Upload 23 files

config.json

@@ -0,0 +1,29 @@
+{
+  "_from_model_config": true,
+  "_name_or_path": "models/Baichuan2-13B-Base",
+  "architectures": [
+    "BaichuanForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_baichuan.BaichuanConfig",
+    "AutoModelForCausalLM": "modeling_baichuan.BaichuanForCausalLM"
+  },
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13696,
+  "model_max_length": 4096,
+  "model_type": "baichuan",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "pad_token_id": 0,
+  "rms_norm_eps": 1e-06,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.30.2",
+  "use_cache": true,
+  "vocab_size": 125696,
+  "z_loss_weight": 0
+}

configuration_baichuan.py

@@ -0,0 +1,48 @@
+# Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class BaichuanConfig(PretrainedConfig):
+    model_type = "baichuan"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=64000,
+        hidden_size=5120,
+        intermediate_size=13696,
+        num_hidden_layers=40,
+        num_attention_heads=40,
+        hidden_act="silu",
+        model_max_length=4096,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        tie_word_embeddings=False,
+        gradient_checkpointing=False,
+        z_loss_weight=0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.model_max_length = model_max_length
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.z_loss_weight = z_loss_weight
+        self.gradient_checkpointing = (gradient_checkpointing,)
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "pad_token_id": 0,
+  "transformers_version": "4.30.2"
+}

generation_utils.py

@@ -0,0 +1,83 @@
+from typing import List
+from queue import Queue
+
+import torch
+
+
+def build_chat_input(model, tokenizer, messages: List[dict], max_new_tokens: int=0):
+    def _parse_messages(messages, split_role="user"):
+        system, rounds = "", []
+        round = []
+        for i, message in enumerate(messages):
+            if message["role"] == "system":
+                assert i == 0
+                system = message["content"]
+                continue
+            if message["role"] == split_role and round:
+                rounds.append(round)
+                round = []
+            round.append(message)
+        if round:
+            rounds.append(round)
+        return system, rounds
+
+    max_new_tokens = max_new_tokens or model.generation_config.max_new_tokens
+    max_input_tokens = model.config.model_max_length - max_new_tokens
+    system, rounds = _parse_messages(messages, split_role="user")
+    system_tokens = tokenizer.encode(system)
+    max_history_tokens = max_input_tokens - len(system_tokens)
+
+    history_tokens = []
+    for round in rounds[::-1]:
+        round_tokens = []
+        for message in round:
+            if message["role"] == "user":
+                round_tokens.append(model.generation_config.user_token_id)
+            else:
+                round_tokens.append(model.generation_config.assistant_token_id)
+            round_tokens.extend(tokenizer.encode(message["content"]))
+        if len(history_tokens) == 0 or len(history_tokens) + len(round_tokens) <= max_history_tokens:
+            history_tokens = round_tokens + history_tokens  # concat left
+            if len(history_tokens) < max_history_tokens:
+                continue
+        break
+
+    input_tokens = system_tokens + history_tokens
+    if messages[-1]["role"] != "assistant":
+        input_tokens.append(model.generation_config.assistant_token_id)
+    input_tokens = input_tokens[-max_input_tokens:]  # truncate left
+    return torch.LongTensor([input_tokens]).to(model.device)
+
+
+class TextIterStreamer:
+    def __init__(self, tokenizer, skip_prompt=False, skip_special_tokens=False):
+        self.tokenizer = tokenizer
+        self.skip_prompt = skip_prompt
+        self.skip_special_tokens = skip_special_tokens
+        self.tokens = []
+        self.text_queue = Queue()
+        self.next_tokens_are_prompt = True
+
+    def put(self, value):
+        if self.skip_prompt and self.next_tokens_are_prompt:
+            self.next_tokens_are_prompt = False
+        else:
+            if len(value.shape) > 1:
+                value = value[0]
+            self.tokens.extend(value.tolist())
+            self.text_queue.put(
+                self.tokenizer.decode(self.tokens, skip_special_tokens=self.skip_special_tokens))
+
+    def end(self):
+        self.text_queue.put(None)
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        value = self.text_queue.get()
+        if value is None:
+            raise StopIteration()
+        else:
+            return value
+

