Upload 27 files

all_results.json

@@ -0,0 +1,7 @@
+{
+    "epoch": 6.0,
+    "train_loss": 0.8953271216487972,
+    "train_runtime": 33942.058,
+    "train_samples_per_second": 12.326,
+    "train_steps_per_second": 0.128
+}

config.json

@@ -0,0 +1,45 @@
+{
+  "_name_or_path": "/home/ubuntu/chatglm2-6b",
+  "add_bias_linear": false,
+  "add_qkv_bias": true,
+  "apply_query_key_layer_scaling": true,
+  "apply_residual_connection_post_layernorm": false,
+  "architectures": [
+    "ChatGLMForConditionalGeneration"
+  ],
+  "attention_dropout": 0.0,
+  "attention_softmax_in_fp32": true,
+  "auto_map": {
+    "AutoConfig": "configuration_chatglm.ChatGLMConfig",
+    "AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration",
+    "AutoModelForCausalLM": "modeling_chatglm.ChatGLMForConditionalGeneration",
+    "AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration"
+  },
+  "bias_dropout_fusion": true,
+  "eos_token_id": 2,
+  "ffn_hidden_size": 13696,
+  "fp32_residual_connection": false,
+  "hidden_dropout": 0.0,
+  "hidden_size": 4096,
+  "kv_channels": 128,
+  "layernorm_epsilon": 1e-05,
+  "model_type": "chatglm",
+  "multi_query_attention": true,
+  "multi_query_group_num": 2,
+  "num_attention_heads": 32,
+  "num_layers": 28,
+  "original_rope": true,
+  "pad_token_id": 0,
+  "padded_vocab_size": 65024,
+  "post_layer_norm": true,
+  "pre_seq_len": null,
+  "prefix_projection": false,
+  "quantization_bit": 0,
+  "rmsnorm": true,
+  "seq_length": 32768,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.30.2",
+  "use_cache": true,
+  "vocab_size": 65024
+}

configuration_chatglm.py

@@ -0,0 +1,59 @@
+from transformers import PretrainedConfig
+
+
+class ChatGLMConfig(PretrainedConfig):
+    model_type = "chatglm"
+    def __init__(
+        self,
+        num_layers=28,
+        padded_vocab_size=65024,
+        hidden_size=4096,
+        ffn_hidden_size=13696,
+        kv_channels=128,
+        num_attention_heads=32,
+        seq_length=2048,
+        hidden_dropout=0.0,
+        attention_dropout=0.0,
+        layernorm_epsilon=1e-5,
+        rmsnorm=True,
+        apply_residual_connection_post_layernorm=False,
+        post_layer_norm=True,
+        add_bias_linear=False,
+        add_qkv_bias=False,
+        bias_dropout_fusion=True,
+        multi_query_attention=False,
+        multi_query_group_num=1,
+        apply_query_key_layer_scaling=True,
+        attention_softmax_in_fp32=True,
+        fp32_residual_connection=False,
+        quantization_bit=0,
+        pre_seq_len=None,
+        prefix_projection=False,
+        **kwargs
+    ):
+        self.num_layers = num_layers
+        self.vocab_size = padded_vocab_size
+        self.padded_vocab_size = padded_vocab_size
+        self.hidden_size = hidden_size
+        self.ffn_hidden_size = ffn_hidden_size
+        self.kv_channels = kv_channels
+        self.num_attention_heads = num_attention_heads
+        self.seq_length = seq_length
+        self.hidden_dropout = hidden_dropout
+        self.attention_dropout = attention_dropout
+        self.layernorm_epsilon = layernorm_epsilon
+        self.rmsnorm = rmsnorm
+        self.apply_residual_connection_post_layernorm = apply_residual_connection_post_layernorm
+        self.post_layer_norm = post_layer_norm
+        self.add_bias_linear = add_bias_linear
+        self.add_qkv_bias = add_qkv_bias
+        self.bias_dropout_fusion = bias_dropout_fusion
+        self.multi_query_attention = multi_query_attention
+        self.multi_query_group_num = multi_query_group_num
+        self.apply_query_key_layer_scaling = apply_query_key_layer_scaling
+        self.attention_softmax_in_fp32 = attention_softmax_in_fp32
+        self.fp32_residual_connection = fp32_residual_connection
+        self.quantization_bit = quantization_bit
+        self.pre_seq_len = pre_seq_len
+        self.prefix_projection = prefix_projection
+        super().__init__(**kwargs)

finetuning_args.json

@@ -0,0 +1,13 @@
+{
+  "finetuning_type": "freeze",
+  "lora_alpha": 9.0,
+  "lora_dropout": 0.05,
+  "lora_rank": 100,
+  "lora_target": [
+    "query_key_value"
+  ],
+  "name_module_trainable": "mlp",
+  "num_layer_trainable": 1,
+  "pre_seq_len": 64,
+  "prefix_projection": false
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": 2,
+  "pad_token_id": 0,
+  "transformers_version": "4.30.2"
+}

