,
+      "step": 950
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023545454545454546,
+      "loss": 4.6373,
+      "step": 960
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002353030303030303,
+      "loss": 4.1047,
+      "step": 970
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023515151515151513,
+      "loss": 4.2828,
+      "step": 980
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00235,
+      "loss": 4.559,
+      "step": 990
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002348484848484849,
+      "loss": 4.6342,
+      "step": 1000
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002346969696969697,
+      "loss": 4.5187,
+      "step": 1010
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023454545454545455,
+      "loss": 4.242,
+      "step": 1020
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023439393939393943,
+      "loss": 4.442,
+      "step": 1030
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023424242424242426,
+      "loss": 4.3424,
+      "step": 1040
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.002340909090909091,
+      "loss": 4.191,
+      "step": 1050
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0023393939393939393,
+      "loss": 4.4494,
+      "step": 1060
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.002337878787878788,
+      "loss": 4.3324,
+      "step": 1070
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023363636363636364,
+      "loss": 4.1201,
+      "step": 1080
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023348484848484848,
+      "loss": 4.2084,
+      "step": 1090
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023333333333333335,
+      "loss": 4.3176,
+      "step": 1100
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.002331818181818182,
+      "loss": 4.4791,
+      "step": 1110
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00233030303030303,
+      "loss": 3.9592,
+      "step": 1120
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.002328787878787879,
+      "loss": 4.3406,
+      "step": 1130
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023272727272727273,
+      "loss": 4.1566,
+      "step": 1140
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023257575757575756,
+      "loss": 4.4951,
+      "step": 1150
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023242424242424244,
+      "loss": 3.873,
+      "step": 1160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023227272727272727,
+      "loss": 4.4193,
+      "step": 1170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.002321212121212121,
+      "loss": 4.2574,
+      "step": 1180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00231969696969697,
+      "loss": 4.7572,
+      "step": 1190
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.002318181818181818,
+      "loss": 4.3244,
+      "step": 1200
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023166666666666665,
+      "loss": 4.2613,
+      "step": 1210
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0023151515151515153,
+      "loss": 4.5949,
+      "step": 1220
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023136363636363636,
+      "loss": 4.1111,
+      "step": 1230
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.002312121212121212,
+      "loss": 4.3355,
+      "step": 1240
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023106060606060607,
+      "loss": 4.5539,
+      "step": 1250
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.002309090909090909,
+      "loss": 4.221,
+      "step": 1260
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023075757575757574,
+      "loss": 4.0617,
+      "step": 1270
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.002306060606060606,
+      "loss": 4.2396,
+      "step": 1280
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023045454545454545,
+      "loss": 4.6004,
+      "step": 1290
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023030303030303033,
+      "loss": 4.533,
+      "step": 1300
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023015151515151516,
+      "loss": 4.3018,
+      "step": 1310
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0023,
+      "loss": 3.9908,
+      "step": 1320
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0022984848484848487,
+      "loss": 4.4869,
+      "step": 1330
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.002296969696969697,
+      "loss": 4.1631,
+      "step": 1340
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0022954545454545454,
+      "loss": 4.3191,
+      "step": 1350
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.002293939393939394,
+      "loss": 4.4717,
+      "step": 1360
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0022924242424242425,
+      "loss": 4.0645,
+      "step": 1370
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.002290909090909091,
+      "loss": 4.2992,
+      "step": 1380
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0022893939393939396,
+      "loss": 4.2506,
+      "step": 1390
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002287878787878788,
+      "loss": 4.1289,
+      "step": 1400
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022863636363636363,
+      "loss": 4.5553,
+      "step": 1410
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002284848484848485,
+      "loss": 4.1268,
+      "step": 1420
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022833333333333334,
+      "loss": 4.3211,
+      "step": 1430
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022818181818181817,
+      "loss": 3.8847,
+      "step": 1440
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022803030303030305,
+      "loss": 3.9451,
+      "step": 1450
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002278787878787879,
+      "loss": 3.6887,
+      "step": 1460