latest

@@ -0,0 +1 @@
+global_step5930

modeling_baichuan.py

@@ -0,0 +1,826 @@
+# Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
+
+from .configuration_baichuan import BaichuanConfig
+from .generation_utils import build_chat_input, TextIterStreamer
+
+import math
+from threading import Thread
+from typing import List, Optional, Tuple, Union
+
+import torch
+from torch import nn
+from torch.nn import CrossEntropyLoss
+from torch.nn import functional as F
+from transformers import PreTrainedModel, PretrainedConfig
+from transformers.activations import ACT2FN
+from transformers.generation.utils import GenerationConfig
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import logging, ContextManagers
+
+import os
+from contextlib import contextmanager
+from accelerate import init_empty_weights
+
+logger = logging.get_logger(__name__)
+
+try:
+    from xformers import ops as xops
+except ImportError:
+    xops = None
+    logger.warning(
+        "Xformers is not installed correctly. If you want to use memory_efficient_attention to accelerate training use the following command to install Xformers\npip install xformers."
+    )
+
+
+def _get_interleave(n):
+    def _get_interleave_power_of_2(n):
+        start = 2 ** (-(2 ** -(math.log2(n) - 3)))
+        ratio = start
+        return [start * ratio**i for i in range(n)]
+
+    if math.log2(n).is_integer():
+        return _get_interleave_power_of_2(n)
+    else:
+        closest_power_of_2 = 2 ** math.floor(math.log2(n))
+        return (
+            _get_interleave_power_of_2(closest_power_of_2)
+            + _get_interleave(2 * closest_power_of_2)[0::2][: n - closest_power_of_2]
+        )
+
+
+def _fill_with_neg_inf(t):
+    """FP16-compatible function that fills a tensor with -inf."""
+    return t.float().fill_(float("-inf")).type_as(t)
+
+
+def _buffered_future_mask(tensor, maxpos, alibi, attn_heads):
+    _future_mask = torch.triu(_fill_with_neg_inf(torch.zeros([maxpos, maxpos])), 1)
+    _future_mask = _future_mask.unsqueeze(0) + alibi
+    new_future_mask = _future_mask.to(tensor)
+    return new_future_mask[: tensor.shape[0] * attn_heads, :maxpos, :maxpos]
+
+
+def _gen_alibi_mask(tensor, n_head, max_pos):
+    slopes = torch.Tensor(_get_interleave(n_head))
+    position_point = torch.arange(max_pos) - max_pos + 1
+    position_point = position_point.unsqueeze(0).unsqueeze(0).expand(n_head, -1, -1)
+    diag = torch.diag(position_point[0])
+    position_point = position_point - diag.unsqueeze(0).unsqueeze(0).transpose(-1, -2)
+    alibi = slopes.unsqueeze(1).unsqueeze(1) * position_point
+    alibi = alibi.view(n_head, 1, max_pos)
+    alibi_mask = torch.triu(_fill_with_neg_inf(torch.zeros([max_pos, max_pos])), 1)
+    alibi_mask = alibi_mask.unsqueeze(0) + alibi
+    return alibi_mask
+
+
+class RMSNorm(torch.nn.Module):
+    def __init__(self, hidden_size, epsilon=1e-6):
+        super().__init__()
+        self.weight = torch.nn.Parameter(torch.empty(hidden_size))
+        self.epsilon = epsilon
+
+    def forward(self, hidden_states):
+        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.epsilon)
+
+        # convert into half-precision
+        if self.weight.dtype in [torch.float16, torch.bfloat16]:
+            hidden_states = hidden_states.to(self.weight.dtype)
+
+        return self.weight * hidden_states
+
+
+class MLP(torch.nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        intermediate_size: int,
+        hidden_act: str,
+    ):
+        super().__init__()
+        self.gate_proj = torch.nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.down_proj = torch.nn.Linear(intermediate_size, hidden_size, bias=False)
+        self.up_proj = torch.nn.Linear(hidden_size, intermediate_size, bias=False)
+        self.act_fn = ACT2FN[hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+class BaichuanAttention(torch.nn.Module):
+    def __init__(self, config: BaichuanConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.max_position_embeddings = config.model_max_length
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size {self.hidden_size} is not divisible by num_heads {self.num_heads}"
+            )
+        self.W_pack = torch.nn.Linear(
+            self.hidden_size, 3 * self.hidden_size, bias=False
+        )
+        self.o_proj = torch.nn.Linear(
+            self.num_heads * self.head_dim, self.hidden_size, bias=False
+        )
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return (
+            tensor.view(bsz, seq_len, self.num_heads, self.head_dim)
+            .transpose(1, 2)
+            .contiguous()
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        proj = self.W_pack(hidden_states)
+        proj = (
+            proj.unflatten(-1, (3, self.hidden_size))
+            .unsqueeze(0)
+            .transpose(0, -2)
+            .squeeze(-2)
+        )
+        query_states = (
+            proj[0].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        )
+        key_states = (
+            proj[1].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        )
+        value_states = (
+            proj[2].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        )
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+        if xops is not None and self.training:
+            attn_weights = None
+            # query_states = query_states.transpose(1, 2)
+            # key_states = key_states.transpose(1, 2)
+            # value_states = value_states.transpose(1, 2)
+            # attn_output = xops.memory_efficient_attention(
+            #     query_states, key_states, value_states, attn_bias=attention_mask
+            # )
+            with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=True):
+                attn_output = F.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask = attention_mask)
+            attn_output = attn_output.transpose(1, 2)
+        else:
+            attn_weights = torch.matmul(
+                query_states, key_states.transpose(2, 3)
+            ) / math.sqrt(self.head_dim)
+
+            if attention_mask is not None:
+                if q_len == 1:  # inference with cache
+                    if len(attention_mask.size()) == 4:
+                        attention_mask = attention_mask[:, :, -1:, :]
+                    else:
+                        attention_mask = attention_mask[:, -1:, :]
+                attn_weights = attn_weights + attention_mask
+                attn_weights = torch.max(
+                    attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min)
+                )
+
+            attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1)
+            attn_output = torch.matmul(attn_weights, value_states)
+
+            attn_output = attn_output.transpose(1, 2)
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class BaichuanLayer(torch.nn.Module):
+    def __init__(self, config: BaichuanConfig):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = BaichuanAttention(config=config)
+        self.mlp = MLP(
+            hidden_size=self.hidden_size,
+            intermediate_size=config.intermediate_size,
+            hidden_act=config.hidden_act,
+        )
+        self.input_layernorm = RMSNorm(config.hidden_size, epsilon=config.rms_norm_eps)
+        self.post_attention_layernorm = RMSNorm(
+            config.hidden_size, epsilon=config.rms_norm_eps
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+    ) -> Tuple[
+        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
+    ]:
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+class BaichuanPreTrainedModel(PreTrainedModel):
+    config_class = BaichuanConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["BaichuanLayer"]
+    _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, torch.nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, torch.nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, BaichuanModel):
+            module.gradient_checkpointing = value
+
+
+class BaichuanModel(BaichuanPreTrainedModel):
+    def __init__(self, config: BaichuanConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+        self.n_head = config.num_attention_heads
+        self.embed_tokens = torch.nn.Embedding(
+            config.vocab_size, config.hidden_size, self.padding_idx
+        )
+        self.layers = torch.nn.ModuleList(
+            [BaichuanLayer(config) for _ in range(config.num_hidden_layers)]
+        )
+        self.norm = RMSNorm(config.hidden_size, epsilon=config.rms_norm_eps)
+
+        self.gradient_checkpointing = config.gradient_checkpointing
+        self.post_init()
+        self.max_cache_pos = config.model_max_length
+        self.first_run = True
+        self.alibi_mask = None
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    def get_alibi_mask(self, tensor, seq_length_with_past):
+        if self.training:
+            slopes = torch.Tensor(_get_interleave(self.n_head))
+            position_point = (
+                torch.arange(seq_length_with_past) - seq_length_with_past + 1
+            )
+            position_point = (
+                position_point.unsqueeze(0)
+                .unsqueeze(0)
+                .expand(self.n_head, seq_length_with_past, -1)
+            )
+            diag = torch.diag(position_point[0])
+            position_point = position_point - diag.unsqueeze(0).unsqueeze(0).transpose(
+                -1, -2
+            )
+            alibi = slopes.unsqueeze(1).unsqueeze(1) * position_point
+            mask = _buffered_future_mask(
+                tensor, seq_length_with_past, alibi, self.n_head
+            )
+        else:
+            if self.first_run:
+                self.first_run = False
+                self.register_buffer(
+                    "future_mask",
+                    _gen_alibi_mask(tensor, self.n_head, self.max_cache_pos).to(
+                        tensor
+                    ),
+                    persistent=False,
+                )
+            if seq_length_with_past > self.max_cache_pos:
+                self.max_cache_pos = seq_length_with_past
+                self.register_buffer(
+                    "future_mask",
+                    _gen_alibi_mask(tensor, self.n_head, self.max_cache_pos).to(
+                        tensor
+                    ),
+                    persistent=False,
+                )
+            mask = self.future_mask[
+                : self.n_head, :seq_length_with_past, :seq_length_with_past
+            ]
+        return mask
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        output_hidden_states: Optional[bool] = False,
+        return_dict: Optional[bool] = True,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot provide both input_ids and inputs_embeds simultaneously"
+            )
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError("You need to provide input_ids or inputs_embeds")
+
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        seq_length_with_past = seq_length
+
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if self.training:
+            if (
+                self.alibi_mask is None
+                or self.alibi_mask.shape[-1] != seq_length_with_past
+            ):
+                self.alibi_mask = self.get_alibi_mask(
+                    inputs_embeds, seq_length_with_past
+                )
+            alibi_mask = self.alibi_mask
+        else:
+            alibi_mask = self.get_alibi_mask(inputs_embeds, seq_length_with_past)
+
+        if attention_mask is not None:
+            if len(attention_mask.shape) == 2:
+                expanded_mask = attention_mask.to(alibi_mask.dtype)
+                expanded_mask = torch.tril(
+                    torch.gt(expanded_mask[:, :, None] * expanded_mask[:, None, :], 0)
+                ) * torch.eq(expanded_mask[:, :, None] - expanded_mask[:, None, :], 0)