modeling_chatglm.py

@@ -0,0 +1,1193 @@
+""" PyTorch ChatGLM model. """
+
+import math
+import copy
+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.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+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/ChatGLM2-6B"
+_CONFIG_FOR_DOC = "ChatGLM6BConfig"
+
+CHATGLM_6B_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "THUDM/chatglm2-6b",
+    # See all ChatGLM models at https://huggingface.co/models?filter=chatglm
+]
+
+
+def default_init(cls, *args, **kwargs):
+    return cls(*args, **kwargs)
+
+
+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
+
+
+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: ChatGLMConfig):
+        super().__init__()
+        self.prefix_projection = config.prefix_projection
+        if self.prefix_projection:
+            # Use a two-layer MLP to encode the prefix
+            kv_size = config.num_layers * config.kv_channels * config.multi_query_group_num * 2
+            self.embedding = torch.nn.Embedding(config.pre_seq_len, kv_size)
+            self.trans = torch.nn.Sequential(
+                torch.nn.Linear(kv_size, config.hidden_size),
+                torch.nn.Tanh(),
+                torch.nn.Linear(config.hidden_size, kv_size)
+            )
+        else:
+            self.embedding = torch.nn.Embedding(config.pre_seq_len,
+                                                config.num_layers * config.kv_channels * config.multi_query_group_num * 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
+
+
+def split_tensor_along_last_dim(
+        tensor: torch.Tensor,
+        num_partitions: int,
+        contiguous_split_chunks: bool = False,
+) -> List[torch.Tensor]:
+    """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.
+
+    Returns:
+        A list of Tensors
+    """
+    # 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
+
+
+class RotaryEmbedding(nn.Module):
+    def __init__(self, dim, original_impl=False, device=None, dtype=None):
+        super().__init__()
+        inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2, device=device).to(dtype=dtype) / dim))
+        self.register_buffer("inv_freq", inv_freq)
+        self.dim = dim
+        self.original_impl = original_impl
+
+    def forward_impl(
+            self, seq_len: int, n_elem: int, dtype: torch.dtype, device: torch.device, base: int = 10000
+    ):
+        """Enhanced Transformer with Rotary Position Embedding.
+
+        Derived from: https://github.com/labmlai/annotated_deep_learning_paper_implementations/blob/master/labml_nn/
+        transformers/rope/__init__.py. MIT License:
+        https://github.com/labmlai/annotated_deep_learning_paper_implementations/blob/master/license.
+        """
+        # $\Theta = {\theta_i = 10000^{\frac{2(i-1)}{d}}, i \in [1, 2, ..., \frac{d}{2}]}$
+        theta = 1.0 / (base ** (torch.arange(0, n_elem, 2, dtype=dtype, device=device) / n_elem))
+
+        # Create position indexes `[0, 1, ..., seq_len - 1]`
+        seq_idx = torch.arange(seq_len, dtype=dtype, device=device)
+
+        # Calculate the product of position index and $\theta_i$
+        idx_theta = torch.outer(seq_idx, theta).float()
+
+        cache = torch.stack([torch.cos(idx_theta), torch.sin(idx_theta)], dim=-1)
+
+        # this is to mimic the behaviour of complex32, else we will get different results
+        if dtype in (torch.float16, torch.bfloat16, torch.int8):
+            cache = cache.bfloat16() if dtype == torch.bfloat16 else cache.half()
+        return cache
+
+    def forward(self, max_seq_len, offset=0):
+        return self.forward_impl(
+            max_seq_len, self.dim, dtype=self.inv_freq.dtype, device=self.inv_freq.device
+        )
+
+
+@torch.jit.script
+def apply_rotary_pos_emb(x: torch.Tensor, rope_cache: torch.Tensor) -> torch.Tensor:
+    # x: [sq, b, np, hn]
+    sq, b, np, hn = x.size(0), x.size(1), x.size(2), x.size(3)
+    rot_dim = rope_cache.shape[-2] * 2
+    x, x_pass = x[..., :rot_dim], x[..., rot_dim:]
+    # truncate to support variable sizes
+    rope_cache = rope_cache[:sq]
+    xshaped = x.reshape(sq, -1, np, rot_dim // 2, 2)
+    rope_cache = rope_cache.view(sq, -1, 1, xshaped.size(3), 2)
+    x_out2 = torch.stack(
+        [
+            xshaped[..., 0] * rope_cache[..., 0] - xshaped[..., 1] * rope_cache[..., 1],
+            xshaped[..., 1] * rope_cache[..., 0] + xshaped[..., 0] * rope_cache[..., 1],
+        ],
+        -1,
+    )
+    x_out2 = x_out2.flatten(3)
+    return torch.cat((x_out2, x_pass), dim=-1)
+
+
+class RMSNorm(torch.nn.Module):
+    def __init__(self, normalized_shape, eps=1e-5, device=None, dtype=None, **kwargs):
+        super().__init__()
+        self.weight = torch.nn.Parameter(torch.empty(normalized_shape, device=device, dtype=dtype))
+        self.eps = eps
+
+    def forward(self, hidden_states: torch.Tensor):
+        input_dtype = hidden_states.dtype
+        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.eps)
+
+        return (self.weight * hidden_states).to(input_dtype)
+
+
+class CoreAttention(torch.nn.Module):
+    def __init__(self, config: ChatGLMConfig, layer_number):
+        super(CoreAttention, self).__init__()
+
+        self.apply_query_key_layer_scaling = config.apply_query_key_layer_scaling
+        self.attention_softmax_in_fp32 = config.attention_softmax_in_fp32
+        if self.apply_query_key_layer_scaling:
+            self.attention_softmax_in_fp32 = True
+        self.layer_number = max(1, layer_number)
+
+        projection_size = config.kv_channels * config.num_attention_heads
+
+        # Per attention head and per partition values.
+        self.hidden_size_per_partition = projection_size
+        self.hidden_size_per_attention_head = projection_size // config.num_attention_heads
+        self.num_attention_heads_per_partition = config.num_attention_heads
+
+        coeff = None
+        self.norm_factor = math.sqrt(self.hidden_size_per_attention_head)
+        if self.apply_query_key_layer_scaling:
+            coeff = self.layer_number
+            self.norm_factor *= coeff
+        self.coeff = coeff
+
+        self.attention_dropout = torch.nn.Dropout(config.attention_dropout)
+
+    def forward(self, query_layer, key_layer, value_layer, attention_mask):
+        pytorch_major_version = int(torch.__version__.split('.')[0])
+        if pytorch_major_version >= 2:
+            query_layer, key_layer, value_layer = [k.permute(1, 2, 0, 3) for k in [query_layer, key_layer, value_layer]]
+            if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
+                context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
+                                                                                 is_causal=True)
+            else:
+                if attention_mask is not None:
+                    attention_mask = ~attention_mask
+                context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
+                                                                                 attention_mask)
+            context_layer = context_layer.permute(2, 0, 1, 3)
+            new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
+            context_layer = context_layer.reshape(*new_context_layer_shape)
+        else:
+            # Raw attention scores
+
+            # [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)
+
+            # preallocting input tensor: [b * np, sq, sk]
+            matmul_input_buffer = torch.empty(
+                output_size[0] * output_size[1], output_size[2], output_size[3], dtype=query_layer.dtype,
+                device=query_layer.device
+            )
+
+            # Raw attention scores. [b * np, sq, sk]
+            matmul_result = torch.baddbmm(
+                matmul_input_buffer,
+                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 / self.norm_factor),
+            )
+
+            # change view to [b, np, sq, sk]
+            attention_scores = matmul_result.view(*output_size)
+
+            # ===========================
+            # Attention probs and dropout
+            # ===========================
+
+            # attention scores and attention mask [b, np, sq, sk]
+            if self.attention_softmax_in_fp32:
+                attention_scores = attention_scores.float()
+            if self.coeff is not None:
+                attention_scores = attention_scores * self.coeff
+            if attention_mask is None and attention_scores.shape[2] == attention_scores.shape[3]:
+                attention_mask = torch.ones(output_size[0], 1, output_size[2], output_size[3],
+                                            device=attention_scores.device, dtype=torch.bool)
+                attention_mask.tril_()
+                attention_mask = ~attention_mask
+            if attention_mask is not None:
+                attention_scores = attention_scores.masked_fill(attention_mask, float("-inf"))
+            attention_probs = F.softmax(attention_scores, dim=-1)
+            attention_probs = attention_probs.type_as(value_layer)
+
+            # This is actually dropping out entire tokens to attend to, which might
+            # seem a bit unusual, but is taken from the original Transformer paper.
+            attention_probs = self.attention_dropout(attention_probs)
+            # =========================
+            # 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] + (self.hidden_size_per_partition,)
+            context_layer = context_layer.view(*new_context_layer_shape)
+
+        return context_layer
+
+
+class SelfAttention(torch.nn.Module):
+    """Parallel self-attention layer abstract class.
+
+    Self-attention layer takes input with size [s, b, h]
+    and returns output of the same size.
+    """
+
+    def __init__(self, config: ChatGLMConfig, layer_number, device=None):
+        super(SelfAttention, self).__init__()
+        self.layer_number = max(1, layer_number)
+
+        self.projection_size = config.kv_channels * config.num_attention_heads
+
+        # Per attention head and per partition values.
+        self.hidden_size_per_attention_head = self.projection_size // config.num_attention_heads
+        self.num_attention_heads_per_partition = config.num_attention_heads
+
+        self.multi_query_attention = config.multi_query_attention
+        self.qkv_hidden_size = 3 * self.projection_size
+        if self.multi_query_attention:
+            self.num_multi_query_groups_per_partition = config.multi_query_group_num
+            self.qkv_hidden_size = (
+                    self.projection_size + 2 * self.hidden_size_per_attention_head * config.multi_query_group_num
+            )
+        self.query_key_value = nn.Linear(config.hidden_size, self.qkv_hidden_size,
+                                         bias=config.add_bias_linear or config.add_qkv_bias,
+                                         device=device, **_config_to_kwargs(config)
+                                         )
+
+        self.core_attention = CoreAttention(config, self.layer_number)
+
+        # Output.
+        self.dense = nn.Linear(self.projection_size, config.hidden_size, bias=config.add_bias_linear,
+                               device=device, **_config_to_kwargs(config)
+                               )
+
+    def _allocate_memory(self, inference_max_sequence_len, batch_size, device=None, dtype=None):
+        if self.multi_query_attention:
+            num_attention_heads = self.num_multi_query_groups_per_partition
+        else:
+            num_attention_heads = self.num_attention_heads_per_partition
+        return torch.empty(
+            inference_max_sequence_len,
+            batch_size,
+            num_attention_heads,
+            self.hidden_size_per_attention_head,
+            dtype=dtype,
+            device=device,
+        )
+
+    def forward(
+            self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True
+    ):
+        # hidden_states: [sq, b, h]
+
+        # =================================================
+        # Pre-allocate memory for key-values for inference.
+        # =================================================
+        # =====================
+        # Query, Key, and Value
+        # =====================
+
+        # Attention heads [sq, b, h] --> [sq, b, (np * 3 * hn)]
+        mixed_x_layer = self.query_key_value(hidden_states)
+
+        if self.multi_query_attention:
+            (query_layer, key_layer, value_layer) = mixed_x_layer.split(
+                [
+                    self.num_attention_heads_per_partition * self.hidden_size_per_attention_head,
+                    self.num_multi_query_groups_per_partition * self.hidden_size_per_attention_head,
+                    self.num_multi_query_groups_per_partition * self.hidden_size_per_attention_head,
+                ],
+                dim=-1,
+            )
+            query_layer = query_layer.view(
+                query_layer.size()[:-1] + (self.num_attention_heads_per_partition, self.hidden_size_per_attention_head)
+            )
+            key_layer = key_layer.view(
+                key_layer.size()[:-1] + (self.num_multi_query_groups_per_partition, self.hidden_size_per_attention_head)
+            )
+            value_layer = value_layer.view(
+                value_layer.size()[:-1]
+                + (self.num_multi_query_groups_per_partition, self.hidden_size_per_attention_head)
+            )
+        else:
+            new_tensor_shape = mixed_x_layer.size()[:-1] + \
+                               (self.num_attention_heads_per_partition,
+                                3 * self.hidden_size_per_attention_head)
+            mixed_x_layer = mixed_x_layer.view(*new_tensor_shape)
+
+            # [sq, b, np, 3 * hn] --> 3 [sq, b, np, hn]
+            (query_layer, key_layer, value_layer) = split_tensor_along_last_dim(mixed_x_layer, 3)
+
+        # apply relative positional encoding (rotary embedding)
+        if rotary_pos_emb is not None:
+            query_layer = apply_rotary_pos_emb(query_layer, rotary_pos_emb)
+            key_layer = apply_rotary_pos_emb(key_layer, rotary_pos_emb)
+
+        # adjust key and value for inference
+        if kv_cache is not None:
+            cache_k, cache_v = kv_cache
+            key_layer = torch.cat((cache_k, key_layer), dim=0)
+            value_layer = torch.cat((cache_v, value_layer), dim=0)
+        if use_cache:
+            kv_cache = (key_layer, value_layer)
+        else:
+            kv_cache = None
+
+        if self.multi_query_attention:
+            key_layer = key_layer.unsqueeze(-2)
+            key_layer = key_layer.expand(
+                -1, -1, -1, self.num_attention_heads_per_partition // self.num_multi_query_groups_per_partition, -1
+            )
+            key_layer = key_layer.contiguous().view(
+                key_layer.size()[:2] + (self.num_attention_heads_per_partition, self.hidden_size_per_attention_head)
+            )
+            value_layer = value_layer.unsqueeze(-2)
+            value_layer = value_layer.expand(
+                -1, -1, -1, self.num_attention_heads_per_partition // self.num_multi_query_groups_per_partition, -1
+            )
+            value_layer = value_layer.contiguous().view(
+                value_layer.size()[:2] + (self.num_attention_heads_per_partition, self.hidden_size_per_attention_head)
+            )
+
+        # ==================================
+        # core attention computation
+        # ==================================
+
+        context_layer = self.core_attention(query_layer, key_layer, value_layer, attention_mask)
+
+        # =================
+        # Output. [sq, b, h]
+        # =================
+
+        output = self.dense(context_layer)
+
+        return output, kv_cache
+
+
+def _config_to_kwargs(args):
+    common_kwargs = {
+        "dtype": args.torch_dtype,
+    }
+    return common_kwargs
+
+
+class MLP(torch.nn.Module):
+    """MLP.
+
+    MLP will take the input with h hidden state, project it to 4*h
+    hidden dimension, perform nonlinear transformation, and project the
+    state back into h hidden dimension.
+    """
+
+    def __init__(self, config: ChatGLMConfig, device=None):
+        super(MLP, self).__init__()
+
+        self.add_bias = config.add_bias_linear
+
+        # Project to 4h. If using swiglu double the output width, see https://arxiv.org/pdf/2002.05202.pdf
+        self.dense_h_to_4h = nn.Linear(
+            config.hidden_size,
+            config.ffn_hidden_size * 2,
+            bias=self.add_bias,
+            device=device,
+            **_config_to_kwargs(config)
+        )
+
+        def swiglu(x):
+            x = torch.chunk(x, 2, dim=-1)
+            return F.silu(x[0]) * x[1]
+
+        self.activation_func = swiglu
+
+        # Project back to h.
+        self.dense_4h_to_h = nn.Linear(
+            config.ffn_hidden_size,
+            config.hidden_size,
+            bias=self.add_bias,
+            device=device,
+            **_config_to_kwargs(config)
+        )
+
+    def forward(self, hidden_states):
+        # [s, b, 4hp]
+        intermediate_parallel = self.dense_h_to_4h(hidden_states)
+        intermediate_parallel = self.activation_func(intermediate_parallel)
+        # [s, b, h]
+        output = self.dense_4h_to_h(intermediate_parallel)
+        return output
+
+
+class GLMBlock(torch.nn.Module):
+    """A single transformer layer.
+
+    Transformer layer takes input with size [s, b, h] and returns an
+    output of the same size.
+    """
+
+    def __init__(self, config: ChatGLMConfig, layer_number, device=None):
+        super(GLMBlock, self).__init__()
+        self.layer_number = layer_number
+