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002277272727272727,
+      "loss": 4.0348,
+      "step": 1470
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002275757575757576,
+      "loss": 4.1012,
+      "step": 1480
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022742424242424243,
+      "loss": 4.2861,
+      "step": 1490
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022727272727272726,
+      "loss": 4.2572,
+      "step": 1500
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022712121212121214,
+      "loss": 4.5604,
+      "step": 1510
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022696969696969697,
+      "loss": 4.2148,
+      "step": 1520
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002268181818181818,
+      "loss": 4.3568,
+      "step": 1530
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0022666666666666664,
+      "loss": 4.293,
+      "step": 1540
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.002265151515151515,
+      "loss": 4.658,
+      "step": 1550
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022636363636363635,
+      "loss": 4.4127,
+      "step": 1560
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022621212121212123,
+      "loss": 4.3584,
+      "step": 1570
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022606060606060606,
+      "loss": 4.3832,
+      "step": 1580
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022590909090909094,
+      "loss": 4.1822,
+      "step": 1590
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022575757575757577,
+      "loss": 4.4045,
+      "step": 1600
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.002256060606060606,
+      "loss": 4.4514,
+      "step": 1610
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.002254545454545455,
+      "loss": 3.8766,
+      "step": 1620
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.002253030303030303,
+      "loss": 4.6047,
+      "step": 1630
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022515151515151515,
+      "loss": 4.1969,
+      "step": 1640
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022500000000000003,
+      "loss": 4.1688,
+      "step": 1650
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022484848484848486,
+      "loss": 4.5,
+      "step": 1660
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.002246969696969697,
+      "loss": 4.1338,
+      "step": 1670
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022454545454545457,
+      "loss": 4.4885,
+      "step": 1680
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.002243939393939394,
+      "loss": 3.8264,
+      "step": 1690
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022424242424242424,
+      "loss": 4.1896,
+      "step": 1700
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.002240909090909091,
+      "loss": 4.1893,
+      "step": 1710
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0022393939393939395,
+      "loss": 4.3912,
+      "step": 1720
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.002237878787878788,
+      "loss": 4.2867,
+      "step": 1730
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022363636363636366,
+      "loss": 4.3912,
+      "step": 1740
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.002234848484848485,
+      "loss": 4.1045,
+      "step": 1750
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022333333333333333,
+      "loss": 4.1043,
+      "step": 1760
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022318181818181816,
+      "loss": 4.2686,
+      "step": 1770
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022303030303030304,
+      "loss": 4.4639,
+      "step": 1780
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022287878787878787,
+      "loss": 4.0201,
+      "step": 1790
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.002227272727272727,
+      "loss": 4.0189,
+      "step": 1800
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.002225757575757576,
+      "loss": 4.0787,
+      "step": 1810
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.002224242424242424,
+      "loss": 3.8324,
+      "step": 1820
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022227272727272725,
+      "loss": 4.4113,
+      "step": 1830
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022212121212121213,
+      "loss": 4.117,
+      "step": 1840
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00221969696969697,
+      "loss": 4.1695,
+      "step": 1850
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022181818181818184,
+      "loss": 4.2963,
+      "step": 1860
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022166666666666667,
+      "loss": 4.4145,
+      "step": 1870
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0022151515151515155,
+      "loss": 4.076,
+      "step": 1880
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002213636363636364,
+      "loss": 4.3945,
+      "step": 1890
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002212121212121212,
+      "loss": 4.042,
+      "step": 1900
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002210606060606061,
+      "loss": 3.8959,
+      "step": 1910
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0022090909090909092,
+      "loss": 3.8939,
+      "step": 1920
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0022075757575757576,
+      "loss": 4.3057,
+      "step": 1930
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0022060606060606064,
+      "loss": 4.5699,
+      "step": 1940
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0022045454545454547,
+      "loss": 3.8951,
+      "step": 1950
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002203030303030303,
+      "loss": 4.041,
+      "step": 1960
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002201515151515152,
+      "loss": 4.4762,