+            else:
+                expanded_mask = attention_mask
+            bsz = inputs_embeds.size(0)
+            src_len, tgt_len = alibi_mask.size()[-2:]
+            expanded_mask = (
+                expanded_mask.unsqueeze(1)
+                .expand(bsz, 1, src_len, tgt_len)
+                .to(alibi_mask.dtype)
+            )
+            inverted_mask = 1.0 - expanded_mask
+            inverted_mask = inverted_mask.masked_fill(
+                inverted_mask.to(torch.bool), torch.finfo(alibi_mask.dtype).min
+            )
+            attention_mask = inverted_mask + alibi_mask.unsqueeze(0)
+        else:
+            attention_mask = alibi_mask
+
+        hidden_states = inputs_embeds
+
+        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
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = (
+                past_key_values[idx] if past_key_values is not None else None
+            )
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, output_attentions, None)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(decoder_layer),
+                    hidden_states,
+                    attention_mask,
+                    None,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
+                if v is not None
+            )
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+class NormHead(nn.Module):
+    def __init__(self, hidden_size, vocab_size, bias=False):
+        super().__init__()
+        self.weight = nn.Parameter(torch.empty((vocab_size, hidden_size)))
+        nn.init.kaiming_uniform_(self.weight, a=math.sqrt(5))
+        self.first_flag = True
+
+    def forward(self, hidden_states):
+        if self.training:
+            norm_weight = nn.functional.normalize(self.weight)
+        elif self.first_flag:
+            self.first_flag = False
+            self.weight = nn.Parameter(nn.functional.normalize(self.weight))
+            norm_weight = self.weight
+        else:
+            norm_weight = self.weight
+        return nn.functional.linear(hidden_states, norm_weight)
+
+_init_weights = True
+@contextmanager
+def no_init_weights(_enable=True):
+    global _init_weights
+    old_init_weights = _init_weights
+    if _enable:
+        _init_weights = False
+    try:
+        yield
+    finally:
+        _init_weights = old_init_weights
+
+
+class BaichuanForCausalLM(BaichuanPreTrainedModel):
+    def __init__(self, config, *model_args, **model_kwargs):
+        super().__init__(config, *model_args, **model_kwargs)
+        self.model = BaichuanModel(config)
+        self.lm_head = NormHead(config.hidden_size, config.vocab_size, bias=False)
+        #if hasattr(config, "quantization_config") and config.quantization_config['load_in_4bit']:
+        if hasattr(config, "quantization_config") and isinstance(config.quantization_config, dict) and config.quantization_config.get('load_in_4bit', False):
+            try:
+                from .quantizer import quantize_offline, init_model_weight_int4
+            except ImportError:
+                raise ImportError(f"Needs quantize_offline to run quantize.")
+            quantize_offline(self, 4)
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+    
+    @classmethod
+    def from_pretrained(
+        cls,
+        pretrained_model_name_or_path: Optional[Union[str, os.PathLike]],
+        *model_args,
+        config: Optional[Union[PretrainedConfig, str, os.PathLike]] = None,
+        cache_dir: Optional[Union[str, os.PathLike]] = None,
+        ignore_mismatched_sizes: bool = False,
+        force_download: bool = False,
+        local_files_only: bool = False,
+        token: Optional[Union[str, bool]] = None,
+        revision: str = "main",
+        use_safetensors: bool = None,
+        **kwargs,
+    ):
+    
+        # Load config if we don't provide a configuration
+        if not isinstance(config, PretrainedConfig):
+            config_path = config if config is not None else pretrained_model_name_or_path
+            config, model_kwargs = cls.config_class.from_pretrained(
+                config_path,
+                cache_dir=cache_dir,
+                return_unused_kwargs=True,
+                force_download=force_download,
+                resume_download=False,
+                proxies=None,
+                local_files_only=local_files_only,
+                token=token,
+                revision=revision,
+                subfolder="",
+                _from_auto=False,
+                _from_pipeline=None,
+                **kwargs,
+            )
+        else:
+            model_kwargs = kwargs
+        
+        if hasattr(config, "quantization_config") and config.quantization_config['load_in_4bit']:
+            try:
+                from .quantizer import init_model_weight_int4
+                from accelerate import init_empty_weights, dispatch_model, infer_auto_device_map
+                from accelerate.utils import CustomDtype
+                from accelerate.utils import get_balanced_memory
+            except ImportError:
+                raise ImportError(f"Needs import model weight init func to run quantize.") 
+            # Instantiate model.
+            init_contexts = [no_init_weights(_enable=True)]
+            init_contexts.append(init_empty_weights())
+            with ContextManagers(init_contexts):
+                model = cls(config)
+            
+            model_file = os.path.join(pretrained_model_name_or_path, 'pytorch_model.bin')
+            state_dict = torch.load(model_file, map_location="cpu") 
+            model.is_quantized = True
+                        
+            device_map = kwargs.pop("device_map", None)
+            torch_dtype = kwargs.pop("torch_dtype", None)
+            if device_map is not None:
+                kwargs = {"no_split_module_classes": model._no_split_modules}
+                target_dtype = CustomDtype.INT4
+                max_memory = get_balanced_memory(
+                    model,
+                    dtype=target_dtype,
+                    low_zero=(device_map == "balanced_low_0"),
+                    max_memory=None,
+                    **kwargs,
+                )
+                kwargs["max_memory"] = max_memory
+                device_map = infer_auto_device_map(model, dtype=target_dtype, **kwargs)
+            model = init_model_weight_int4(config, model, state_dict)
+            
+            # Set model in evaluation mode to deactivate DropOut modules by default
+            model.eval()
+            # If it is a model with generation capabilities, attempt to load the generation config
+            if model.can_generate():
+                try:
+                    model.generation_config = GenerationConfig.from_pretrained(
+                        pretrained_model_name_or_path,
+                        cache_dir=cache_dir,
+                        force_download=force_download,
+                        resume_download=False,
+                        proxies=None,
+                        local_files_only=local_files_only,
+                        token=token,
+                        revision=revision,
+                        subfolder="",
+                        _from_auto=False,
+                        _from_pipeline=None,
+                        **kwargs,
+                    )
+                except (OSError, TypeError):
+                    logger.info(
+                        "Generation config file not found, using a generation config created from the model config."
+                    )
+                    pass
+            
+            if device_map is not None:
+                dispatch_model(model, device_map=device_map)
+            
+            return model
+
+        return super(BaichuanForCausalLM, cls).from_pretrained(pretrained_model_name_or_path, *model_args, 
+                config=config, cache_dir=cache_dir, ignore_mismatched_sizes=ignore_mismatched_sizes, 
+                force_download=force_download, local_files_only=local_files_only, token=token, revision=revision, 
+                use_safetensors=use_safetensors, **kwargs)
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = False,
+        output_hidden_states: Optional[bool] = False,
+        return_dict: Optional[bool] = True,
+        **kwargs,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            softmax_normalizer = shift_logits.max(-1).values ** 2
+            z_loss = self.config.z_loss_weight * softmax_normalizer.mean()
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels) + z_loss
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def quantize(self, bits: int):
+        try:
+            from .quantizer import quantize_online
+        except ImportError:
+            raise ImportError(f"Needs QLinear to run quantize.")
+        return quantize_online(self, bits)
+        
+    def prepare_inputs_for_generation(
+        self,
+        input_ids: torch.LongTensor,
+        past_key_values: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        **kwargs,
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        return tuple(
+            tuple(past_state.index_select(0, beam_idx) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+
+    def _build_chat_input(
+        self, tokenizer, messages: List[dict], max_new_tokens: int = 0
+    ):
+        max_new_tokens = max_new_tokens or self.generation_config.max_new_tokens
+        max_input_tokens = self.config.model_max_length - max_new_tokens
+        max_input_tokens = max(self.config.model_max_length // 2, max_input_tokens)
+        total_input, round_input = [], []
+        for i, message in enumerate(messages[::-1]):
+            content_tokens = tokenizer.encode(message["content"])
+            if message["role"] == "user":
+                round_input = (
+                    [self.generation_config.user_token_id]
+                    + content_tokens
+                    + round_input
+                )
+                if (
+                    total_input
+                    and len(total_input) + len(round_input) > max_input_tokens
+                ):
+                    break
+                else:
+                    total_input = round_input + total_input
+                    if len(total_input) >= max_input_tokens:
+                        break
+                    else:
+                        round_input = []
+            elif message["role"] == "assistant":
+                round_input = (
+                    [self.generation_config.assistant_token_id]
+                    + content_tokens
+                    + [self.generation_config.eos_token_id]
+                    + round_input
+                )
+            else:
+                raise ValueError(f"message role not supported yet: {message['role']}")
+        total_input = total_input[-max_input_tokens:]  # truncate left
+        total_input.append(self.generation_config.assistant_token_id)
+        total_input = torch.LongTensor([total_input]).to(self.device)
+        return total_input
+
+    def chat(self, tokenizer, messages: List[dict], stream=False,
+             generation_config: Optional[GenerationConfig]=None):
+        generation_config = generation_config or self.generation_config
+        input_ids = build_chat_input(self, tokenizer, messages, generation_config.max_new_tokens)
+        if stream:
+            streamer = TextIterStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
+            Thread(target=self.generate, kwargs=dict(
+                inputs=input_ids, streamer=streamer,
+                generation_config=generation_config,
+            )).start()
+            return streamer
+        else:
+            outputs = self.generate(input_ids, generation_config=generation_config)
+            response = tokenizer.decode(outputs[0][len(input_ids[0]):], skip_special_tokens=True)
+            return response