+        self.apply_residual_connection_post_layernorm = config.apply_residual_connection_post_layernorm
+
+        self.fp32_residual_connection = config.fp32_residual_connection
+
+        LayerNormFunc = RMSNorm if config.rmsnorm else LayerNorm
+        # Layernorm on the input data.
+        self.input_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
+                                             dtype=config.torch_dtype)
+
+        # Self attention.
+        self.self_attention = SelfAttention(config, layer_number, device=device)
+        self.hidden_dropout = config.hidden_dropout
+
+        # Layernorm on the attention output
+        self.post_attention_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
+                                                      dtype=config.torch_dtype)
+
+        # MLP
+        self.mlp = MLP(config, device=device)
+
+    def forward(
+            self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True,
+    ):
+        # hidden_states: [s, b, h]
+
+        # Layer norm at the beginning of the transformer layer.
+        layernorm_output = self.input_layernorm(hidden_states)
+        # Self attention.
+        attention_output, kv_cache = self.self_attention(
+            layernorm_output,
+            attention_mask,
+            rotary_pos_emb,
+            kv_cache=kv_cache,
+            use_cache=use_cache
+        )
+
+        # Residual connection.
+        if self.apply_residual_connection_post_layernorm:
+            residual = layernorm_output
+        else:
+            residual = hidden_states
+
+        layernorm_input = torch.nn.functional.dropout(attention_output, p=self.hidden_dropout, training=self.training)
+        layernorm_input = residual + layernorm_input
+
+        # Layer norm post the self attention.
+        layernorm_output = self.post_attention_layernorm(layernorm_input)
+
+        # MLP.
+        mlp_output = self.mlp(layernorm_output)
+
+        # Second residual connection.
+        if self.apply_residual_connection_post_layernorm:
+            residual = layernorm_output
+        else:
+            residual = layernorm_input
+
+        output = torch.nn.functional.dropout(mlp_output, p=self.hidden_dropout, training=self.training)
+        output = residual + output
+
+        return output, kv_cache
+
+
+class GLMTransformer(torch.nn.Module):
+    """Transformer class."""
+
+    def __init__(self, config: ChatGLMConfig, device=None):
+        super(GLMTransformer, self).__init__()
+
+        self.fp32_residual_connection = config.fp32_residual_connection
+        self.post_layer_norm = config.post_layer_norm
+
+        # Number of layers.
+        self.num_layers = config.num_layers
+
+        # Transformer layers.
+        def build_layer(layer_number):
+            return GLMBlock(config, layer_number, device=device)
+
+        self.layers = torch.nn.ModuleList([build_layer(i + 1) for i in range(self.num_layers)])
+
+        if self.post_layer_norm:
+            LayerNormFunc = RMSNorm if config.rmsnorm else LayerNorm
+            # Final layer norm before output.
+            self.final_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
+                                                 dtype=config.torch_dtype)
+
+        self.gradient_checkpointing = False
+
+    def _get_layer(self, layer_number):
+        return self.layers[layer_number]
+
+    def forward(
+            self, hidden_states, attention_mask, rotary_pos_emb, kv_caches=None,
+            use_cache: Optional[bool] = True,
+            output_hidden_states: Optional[bool] = False,
+    ):
+        if not kv_caches:
+            kv_caches = [None for _ in range(self.num_layers)]
+        presents = () if use_cache else None
+        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
+
+        all_self_attentions = None
+        all_hidden_states = () if output_hidden_states else None
+        for index in range(self.num_layers):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            layer = self._get_layer(index)
+            if self.gradient_checkpointing and self.training:
+                layer_ret = torch.utils.checkpoint.checkpoint(
+                    layer,
+                    hidden_states,
+                    attention_mask,
+                    rotary_pos_emb,
+                    kv_caches[index],
+                    use_cache
+                )
+            else:
+                layer_ret = layer(
+                    hidden_states,
+                    attention_mask,
+                    rotary_pos_emb,
+                    kv_cache=kv_caches[index],
+                    use_cache=use_cache
+                )
+            hidden_states, kv_cache = layer_ret
+            if use_cache:
+                presents = presents + (kv_cache,)
+
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        # Final layer norm.
+        if self.post_layer_norm:
+            hidden_states = self.final_layernorm(hidden_states)
+
+        return hidden_states, presents, all_hidden_states, all_self_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_weights(self, module: nn.Module):
+        """Initialize the weights."""
+        return
+
+    def get_masks(self, input_ids, past_key_values, padding_mask=None):
+        batch_size, seq_length = input_ids.shape
+        full_attention_mask = torch.ones(batch_size, seq_length, seq_length, device=input_ids.device)
+        full_attention_mask.tril_()
+        past_length = 0
+        if past_key_values:
+            past_length = past_key_values[0][0].shape[0]
+        if past_length:
+            full_attention_mask = torch.cat((torch.ones(batch_size, seq_length, past_length,
+                                                        device=input_ids.device), full_attention_mask), dim=-1)
+        if padding_mask is not None:
+            full_attention_mask = full_attention_mask * padding_mask.unsqueeze(1)
+        if not past_length and padding_mask is not None:
+            full_attention_mask -= padding_mask.unsqueeze(-1) - 1
+        full_attention_mask = (full_attention_mask < 0.5).bool()
+        full_attention_mask.unsqueeze_(1)
+        return full_attention_mask
+
+    def get_position_ids(self, input_ids, device):
+        batch_size, seq_length = input_ids.shape
+        position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
+        return position_ids
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, GLMTransformer):
+            module.gradient_checkpointing = value
+
+
+class Embedding(torch.nn.Module):
+    """Language model embeddings."""
+
+    def __init__(self, config: ChatGLMConfig, device=None):
+        super(Embedding, self).__init__()
+
+        self.hidden_size = config.hidden_size
+        # Word embeddings (parallel).
+        self.word_embeddings = nn.Embedding(
+            config.padded_vocab_size,
+            self.hidden_size,
+            dtype=config.torch_dtype,
+            device=device
+        )
+        self.fp32_residual_connection = config.fp32_residual_connection
+
+    def forward(self, input_ids):
+        # Embeddings.
+        words_embeddings = self.word_embeddings(input_ids)
+        embeddings = words_embeddings
+        # Data format change to avoid explicit tranposes : [b s h] --> [s b h].
+        embeddings = embeddings.transpose(0, 1).contiguous()
+        # If the input flag for fp32 residual connection is set, convert for float.
+        if self.fp32_residual_connection:
+            embeddings = embeddings.float()
+        return embeddings
+
+
+class ChatGLMModel(ChatGLMPreTrainedModel):
+    def __init__(self, config: ChatGLMConfig, device=None, empty_init=True):
+        super().__init__(config)
+        if empty_init:
+            init_method = skip_init
+        else:
+            init_method = default_init
+        init_kwargs = {}
+        if device is not None:
+            init_kwargs["device"] = device
+        self.embedding = init_method(Embedding, config, **init_kwargs)
+        self.num_layers = config.num_layers
+        self.multi_query_group_num = config.multi_query_group_num
+        self.kv_channels = config.kv_channels
+
+        # Rotary positional embeddings
+        self.seq_length = config.seq_length
+        rotary_dim = (
+            config.hidden_size // config.num_attention_heads if config.kv_channels is None else config.kv_channels
+        )
+
+        self.rotary_pos_emb = RotaryEmbedding(rotary_dim // 2, original_impl=config.original_rope, device=device,
+                                              dtype=config.torch_dtype)
+        self.encoder = init_method(GLMTransformer, config, **init_kwargs)
+        self.output_layer = init_method(nn.Linear, config.hidden_size, config.padded_vocab_size, bias=False,
+                                        dtype=config.torch_dtype, **init_kwargs)
+        self.pre_seq_len = config.pre_seq_len
+        self.prefix_projection = config.prefix_projection
+        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)
+
+    def get_input_embeddings(self):
+        return self.embedding.word_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.multi_query_group_num,
+            self.kv_channels
+        )
+        # 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)
+        return past_key_values
+
+    def forward(
+            self,
+            input_ids,
+            position_ids: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.BoolTensor] = None,
+            full_attention_mask: Optional[torch.BoolTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
+            inputs_embeds: Optional[torch.Tensor] = None,
+            use_cache: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ):
+        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
+
+        batch_size, seq_length = input_ids.shape
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embedding(input_ids)
+
+        if self.pre_seq_len is not None:
+            if past_key_values is None:
+                past_key_values = self.get_prompt(batch_size=batch_size, device=input_ids.device,
+                                                  dtype=inputs_embeds.dtype)
+            if attention_mask is not None:
+                attention_mask = torch.cat([attention_mask.new_ones((batch_size, self.pre_seq_len)),
+                                            attention_mask], dim=-1)
+
+        if full_attention_mask is None:
+            if (attention_mask is not None and not attention_mask.all()) or (past_key_values and seq_length != 1):
+                full_attention_mask = self.get_masks(input_ids, past_key_values, padding_mask=attention_mask)
+
+        # Rotary positional embeddings
+        rotary_pos_emb = self.rotary_pos_emb(self.seq_length)
+        if position_ids is not None:
+            rotary_pos_emb = rotary_pos_emb[position_ids]
+        else:
+            rotary_pos_emb = rotary_pos_emb[None, :seq_length]
+        rotary_pos_emb = rotary_pos_emb.transpose(0, 1).contiguous()
+
+        # Run encoder.
+        hidden_states, presents, all_hidden_states, all_self_attentions = self.encoder(
+            inputs_embeds, full_attention_mask, rotary_pos_emb=rotary_pos_emb,
+            kv_caches=past_key_values, use_cache=use_cache, output_hidden_states=output_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,
+        )
+
+    def quantize(self, weight_bit_width: int):
+        from .quantization import quantize
+        quantize(self.encoder, weight_bit_width)
+        return self
+
+
+class ChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):
+    def __init__(self, config: ChatGLMConfig, empty_init=True, device=None):
+        super().__init__(config)
+
+        self.max_sequence_length = config.max_length
+        self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
+        self.config = config
+        self.quantized = False
+
+        if self.config.quantization_bit:
+            self.quantize(self.config.quantization_bit, empty_init=True)
+
+    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"]
+            model_kwargs["attention_mask"] = torch.cat(
+                [attention_mask, attention_mask.new_ones((attention_mask.shape[0], 1))], dim=-1
+            )
+
+        # 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
+            model_kwargs["position_ids"] = torch.cat(
+                [position_ids, new_position_id], dim=-1
+            )
+
+        model_kwargs["is_first_forward"] = False
+        return model_kwargs
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids: torch.LongTensor,
+            past_key_values: Optional[torch.Tensor] = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.Tensor] = None,
+            is_first_forward: bool = True,
+            **kwargs
+    ) -> dict:
+        # only last token for input_ids if past is not None
+        if position_ids is None:
+            position_ids = self.get_position_ids(input_ids, device=input_ids.device)
+        if not is_first_forward:
+            position_ids = position_ids[..., -1:]
+            input_ids = input_ids[:, -1:]
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past_key_values,
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "return_last_logit": True
+        }
+
+    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,
+            return_last_logit: Optional[bool] = False,
+    ):
+        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_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = transformer_outputs[0]
+        if return_last_logit:
+            hidden_states = hidden_states[-1:]
+        lm_logits = self.transformer.output_layer(hidden_states)
+        lm_logits = lm_logits.transpose(0, 1).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")
+        return response
+
+    def build_inputs(self, tokenizer, query: str, history: List[Tuple[str, str]] = None):
+        prompt = tokenizer.build_prompt(query, history=history)
+        inputs = tokenizer([prompt], return_tensors="pt")
+        inputs = inputs.to(self.device)
+        return inputs
+
+    def build_stream_inputs(self, tokenizer, query: str, history: List[Tuple[str, str]] = None):
+        if history:
+            prompt = "\n\n[Round {}]\n\n问：{}\n\n答：".format(len(history) + 1, query)
+            input_ids = tokenizer.encode(prompt, add_special_tokens=False)
+            input_ids = input_ids[1:]
+            inputs = tokenizer.batch_encode_plus([(input_ids, None)], return_tensors="pt", add_special_tokens=False)
+        else:
+            prompt = "[Round {}]\n\n问：{}\n\n答：".format(len(history) + 1, query)
+            inputs = tokenizer([prompt], return_tensors="pt")
+        inputs = inputs.to(self.device)
+        return inputs
+
+    @torch.no_grad()
+    def chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = None, max_length: int = 8192, num_beams=1,
+             do_sample=True, top_p=0.8, temperature=0.8, 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}
+        inputs = self.build_inputs(tokenizer, query, history=history)
+        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, past_key_values=None,
+                    max_length: int = 8192, do_sample=True, top_p=0.8, temperature=0.8, logits_processor=None,
+                    return_past_key_values=False, **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 past_key_values is None and not return_past_key_values:
+            inputs = self.build_inputs(tokenizer, query, history=history)
+        else:
+            inputs = self.build_stream_inputs(tokenizer, query, history=history)
+        if past_key_values is not None:
+            past_length = past_key_values[0][0].shape[0]
+            if self.transformer.pre_seq_len is not None:
+                past_length -= self.transformer.pre_seq_len
+            inputs.position_ids += past_length
+            attention_mask = inputs.attention_mask
+            attention_mask = torch.cat((attention_mask.new_ones(1, past_length), attention_mask), dim=1)
+            inputs['attention_mask'] = attention_mask
+        for outputs in self.stream_generate(**inputs, past_key_values=past_key_values,
+                                            return_past_key_values=return_past_key_values, **gen_kwargs):
+            if return_past_key_values:
+                outputs, past_key_values = outputs
+            outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):]
+            response = tokenizer.decode(outputs)
+            if response and response[-1] != "�":
+                response = self.process_response(response)
+                new_history = history + [(query, response)]
+                if return_past_key_values:
+                    yield response, new_history, past_key_values
+                else:
+                    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,
+            return_past_key_values=False,
+            **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())
+            if return_past_key_values:
+                yield input_ids, outputs.past_key_values
+            else:
+                yield input_ids
+            # 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
+
+    def quantize(self, bits: int, empty_init=False, device=None, **kwargs):
+        if bits == 0:
+            return
+
+        from .quantization import quantize
+
+        if self.quantized:
+            logger.info("Already quantized.")
+            return self
+
+        self.quantized = True
+
+        self.config.quantization_bit = bits
+
+        self.transformer.encoder = quantize(self.transformer.encoder, bits, empty_init=empty_init, device=device,
+                                            **kwargs)
+        return self