+      "step": 1970
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0022,
+      "loss": 4.225,
+      "step": 1980
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0021984848484848485,
+      "loss": 4.4705,
+      "step": 1990
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002196969696969697,
+      "loss": 4.2803,
+      "step": 2000
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0021954545454545456,
+      "loss": 4.0248,
+      "step": 2010
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002193939393939394,
+      "loss": 4.09,
+      "step": 2020
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0021924242424242422,
+      "loss": 4.2914,
+      "step": 2030
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.002190909090909091,
+      "loss": 4.1557,
+      "step": 2040
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021893939393939394,
+      "loss": 4.041,
+      "step": 2050
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021878787878787877,
+      "loss": 4.6293,
+      "step": 2060
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021863636363636365,
+      "loss": 4.2543,
+      "step": 2070
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.002184848484848485,
+      "loss": 4.3525,
+      "step": 2080
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.002183333333333333,
+      "loss": 4.2623,
+      "step": 2090
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.002181818181818182,
+      "loss": 4.3332,
+      "step": 2100
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021803030303030302,
+      "loss": 4.4281,
+      "step": 2110
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021787878787878786,
+      "loss": 4.3771,
+      "step": 2120
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021772727272727273,
+      "loss": 4.4289,
+      "step": 2130
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.002175757575757576,
+      "loss": 3.9854,
+      "step": 2140
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021742424242424245,
+      "loss": 4.7166,
+      "step": 2150
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.002172727272727273,
+      "loss": 4.2908,
+      "step": 2160
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021712121212121216,
+      "loss": 4.5039,
+      "step": 2170
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00216969696969697,
+      "loss": 4.2756,
+      "step": 2180
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021681818181818182,
+      "loss": 4.1209,
+      "step": 2190
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.002166666666666667,
+      "loss": 3.9537,
+      "step": 2200
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0021651515151515153,
+      "loss": 4.2012,
+      "step": 2210
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021636363636363637,
+      "loss": 4.1775,
+      "step": 2220
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.002162121212121212,
+      "loss": 4.1004,
+      "step": 2230
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021606060606060608,
+      "loss": 4.5125,
+      "step": 2240
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.002159090909090909,
+      "loss": 4.1016,
+      "step": 2250
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021575757575757575,
+      "loss": 4.1955,
+      "step": 2260
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021560606060606062,
+      "loss": 4.3717,
+      "step": 2270
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021545454545454546,
+      "loss": 4.6578,
+      "step": 2280
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.002153030303030303,
+      "loss": 4.5115,
+      "step": 2290
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021515151515151517,
+      "loss": 4.4395,
+      "step": 2300
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00215,
+      "loss": 4.1674,
+      "step": 2310
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021484848484848483,
+      "loss": 4.1488,
+      "step": 2320
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.002146969696969697,
+      "loss": 4.0652,
+      "step": 2330
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021454545454545454,
+      "loss": 4.2736,
+      "step": 2340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021439393939393938,
+      "loss": 4.2018,
+      "step": 2350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0021424242424242426,
+      "loss": 4.3178,
+      "step": 2360
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.002140909090909091,
+      "loss": 4.5182,
+      "step": 2370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021393939393939392,
+      "loss": 4.4102,
+      "step": 2380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002137878787878788,
+      "loss": 4.392,
+      "step": 2390
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021363636363636363,
+      "loss": 4.2674,
+      "step": 2400
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021348484848484847,
+      "loss": 3.9971,
+      "step": 2410
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021333333333333334,
+      "loss": 4.4205,
+      "step": 2420
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002131818181818182,
+      "loss": 4.0297,
+      "step": 2430
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021303030303030305,
+      "loss": 4.0076,
+      "step": 2440
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002128787878787879,
+      "loss": 4.1734,
+      "step": 2450
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002127272727272727,
+      "loss": 4.0545,
+      "step": 2460
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002125757575757576,
+      "loss": 3.9951,
+      "step": 2470
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021242424242424243,
+      "loss": 4.2332,