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5070f02c0778e02cdd67d351df660662ae062a92879858b0dadcee06128a07ca
+size 29080502643

quantizer.py

@@ -0,0 +1,211 @@
+import bitsandbytes as bnb
+from accelerate import init_empty_weights
+from bitsandbytes.nn.modules import Params4bit, Int8Params
+import torch 
+
+def Params4bitCuda(self, device):
+    self.data = self.data.cuda(device)
+    self.quant_state[0] = self.quant_state[0].cuda(device)
+    self.quant_state[4][0] = self.quant_state[4][0].cuda(device)
+    self.quant_state[4][1][0] = self.quant_state[4][1][0].cuda(device)
+    self.quant_state[4][1][1] = self.quant_state[4][1][1].cuda(device)
+
+    self.quant_state[6] = self.quant_state[6].cuda(device)
+    return self
+
+class Linear4bitOnline(torch.nn.Module):
+    def __init__(self, weight, bias, quant_type):
+        super().__init__()
+        self.weight = Params4bit(
+            weight.data, requires_grad=False, compress_statistics=True, quant_type=quant_type
+        )
+        self.compute_dtype = None
+        #self.weight.cuda(weight.device)
+        self.bias = bias
+
+    def forward(self, x: torch.Tensor):
+        # weights are cast automatically as Int8Params, but the bias has to be cast manually
+        if self.bias is not None and self.bias.dtype != x.dtype:
+            self.bias.data = self.bias.data.to(x.dtype)
+
+        if getattr(self.weight, "quant_state", None) is None:
+            print(
+                "FP4 quantization state not initialized. Please call .cuda() or .to(device) on the LinearFP4 layer first."
+            )
+        inp_dtype = x.dtype
+        if self.compute_dtype is not None:
+            x = x.to(self.compute_dtype)
+
+        bias = None if self.bias is None else self.bias.to(self.compute_dtype)
+        out = bnb.matmul_4bit(
+            x, self.weight.t(), bias=bias, quant_state=self.weight.quant_state
+        )
+
+        out = out.to(inp_dtype)
+
+        return out
+    
+class Linear8bitLtOnline(torch.nn.Module):
+    def __init__(
+        self,
+        weight,
+        bias,
+        has_fp16_weights=True,
+        memory_efficient_backward=False,
+        threshold=0.0,
+        index=None,
+    ):
+        super().__init__()
+        assert (
+            not memory_efficient_backward
+        ), "memory_efficient_backward is no longer required and the argument is deprecated in 0.37.0 and will be removed in 0.39.0"
+        self.state = bnb.MatmulLtState()
+        self.index = index
+
+        # Necessary for stacked layers
+        self.state.threshold = threshold
+        self.state.has_fp16_weights = has_fp16_weights
+        self.state.memory_efficient_backward = memory_efficient_backward
+        if threshold > 0.0 and not has_fp16_weights:
+            self.state.use_pool = True
+
+        self.weight = Int8Params(
+            weight.data,
+            has_fp16_weights=has_fp16_weights,
+            requires_grad=has_fp16_weights,
+        )
+        self.bias = bias
+
+    def init_8bit_state(self):
+        self.state.CB = self.weight.CB
+        self.state.SCB = self.weight.SCB
+        self.weight.CB = None
+        self.weight.SCB = None
+
+    def forward(self, x: torch.Tensor):
+        self.state.is_training = self.training
+        if self.weight.CB is not None:
+            self.init_8bit_state()
+
+        # weights are cast automatically as Int8Params, but the bias has to be cast manually
+        if self.bias is not None and self.bias.dtype != x.dtype:
+            self.bias.data = self.bias.data.to(x.dtype)
+        
+        out = bnb.matmul(x, self.weight, bias=self.bias, state=self.state)
+
+        if not self.state.has_fp16_weights:
+            if self.state.CB is not None and self.state.CxB is not None:
+                # we converted 8-bit row major to turing/ampere format in the first inference pass
+                # we no longer need the row-major weight
+                del self.state.CB
+                self.weight.data = self.state.CxB
+        return out
+    
+def quantize_offline(model, bits: int):
+    assert (bits == 4), f'bits: {bits} is not supported'
+    
+    for i, layer in enumerate(model.model.layers):
+        layer.self_attn.W_pack = bnb.nn.Linear4bit(
+                            layer.self_attn.W_pack.weight.shape[1],
+                            layer.self_attn.W_pack.weight.shape[0],
+                            False,
+                            torch.float16,
+                            compress_statistics=True,
+                            quant_type="nf4",
+                        )
+        layer.self_attn.o_proj = bnb.nn.Linear4bit(
+                            layer.self_attn.o_proj.weight.shape[1],
+                            layer.self_attn.o_proj.weight.shape[0],
+                            False,
+                            torch.float16,
+                            compress_statistics=True,
+                            quant_type="nf4",
+                        )
+
+        layer.mlp.gate_proj = bnb.nn.Linear4bit(
+                            layer.mlp.gate_proj.weight.shape[1],
+                            layer.mlp.gate_proj.weight.shape[0],
+                            False,
+                            torch.float16,
+                            compress_statistics=True,
+                            quant_type="nf4",
+                        )
+        layer.mlp.down_proj = bnb.nn.Linear4bit(
+                            layer.mlp.down_proj.weight.shape[1],
+                            layer.mlp.down_proj.weight.shape[0],
+                            False,
+                            torch.float16,
+                            compress_statistics=True,
+                            quant_type="nf4",
+                        )
+        layer.mlp.up_proj = bnb.nn.Linear4bit(
+                            layer.mlp.up_proj.weight.shape[1],
+                            layer.mlp.up_proj.weight.shape[0],
+                            False,
+                            torch.float16,
+                            compress_statistics=True,
+                            quant_type="nf4",
+                        )
+    return model
+
+def quantize_online(model, bits: int):
+    def quant(weight, bias=None):
+        if bits == 8:
+            linear = Linear8bitLtOnline(
+                weight,
+                bias,
+                has_fp16_weights=False,
+                threshold=6.0,
+            )
+            if bias is not None:
+                linear.bias = torch.nn.Parameter(bias)
+        elif bits == 4:
+            linear = Linear4bitOnline(
+                weight,
+                bias,
+                quant_type="nf4", #fp4/nf4
+            )
+        else:
+            raise ValueError("quantize only support 4/8 bit")
+        return linear
+
+    for i, layer in enumerate(model.model.layers):
+        layer.self_attn.W_pack = quant(layer.self_attn.W_pack.weight)
+        layer.self_attn.o_proj = quant(layer.self_attn.o_proj.weight)
+        layer.mlp.gate_proj = quant(layer.mlp.gate_proj.weight)
+        layer.mlp.down_proj = quant(layer.mlp.down_proj.weight)
+        layer.mlp.up_proj = quant(layer.mlp.up_proj.weight)
+    return model
+
+def init_model_weight_int4(config, model, state_dict):
+    #replace Params4bit.cuda with Params4bitCuda
+    Params4bit.cuda = Params4bitCuda
+
+    for i in range(config.num_hidden_layers):
+        weight_data = state_dict[f'model.layers.{i}.self_attn.W_pack.weight.data']
+        weight_quant_state = state_dict[f'model.layers.{i}.self_attn.W_pack.weight.quant_state']
+        model.model.layers[i].self_attn.W_pack.weight = Params4bit(weight_data, requires_grad=False, quant_state=weight_quant_state)
+        
+        weight_data = state_dict[f'model.layers.{i}.self_attn.o_proj.weight.data']
+        weight_quant_state = state_dict[f'model.layers.{i}.self_attn.o_proj.weight.quant_state']
+        model.model.layers[i].self_attn.o_proj.weight = Params4bit(weight_data, requires_grad=False, quant_state=weight_quant_state)
+        
+        weight_data = state_dict[f'model.layers.{i}.mlp.gate_proj.weight.data']
+        weight_quant_state = state_dict[f'model.layers.{i}.mlp.gate_proj.weight.quant_state']
+        model.model.layers[i].mlp.gate_proj.weight = Params4bit(weight_data, requires_grad=False, quant_state=weight_quant_state)
+        
+        weight_data = state_dict[f'model.layers.{i}.mlp.up_proj.weight.data']
+        weight_quant_state = state_dict[f'model.layers.{i}.mlp.up_proj.weight.quant_state']
+        model.model.layers[i].mlp.up_proj.weight = Params4bit(weight_data, requires_grad=False, quant_state=weight_quant_state)
+        
+        weight_data = state_dict[f'model.layers.{i}.mlp.down_proj.weight.data']
+        weight_quant_state = state_dict[f'model.layers.{i}.mlp.down_proj.weight.quant_state']
+        model.model.layers[i].mlp.down_proj.weight = Params4bit(weight_data, requires_grad=False, quant_state=weight_quant_state)
+        
+        model.model.layers[i].input_layernorm.weight = state_dict[f'model.layers.{i}.input_layernorm.weight']
+        model.model.layers[i].post_attention_layernorm.weight = state_dict[f'model.layers.{i}.post_attention_layernorm.weight']
+    
+    model.model.embed_tokens.weight = state_dict['model.embed_tokens.weight']
+    model.model.norm.weight = state_dict['model.norm.weight']
+    model.lm_head.weight = state_dict['lm_head.weight'] 
+    return model