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37f54a73a90d73f08d35fd1de3eb9ff4b09acf4e579a779840e7a761514aef20
+size 673186943

quantization.py

@@ -0,0 +1,188 @@
+from torch.nn import Linear
+from torch.nn.parameter import Parameter
+
+import bz2
+import torch
+import base64
+import ctypes
+from transformers.utils import logging
+
+from typing import List
+from functools import partial
+
+logger = logging.get_logger(__name__)
+
+try:
+    from cpm_kernels.kernels.base import LazyKernelCModule, KernelFunction, round_up
+
+    class Kernel:
+        def __init__(self, code: bytes, function_names: List[str]):
+            self.code = code
+            self._function_names = function_names
+            self._cmodule = LazyKernelCModule(self.code)
+
+            for name in self._function_names:
+                setattr(self, name, KernelFunction(self._cmodule, name))
+
+    quantization_code = "$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"
+
+    kernels = Kernel(
+        bz2.decompress(base64.b64decode(quantization_code)),
+        [
+            "int4WeightCompression",
+            "int4WeightExtractionFloat",
+            "int4WeightExtractionHalf",
+            "int8WeightExtractionFloat",
+            "int8WeightExtractionHalf",
+        ],
+    )
+except Exception as exception:
+    kernels = None
+    logger.warning("Failed to load cpm_kernels:" + str(exception))
+
+
+class W8A16Linear(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, inp: torch.Tensor, quant_w: torch.Tensor, scale_w: torch.Tensor, weight_bit_width):
+        ctx.inp_shape = inp.size()
+        ctx.weight_bit_width = weight_bit_width
+        out_features = quant_w.size(0)
+        inp = inp.contiguous().view(-1, inp.size(-1))
+        weight = extract_weight_to_half(quant_w, scale_w, weight_bit_width)
+        ctx.weight_shape = weight.size()
+        output = inp.mm(weight.t())
+        ctx.save_for_backward(inp, quant_w, scale_w)
+        return output.view(*(ctx.inp_shape[:-1] + (out_features,)))
+
+    @staticmethod
+    def backward(ctx, grad_output: torch.Tensor):
+        inp, quant_w, scale_w = ctx.saved_tensors
+        weight = extract_weight_to_half(quant_w, scale_w, ctx.weight_bit_width)
+        grad_output = grad_output.contiguous().view(-1, weight.size(0))
+        grad_input = grad_output.mm(weight)
+        grad_weight = grad_output.t().mm(inp)
+        return grad_input.view(ctx.inp_shape), grad_weight.view(ctx.weight_shape), None, None
+
+
+def compress_int4_weight(weight: torch.Tensor):  # (n, m)
+    with torch.cuda.device(weight.device):
+        n, m = weight.size(0), weight.size(1)
+        assert m % 2 == 0
+        m = m // 2
+        out = torch.empty(n, m, dtype=torch.int8, device="cuda")
+        stream = torch.cuda.current_stream()
+
+        gridDim = (n, 1, 1)
+        blockDim = (min(round_up(m, 32), 1024), 1, 1)
+
+        kernels.int4WeightCompression(
+            gridDim,
+            blockDim,
+            0,
+            stream,
+            [ctypes.c_void_p(weight.data_ptr()), ctypes.c_void_p(out.data_ptr()), ctypes.c_int32(n), ctypes.c_int32(m)],
+        )
+        return out
+
+
+def extract_weight_to_half(weight: torch.Tensor, scale_list: torch.Tensor, source_bit_width: int):
+    assert scale_list.dtype in [torch.half, torch.bfloat16]
+    assert weight.dtype in [torch.int8]
+    if source_bit_width == 8:
+        return weight.to(scale_list.dtype) * scale_list[:, None]
+    elif source_bit_width == 4:
+        func = (
+            kernels.int4WeightExtractionHalf if scale_list.dtype == torch.half else kernels.int4WeightExtractionBFloat16
+        )
+    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=scale_list.dtype, device="cuda")
+        stream = torch.cuda.current_stream()
+
+        gridDim = (n, 1, 1)
+        blockDim = (min(round_up(m, 32), 1024), 1, 1)
+
+        func(
+            gridDim,
+            blockDim,
+            0,
+            stream,
+            [
+                ctypes.c_void_p(weight.data_ptr()),
+                ctypes.c_void_p(scale_list.data_ptr()),
+                ctypes.c_void_p(out.data_ptr()),
+                ctypes.c_int32(n),
+                ctypes.c_int32(m),
+            ],
+        )
+        return out
+
+
+class QuantizedLinear(torch.nn.Module):
+    def __init__(self, weight_bit_width: int, weight, bias=None, device="cpu", dtype=None, empty_init=False, *args,
+                 **kwargs):
+        super().__init__()
+        self.weight_bit_width = weight_bit_width
+
+        shape = weight.shape
+
+        if weight is None or empty_init:
+            self.weight = torch.empty(shape[0], shape[1] * weight_bit_width // 8, dtype=torch.int8, device=device)
+            self.weight_scale = torch.empty(shape[0], dtype=dtype, device=device)
+        else:
+            self.weight_scale = weight.abs().max(dim=-1).values / ((2 ** (weight_bit_width - 1)) - 1)
+            self.weight = torch.round(weight / 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(device), requires_grad=False)
+        self.weight_scale = Parameter(self.weight_scale.to(device), requires_grad=False)
+        self.bias = Parameter(bias.to(device), requires_grad=False) if bias is not None else None
+
+    def forward(self, input):
+        output = W8A16Linear.apply(input, self.weight, self.weight_scale, self.weight_bit_width)
+        if self.bias is not None:
+            output = output + self.bias
+        return output
+
+
+def quantize(model, weight_bit_width, empty_init=False, device=None):
+    """Replace fp16 linear with quantized linear"""
+    for layer in model.layers:
+        layer.self_attention.query_key_value = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight=layer.self_attention.query_key_value.weight.to(torch.cuda.current_device()),
+            bias=layer.self_attention.query_key_value.bias,
+            dtype=layer.self_attention.query_key_value.weight.dtype,
+            device=layer.self_attention.query_key_value.weight.device if device is None else device,
+            empty_init=empty_init
+        )
+        layer.self_attention.dense = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight=layer.self_attention.dense.weight.to(torch.cuda.current_device()),
+            bias=layer.self_attention.dense.bias,
+            dtype=layer.self_attention.dense.weight.dtype,
+            device=layer.self_attention.dense.weight.device if device is None else device,
+            empty_init=empty_init
+        )
+        layer.mlp.dense_h_to_4h = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight=layer.mlp.dense_h_to_4h.weight.to(torch.cuda.current_device()),
+            bias=layer.mlp.dense_h_to_4h.bias,
+            dtype=layer.mlp.dense_h_to_4h.weight.dtype,
+            device=layer.mlp.dense_h_to_4h.weight.device if device is None else device,
+            empty_init=empty_init
+        )
+        layer.mlp.dense_4h_to_h = QuantizedLinear(
+            weight_bit_width=weight_bit_width,
+            weight=layer.mlp.dense_4h_to_h.weight.to(torch.cuda.current_device()),
+            bias=layer.mlp.dense_4h_to_h.bias,
+            dtype=layer.mlp.dense_4h_to_h.weight.dtype,
+            device=layer.mlp.dense_4h_to_h.weight.device if device is None else device,
+            empty_init=empty_init
+        )
+
+    return model

runs/Aug01_10-45-37_ip-172-31-72-127/events.out.tfevents.1690886874.ip-172-31-72-127.2313011.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77532330957084257016a692d22c3ec1354b8180f77f64ddeae4e556ae998ad2
+size 4184

runs/Aug01_10-48-39_ip-172-31-72-127/events.out.tfevents.1690887057.ip-172-31-72-127.2314553.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b37ab0e36319f38ef13032cb11c034f6cf9c8734617801f446e8772d98694e3c
+size 4184

runs/Aug01_10-55-01_ip-172-31-72-127/events.out.tfevents.1690887438.ip-172-31-72-127.2317680.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c9ba3fd78cc012bc5cc0c3a2212140bb9c0bf082856b359ee5e0d9adfc2978e
+size 5789

runs/Aug01_10-58-34_ip-172-31-72-127/events.out.tfevents.1690887651.ip-172-31-72-127.2319667.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ded0dd2f9a2a7334504d97fc6b2fff723b1aadc61a70d904e9231259c3a00832
+size 4184

runs/Aug01_11-04-29_ip-172-31-72-127/events.out.tfevents.1690888007.ip-172-31-72-127.2322272.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83714359dcd4f14768e252474d2aff10a8d12253ead4231a5ecb880737e70b13
+size 4865

runs/Aug01_11-07-41_ip-172-31-72-127/events.out.tfevents.1690888199.ip-172-31-72-127.2323921.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e04fca0ae81ae29fb81e0144a77d86d97106cc6c8169f62c3d205294b858210d
+size 4184

runs/Aug01_11-11-10_ip-172-31-72-127/events.out.tfevents.1690888407.ip-172-31-72-127.2325679.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8c0786cab806ed9aaba360bfa72a969399513a08b1fb0ec171f3d0957d9ebb0
+size 19902

runs/Aug02_02-22-15_ip-172-31-72-127/events.out.tfevents.1690943071.ip-172-31-72-127.2474016.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a969fa0db06769ce3bf412f4a434418954b69551ec1771e5a43c54cace8236d
+size 5943

runs/Aug02_02-49-56_ip-172-31-72-127/events.out.tfevents.1690944733.ip-172-31-72-127.2481410.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26e31372af6357ff146c092ba23fdd6777c9aa23d44795240c24e15da3e70e24
+size 4865

runs/Aug02_03-28-30_ip-172-31-72-127/events.out.tfevents.1690947047.ip-172-31-72-127.2482663.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c8e13518d69987dff950e4ccb02b2833478c57fb351f3ba61febc12a0482677
+size 73478