+      "step": 2480
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021227272727272727,
+      "loss": 3.9803,
+      "step": 2490
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021212121212121214,
+      "loss": 4.5166,
+      "step": 2500
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0021196969696969698,
+      "loss": 4.1238,
+      "step": 2510
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002118181818181818,
+      "loss": 3.9596,
+      "step": 2520
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002116666666666667,
+      "loss": 4.0205,
+      "step": 2530
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.002115151515151515,
+      "loss": 4.4902,
+      "step": 2540
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021136363636363635,
+      "loss": 4.3045,
+      "step": 2550
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021121212121212123,
+      "loss": 3.9855,
+      "step": 2560
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021106060606060606,
+      "loss": 4.216,
+      "step": 2570
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.002109090909090909,
+      "loss": 4.4146,
+      "step": 2580
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021075757575757578,
+      "loss": 3.9723,
+      "step": 2590
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.002106060606060606,
+      "loss": 4.2469,
+      "step": 2600
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021045454545454544,
+      "loss": 3.9295,
+      "step": 2610
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.002103030303030303,
+      "loss": 4.2799,
+      "step": 2620
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021015151515151515,
+      "loss": 4.1453,
+      "step": 2630
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0021,
+      "loss": 4.1547,
+      "step": 2640
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0020984848484848486,
+      "loss": 4.2727,
+      "step": 2650
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.002096969696969697,
+      "loss": 4.1533,
+      "step": 2660
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0020954545454545453,
+      "loss": 3.9719,
+      "step": 2670
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.002093939393939394,
+      "loss": 4.3705,
+      "step": 2680
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0020924242424242424,
+      "loss": 3.7813,
+      "step": 2690
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0020909090909090908,
+      "loss": 4.1619,
+      "step": 2700
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020893939393939395,
+      "loss": 3.8096,
+      "step": 2710
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.002087878787878788,
+      "loss": 4.2779,
+      "step": 2720
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020863636363636366,
+      "loss": 4.152,
+      "step": 2730
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.002084848484848485,
+      "loss": 4.6047,
+      "step": 2740
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020833333333333333,
+      "loss": 3.9848,
+      "step": 2750
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.002081818181818182,
+      "loss": 4.068,
+      "step": 2760
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020803030303030304,
+      "loss": 4.0783,
+      "step": 2770
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020787878787878787,
+      "loss": 3.8646,
+      "step": 2780
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020772727272727275,
+      "loss": 4.4756,
+      "step": 2790
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.002075757575757576,
+      "loss": 4.0518,
+      "step": 2800
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.002074242424242424,
+      "loss": 4.1688,
+      "step": 2810
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.002072727272727273,
+      "loss": 4.3254,
+      "step": 2820
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020712121212121213,
+      "loss": 4.3361,
+      "step": 2830
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020696969696969696,
+      "loss": 4.5381,
+      "step": 2840
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020681818181818184,
+      "loss": 4.1404,
+      "step": 2850
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0020666666666666667,
+      "loss": 4.3307,
+      "step": 2860
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002065151515151515,
+      "loss": 4.2369,
+      "step": 2870
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002063636363636364,
+      "loss": 3.973,
+      "step": 2880
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002062121212121212,
+      "loss": 4.0045,
+      "step": 2890
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020606060606060605,
+      "loss": 3.717,
+      "step": 2900
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002059090909090909,
+      "loss": 4.3682,
+      "step": 2910
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020575757575757576,
+      "loss": 4.2744,
+      "step": 2920
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002056060606060606,
+      "loss": 4.0842,
+      "step": 2930
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020545454545454543,
+      "loss": 4.4441,
+      "step": 2940
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002053030303030303,
+      "loss": 3.993,
+      "step": 2950
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020515151515151514,
+      "loss": 3.9664,
+      "step": 2960
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020499999999999997,
+      "loss": 4.2275,
+      "step": 2970
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020484848484848485,
+      "loss": 3.7014,
+      "step": 2980
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020469696969696973,