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4547d25b82ba64d9faa267de2e504b4ea84528cc19f0cc94c349c2b971e9872
+size 21623

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f22fa4a1570431c4194cf7b615c4d2f7a8906210d284301d228215bee88181b3
+size 21623

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da027964f40fd57685b85fd951af2b40686c88769ac372724aef8f4394c30560
+size 21623

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adb3731b987e36248fb79ac21dbe0840d4cb8ae55603151838c7bbf0ff64dc3c
+size 21623

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8762ee183e1560a34b62be47800a29fc56ff6a290d451e8ebe90eecd4a38199e
+size 21623

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63dfa85f80f60513066f18021a3be35cf5bbd7789513512a5c690256d110a741
+size 21623

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:454f8ce5599b4fcef2a91bbccbd18127a853e3ef0c511b43360421006b5deb96
+size 21623

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88b8ed012685f65992f02156e2bbdddae6eadf3754967bcaa9b0b7897e9f1c4f
+size 21623

special_tokens_map.json

@@ -0,0 +1,35 @@
+{
+  "additional_special_tokens": [
+    "<reserved_108>",
+    "<reserved_109>",
+    "<reserved_110>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "pad_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  }
+}

tokenization_baichuan.py

@@ -0,0 +1,258 @@
+# Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
+
+import os
+from shutil import copyfile
+from typing import Any, Dict, List, Optional, Tuple
+
+import sentencepiece as spm
+from transformers.tokenization_utils import AddedToken, PreTrainedTokenizer
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.model"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {},
+    "tokenizer_file": {},
+}
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {}
+
+
+class BaichuanTokenizer(PreTrainedTokenizer):
+    """
+    Construct a Baichuan tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["input_ids", "attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file,
+        unk_token="<unk>",
+        bos_token="<s>",
+        eos_token="</s>",
+        pad_token=None,
+        sp_model_kwargs: Optional[Dict[str, Any]] = None,
+        add_bos_token=True,
+        add_eos_token=False,
+        clean_up_tokenization_spaces=False,
+        **kwargs,
+    ):
+        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
+        bos_token = (
+            AddedToken(bos_token, lstrip=False, rstrip=False)
+            if isinstance(bos_token, str)
+            else bos_token
+        )
+        eos_token = (
+            AddedToken(eos_token, lstrip=False, rstrip=False)
+            if isinstance(eos_token, str)
+            else eos_token
+        )
+        unk_token = (
+            AddedToken(unk_token, lstrip=False, rstrip=False)
+            if isinstance(unk_token, str)
+            else unk_token
+        )
+        pad_token = (
+            AddedToken(pad_token, lstrip=False, rstrip=False)
+            if isinstance(pad_token, str)
+            else pad_token
+        )
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            pad_token=pad_token,
+            add_bos_token=add_bos_token,
+            add_eos_token=add_eos_token,
+            sp_model_kwargs=self.sp_model_kwargs,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            **kwargs,
+        )
+        self.vocab_file = vocab_file
+        self.add_bos_token = add_bos_token
+        self.add_eos_token = add_eos_token
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(vocab_file)
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(self.vocab_file)
+
+    @property
+    def vocab_size(self):
+        """Returns vocab size"""
+        return self.sp_model.get_piece_size()
+
+    def get_vocab(self):
+        """Returns vocab as a dict"""
+        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def _tokenize(self, text):
+        """Returns a tokenized string."""
+        return self.sp_model.encode(text, out_type=str)
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.sp_model.piece_to_id(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        token = self.sp_model.IdToPiece(index)
+        return token
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        current_sub_tokens = []
+        out_string = ""
+        prev_is_special = False
+        for i, token in enumerate(tokens):
+            # make sure that special tokens are not decoded using sentencepiece model
+            if token in self.all_special_tokens:
+                if not prev_is_special and i != 0:
+                    out_string += " "
+                out_string += self.sp_model.decode(current_sub_tokens) + token
+                prev_is_special = True
+                current_sub_tokens = []
+            else:
+                current_sub_tokens.append(token)
+                prev_is_special = False
+        out_string += self.sp_model.decode(current_sub_tokens)
+        return out_string
+
+    def save_vocabulary(
+        self, save_directory, filename_prefix: Optional[str] = None
+    ) -> Tuple[str]:
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        out_vocab_file = os.path.join(
+            save_directory,
+            (filename_prefix + "-" if filename_prefix else "")
+            + VOCAB_FILES_NAMES["vocab_file"],
+        )
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(
+            out_vocab_file
+        ) and os.path.isfile(self.vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
+
+        return (out_vocab_file,)
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = bos_token_id + token_ids_0 + eos_token_id
+
+        if token_ids_1 is not None:
+            output = output + bos_token_id + token_ids_1 + eos_token_id
+
+        return output
+
+    def get_special_tokens_mask(
+        self,
+        token_ids_0: List[int],
+        token_ids_1: Optional[List[int]] = None,
+        already_has_special_tokens: bool = False,
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer `prepare_for_model` method.
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0,
+                token_ids_1=token_ids_1,
+                already_has_special_tokens=True,
+            )
+
+        bos_token_id = [1] if self.add_bos_token else []
+        eos_token_id = [1] if self.add_eos_token else []
+
+        if token_ids_1 is None:
+            return bos_token_id + ([0] * len(token_ids_0)) + eos_token_id
+        return (
+            bos_token_id
+            + ([0] * len(token_ids_0))
+            + eos_token_id
+            + bos_token_id
+            + ([0] * len(token_ids_1))
+            + eos_token_id
+        )
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task. An ALBERT
+        sequence pair mask has the following format:
+
+        ```
+        0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence |
+        ```
+
+        if token_ids_1 is None, only returns the first portion of the mask (0s).
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of ids.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
+        """
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = [0] * len(bos_token_id + token_ids_0 + eos_token_id)
+
+        if token_ids_1 is not None:
+            output += [1] * len(bos_token_id + token_ids_1 + eos_token_id)
+
+        return output