special_tokens_map.json

@@ -0,0 +1 @@
+{}

tokenization_chatglm.py

@@ -0,0 +1,253 @@
+import os
+import torch
+from typing import List, Optional, Union, Dict
+from sentencepiece import SentencePieceProcessor
+from transformers import PreTrainedTokenizer
+from transformers.utils import logging, PaddingStrategy
+from transformers.tokenization_utils_base import EncodedInput, BatchEncoding
+
+
+class SPTokenizer:
+    def __init__(self, model_path: str):
+        # reload tokenizer
+        assert os.path.isfile(model_path), model_path
+        self.sp_model = SentencePieceProcessor(model_file=model_path)
+
+        # BOS / EOS token IDs
+        self.n_words: int = self.sp_model.vocab_size()
+        self.bos_id: int = self.sp_model.bos_id()
+        self.eos_id: int = self.sp_model.eos_id()
+        self.pad_id: int = self.sp_model.unk_id()
+        assert self.sp_model.vocab_size() == self.sp_model.get_piece_size()
+
+        special_tokens = ["[MASK]", "[gMASK]", "[sMASK]", "sop", "eop"]
+        self.special_tokens = {}
+        self.index_special_tokens = {}
+        for token in special_tokens:
+            self.special_tokens[token] = self.n_words
+            self.index_special_tokens[self.n_words] = token
+            self.n_words += 1
+
+    def tokenize(self, s: str):
+        return self.sp_model.EncodeAsPieces(s)
+
+    def encode(self, s: str, bos: bool = False, eos: bool = False) -> List[int]:
+        assert type(s) is str
+        t = self.sp_model.encode(s)
+        if bos:
+            t = [self.bos_id] + t
+        if eos:
+            t = t + [self.eos_id]
+        return t
+
+    def decode(self, t: List[int]) -> str:
+        return self.sp_model.decode(t)
+
+    def decode_tokens(self, tokens: List[str]) -> str:
+        text = self.sp_model.DecodePieces(tokens)
+        return text
+
+    def convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+        if token in self.special_tokens:
+            return self.special_tokens[token]
+        return self.sp_model.PieceToId(token)
+
+    def convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        if index in self.index_special_tokens or index in [self.eos_id, self.bos_id, self.pad_id] or index < 0:
+            return ""
+        return self.sp_model.IdToPiece(index)
+
+
+class ChatGLMTokenizer(PreTrainedTokenizer):
+    vocab_files_names = {"vocab_file": "tokenizer.model"}
+
+    model_input_names = ["input_ids", "attention_mask", "position_ids"]
+
+    def __init__(self, vocab_file, padding_side="left", **kwargs):
+        super().__init__(padding_side=padding_side, **kwargs)
+        self.name = "GLMTokenizer"
+
+        self.vocab_file = vocab_file
+        self.tokenizer = SPTokenizer(vocab_file)
+        self.special_tokens = {
+            "<bos>": self.tokenizer.bos_id,
+            "<eos>": self.tokenizer.eos_id,
+            "<pad>": self.tokenizer.pad_id
+        }
+
+    def get_command(self, token):
+        if token in self.special_tokens:
+            return self.special_tokens[token]
+        assert token in self.tokenizer.special_tokens, f"{token} is not a special token for {self.name}"
+        return self.tokenizer.special_tokens[token]
+
+    @property
+    def pad_token(self) -> str:
+        return "<unk>"
+
+    @property
+    def pad_token_id(self):
+        return self.get_command("<pad>")
+
+    @property
+    def eos_token(self) -> str:
+        return "</s>"
+
+    @property
+    def eos_token_id(self):
+        return self.get_command("<eos>")
+
+    @property
+    def vocab_size(self):
+        return self.tokenizer.n_words
+
+    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 _tokenize(self, text, **kwargs):
+        return self.tokenizer.tokenize(text)
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str) in an id using the vocab. """
+        return self.tokenizer.convert_token_to_id(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        return self.tokenizer.convert_id_to_token(index)
+
+    def convert_tokens_to_string(self, tokens: List[str]) -> str:
+        return self.tokenizer.decode_tokens(tokens)
+
+    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 get_prefix_tokens(self):
+        prefix_tokens = [self.get_command("[gMASK]"), self.get_command("sop")]
+        return prefix_tokens
+
+    def build_prompt(self, query, history=None):
+        if history is None:
+            history = []
+        prompt = ""
+        for i, (old_query, response) in enumerate(history):
+            prompt += "[Round {}]\n\n问：{}\n\n答：{}\n\n".format(i + 1, old_query, response)
+        prompt += "[Round {}]\n\n问：{}\n\n答：".format(len(history) + 1, query)
+        return prompt
+
+    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.
+        """
+        prefix_tokens = self.get_prefix_tokens()
+        token_ids_0 = prefix_tokens + token_ids_0
+        if token_ids_1 is not None:
+            token_ids_0 = token_ids_0 + token_ids_1 + [self.get_command("<eos>")]
+        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
+        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 "attention_mask" not in encoded_inputs:
+            encoded_inputs["attention_mask"] = [1] * seq_length
+
+        if "position_ids" not in encoded_inputs:
+            encoded_inputs["position_ids"] = list(range(seq_length))
+
+        if needs_to_be_padded:
+            difference = max_length - len(required_input)
+
+            if "attention_mask" in encoded_inputs:
+                encoded_inputs["attention_mask"] = [0] * difference + encoded_inputs["attention_mask"]
+            if "position_ids" in encoded_inputs:
+                encoded_inputs["position_ids"] = [0] * difference + encoded_inputs["position_ids"]
+            encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input
+
+        return encoded_inputs

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7dc4c393423b76e4373e5157ddc34803a0189ba96b21ddbb40269d31468a6f2
+size 1018370

tokenizer_config.json

@@ -0,0 +1,14 @@
+{
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_chatglm.ChatGLMTokenizer",
+      null
+    ]
+  },
+  "clean_up_tokenization_spaces": true,
+  "do_lower_case": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "padding_side": "left",
+  "remove_space": false,
+  "tokenizer_class": "ChatGLMTokenizer"
+}

train_results.json

@@ -0,0 +1,7 @@
+{
+    "epoch": 6.0,
+    "train_loss": 0.8953271216487972,
+    "train_runtime": 33942.058,
+    "train_samples_per_second": 12.326,
+    "train_steps_per_second": 0.128
+}