+      "loss": 3.8799,
+      "step": 2990
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020454545454545456,
+      "loss": 3.9643,
+      "step": 3000
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002043939393939394,
+      "loss": 4.0553,
+      "step": 3010
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0020424242424242427,
+      "loss": 4.0256,
+      "step": 3020
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.002040909090909091,
+      "loss": 4.1273,
+      "step": 3030
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020393939393939394,
+      "loss": 4.0736,
+      "step": 3040
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002037878787878788,
+      "loss": 4.1354,
+      "step": 3050
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020363636363636365,
+      "loss": 4.2006,
+      "step": 3060
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002034848484848485,
+      "loss": 4.284,
+      "step": 3070
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020333333333333336,
+      "loss": 4.191,
+      "step": 3080
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002031818181818182,
+      "loss": 4.3771,
+      "step": 3090
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020303030303030303,
+      "loss": 3.9966,
+      "step": 3100
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002028787878787879,
+      "loss": 4.1867,
+      "step": 3110
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020272727272727274,
+      "loss": 3.9691,
+      "step": 3120
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020257575757575757,
+      "loss": 3.9979,
+      "step": 3130
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002024242424242424,
+      "loss": 4.3219,
+      "step": 3140
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002022727272727273,
+      "loss": 3.9918,
+      "step": 3150
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.002021212121212121,
+      "loss": 4.3541,
+      "step": 3160
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020196969696969695,
+      "loss": 4.2801,
+      "step": 3170
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020181818181818183,
+      "loss": 3.8957,
+      "step": 3180
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0020166666666666666,
+      "loss": 4.8797,
+      "step": 3190
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.002015151515151515,
+      "loss": 4.1672,
+      "step": 3200
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0020136363636363637,
+      "loss": 4.1686,
+      "step": 3210
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.002012121212121212,
+      "loss": 4.3756,
+      "step": 3220
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0020106060606060604,
+      "loss": 4.307,
+      "step": 3230
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.002009090909090909,
+      "loss": 4.0963,
+      "step": 3240
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0020075757575757575,
+      "loss": 3.9398,
+      "step": 3250
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.002006060606060606,
+      "loss": 4.008,
+      "step": 3260
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0020045454545454546,
+      "loss": 3.7615,
+      "step": 3270
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0020030303030303034,
+      "loss": 4.1881,
+      "step": 3280
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0020015151515151517,
+      "loss": 3.8508,
+      "step": 3290
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.002,
+      "loss": 4.3887,
+      "step": 3300
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.001998484848484849,
+      "loss": 4.0209,
+      "step": 3310
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.001996969696969697,
+      "loss": 4.1906,
+      "step": 3320
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0019954545454545455,
+      "loss": 4.1512,
+      "step": 3330
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0019939393939393943,
+      "loss": 4.2525,
+      "step": 3340
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0019924242424242426,
+      "loss": 4.3814,
+      "step": 3350
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.001990909090909091,
+      "loss": 4.0068,
+      "step": 3360
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019893939393939393,
+      "loss": 4.3184,
+      "step": 3370
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.001987878787878788,
+      "loss": 4.4627,
+      "step": 3380
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019863636363636364,
+      "loss": 4.1746,
+      "step": 3390
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019848484848484847,
+      "loss": 3.9419,
+      "step": 3400
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019833333333333335,
+      "loss": 4.0953,
+      "step": 3410
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.001981818181818182,
+      "loss": 4.3074,
+      "step": 3420
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00198030303030303,
+      "loss": 4.2008,
+      "step": 3430
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.001978787878787879,
+      "loss": 4.1914,
+      "step": 3440
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019772727272727273,
+      "loss": 4.0264,
+      "step": 3450
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019757575757575756,
+      "loss": 4.1937,
+      "step": 3460
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019742424242424244,
+      "loss": 4.067,
+      "step": 3470
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019727272727272727,
+      "loss": 4.2822,
+      "step": 3480
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.001971212121212121,
+      "loss": 3.798,
+      "step": 3490
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00196969696969697,