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79452955be6b419a65984273a9f08af86042e1c2a75ee3ba989cbf620a133cc2
+size 2001107

tokenizer_config.json

@@ -0,0 +1,46 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_baichuan.BaichuanTokenizer",
+      null
+    ]
+  },
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "model_max_length": 4096,
+  "pad_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  },
+  "sp_model_kwargs": {},
+  "tokenizer_class": "BaichuanTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": true
+  }
+}

trainer_state.json

@@ -0,0 +1,3606 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "global_step": 5930,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.995951417004049e-06,
+      "loss": 4.0911,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.991902834008098e-06,
+      "loss": 2.8012,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.987854251012146e-06,
+      "loss": 2.389,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.983805668016194e-06,
+      "loss": 2.1658,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 9.979757085020244e-06,
+      "loss": 2.026,
+      "step": 50
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.975708502024292e-06,
+      "loss": 1.7846,
+      "step": 60
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.971659919028341e-06,
+      "loss": 1.703,
+      "step": 70
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 9.96761133603239e-06,
+      "loss": 1.5691,
+      "step": 80
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 9.963562753036437e-06,
+      "loss": 1.4297,
+      "step": 90
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.959514170040487e-06,
+      "loss": 1.4341,
+      "step": 100
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 9.955465587044535e-06,
+      "loss": 1.3304,
+      "step": 110
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 9.951417004048583e-06,
+      "loss": 1.3988,
+      "step": 120
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.947368421052632e-06,
+      "loss": 1.3272,
+      "step": 130
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 9.943319838056682e-06,
+      "loss": 1.2999,
+      "step": 140
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 9.93927125506073e-06,
+      "loss": 1.2997,
+      "step": 150
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 9.935222672064778e-06,
+      "loss": 1.3704,
+      "step": 160
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 9.931174089068828e-06,
+      "loss": 1.3313,
+      "step": 170
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 9.927125506072876e-06,
+      "loss": 1.3554,
+      "step": 180
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 9.923076923076923e-06,
+      "loss": 1.41,
+      "step": 190
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 9.919028340080973e-06,
+      "loss": 1.3005,
+      "step": 200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 9.914979757085021e-06,
+      "loss": 1.2311,
+      "step": 210
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 9.910931174089069e-06,
+      "loss": 1.3162,
+      "step": 220
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 9.906882591093119e-06,
+      "loss": 1.2895,
+      "step": 230
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 9.902834008097167e-06,
+      "loss": 1.2812,
+      "step": 240
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 9.898785425101216e-06,
+      "loss": 1.2682,
+      "step": 250
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 9.894736842105264e-06,
+      "loss": 1.2992,
+      "step": 260
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 9.890688259109312e-06,
+      "loss": 1.2832,
+      "step": 270
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 9.886639676113362e-06,
+      "loss": 1.2912,
+      "step": 280
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 9.88259109311741e-06,
+      "loss": 1.2949,
+      "step": 290
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 9.878542510121458e-06,
+      "loss": 1.2557,
+      "step": 300
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 9.874493927125507e-06,
+      "loss": 1.3119,
+      "step": 310
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 9.870445344129555e-06,
+      "loss": 1.3092,
+      "step": 320
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 9.866396761133603e-06,
+      "loss": 1.2441,
+      "step": 330
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 9.862348178137653e-06,
+      "loss": 1.3288,
+      "step": 340
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 9.8582995951417e-06,
+      "loss": 1.2382,
+      "step": 350
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 9.85425101214575e-06,
+      "loss": 1.2819,
+      "step": 360
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 9.850202429149798e-06,
+      "loss": 1.3011,
+      "step": 370
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 9.846153846153846e-06,
+      "loss": 1.2421,
+      "step": 380
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 9.842105263157896e-06,
+      "loss": 1.2071,
+      "step": 390
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 9.838056680161945e-06,
+      "loss": 1.2838,
+      "step": 400
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 9.834008097165993e-06,
+      "loss": 1.298,
+      "step": 410
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 9.829959514170041e-06,
+      "loss": 1.2867,
+      "step": 420
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 9.825910931174091e-06,
+      "loss": 1.1868,
+      "step": 430
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 9.821862348178139e-06,
+      "loss": 1.2263,
+      "step": 440
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 9.817813765182187e-06,
+      "loss": 1.2701,
+      "step": 450
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 9.813765182186236e-06,
+      "loss": 1.2542,
+      "step": 460
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 9.809716599190284e-06,
+      "loss": 1.2833,
+      "step": 470
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 9.805668016194332e-06,
+      "loss": 1.2109,
+      "step": 480
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 9.801619433198382e-06,
+      "loss": 1.2657,
+      "step": 490
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 9.79757085020243e-06,
+      "loss": 1.2587,
+      "step": 500
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 9.79352226720648e-06,
+      "loss": 1.2647,
+      "step": 510
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 9.789473684210527e-06,
+      "loss": 1.1917,
+      "step": 520
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 9.785425101214575e-06,
+      "loss": 1.2281,
+      "step": 530
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 9.781376518218625e-06,
+      "loss": 1.2677,
+      "step": 540
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 9.777327935222673e-06,
+      "loss": 1.2299,
+      "step": 550
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 9.77327935222672e-06,
+      "loss": 1.2524,
+      "step": 560
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 9.76923076923077e-06,
+      "loss": 1.2902,
+      "step": 570
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 9.765182186234818e-06,
+      "loss": 1.2603,
+      "step": 580
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 9.761133603238866e-06,
+      "loss": 1.2572,
+      "step": 590
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 9.757085020242916e-06,
+      "loss": 1.2223,
+      "step": 600
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 9.753036437246964e-06,
+      "loss": 1.2284,
+      "step": 610
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 9.748987854251012e-06,
+      "loss": 1.2331,
+      "step": 620
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 9.744939271255061e-06,
+      "loss": 1.2601,
+      "step": 630
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 9.74089068825911e-06,
+      "loss": 1.195,
+      "step": 640
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 9.736842105263159e-06,
+      "loss": 1.2176,
+      "step": 650
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 9.732793522267207e-06,
+      "loss": 1.2684,
+      "step": 660
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 9.728744939271257e-06,
+      "loss": 1.2575,
+      "step": 670
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 9.724696356275305e-06,
+      "loss": 1.2187,
+      "step": 680
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 9.720647773279354e-06,
+      "loss": 1.2184,
+      "step": 690
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 9.716599190283402e-06,
+      "loss": 1.282,
+      "step": 700
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 9.71255060728745e-06,
+      "loss": 1.2464,
+      "step": 710
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 9.7085020242915e-06,
+      "loss": 1.2041,
+      "step": 720
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 9.704453441295548e-06,
+      "loss": 1.26,
+      "step": 730
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 9.700404858299596e-06,
+      "loss": 1.2401,
+      "step": 740
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 9.696356275303645e-06,
+      "loss": 1.2289,
+      "step": 750
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 9.692307692307693e-06,
+      "loss": 1.2587,
+      "step": 760
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 9.688259109311741e-06,
+      "loss": 1.2078,
+      "step": 770
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 9.68421052631579e-06,
+      "loss": 1.192,
+      "step": 780
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 9.680161943319839e-06,
+      "loss": 1.2981,
+      "step": 790
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 9.676113360323888e-06,
+      "loss": 1.1942,
+      "step": 800
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 9.672064777327936e-06,
+      "loss": 1.247,
+      "step": 810
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 9.668016194331984e-06,
+      "loss": 1.1902,
+      "step": 820
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 9.663967611336034e-06,
+      "loss": 1.2271,
+      "step": 830
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 9.659919028340082e-06,
+      "loss": 1.2407,
+      "step": 840
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 9.65587044534413e-06,
+      "loss": 1.2067,
+      "step": 850
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 9.65182186234818e-06,
+      "loss": 1.2284,
+      "step": 860
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 9.647773279352227e-06,
+      "loss": 1.2676,
+      "step": 870
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 9.643724696356275e-06,
+      "loss": 1.2542,
+      "step": 880
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 9.639676113360325e-06,
+      "loss": 1.2364,
+      "step": 890
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 9.635627530364373e-06,
+      "loss": 1.3608,
+      "step": 900
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 9.63157894736842e-06,
+      "loss": 1.2739,
+      "step": 910
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 9.62753036437247e-06,
+      "loss": 1.1984,
+      "step": 920
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 9.62348178137652e-06,
+      "loss": 1.1875,
+      "step": 930
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 9.619433198380568e-06,
+      "loss": 1.2447,
+      "step": 940
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 1.26,
+      "step": 950
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 9.611336032388665e-06,
+      "loss": 1.2059,
+      "step": 960
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 9.607287449392713e-06,
+      "loss": 1.233,
+      "step": 970
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 9.603238866396763e-06,
+      "loss": 1.244,
+      "step": 980
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 9.59919028340081e-06,
+      "loss": 1.2613,
+      "step": 990
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 9.595141700404859e-06,
+      "loss": 1.2274,
+      "step": 1000
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 9.591093117408908e-06,
+      "loss": 1.2727,
+      "step": 1010
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 9.587044534412956e-06,
+      "loss": 1.2834,
+      "step": 1020
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 9.582995951417004e-06,
+      "loss": 1.259,
+      "step": 1030
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 9.578947368421054e-06,
+      "loss": 1.2742,
+      "step": 1040
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 9.574898785425102e-06,
+      "loss": 1.2658,
+      "step": 1050
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.57085020242915e-06,
+      "loss": 1.2203,
+      "step": 1060
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 9.5668016194332e-06,
+      "loss": 1.1757,
+      "step": 1070
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 9.562753036437247e-06,
+      "loss": 1.2296,
+      "step": 1080
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 9.558704453441297e-06,
+      "loss": 1.2714,
+      "step": 1090
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 9.554655870445345e-06,
+      "loss": 1.2265,
+      "step": 1100
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 9.550607287449393e-06,
+      "loss": 1.1984,
+      "step": 1110
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.546558704453442e-06,
+      "loss": 1.2706,
+      "step": 1120
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 9.54251012145749e-06,
+      "loss": 1.2611,
+      "step": 1130
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 9.538461538461538e-06,
+      "loss": 1.2584,
+      "step": 1140
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 9.534412955465588e-06,
+      "loss": 1.2849,
+      "step": 1150
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 9.530364372469636e-06,
+      "loss": 1.2254,
+      "step": 1160
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 9.526315789473686e-06,
+      "loss": 1.2698,
+      "step": 1170
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.522267206477733e-06,
+      "loss": 1.2037,
+      "step": 1180
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 9.518218623481783e-06,
+      "loss": 1.2408,
+      "step": 1190
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 9.514170040485831e-06,
+      "loss": 1.2116,
+      "step": 1200
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.510121457489879e-06,
+      "loss": 1.2393,
+      "step": 1210
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 9.506072874493929e-06,
+      "loss": 1.2177,
+      "step": 1220
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 9.502024291497977e-06,
+      "loss": 1.23,
+      "step": 1230
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 9.497975708502026e-06,
+      "loss": 1.2245,
+      "step": 1240
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 9.493927125506074e-06,
+      "loss": 1.2602,
+      "step": 1250
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 9.489878542510122e-06,
+      "loss": 1.2017,
+      "step": 1260
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 9.485829959514172e-06,
+      "loss": 1.2281,
+      "step": 1270
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 9.48178137651822e-06,
+      "loss": 1.2808,
+      "step": 1280
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 9.477732793522268e-06,
+      "loss": 1.2017,
+      "step": 1290
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 9.473684210526317e-06,
+      "loss": 1.2557,
+      "step": 1300
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 9.469635627530365e-06,
+      "loss": 1.2952,
+      "step": 1310
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 9.465587044534413e-06,
+      "loss": 1.2477,
+      "step": 1320
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 9.461538461538463e-06,
+      "loss": 1.2397,
+      "step": 1330
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 9.45748987854251e-06,
+      "loss": 1.1556,
+      "step": 1340
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 9.453441295546559e-06,
+      "loss": 1.217,
+      "step": 1350
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 9.449392712550608e-06,
+      "loss": 1.2725,
+      "step": 1360
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 9.445344129554656e-06,
+      "loss": 1.2079,
+      "step": 1370
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 9.441295546558706e-06,
+      "loss": 1.2646,
+      "step": 1380
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 9.437246963562754e-06,
+      "loss": 1.2034,
+      "step": 1390
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 9.433198380566802e-06,
+      "loss": 1.2442,
+      "step": 1400
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 9.429149797570851e-06,
+      "loss": 1.2463,
+      "step": 1410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 9.4251012145749e-06,
+      "loss": 1.2257,
+      "step": 1420
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 9.421052631578947e-06,
+      "loss": 1.2969,
+      "step": 1430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 9.417004048582997e-06,
+      "loss": 1.2288,
+      "step": 1440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.412955465587045e-06,
+      "loss": 1.2307,
+      "step": 1450
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 9.408906882591094e-06,