trainer_state.json

@@ -0,0 +1,2635 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.997246443322625,
+  "global_step": 4356,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 0.00019999739928881725,
+      "loss": 2.1906,
+      "step": 10
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00019998959729054295,
+      "loss": 1.8919,
+      "step": 20
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019997659441099206,
+      "loss": 1.7855,
+      "step": 30
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00019995839132649917,
+      "loss": 1.7235,
+      "step": 40
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0001999349889838836,
+      "loss": 1.7027,
+      "step": 50
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00019990638860040006,
+      "loss": 1.7118,
+      "step": 60
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00019987259166367533,
+      "loss": 1.689,
+      "step": 70
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00019983359993163078,
+      "loss": 1.6845,
+      "step": 80
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0001997894154323911,
+      "loss": 1.6655,
+      "step": 90
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0001997400404641787,
+      "loss": 1.6518,
+      "step": 100
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00019968547759519425,
+      "loss": 1.6571,
+      "step": 110
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.000199625729663483,
+      "loss": 1.6559,
+      "step": 120
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00019956079977678722,
+      "loss": 1.6417,
+      "step": 130
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0001994906913123846,
+      "loss": 1.6313,
+      "step": 140
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00019941540791691245,
+      "loss": 1.6243,
+      "step": 150
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00019933495350617813,
+      "loss": 1.6189,
+      "step": 160
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0001992493322649554,
+      "loss": 1.628,
+      "step": 170
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00019915854864676664,
+      "loss": 1.6149,
+      "step": 180
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00019906260737365122,
+      "loss": 1.6032,
+      "step": 190
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00019896151343592008,
+      "loss": 1.626,
+      "step": 200
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0001988552720918958,
+      "loss": 1.6369,
+      "step": 210
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00019874388886763944,
+      "loss": 1.5757,
+      "step": 220
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00019862736955666296,
+      "loss": 1.5873,
+      "step": 230
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00019850572021962788,
+      "loss": 1.6041,
+      "step": 240
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00019837894718402997,
+      "loss": 1.5905,
+      "step": 250
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00019824705704387028,
+      "loss": 1.6131,
+      "step": 260
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00019811005665931205,
+      "loss": 1.6045,
+      "step": 270
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00019796795315632395,
+      "loss": 1.5942,
+      "step": 280
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00019782075392630935,
+      "loss": 1.5768,
+      "step": 290
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00019766846662572191,
+      "loss": 1.5967,
+      "step": 300
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00019751109917566737,
+      "loss": 1.5606,
+      "step": 310
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00019734865976149145,
+      "loss": 1.58,
+      "step": 320
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00019718115683235417,
+      "loss": 1.5809,
+      "step": 330
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00019700859910079036,
+      "loss": 1.5758,
+      "step": 340
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00019683099554225649,
+      "loss": 1.57,
+      "step": 350
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0001966483553946637,
+      "loss": 1.5783,
+      "step": 360
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00019646068815789755,
+      "loss": 1.5699,
+      "step": 370
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00019626800359332362,
+      "loss": 1.5767,
+      "step": 380
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00019607031172327996,
+      "loss": 1.5635,
+      "step": 390
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00019586762283055573,
+      "loss": 1.5763,
+      "step": 400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0001956599474578563,
+      "loss": 1.5719,
+      "step": 410
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00019544729640725498,
+      "loss": 1.5712,
+      "step": 420
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00019522968073963106,
+      "loss": 1.5699,
+      "step": 430
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00019500711177409454,
+      "loss": 1.539,
+      "step": 440
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0001947796010873974,
+      "loss": 1.5515,
+      "step": 450
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00019454716051333135,
+      "loss": 1.5691,
+      "step": 460
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0001943098021421124,
+      "loss": 1.573,
+      "step": 470
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00019406753831975203,
+      "loss": 1.5532,
+      "step": 480
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00019382038164741477,
+      "loss": 1.5778,
+      "step": 490
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0001935683449807631,
+      "loss": 1.5514,
+      "step": 500
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00019331144142928854,
+      "loss": 1.5493,
+      "step": 510
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019304968435562993,
+      "loss": 1.5698,
+      "step": 520
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019278308737487822,
+      "loss": 1.5699,
+      "step": 530
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001925116643538684,
+      "loss": 1.554,
+      "step": 540
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019223542941045817,
+      "loss": 1.5527,
+      "step": 550
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019195439691279363,
+      "loss": 1.5517,
+      "step": 560
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00019166858147856203,
+      "loss": 1.5673,
+      "step": 570
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00019137799797423126,
+      "loss": 1.5469,
+      "step": 580
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00019108266151427673,
+      "loss": 1.5499,
+      "step": 590
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00019078258746039507,
+      "loss": 1.5502,
+      "step": 600
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00019047779142070527,
+      "loss": 1.5402,
+      "step": 610
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001901682892489367,
+      "loss": 1.5444,
+      "step": 620
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00018985409704360456,
+      "loss": 1.5524,
+      "step": 630
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00018953523114717245,
+      "loss": 1.5338,
+      "step": 640
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00018921170814520247,
+      "loss": 1.5416,
+      "step": 650
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00018888354486549237,
+      "loss": 1.5524,
+      "step": 660
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00018855075837720034,
+      "loss": 1.544,
+      "step": 670
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001882133659899573,
+      "loss": 1.5317,
+      "step": 680
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0001878713852529663,
+      "loss": 1.5516,
+      "step": 690
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00018752483395408987,
+      "loss": 1.5479,
+      "step": 700
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00018717373011892474,
+      "loss": 1.5328,
+      "step": 710
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001868180920098644,
+      "loss": 1.524,
+      "step": 720
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00018645793812514894,
+      "loss": 1.4712,
+      "step": 730
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0001860932871979031,
+      "loss": 1.3509,
+      "step": 740
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00018572415819516174,
+      "loss": 1.3227,
+      "step": 750
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00018535057031688335,
+      "loss": 1.3216,
+      "step": 760
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00018497254299495146,
+      "loss": 1.3022,
+      "step": 770
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00018459009589216364,
+      "loss": 1.308,
+      "step": 780
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00018420324890120916,
+      "loss": 1.3295,
+      "step": 790
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001838120221436338,
+      "loss": 1.3249,
+      "step": 800
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00018341643596879367,
+      "loss": 1.3109,
+      "step": 810
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00018301651095279655,
+      "loss": 1.328,
+      "step": 820
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00018261226789743172,
+      "loss": 1.3215,
+      "step": 830
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00018220372782908777,
+      "loss": 1.3378,
+      "step": 840
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00018179091199765926,
+      "loss": 1.3289,
+      "step": 850
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00018137384187544116,
+      "loss": 1.3362,
+      "step": 860
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00018095253915601206,
+      "loss": 1.3263,
+      "step": 870
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00018052702575310588,
+      "loss": 1.3238,
+      "step": 880
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00018009732379947188,
+      "loss": 1.3278,
+      "step": 890
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0001796634556457236,
+      "loss": 1.3277,
+      "step": 900
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00017922544385917628,
+      "loss": 1.3332,
+      "step": 910
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00017878331122267284,
+      "loss": 1.3268,
+      "step": 920
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00017833708073339922,
+      "loss": 1.3316,
+      "step": 930
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00017788677560168784,
+      "loss": 1.337,
+      "step": 940
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0001774324192498105,
+      "loss": 1.3307,
+      "step": 950
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0001769740353107602,
+      "loss": 1.3295,
+      "step": 960
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0001765116476270216,
+      "loss": 1.3148,
+      "step": 970
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00017604528024933115,
+      "loss": 1.3269,
+      "step": 980
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 0.00017557495743542585,
+      "loss": 1.3488,
+      "step": 990
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 0.00017510070364878177,
+      "loss": 1.3377,
+      "step": 1000
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 0.0001746225435573415,
+      "loss": 1.3396,
+      "step": 1010
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 0.0001741405020322309,
+      "loss": 1.3375,
+      "step": 1020
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 0.00017365460414646574,
+      "loss": 1.326,
+      "step": 1030
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 0.00017316487517364721,
+      "loss": 1.3297,
+      "step": 1040
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 0.00017267134058664775,
+      "loss": 1.3429,
+      "step": 1050
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 0.00017217402605628572,
+      "loss": 1.3339,
+      "step": 1060
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 0.00017167295744999027,
+      "loss": 1.3375,
+      "step": 1070
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 0.00017116816083045602,
+      "loss": 1.3161,
+      "step": 1080
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 0.00017065966245428723,
+      "loss": 1.3281,
+      "step": 1090
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 0.00017014748877063214,
+      "loss": 1.322,
+      "step": 1100
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 0.0001696316664198073,
+      "loss": 1.3344,
+      "step": 1110
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 0.00016911222223191182,
+      "loss": 1.3607,
+      "step": 1120
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 0.00016858918322543186,
+      "loss": 1.34,
+      "step": 1130
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 0.00016806257660583534,
+      "loss": 1.3324,
+      "step": 1140
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 0.00016753242976415666,
+      "loss": 1.3249,
+      "step": 1150
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 0.00016699877027557226,
+      "loss": 1.3363,
+      "step": 1160
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 0.00016646162589796615,
+      "loss": 1.3403,
+      "step": 1170
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 0.0001659210245704861,
+      "loss": 1.3283,
+      "step": 1180
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 0.00016537699441209047,
+      "loss": 1.3258,
+      "step": 1190
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 0.0001648295637200856,
+      "loss": 1.3258,
+      "step": 1200
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 0.00016427876096865394,
+      "loss": 1.3581,
+      "step": 1210
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 0.00016372461480737297,
+      "loss": 1.3337,
+      "step": 1220
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 0.00016316715405972508,
+      "loss": 1.3432,
+      "step": 1230
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 0.0001626064077215983,
+      "loss": 1.3385,
+      "step": 1240
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 0.00016204240495977805,
+      "loss": 1.3474,
+      "step": 1250
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 0.0001614751751104301,
+      "loss": 1.3198,
+      "step": 1260
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 0.00016090474767757474,
+      "loss": 1.3407,
+      "step": 1270
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 0.00016033115233155202,
+      "loss": 1.3203,
+      "step": 1280
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 0.00015975441890747855,
+      "loss": 1.3311,
+      "step": 1290
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 0.00015917457740369565,
+      "loss": 1.3271,
+      "step": 1300
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 0.000158591657980209,
+      "loss": 1.3432,
+      "step": 1310
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 0.00015800569095711982,
+      "loss": 1.3302,
+      "step": 1320
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 0.00015741670681304796,
+      "loss": 1.3326,
+      "step": 1330
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 0.00015682473618354635,
+      "loss": 1.3358,
+      "step": 1340
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 0.0001562298098595078,
+      "loss": 1.3343,
+      "step": 1350
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 0.0001556319587855631,
+      "loss": 1.3214,
+      "step": 1360
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 0.0001550312140584718,
+      "loss": 1.3342,
+      "step": 1370
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 0.00015442760692550443,
+      "loss": 1.3253,
+      "step": 1380
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 0.0001538211687828174,
+      "loss": 1.3494,
+      "step": 1390
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 0.00015321193117381996,
+      "loss": 1.324,
+      "step": 1400
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 0.00015259992578753334,
+      "loss": 1.3267,
+      "step": 1410
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 0.00015198518445694255,
+      "loss": 1.343,
+      "step": 1420
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 0.00015136773915734066,
+      "loss": 1.3209,
+      "step": 1430
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 0.00015074762200466556,
+      "loss": 1.3224,
+      "step": 1440
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 0.0001501248652538296,
+      "loss": 1.3481,
+      "step": 1450
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 0.00014949950129704162,
+      "loss": 1.1061,
+      "step": 1460
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 0.00014887156266212237,
+      "loss": 0.9927,
+      "step": 1470
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 0.00014824108201081247,
+      "loss": 0.9946,
+      "step": 1480
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 0.00014760809213707344,