+      "loss": 4.3059,
+      "step": 3500
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.001968181818181818,
+      "loss": 3.8936,
+      "step": 3510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0019666666666666665,
+      "loss": 4.3832,
+      "step": 3520
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019651515151515152,
+      "loss": 4.3246,
+      "step": 3530
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019636363636363636,
+      "loss": 4.0193,
+      "step": 3540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.001962121212121212,
+      "loss": 4.2828,
+      "step": 3550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019606060606060607,
+      "loss": 4.468,
+      "step": 3560
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019590909090909095,
+      "loss": 4.2574,
+      "step": 3570
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.001957575757575758,
+      "loss": 4.2453,
+      "step": 3580
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.001956060606060606,
+      "loss": 4.1461,
+      "step": 3590
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019545454545454545,
+      "loss": 4.1102,
+      "step": 3600
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019530303030303032,
+      "loss": 4.5449,
+      "step": 3610
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019515151515151516,
+      "loss": 3.9111,
+      "step": 3620
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019500000000000001,
+      "loss": 3.9598,
+      "step": 3630
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019484848484848487,
+      "loss": 4.2061,
+      "step": 3640
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.001946969696969697,
+      "loss": 4.4254,
+      "step": 3650
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019454545454545456,
+      "loss": 3.7582,
+      "step": 3660
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.001943939393939394,
+      "loss": 4.152,
+      "step": 3670
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0019424242424242425,
+      "loss": 4.0951,
+      "step": 3680
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.001940909090909091,
+      "loss": 4.1152,
+      "step": 3690
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019393939393939393,
+      "loss": 4.3834,
+      "step": 3700
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.001937878787878788,
+      "loss": 4.3691,
+      "step": 3710
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019363636363636365,
+      "loss": 4.1271,
+      "step": 3720
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019348484848484848,
+      "loss": 4.2613,
+      "step": 3730
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019333333333333333,
+      "loss": 3.8348,
+      "step": 3740
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.001931818181818182,
+      "loss": 4.1611,
+      "step": 3750
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019303030303030302,
+      "loss": 4.2691,
+      "step": 3760
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019287878787878788,
+      "loss": 4.2182,
+      "step": 3770
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019272727272727273,
+      "loss": 4.1658,
+      "step": 3780
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019257575757575757,
+      "loss": 3.9271,
+      "step": 3790
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019242424242424242,
+      "loss": 3.7492,
+      "step": 3800
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019227272727272726,
+      "loss": 4.0193,
+      "step": 3810
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019212121212121211,
+      "loss": 4.4439,
+      "step": 3820
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019196969696969697,
+      "loss": 4.0443,
+      "step": 3830
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.001918181818181818,
+      "loss": 4.2902,
+      "step": 3840
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0019166666666666668,
+      "loss": 4.0879,
+      "step": 3850
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019151515151515153,
+      "loss": 4.509,
+      "step": 3860
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019136363636363639,
+      "loss": 4.2645,
+      "step": 3870
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019121212121212122,
+      "loss": 3.8994,
+      "step": 3880
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019106060606060608,
+      "loss": 3.8268,
+      "step": 3890
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019090909090909091,
+      "loss": 4.0371,
+      "step": 3900
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019075757575757577,
+      "loss": 4.2105,
+      "step": 3910
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019060606060606062,
+      "loss": 4.0234,
+      "step": 3920
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019045454545454546,
+      "loss": 3.8813,
+      "step": 3930
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.001903030303030303,
+      "loss": 4.4242,
+      "step": 3940
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019015151515151517,
+      "loss": 3.7244,
+      "step": 3950
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0019,
+      "loss": 4.2068,
+      "step": 3960
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0018984848484848485,
+      "loss": 4.008,
+      "step": 3970
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.001896969696969697,
+      "loss": 4.1963,
+      "step": 3980
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0018954545454545454,
+      "loss": 4.5043,
+      "step": 3990
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.001893939393939394,
+      "loss": 4.0703,
+      "step": 4000
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0018924242424242425,