+      "loss": 1.2101,
+      "step": 1460
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 9.404858299595142e-06,
+      "loss": 1.2009,
+      "step": 1470
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 9.400809716599192e-06,
+      "loss": 1.2698,
+      "step": 1480
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 1.1532880067825317,
+      "eval_runtime": 11.4194,
+      "eval_samples_per_second": 96.327,
+      "eval_steps_per_second": 12.085,
+      "step": 1482
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 9.39676113360324e-06,
+      "loss": 1.1808,
+      "step": 1490
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 9.392712550607288e-06,
+      "loss": 1.1629,
+      "step": 1500
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 9.388663967611337e-06,
+      "loss": 1.1656,
+      "step": 1510
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 9.384615384615385e-06,
+      "loss": 1.1163,
+      "step": 1520
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 9.380566801619435e-06,
+      "loss": 1.1403,
+      "step": 1530
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 9.376518218623483e-06,
+      "loss": 1.122,
+      "step": 1540
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 9.37246963562753e-06,
+      "loss": 1.1426,
+      "step": 1550
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 9.36842105263158e-06,
+      "loss": 1.1142,
+      "step": 1560
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 9.364372469635628e-06,
+      "loss": 1.1143,
+      "step": 1570
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 9.360323886639676e-06,
+      "loss": 1.1798,
+      "step": 1580
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 9.356275303643726e-06,
+      "loss": 1.1575,
+      "step": 1590
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 9.352226720647774e-06,
+      "loss": 1.1459,
+      "step": 1600
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 9.348178137651822e-06,
+      "loss": 1.199,
+      "step": 1610
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 9.344129554655871e-06,
+      "loss": 1.1568,
+      "step": 1620
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 9.34008097165992e-06,
+      "loss": 1.1577,
+      "step": 1630
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 9.336032388663967e-06,
+      "loss": 1.0549,
+      "step": 1640
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 9.331983805668017e-06,
+      "loss": 1.1889,
+      "step": 1650
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 9.327935222672065e-06,
+      "loss": 1.1355,
+      "step": 1660
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 9.323886639676115e-06,
+      "loss": 1.1725,
+      "step": 1670
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 9.319838056680162e-06,
+      "loss": 1.1096,
+      "step": 1680
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 9.315789473684212e-06,
+      "loss": 1.2392,
+      "step": 1690
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 9.31174089068826e-06,
+      "loss": 1.1508,
+      "step": 1700
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 9.30769230769231e-06,
+      "loss": 1.1437,
+      "step": 1710
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 9.303643724696358e-06,
+      "loss": 1.204,
+      "step": 1720
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 9.299595141700406e-06,
+      "loss": 1.1795,
+      "step": 1730
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 9.295546558704455e-06,
+      "loss": 1.112,
+      "step": 1740
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 9.291497975708503e-06,
+      "loss": 1.2072,
+      "step": 1750
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 9.287449392712551e-06,
+      "loss": 1.1518,
+      "step": 1760
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 9.2834008097166e-06,
+      "loss": 1.1479,
+      "step": 1770
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 9.279352226720649e-06,
+      "loss": 1.1183,
+      "step": 1780
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 9.275303643724697e-06,
+      "loss": 1.1493,
+      "step": 1790
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 9.271255060728746e-06,
+      "loss": 1.1603,
+      "step": 1800
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 9.267206477732794e-06,
+      "loss": 1.186,
+      "step": 1810
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 9.263157894736844e-06,
+      "loss": 1.1857,
+      "step": 1820
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 9.259109311740892e-06,
+      "loss": 1.1957,
+      "step": 1830
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 9.25506072874494e-06,
+      "loss": 1.1102,
+      "step": 1840
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 9.25101214574899e-06,
+      "loss": 1.1439,
+      "step": 1850
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 9.246963562753037e-06,
+      "loss": 1.1252,
+      "step": 1860
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 9.242914979757085e-06,
+      "loss": 1.1238,
+      "step": 1870
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 9.238866396761135e-06,
+      "loss": 1.135,
+      "step": 1880
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 9.234817813765183e-06,
+      "loss": 1.1542,
+      "step": 1890
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 9.23076923076923e-06,
+      "loss": 1.1481,
+      "step": 1900
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 9.22672064777328e-06,
+      "loss": 1.0846,
+      "step": 1910
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 9.222672064777328e-06,
+      "loss": 1.1346,
+      "step": 1920
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 9.218623481781376e-06,
+      "loss": 1.0831,
+      "step": 1930
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 9.214574898785426e-06,
+      "loss": 1.2031,
+      "step": 1940
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 9.210526315789474e-06,
+      "loss": 1.0651,
+      "step": 1950
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 9.206477732793523e-06,
+      "loss": 1.1237,
+      "step": 1960
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 9.202429149797571e-06,
+      "loss": 1.0612,
+      "step": 1970
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 9.198380566801621e-06,
+      "loss": 1.1377,
+      "step": 1980
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 9.194331983805669e-06,
+      "loss": 1.1692,
+      "step": 1990
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 9.190283400809718e-06,
+      "loss": 1.1136,
+      "step": 2000
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.186234817813766e-06,
+      "loss": 1.1361,
+      "step": 2010
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.182186234817814e-06,
+      "loss": 1.1663,
+      "step": 2020
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 9.178137651821864e-06,
+      "loss": 1.1054,
+      "step": 2030
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.174089068825912e-06,
+      "loss": 1.1427,
+      "step": 2040
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.17004048582996e-06,
+      "loss": 1.1392,
+      "step": 2050
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 9.16599190283401e-06,
+      "loss": 1.191,
+      "step": 2060
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 9.161943319838057e-06,
+      "loss": 1.1006,
+      "step": 2070
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 9.157894736842105e-06,
+      "loss": 1.1372,
+      "step": 2080
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 9.153846153846155e-06,
+      "loss": 1.1428,
+      "step": 2090
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 9.149797570850203e-06,
+      "loss": 1.1433,
+      "step": 2100
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 9.145748987854253e-06,
+      "loss": 1.0997,
+      "step": 2110
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 9.1417004048583e-06,
+      "loss": 1.1478,
+      "step": 2120
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 9.137651821862348e-06,
+      "loss": 1.0928,
+      "step": 2130
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 9.133603238866398e-06,
+      "loss": 1.0993,
+      "step": 2140
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 9.129554655870446e-06,
+      "loss": 1.1306,
+      "step": 2150
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 9.125506072874494e-06,
+      "loss": 1.1527,
+      "step": 2160
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 9.121457489878544e-06,
+      "loss": 1.1394,
+      "step": 2170
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 9.117408906882591e-06,
+      "loss": 1.1787,
+      "step": 2180
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 9.11336032388664e-06,
+      "loss": 1.1295,
+      "step": 2190
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 9.109311740890689e-06,
+      "loss": 1.1843,
+      "step": 2200
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 9.105263157894739e-06,
+      "loss": 1.1243,
+      "step": 2210
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 9.101214574898785e-06,
+      "loss": 1.1748,
+      "step": 2220
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 9.097165991902835e-06,
+      "loss": 1.1282,
+      "step": 2230
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 9.093117408906884e-06,
+      "loss": 1.0969,
+      "step": 2240
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 9.089068825910932e-06,
+      "loss": 1.158,
+      "step": 2250
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 9.085020242914982e-06,
+      "loss": 1.1417,
+      "step": 2260
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 9.08097165991903e-06,
+      "loss": 1.1397,
+      "step": 2270
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 9.076923076923078e-06,
+      "loss": 1.1622,
+      "step": 2280
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 9.072874493927127e-06,
+      "loss": 1.1217,
+      "step": 2290
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 9.068825910931175e-06,
+      "loss": 1.1527,
+      "step": 2300
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 9.064777327935223e-06,
+      "loss": 1.1927,
+      "step": 2310
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 9.060728744939273e-06,
+      "loss": 1.1505,
+      "step": 2320
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 9.05668016194332e-06,
+      "loss": 1.123,
+      "step": 2330
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 9.052631578947369e-06,
+      "loss": 1.1453,
+      "step": 2340
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 9.048582995951418e-06,
+      "loss": 1.1903,
+      "step": 2350
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 9.044534412955466e-06,
+      "loss": 1.144,
+      "step": 2360
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 9.040485829959514e-06,
+      "loss": 1.1469,
+      "step": 2370
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 9.036437246963564e-06,
+      "loss": 1.1463,
+      "step": 2380
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 9.032388663967612e-06,
+      "loss": 1.1702,
+      "step": 2390
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 9.028340080971661e-06,
+      "loss": 1.078,
+      "step": 2400
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 9.02429149797571e-06,
+      "loss": 1.0907,
+      "step": 2410
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 9.020242914979757e-06,
+      "loss": 1.1621,
+      "step": 2420
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 9.016194331983807e-06,
+      "loss": 1.0947,
+      "step": 2430
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 9.012145748987855e-06,
+      "loss": 1.1301,
+      "step": 2440
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 9.008097165991903e-06,
+      "loss": 1.1232,
+      "step": 2450
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 9.004048582995952e-06,
+      "loss": 1.1737,
+      "step": 2460
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 9e-06,
+      "loss": 1.1475,
+      "step": 2470
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 8.99595141700405e-06,
+      "loss": 1.1242,
+      "step": 2480
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 8.991902834008098e-06,
+      "loss": 1.1321,
+      "step": 2490
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 8.987854251012147e-06,
+      "loss": 1.1131,
+      "step": 2500
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 8.983805668016195e-06,
+      "loss": 1.1456,
+      "step": 2510
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 8.979757085020243e-06,
+      "loss": 1.0983,
+      "step": 2520
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 8.975708502024293e-06,
+      "loss": 1.1598,
+      "step": 2530
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 8.971659919028341e-06,
+      "loss": 1.1861,
+      "step": 2540
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 8.96761133603239e-06,
+      "loss": 1.1334,
+      "step": 2550
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.963562753036438e-06,
+      "loss": 1.1605,
+      "step": 2560
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 8.959514170040486e-06,
+      "loss": 1.136,
+      "step": 2570
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 8.955465587044536e-06,
+      "loss": 1.1162,
+      "step": 2580
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 8.951417004048584e-06,
+      "loss": 1.1115,
+      "step": 2590
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 8.947368421052632e-06,
+      "loss": 1.1025,
+      "step": 2600
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 8.943319838056681e-06,
+      "loss": 1.1616,
+      "step": 2610
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 8.93927125506073e-06,
+      "loss": 1.1407,
+      "step": 2620
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 8.935222672064777e-06,
+      "loss": 1.1016,
+      "step": 2630
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 8.931174089068827e-06,
+      "loss": 1.111,
+      "step": 2640
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 8.927125506072875e-06,
+      "loss": 1.1226,
+      "step": 2650
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 8.923076923076923e-06,
+      "loss": 1.1105,
+      "step": 2660
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 8.919028340080972e-06,
+      "loss": 1.1227,
+      "step": 2670
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 8.91497975708502e-06,
+      "loss": 1.1462,
+      "step": 2680
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 8.91093117408907e-06,
+      "loss": 1.1119,
+      "step": 2690
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 8.906882591093118e-06,
+      "loss": 1.1567,
+      "step": 2700
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 8.902834008097166e-06,
+      "loss": 1.1565,
+      "step": 2710
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 8.898785425101216e-06,
+      "loss": 1.1116,
+      "step": 2720
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 8.894736842105265e-06,
+      "loss": 1.1139,
+      "step": 2730
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 8.890688259109311e-06,
+      "loss": 1.1454,
+      "step": 2740
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 8.886639676113361e-06,
+      "loss": 1.1084,
+      "step": 2750
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 8.882591093117409e-06,
+      "loss": 1.1829,
+      "step": 2760
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 8.878542510121459e-06,
+      "loss": 1.1262,
+      "step": 2770
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 8.874493927125507e-06,
+      "loss": 1.1754,
+      "step": 2780
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 8.870445344129556e-06,
+      "loss": 1.1269,
+      "step": 2790
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 8.866396761133604e-06,
+      "loss": 1.1663,
+      "step": 2800
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 8.862348178137652e-06,
+      "loss": 1.1457,
+      "step": 2810
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 8.858299595141702e-06,
+      "loss": 1.1143,
+      "step": 2820
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 8.85425101214575e-06,
+      "loss": 1.1956,
+      "step": 2830
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 8.8502024291498e-06,
+      "loss": 1.152,
+      "step": 2840
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 8.846153846153847e-06,
+      "loss": 1.1365,
+      "step": 2850
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 8.842105263157895e-06,
+      "loss": 1.1519,
+      "step": 2860
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 8.838056680161945e-06,
+      "loss": 1.126,
+      "step": 2870
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 8.834008097165993e-06,
+      "loss": 1.1408,
+      "step": 2880
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 8.82995951417004e-06,
+      "loss": 1.1522,
+      "step": 2890
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 8.82591093117409e-06,
+      "loss": 1.1236,
+      "step": 2900
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 8.821862348178138e-06,
+      "loss": 1.0927,
+      "step": 2910
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 8.817813765182186e-06,
+      "loss": 1.21,
+      "step": 2920
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 8.813765182186236e-06,
+      "loss": 1.1119,