+      "loss": 0.9764,
+      "step": 1490
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 0.00014697262596538227,
+      "loss": 0.9851,
+      "step": 1500
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 0.00014633471654901842,
+      "loss": 0.9941,
+      "step": 1510
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 0.000145694397068345,
+      "loss": 0.9875,
+      "step": 1520
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 0.0001450517008290827,
+      "loss": 0.9828,
+      "step": 1530
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 0.00014440666126057744,
+      "loss": 0.9913,
+      "step": 1540
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 0.00014375931191406159,
+      "loss": 0.9881,
+      "step": 1550
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 0.00014310968646090883,
+      "loss": 0.9866,
+      "step": 1560
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 0.0001424578186908828,
+      "loss": 0.9934,
+      "step": 1570
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 0.0001418037425103795,
+      "loss": 0.9963,
+      "step": 1580
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 0.00014114749194066363,
+      "loss": 0.9791,
+      "step": 1590
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 0.00014048910111609915,
+      "loss": 0.9851,
+      "step": 1600
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 0.0001398286042823736,
+      "loss": 0.9829,
+      "step": 1610
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 0.00013916603579471705,
+      "loss": 0.9995,
+      "step": 1620
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 0.00013850143011611497,
+      "loss": 0.9843,
+      "step": 1630
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 0.0001378348218155158,
+      "loss": 1.0025,
+      "step": 1640
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 0.00013716624556603274,
+      "loss": 1.0135,
+      "step": 1650
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 0.00013649573614314044,
+      "loss": 1.0007,
+      "step": 1660
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 0.00013582332842286592,
+      "loss": 1.0117,
+      "step": 1670
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 0.00013514905737997473,
+      "loss": 1.0068,
+      "step": 1680
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 0.0001344729580861517,
+      "loss": 0.998,
+      "step": 1690
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 0.0001337950657081768,
+      "loss": 0.9995,
+      "step": 1700
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 0.00013311541550609565,
+      "loss": 0.9975,
+      "step": 1710
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 0.00013243404283138597,
+      "loss": 1.014,
+      "step": 1720
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 0.0001317509831251184,
+      "loss": 0.9954,
+      "step": 1730
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 0.00013106627191611332,
+      "loss": 1.0051,
+      "step": 1740
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 0.00013037994481909264,
+      "loss": 1.0156,
+      "step": 1750
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 0.0001296920375328275,
+      "loss": 1.0155,
+      "step": 1760
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 0.00012900258583828137,
+      "loss": 1.0223,
+      "step": 1770
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 0.00012831162559674887,
+      "loss": 1.0181,
+      "step": 1780
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 0.00012761919274799054,
+      "loss": 1.0023,
+      "step": 1790
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 0.00012692532330836346,
+      "loss": 0.9917,
+      "step": 1800
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 0.0001262300533689478,
+      "loss": 1.0132,
+      "step": 1810
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 0.00012553341909366978,
+      "loss": 1.0136,
+      "step": 1820
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 0.0001248354567174203,
+      "loss": 0.9961,
+      "step": 1830
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 0.00012413620254417057,
+      "loss": 1.0244,
+      "step": 1840
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 0.0001234356929450835,
+      "loss": 1.0126,
+      "step": 1850
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 0.00012273396435662212,
+      "loss": 1.0121,
+      "step": 1860
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 0.00012203105327865407,
+      "loss": 1.0293,
+      "step": 1870
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 0.00012132699627255347,
+      "loss": 1.0104,
+      "step": 1880
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 0.00012062182995929882,
+      "loss": 1.0119,
+      "step": 1890
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 0.00011991559101756852,
+      "loss": 1.0047,
+      "step": 1900
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 0.00011920831618183282,
+      "loss": 1.0095,
+      "step": 1910
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 0.00011850004224044315,
+      "loss": 1.0112,
+      "step": 1920
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 0.0001177908060337188,
+      "loss": 1.0131,
+      "step": 1930
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 0.00011708064445203042,
+      "loss": 1.0082,
+      "step": 1940
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 0.00011636959443388132,
+      "loss": 1.0246,
+      "step": 1950
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 0.00011565769296398618,
+      "loss": 1.0102,
+      "step": 1960
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 0.00011494497707134731,
+      "loss": 1.0103,
+      "step": 1970
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 0.00011423148382732853,
+      "loss": 1.0081,
+      "step": 1980
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 0.000113517250343727,
+      "loss": 1.0092,
+      "step": 1990
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 0.0001128023137708429,
+      "loss": 0.9983,
+      "step": 2000
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 0.00011208671129554702,
+      "loss": 1.0136,
+      "step": 2010
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 0.00011137048013934656,
+      "loss": 1.0348,
+      "step": 2020
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 0.00011065365755644906,
+      "loss": 0.9996,
+      "step": 2030
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 0.00010993628083182467,
+      "loss": 1.0142,
+      "step": 2040
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 0.00010921838727926681,
+      "loss": 1.0101,
+      "step": 2050
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 0.00010850001423945126,
+      "loss": 1.0126,
+      "step": 2060
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 0.00010778119907799398,
+      "loss": 1.0188,
+      "step": 2070
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 0.00010706197918350758,
+      "loss": 1.0232,
+      "step": 2080
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 0.00010634239196565646,
+      "loss": 1.0161,
+      "step": 2090
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 0.00010562247485321115,
+      "loss": 1.0213,
+      "step": 2100
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 0.0001049022652921013,
+      "loss": 0.9976,
+      "step": 2110
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 0.00010418180074346815,
+      "loss": 1.0324,
+      "step": 2120
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 0.00010346111868171584,
+      "loss": 1.0132,
+      "step": 2130
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 0.00010274025659256232,
+      "loss": 1.0003,
+      "step": 2140
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 0.00010201925197108953,
+      "loss": 1.0046,
+      "step": 2150
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 0.0001012981423197931,
+      "loss": 1.0093,
+      "step": 2160
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 0.00010057696514663169,
+      "loss": 1.0065,
+      "step": 2170
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 9.985575796307615e-05,
+      "loss": 0.9698,
+      "step": 2180
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 9.913455828215814e-05,
+      "loss": 0.6643,
+      "step": 2190
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 9.84134036165192e-05,
+      "loss": 0.6469,
+      "step": 2200
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 9.769233147645943e-05,
+      "loss": 0.6258,
+      "step": 2210
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 9.697137936798634e-05,
+      "loss": 0.6403,
+      "step": 2220
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 9.625058479086418e-05,
+      "loss": 0.6438,
+      "step": 2230
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 9.552998523666326e-05,
+      "loss": 0.6249,
+      "step": 2240
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 9.480961818681004e-05,
+      "loss": 0.6426,
+      "step": 2250
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 9.408952111063727e-05,
+      "loss": 0.652,
+      "step": 2260
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 9.336973146343537e-05,
+      "loss": 0.644,
+      "step": 2270
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 9.265028668450402e-05,
+      "loss": 0.6364,
+      "step": 2280
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 9.193122419520485e-05,
+      "loss": 0.6513,
+      "step": 2290
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 9.121258139701502e-05,
+      "loss": 0.6501,
+      "step": 2300
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 9.049439566958175e-05,
+      "loss": 0.6431,
+      "step": 2310
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 8.977670436877811e-05,
+      "loss": 0.652,
+      "step": 2320
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 8.905954482475991e-05,
+      "loss": 0.6494,
+      "step": 2330
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 8.83429543400241e-05,
+      "loss": 0.6486,
+      "step": 2340
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 8.76269701874684e-05,
+      "loss": 0.6399,
+      "step": 2350
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 8.691162960845264e-05,
+      "loss": 0.6452,
+      "step": 2360
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 8.619696981086172e-05,
+      "loss": 0.6448,
+      "step": 2370
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 8.548302796717019e-05,
+      "loss": 0.6541,
+      "step": 2380
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 8.476984121250875e-05,
+      "loss": 0.6405,
+      "step": 2390
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 8.405744664273278e-05,
+      "loss": 0.6419,
+      "step": 2400
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 8.334588131249277e-05,
+      "loss": 0.6369,
+      "step": 2410
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 8.263518223330697e-05,
+      "loss": 0.6585,
+      "step": 2420
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 8.192538637163621e-05,
+      "loss": 0.6461,
+      "step": 2430
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 8.121653064696118e-05,
+      "loss": 0.6614,
+      "step": 2440
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 8.050865192986211e-05,
+      "loss": 0.6536,
+      "step": 2450
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 7.980178704010089e-05,
+      "loss": 0.6559,
+      "step": 2460
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 7.9095972744706e-05,
+      "loss": 0.6605,
+      "step": 2470
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 7.839124575606004e-05,
+      "loss": 0.6631,
+      "step": 2480
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 7.76876427299903e-05,
+      "loss": 0.6666,
+      "step": 2490
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 7.69852002638618e-05,
+      "loss": 0.6646,
+      "step": 2500
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 7.62839548946742e-05,
+      "loss": 0.6526,
+      "step": 2510
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 7.558394309716088e-05,
+      "loss": 0.657,
+      "step": 2520
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 7.488520128189209e-05,
+      "loss": 0.6546,
+      "step": 2530
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 7.41877657933809e-05,
+      "loss": 0.657,
+      "step": 2540
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 7.349167290819274e-05,
+      "loss": 0.6594,
+      "step": 2550
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 7.279695883305866e-05,
+      "loss": 0.6676,
+      "step": 2560
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 7.210365970299194e-05,
+      "loss": 0.6527,
+      "step": 2570
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 7.141181157940859e-05,
+      "loss": 0.6494,
+      "step": 2580
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 7.072145044825162e-05,
+      "loss": 0.6525,
+      "step": 2590
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 7.003261221811934e-05,
+      "loss": 0.6521,
+      "step": 2600
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 6.934533271839752e-05,
+      "loss": 0.6669,
+      "step": 2610
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 6.865964769739575e-05,
+      "loss": 0.6548,
+      "step": 2620
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 6.797559282048806e-05,
+      "loss": 0.6668,
+      "step": 2630
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 6.729320366825784e-05,
+      "loss": 0.6516,
+      "step": 2640
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 6.661251573464706e-05,
+      "loss": 0.655,
+      "step": 2650
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 6.593356442511015e-05,
+      "loss": 0.6641,
+      "step": 2660
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 6.525638505477231e-05,
+      "loss": 0.6592,
+      "step": 2670
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 6.458101284659286e-05,
+      "loss": 0.6582,
+      "step": 2680
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 6.390748292953284e-05,
+      "loss": 0.6814,
+      "step": 2690
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 6.323583033672799e-05,
+      "loss": 0.658,
+      "step": 2700
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 6.256609000366649e-05,
+      "loss": 0.6653,
+      "step": 2710
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 6.189829676637182e-05,
+      "loss": 0.654,
+      "step": 2720
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 6.123248535959083e-05,
+      "loss": 0.6626,
+      "step": 2730
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 6.056869041498687e-05,
+      "loss": 0.6696,
+      "step": 2740
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 5.9906946459338656e-05,
+      "loss": 0.6635,
+      "step": 2750
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 5.924728791274432e-05,
+      "loss": 0.6593,
+      "step": 2760
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 5.858974908683105e-05,
+      "loss": 0.6622,
+      "step": 2770
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 5.79343641829704e-05,
+      "loss": 0.6605,
+      "step": 2780
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 5.728116729049928e-05,
+      "loss": 0.6664,
+      "step": 2790
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 5.663019238494704e-05,
+      "loss": 0.6575,
+      "step": 2800
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 5.5981473326267976e-05,
+      "loss": 0.6792,
+      "step": 2810
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 5.533504385708024e-05,
+      "loss": 0.6841,
+      "step": 2820
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 5.4690937600910905e-05,
+      "loss": 0.6625,
+      "step": 2830
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 5.404918806044679e-05,
+      "loss": 0.6637,
+      "step": 2840
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 5.340982861579199e-05,
+      "loss": 0.6587,
+      "step": 2850
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 5.277289252273174e-05,
+      "loss": 0.6641,
+      "step": 2860
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 5.213841291100239e-05,
+      "loss": 0.6515,
+      "step": 2870
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 5.1506422782568345e-05,
+      "loss": 0.6596,
+      "step": 2880
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 5.087695500990555e-05,
+      "loss": 0.6508,
+      "step": 2890
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 5.025004233429145e-05,
+      "loss": 0.6552,
+      "step": 2900
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 4.962571736410223e-05,
+      "loss": 0.5383,
+      "step": 2910
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 4.90040125731165e-05,
+      "loss": 0.4324,
+      "step": 2920
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 4.8384960298826274e-05,
+      "loss": 0.4194,
+      "step": 2930
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 4.776859274075506e-05,
+      "loss": 0.4238,
+      "step": 2940
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 4.715494195878285e-05,
+      "loss": 0.419,