+      "loss": 3.9598,
+      "step": 4010
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018909090909090909,
+      "loss": 4.1252,
+      "step": 4020
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018893939393939394,
+      "loss": 4.2848,
+      "step": 4030
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018878787878787878,
+      "loss": 3.9209,
+      "step": 4040
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018863636363636363,
+      "loss": 4.2768,
+      "step": 4050
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018848484848484849,
+      "loss": 4.1338,
+      "step": 4060
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018833333333333332,
+      "loss": 4.1549,
+      "step": 4070
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018818181818181818,
+      "loss": 3.7658,
+      "step": 4080
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018803030303030303,
+      "loss": 3.865,
+      "step": 4090
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018787878787878787,
+      "loss": 3.9766,
+      "step": 4100
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018772727272727272,
+      "loss": 4.2186,
+      "step": 4110
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018757575757575758,
+      "loss": 4.3256,
+      "step": 4120
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018742424242424243,
+      "loss": 4.1537,
+      "step": 4130
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018727272727272729,
+      "loss": 3.8939,
+      "step": 4140
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018712121212121214,
+      "loss": 4.201,
+      "step": 4150
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018696969696969698,
+      "loss": 4.3982,
+      "step": 4160
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018681818181818183,
+      "loss": 4.3611,
+      "step": 4170
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0018666666666666669,
+      "loss": 3.9762,
+      "step": 4180
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018651515151515152,
+      "loss": 4.0193,
+      "step": 4190
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018636363636363638,
+      "loss": 4.507,
+      "step": 4200
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018621212121212123,
+      "loss": 4.2193,
+      "step": 4210
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018606060606060606,
+      "loss": 4.0932,
+      "step": 4220
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018590909090909092,
+      "loss": 3.9719,
+      "step": 4230
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018575757575757575,
+      "loss": 4.4113,
+      "step": 4240
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.001856060606060606,
+      "loss": 4.0824,
+      "step": 4250
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018545454545454546,
+      "loss": 3.5553,
+      "step": 4260
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.001853030303030303,
+      "loss": 3.9251,
+      "step": 4270
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018515151515151515,
+      "loss": 4.3514,
+      "step": 4280
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00185,
+      "loss": 4.2945,
+      "step": 4290
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018484848484848484,
+      "loss": 4.1324,
+      "step": 4300
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.001846969696969697,
+      "loss": 4.3477,
+      "step": 4310
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018454545454545455,
+      "loss": 3.9568,
+      "step": 4320
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018439393939393939,
+      "loss": 4.1842,
+      "step": 4330
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0018424242424242424,
+      "loss": 3.8381,
+      "step": 4340
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.001840909090909091,
+      "loss": 3.9949,
+      "step": 4350
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018393939393939393,
+      "loss": 3.9988,
+      "step": 4360
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018378787878787879,
+      "loss": 3.8552,
+      "step": 4370
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018363636363636362,
+      "loss": 4.0248,
+      "step": 4380
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018348484848484847,
+      "loss": 4.0281,
+      "step": 4390
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018333333333333333,
+      "loss": 3.7404,
+      "step": 4400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.001831818181818182,
+      "loss": 4.2635,
+      "step": 4410
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018303030303030304,
+      "loss": 4.283,
+      "step": 4420
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.001828787878787879,
+      "loss": 4.2484,
+      "step": 4430
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018272727272727275,
+      "loss": 3.4916,
+      "step": 4440
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018257575757575758,
+      "loss": 3.9377,
+      "step": 4450
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018242424242424244,
+      "loss": 3.9158,
+      "step": 4460
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018227272727272727,
+      "loss": 3.6584,
+      "step": 4470
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018212121212121213,
+      "loss": 3.9092,
+      "step": 4480
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018196969696969698,
+      "loss": 4.0258,
+      "step": 4490
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0018181818181818182,
+      "loss": 4.1592,
+      "step": 4500
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018166666666666667,
+      "loss": 3.96,
+      "step": 4510
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018151515151515153,