+      "step": 2950
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 4.654403987147865e-05,
+      "loss": 0.4239,
+      "step": 2960
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 4.593591825444028e-05,
+      "loss": 0.4261,
+      "step": 2970
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 4.5330608738641486e-05,
+      "loss": 0.4415,
+      "step": 2980
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 4.472814280878689e-05,
+      "loss": 0.4236,
+      "step": 2990
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 4.412855180167406e-05,
+      "loss": 0.4334,
+      "step": 3000
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 4.353186690456371e-05,
+      "loss": 0.422,
+      "step": 3010
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 4.293811915355761e-05,
+      "loss": 0.4361,
+      "step": 3020
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 4.234733943198399e-05,
+      "loss": 0.4339,
+      "step": 3030
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 4.175955846879151e-05,
+      "loss": 0.44,
+      "step": 3040
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 4.11748068369506e-05,
+      "loss": 0.4328,
+      "step": 3050
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 4.059311495186338e-05,
+      "loss": 0.4297,
+      "step": 3060
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 4.001451306978174e-05,
+      "loss": 0.4225,
+      "step": 3070
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 3.943903128623335e-05,
+      "loss": 0.4288,
+      "step": 3080
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 3.886669953445637e-05,
+      "loss": 0.4198,
+      "step": 3090
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 3.829754758384262e-05,
+      "loss": 0.4199,
+      "step": 3100
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 3.77316050383889e-05,
+      "loss": 0.4308,
+      "step": 3110
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 3.7168901335157315e-05,
+      "loss": 0.4335,
+      "step": 3120
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 3.660946574274421e-05,
+      "loss": 0.4267,
+      "step": 3130
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 3.6053327359757535e-05,
+      "loss": 0.4348,
+      "step": 3140
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 3.550051511330361e-05,
+      "loss": 0.4403,
+      "step": 3150
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 3.4951057757482205e-05,
+      "loss": 0.4252,
+      "step": 3160
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 3.440498387189111e-05,
+      "loss": 0.4212,
+      "step": 3170
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 3.3862321860139576e-05,
+      "loss": 0.4336,
+      "step": 3180
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 3.332309994837085e-05,
+      "loss": 0.4304,
+      "step": 3190
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 3.278734618379402e-05,
+      "loss": 0.4245,
+      "step": 3200
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 3.225508843322524e-05,
+      "loss": 0.4113,
+      "step": 3210
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 3.172635438163816e-05,
+      "loss": 0.4389,
+      "step": 3220
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 3.120117153072404e-05,
+      "loss": 0.427,
+      "step": 3230
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 3.0679567197461134e-05,
+      "loss": 0.4162,
+      "step": 3240
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 3.016156851269384e-05,
+      "loss": 0.4347,
+      "step": 3250
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 2.9647202419721687e-05,
+      "loss": 0.4259,
+      "step": 3260
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 2.913649567289759e-05,
+      "loss": 0.4399,
+      "step": 3270
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 2.862947483623659e-05,
+      "loss": 0.4278,
+      "step": 3280
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 2.812616628203383e-05,
+      "loss": 0.4344,
+      "step": 3290
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 2.7626596189492983e-05,
+      "loss": 0.4339,
+      "step": 3300
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 2.7130790543364646e-05,
+      "loss": 0.4353,
+      "step": 3310
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 2.6638775132594553e-05,
+      "loss": 0.4333,
+      "step": 3320
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 2.6150575548982292e-05,
+      "loss": 0.4336,
+      "step": 3330
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 2.5666217185850262e-05,
+      "loss": 0.4388,
+      "step": 3340
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 2.5185725236722636e-05,
+      "loss": 0.4322,
+      "step": 3350
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 2.4709124694015116e-05,
+      "loss": 0.4435,
+      "step": 3360
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 2.423644034773498e-05,
+      "loss": 0.4358,
+      "step": 3370
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 2.3767696784191463e-05,
+      "loss": 0.4471,
+      "step": 3380
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 2.3302918384717177e-05,
+      "loss": 0.4227,
+      "step": 3390
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 2.284212932439972e-05,
+      "loss": 0.4269,
+      "step": 3400
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 2.2385353570824308e-05,
+      "loss": 0.4393,
+      "step": 3410
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 2.1932614882827197e-05,
+      "loss": 0.4331,
+      "step": 3420
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 2.148393680925973e-05,
+      "loss": 0.4407,
+      "step": 3430
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 2.1039342687763586e-05,
+      "loss": 0.4335,
+      "step": 3440
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 2.0598855643556826e-05,
+      "loss": 0.4196,
+      "step": 3450
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 2.016249858823106e-05,
+      "loss": 0.4298,
+      "step": 3460
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 1.973029421855981e-05,
+      "loss": 0.4392,
+      "step": 3470
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 1.93022650153178e-05,
+      "loss": 0.4368,
+      "step": 3480
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 1.8878433242111716e-05,
+      "loss": 0.4373,
+      "step": 3490
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 1.8458820944222255e-05,
+      "loss": 0.4409,
+      "step": 3500
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 1.804344994745727e-05,
+      "loss": 0.4263,
+      "step": 3510
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 1.763234185701673e-05,
+      "loss": 0.4119,
+      "step": 3520
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 1.7225518056368785e-05,
+      "loss": 0.4381,
+      "step": 3530
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 1.6822999706137567e-05,
+      "loss": 0.4296,
+      "step": 3540
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 1.6424807743002612e-05,
+      "loss": 0.4212,
+      "step": 3550
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 1.6030962878609725e-05,
+      "loss": 0.4228,
+      "step": 3560
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 1.5641485598493743e-05,
+      "loss": 0.4132,
+      "step": 3570
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 1.5256396161013075e-05,
+      "loss": 0.4433,
+      "step": 3580
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 1.487571459629582e-05,
+      "loss": 0.4401,
+      "step": 3590
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 1.4499460705197998e-05,
+      "loss": 0.4329,
+      "step": 3600
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 1.412765405827372e-05,
+      "loss": 0.43,
+      "step": 3610
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 1.3760313994757001e-05,
+      "loss": 0.4292,
+      "step": 3620
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 1.339745962155613e-05,
+      "loss": 0.4316,
+      "step": 3630
+    },
+    {
+      "epoch": 5.01,
+      "learning_rate": 1.3039109812259598e-05,
+      "loss": 0.353,
+      "step": 3640
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 1.268528320615452e-05,
+      "loss": 0.3517,
+      "step": 3650
+    },
+    {
+      "epoch": 5.04,
+      "learning_rate": 1.2335998207257137e-05,
+      "loss": 0.3451,
+      "step": 3660
+    },
+    {
+      "epoch": 5.05,
+      "learning_rate": 1.1991272983355505e-05,
+      "loss": 0.3536,
+      "step": 3670
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 1.1651125465064516e-05,
+      "loss": 0.3435,
+      "step": 3680
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 1.131557334489326e-05,
+      "loss": 0.3361,
+      "step": 3690
+    },
+    {
+      "epoch": 5.09,
+      "learning_rate": 1.098463407632474e-05,
+      "loss": 0.3415,
+      "step": 3700
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 1.0658324872908121e-05,
+      "loss": 0.3426,
+      "step": 3710
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 1.0336662707363287e-05,
+      "loss": 0.3476,
+      "step": 3720
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 1.0019664310698029e-05,
+      "loss": 0.3474,
+      "step": 3730
+    },
+    {
+      "epoch": 5.15,
+      "learning_rate": 9.707346171337894e-06,
+      "loss": 0.3476,
+      "step": 3740
+    },
+    {
+      "epoch": 5.16,
+      "learning_rate": 9.399724534268384e-06,
+      "loss": 0.3354,
+      "step": 3750
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 9.096815400190172e-06,
+      "loss": 0.3514,
+      "step": 3760
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 8.798634524686699e-06,
+      "loss": 0.3519,
+      "step": 3770
+    },
+    {
+      "epoch": 5.2,
+      "learning_rate": 8.505197417404687e-06,
+      "loss": 0.3387,
+      "step": 3780
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 8.216519341247486e-06,
+      "loss": 0.3458,
+      "step": 3790
+    },
+    {
+      "epoch": 5.23,
+      "learning_rate": 7.932615311581126e-06,
+      "loss": 0.3403,
+      "step": 3800
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 7.653500095453248e-06,
+      "loss": 0.3461,
+      "step": 3810
+    },
+    {
+      "epoch": 5.26,
+      "learning_rate": 7.3791882108251945e-06,
+      "loss": 0.3445,
+      "step": 3820
+    },
+    {
+      "epoch": 5.27,
+      "learning_rate": 7.109693925816651e-06,
+      "loss": 0.3439,
+      "step": 3830
+    },
+    {
+      "epoch": 5.29,
+      "learning_rate": 6.845031257963619e-06,
+      "loss": 0.3446,
+      "step": 3840
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 6.585213973489335e-06,
+      "loss": 0.3577,
+      "step": 3850
+    },
+    {
+      "epoch": 5.31,
+      "learning_rate": 6.3302555865880965e-06,
+      "loss": 0.3368,
+      "step": 3860
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 6.08016935872251e-06,
+      "loss": 0.338,
+      "step": 3870
+    },
+    {
+      "epoch": 5.34,
+      "learning_rate": 5.834968297933541e-06,
+      "loss": 0.3552,
+      "step": 3880
+    },
+    {
+      "epoch": 5.36,
+      "learning_rate": 5.594665158163992e-06,
+      "loss": 0.3489,
+      "step": 3890
+    },
+    {
+      "epoch": 5.37,
+      "learning_rate": 5.359272438595153e-06,
+      "loss": 0.354,
+      "step": 3900
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 5.128802382996567e-06,
+      "loss": 0.3512,
+      "step": 3910
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 4.903266979089249e-06,
+      "loss": 0.3333,
+      "step": 3920
+    },
+    {
+      "epoch": 5.41,
+      "learning_rate": 4.682677957922155e-06,
+      "loss": 0.3523,
+      "step": 3930
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 4.467046793261931e-06,
+      "loss": 0.3462,
+      "step": 3940
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 4.256384700996252e-06,
+      "loss": 0.34,
+      "step": 3950
+    },
+    {
+      "epoch": 5.45,
+      "learning_rate": 4.050702638550275e-06,
+      "loss": 0.3555,
+      "step": 3960
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 3.850011304316781e-06,
+      "loss": 0.357,
+      "step": 3970
+    },
+    {
+      "epoch": 5.48,
+      "learning_rate": 3.6543211370997587e-06,
+      "loss": 0.3549,
+      "step": 3980
+    },
+    {
+      "epoch": 5.49,
+      "learning_rate": 3.4636423155712916e-06,
+      "loss": 0.3455,
+      "step": 3990
+    },
+    {
+      "epoch": 5.51,
+      "learning_rate": 3.2779847577422697e-06,
+      "loss": 0.3513,
+      "step": 4000
+    },
+    {
+      "epoch": 5.52,
+      "learning_rate": 3.0973581204464362e-06,
+      "loss": 0.3554,
+      "step": 4010
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 2.921771798838069e-06,
+      "loss": 0.3433,
+      "step": 4020
+    },
+    {
+      "epoch": 5.55,
+      "learning_rate": 2.751234925903412e-06,
+      "loss": 0.3544,
+      "step": 4030
+    },
+    {
+      "epoch": 5.56,
+      "learning_rate": 2.585756371985493e-06,
+      "loss": 0.3466,
+      "step": 4040
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 2.4253447443228106e-06,
+      "loss": 0.3459,
+      "step": 4050
+    },
+    {
+      "epoch": 5.59,
+      "learning_rate": 2.270008386601685e-06,
+      "loss": 0.3462,
+      "step": 4060
+    },
+    {
+      "epoch": 5.6,
+      "learning_rate": 2.119755378522137e-06,
+      "loss": 0.3398,
+      "step": 4070
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 1.974593535377722e-06,
+      "loss": 0.3488,
+      "step": 4080
+    },
+    {
+      "epoch": 5.63,
+      "learning_rate": 1.83453040764906e-06,
+      "loss": 0.349,
+      "step": 4090
+    },
+    {
+      "epoch": 5.64,
+      "learning_rate": 1.6995732806109554e-06,
+      "loss": 0.3416,
+      "step": 4100
+    },
+    {
+      "epoch": 5.66,
+      "learning_rate": 1.569729173953638e-06,
+      "loss": 0.3556,
+      "step": 4110
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 1.4450048414174854e-06,
+      "loss": 0.3341,
+      "step": 4120
+    },
+    {
+      "epoch": 5.69,
+      "learning_rate": 1.3254067704418283e-06,
+      "loss": 0.3511,
+      "step": 4130
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 1.2109411818274852e-06,
+      "loss": 0.3536,
+      "step": 4140
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 1.1016140294131894e-06,
+      "loss": 0.3318,
+      "step": 4150
+    },
+    {
+      "epoch": 5.73,
+      "learning_rate": 9.974309997658915e-07,
+      "loss": 0.3383,
+      "step": 4160
+    },
+    {
+      "epoch": 5.74,
+      "learning_rate": 8.983975118849852e-07,
+      "loss": 0.3465,
+      "step": 4170
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 8.04518716920466e-07,
+      "loss": 0.3512,
+      "step": 4180
+    },
+    {
+      "epoch": 5.77,
+      "learning_rate": 7.157994979049898e-07,
+      "loss": 0.3377,
+      "step": 4190
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 6.322444694998319e-07,
+      "loss": 0.3473,
+      "step": 4200
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 5.538579777549347e-07,
+      "loss": 0.3423,
+      "step": 4210
+    },
+    {
+      "epoch": 5.81,
+      "learning_rate": 4.80644099882821e-07,
+      "loss": 0.3424,
+      "step": 4220
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 4.126066440464982e-07,
+      "loss": 0.3448,
+      "step": 4230
+    },
+    {
+      "epoch": 5.84,
+      "learning_rate": 3.497491491614158e-07,
+      "loss": 0.3438,
+      "step": 4240
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 2.920748847113686e-07,
+      "loss": 0.3514,
+      "step": 4250
+    },
+    {
+      "epoch": 5.87,
+      "learning_rate": 2.395868505784438e-07,
+      "loss": 0.3395,
+      "step": 4260
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 1.9228777688700127e-07,
+      "loss": 0.353,
+      "step": 4270
+    },
+    {
+      "epoch": 5.89,
+      "learning_rate": 1.5018012386162072e-07,
+      "loss": 0.3474,
+      "step": 4280
+    },
+    {
+      "epoch": 5.91,
+      "learning_rate": 1.1326608169920372e-07,
+      "loss": 0.345,
+      "step": 4290
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 8.154757045497619e-08,
+      "loss": 0.3494,
+      "step": 4300
+    },
+    {
+      "epoch": 5.93,
+      "learning_rate": 5.50262399426904e-08,
+      "loss": 0.3439,
+      "step": 4310
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 3.370346964876036e-08,
+      "loss": 0.3474,
+      "step": 4320
+    },
+    {
+      "epoch": 5.96,
+      "learning_rate": 1.7580368660519152e-08,
+      "loss": 0.335,
+      "step": 4330
+    },
+    {
+      "epoch": 5.98,
+      "learning_rate": 6.657775608553962e-09,
+      "loss": 0.3516,
+      "step": 4340
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 9.362586230632354e-10,
+      "loss": 0.3405,
+      "step": 4350
+    },
+    {
+      "epoch": 6.0,
+      "step": 4356,
+      "total_flos": 5.268602572020646e+18,
+      "train_loss": 0.8953271216487972,
+      "train_runtime": 33942.058,
+      "train_samples_per_second": 12.326,
+      "train_steps_per_second": 0.128
+    }
+  ],
+  "max_steps": 4356,
+  "num_train_epochs": 6,
+  "total_flos": 5.268602572020646e+18,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:326a38cf0f9916b8ede07877300460617478f67a798b1bed6b7d9bf33cb7066b
+size 3289