+      "loss": 4.0961,
+      "step": 4520
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018136363636363636,
+      "loss": 4.3369,
+      "step": 4530
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018121212121212122,
+      "loss": 4.26,
+      "step": 4540
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018106060606060607,
+      "loss": 3.7775,
+      "step": 4550
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.001809090909090909,
+      "loss": 4.2238,
+      "step": 4560
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018075757575757576,
+      "loss": 4.024,
+      "step": 4570
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018060606060606062,
+      "loss": 4.4203,
+      "step": 4580
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018045454545454545,
+      "loss": 3.927,
+      "step": 4590
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.001803030303030303,
+      "loss": 4.0777,
+      "step": 4600
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018015151515151514,
+      "loss": 4.002,
+      "step": 4610
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0018,
+      "loss": 4.4143,
+      "step": 4620
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0017984848484848485,
+      "loss": 3.9836,
+      "step": 4630
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0017969696969696968,
+      "loss": 4.1457,
+      "step": 4640
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0017954545454545454,
+      "loss": 3.8955,
+      "step": 4650
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.001793939393939394,
+      "loss": 4.282,
+      "step": 4660
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0017924242424242423,
+      "loss": 4.1152,
+      "step": 4670
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017909090909090908,
+      "loss": 4.4537,
+      "step": 4680
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017893939393939394,
+      "loss": 4.3918,
+      "step": 4690
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.001787878787878788,
+      "loss": 3.9969,
+      "step": 4700
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017863636363636365,
+      "loss": 4.4594,
+      "step": 4710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.001784848484848485,
+      "loss": 3.8802,
+      "step": 4720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017833333333333334,
+      "loss": 4.3977,
+      "step": 4730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.001781818181818182,
+      "loss": 3.9135,
+      "step": 4740
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017803030303030305,
+      "loss": 4.2225,
+      "step": 4750
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017787878787878788,
+      "loss": 4.1572,
+      "step": 4760
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017772727272727274,
+      "loss": 3.9195,
+      "step": 4770
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.001775757575757576,
+      "loss": 4.1182,
+      "step": 4780
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017742424242424243,
+      "loss": 4.0127,
+      "step": 4790
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017727272727272728,
+      "loss": 3.725,
+      "step": 4800
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017712121212121214,
+      "loss": 3.9258,
+      "step": 4810
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017696969696969697,
+      "loss": 3.6133,
+      "step": 4820
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0017681818181818183,
+      "loss": 3.6561,
+      "step": 4830
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017666666666666666,
+      "loss": 3.9838,
+      "step": 4840
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017651515151515152,
+      "loss": 4.0107,
+      "step": 4850
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017636363636363637,
+      "loss": 4.1814,
+      "step": 4860
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.001762121212121212,
+      "loss": 4.0902,
+      "step": 4870
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017606060606060606,
+      "loss": 4.418,
+      "step": 4880
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017590909090909092,
+      "loss": 4.008,
+      "step": 4890
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017575757575757575,
+      "loss": 4.2344,
+      "step": 4900
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.001756060606060606,
+      "loss": 3.9775,
+      "step": 4910
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017545454545454546,
+      "loss": 4.8027,
+      "step": 4920
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.001753030303030303,
+      "loss": 4.2629,
+      "step": 4930
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017515151515151515,
+      "loss": 4.198,
+      "step": 4940
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017499999999999998,
+      "loss": 4.0887,
+      "step": 4950
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017484848484848484,
+      "loss": 4.1445,
+      "step": 4960
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.001746969696969697,
+      "loss": 3.9627,
+      "step": 4970
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0017454545454545457,
+      "loss": 4.4545,
+      "step": 4980
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.001743939393939394,
+      "loss": 4.0764,
+      "step": 4990
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0017424242424242426,
+      "loss": 4.1914,
+      "step": 5000
+    }
+  ],
+  "max_steps": 16500,
+  "num_train_epochs": 2,
+  "total_flos": 1.7332860616704e+17,
+  "trial_name": null,
+  "trial_params": null
+}

xiaowo/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bab6fefeccf33ecae53947b1c444bd9a1f0f1f040993b12582eba666e374911
+size 3771