+      "step": 2930
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 8.809716599190284e-06,
+      "loss": 1.1383,
+      "step": 2940
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 8.805668016194333e-06,
+      "loss": 1.1438,
+      "step": 2950
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 8.801619433198381e-06,
+      "loss": 1.1747,
+      "step": 2960
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 1.1016342639923096,
+      "eval_runtime": 10.6294,
+      "eval_samples_per_second": 103.487,
+      "eval_steps_per_second": 12.983,
+      "step": 2965
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 8.79757085020243e-06,
+      "loss": 1.083,
+      "step": 2970
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 8.793522267206479e-06,
+      "loss": 1.0542,
+      "step": 2980
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 8.789473684210527e-06,
+      "loss": 1.0974,
+      "step": 2990
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 8.785425101214576e-06,
+      "loss": 1.0074,
+      "step": 3000
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 8.781376518218624e-06,
+      "loss": 1.0825,
+      "step": 3010
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 8.777327935222674e-06,
+      "loss": 1.065,
+      "step": 3020
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 8.77327935222672e-06,
+      "loss": 1.0996,
+      "step": 3030
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 8.76923076923077e-06,
+      "loss": 1.1079,
+      "step": 3040
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 8.76518218623482e-06,
+      "loss": 1.0803,
+      "step": 3050
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 8.761133603238867e-06,
+      "loss": 1.0923,
+      "step": 3060
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 8.757085020242915e-06,
+      "loss": 1.1171,
+      "step": 3070
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 8.753036437246965e-06,
+      "loss": 1.0404,
+      "step": 3080
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 8.748987854251013e-06,
+      "loss": 1.0881,
+      "step": 3090
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 8.744939271255061e-06,
+      "loss": 1.1536,
+      "step": 3100
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 8.74089068825911e-06,
+      "loss": 1.0868,
+      "step": 3110
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 8.736842105263158e-06,
+      "loss": 1.1422,
+      "step": 3120
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 8.732793522267208e-06,
+      "loss": 1.1048,
+      "step": 3130
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 8.728744939271256e-06,
+      "loss": 1.136,
+      "step": 3140
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 8.724696356275304e-06,
+      "loss": 1.07,
+      "step": 3150
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 8.720647773279354e-06,
+      "loss": 1.1336,
+      "step": 3160
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 8.716599190283401e-06,
+      "loss": 1.0946,
+      "step": 3170
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 8.71255060728745e-06,
+      "loss": 1.061,
+      "step": 3180
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 8.708502024291499e-06,
+      "loss": 1.0275,
+      "step": 3190
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 8.704453441295547e-06,
+      "loss": 1.0665,
+      "step": 3200
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 8.700404858299595e-06,
+      "loss": 1.0563,
+      "step": 3210
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 8.696356275303645e-06,
+      "loss": 1.0756,
+      "step": 3220
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 8.692307692307692e-06,
+      "loss": 1.0981,
+      "step": 3230
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 8.688259109311742e-06,
+      "loss": 1.1138,
+      "step": 3240
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 8.68421052631579e-06,
+      "loss": 1.0838,
+      "step": 3250
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 8.680161943319838e-06,
+      "loss": 1.1053,
+      "step": 3260
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 8.676113360323888e-06,
+      "loss": 1.1099,
+      "step": 3270
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 8.672064777327936e-06,
+      "loss": 1.0884,
+      "step": 3280
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 8.668016194331985e-06,
+      "loss": 1.0857,
+      "step": 3290
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 8.663967611336033e-06,
+      "loss": 1.0629,
+      "step": 3300
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 8.659919028340083e-06,
+      "loss": 1.0753,
+      "step": 3310
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 8.65587044534413e-06,
+      "loss": 1.1481,
+      "step": 3320
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 8.651821862348179e-06,
+      "loss": 1.1127,
+      "step": 3330
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 8.647773279352228e-06,
+      "loss": 1.1666,
+      "step": 3340
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 8.643724696356276e-06,
+      "loss": 1.1086,
+      "step": 3350
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 8.639676113360324e-06,
+      "loss": 1.1102,
+      "step": 3360
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 8.635627530364374e-06,
+      "loss": 1.1194,
+      "step": 3370
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 8.631578947368422e-06,
+      "loss": 1.145,
+      "step": 3380
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 8.62753036437247e-06,
+      "loss": 1.1103,
+      "step": 3390
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 8.62348178137652e-06,
+      "loss": 1.0998,
+      "step": 3400
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 8.619433198380567e-06,
+      "loss": 1.1551,
+      "step": 3410
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 8.615384615384615e-06,
+      "loss": 1.0687,
+      "step": 3420
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 8.611336032388665e-06,
+      "loss": 1.0959,
+      "step": 3430
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 8.607287449392713e-06,
+      "loss": 1.0799,
+      "step": 3440
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 8.60323886639676e-06,
+      "loss": 1.0673,
+      "step": 3450
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 8.59919028340081e-06,
+      "loss": 1.0873,
+      "step": 3460
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 8.595141700404858e-06,
+      "loss": 1.1119,
+      "step": 3470
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 8.591093117408906e-06,
+      "loss": 1.1356,
+      "step": 3480
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 8.587044534412956e-06,
+      "loss": 1.0924,
+      "step": 3490
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 8.582995951417005e-06,
+      "loss": 1.0818,
+      "step": 3500
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 8.578947368421053e-06,
+      "loss": 1.1007,
+      "step": 3510
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 8.574898785425103e-06,
+      "loss": 1.0591,
+      "step": 3520
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 8.57085020242915e-06,
+      "loss": 1.1069,
+      "step": 3530
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 8.566801619433199e-06,
+      "loss": 1.1191,
+      "step": 3540
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 8.562753036437247e-06,
+      "loss": 1.077,
+      "step": 3550
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 8.558704453441296e-06,
+      "loss": 1.104,
+      "step": 3560
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 8.554655870445344e-06,
+      "loss": 1.0758,
+      "step": 3570
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 8.550607287449394e-06,
+      "loss": 1.0885,
+      "step": 3580
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 8.546558704453442e-06,
+      "loss": 1.084,
+      "step": 3590
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 8.54251012145749e-06,
+      "loss": 1.0869,
+      "step": 3600
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 8.53846153846154e-06,
+      "loss": 1.073,
+      "step": 3610
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 8.534412955465587e-06,
+      "loss": 1.0766,
+      "step": 3620
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 8.530364372469635e-06,
+      "loss": 1.1085,
+      "step": 3630
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 8.526315789473685e-06,
+      "loss": 1.097,
+      "step": 3640
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 8.522267206477733e-06,
+      "loss": 1.0968,
+      "step": 3650
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 8.518218623481783e-06,
+      "loss": 1.0917,
+      "step": 3660
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 8.51417004048583e-06,
+      "loss": 1.047,
+      "step": 3670
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 8.510121457489878e-06,
+      "loss": 1.0469,
+      "step": 3680
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 8.506072874493928e-06,
+      "loss": 1.1114,
+      "step": 3690
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 8.502024291497976e-06,
+      "loss": 1.0319,
+      "step": 3700
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 8.497975708502024e-06,
+      "loss": 1.0733,
+      "step": 3710
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 8.493927125506074e-06,
+      "loss": 1.1237,
+      "step": 3720
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 8.489878542510121e-06,
+      "loss": 1.0649,
+      "step": 3730
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 8.48582995951417e-06,
+      "loss": 1.0903,
+      "step": 3740
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 8.481781376518219e-06,
+      "loss": 1.0633,
+      "step": 3750
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 8.477732793522267e-06,
+      "loss": 1.1259,
+      "step": 3760
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 8.473684210526317e-06,
+      "loss": 1.0978,
+      "step": 3770
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 8.469635627530365e-06,
+      "loss": 1.1362,
+      "step": 3780
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 8.465587044534414e-06,
+      "loss": 1.1487,
+      "step": 3790
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 8.461538461538462e-06,
+      "loss": 1.0887,
+      "step": 3800
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 8.457489878542512e-06,
+      "loss": 1.1039,
+      "step": 3810
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 8.453441295546558e-06,
+      "loss": 1.0906,
+      "step": 3820
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 8.449392712550608e-06,
+      "loss": 1.1039,
+      "step": 3830
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 8.445344129554657e-06,
+      "loss": 1.0432,
+      "step": 3840
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 8.441295546558705e-06,
+      "loss": 1.1565,
+      "step": 3850
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 8.437246963562753e-06,
+      "loss": 1.1786,
+      "step": 3860
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 8.433198380566803e-06,
+      "loss": 1.0724,
+      "step": 3870
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 8.42914979757085e-06,
+      "loss": 1.0863,
+      "step": 3880
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 8.425101214574899e-06,
+      "loss": 1.0825,
+      "step": 3890
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 8.421052631578948e-06,
+      "loss": 1.0831,
+      "step": 3900
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 8.417004048582996e-06,
+      "loss": 1.0681,
+      "step": 3910
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 8.412955465587044e-06,
+      "loss": 1.0871,
+      "step": 3920
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 8.408906882591094e-06,
+      "loss": 1.0531,
+      "step": 3930
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 8.404858299595142e-06,
+      "loss": 1.1174,
+      "step": 3940
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 8.400809716599191e-06,
+      "loss": 1.117,
+      "step": 3950
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 8.39676113360324e-06,
+      "loss": 1.0938,
+      "step": 3960
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 8.392712550607287e-06,
+      "loss": 1.0736,
+      "step": 3970
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 8.388663967611337e-06,
+      "loss": 1.0872,
+      "step": 3980
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 8.384615384615385e-06,
+      "loss": 1.1,
+      "step": 3990
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 8.380566801619433e-06,
+      "loss": 1.1201,
+      "step": 4000
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 8.376518218623482e-06,
+      "loss": 1.0755,
+      "step": 4010
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 8.37246963562753e-06,
+      "loss": 1.0917,
+      "step": 4020
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 8.368421052631578e-06,
+      "loss": 1.0873,
+      "step": 4030
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 8.364372469635628e-06,
+      "loss": 1.0858,
+      "step": 4040
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 8.360323886639676e-06,
+      "loss": 1.0721,
+      "step": 4050
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 8.356275303643725e-06,
+      "loss": 1.138,
+      "step": 4060
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 8.352226720647773e-06,
+      "loss": 1.1194,
+      "step": 4070
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 8.348178137651823e-06,
+      "loss": 1.1364,
+      "step": 4080
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 8.344129554655871e-06,
+      "loss": 1.1201,
+      "step": 4090
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 8.34008097165992e-06,
+      "loss": 1.0585,
+      "step": 4100
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 8.336032388663968e-06,
+      "loss": 1.0613,
+      "step": 4110
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 8.331983805668016e-06,
+      "loss": 1.1091,
+      "step": 4120
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 8.327935222672066e-06,
+      "loss": 1.1446,
+      "step": 4130
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 8.323886639676114e-06,
+      "loss": 1.0413,
+      "step": 4140
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 8.319838056680162e-06,
+      "loss": 1.1225,
+      "step": 4150
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 8.315789473684212e-06,
+      "loss": 1.1121,
+      "step": 4160
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 8.31174089068826e-06,
+      "loss": 1.0534,
+      "step": 4170
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 8.307692307692307e-06,
+      "loss": 1.1097,
+      "step": 4180
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 8.303643724696357e-06,
+      "loss": 1.1147,
+      "step": 4190
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 8.299595141700405e-06,
+      "loss": 1.1009,
+      "step": 4200
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 8.295546558704453e-06,
+      "loss": 1.0584,
+      "step": 4210
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 8.291497975708503e-06,
+      "loss": 1.1045,
+      "step": 4220
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 8.28744939271255e-06,
+      "loss": 1.0559,
+      "step": 4230
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 8.2834008097166e-06,
+      "loss": 1.063,
+      "step": 4240
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 8.279352226720648e-06,
+      "loss": 1.1671,
+      "step": 4250
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 8.275303643724696e-06,
+      "loss": 1.077,
+      "step": 4260
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.271255060728746e-06,
+      "loss": 1.1119,
+      "step": 4270
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 8.267206477732794e-06,
+      "loss": 1.0773,
+      "step": 4280
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 8.263157894736843e-06,
+      "loss": 1.0874,
+      "step": 4290
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 8.259109311740891e-06,
+      "loss": 1.0793,
+      "step": 4300
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 8.25506072874494e-06,
+      "loss": 1.1011,
+      "step": 4310
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 8.251012145748987e-06,
+      "loss": 1.1232,
+      "step": 4320
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 8.246963562753037e-06,
+      "loss": 1.0953,
+      "step": 4330
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 8.242914979757085e-06,
+      "loss": 1.0841,
+      "step": 4340
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 8.238866396761134e-06,
+      "loss": 1.1036,
+      "step": 4350
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 8.234817813765182e-06,
+      "loss": 1.1207,
+      "step": 4360
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 8.230769230769232e-06,
+      "loss": 1.1033,
+      "step": 4370
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 8.22672064777328e-06,
+      "loss": 1.0719,
+      "step": 4380
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 8.22267206477733e-06,
+      "loss": 1.0539,
+      "step": 4390
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 8.218623481781377e-06,
+      "loss": 1.0818,
+      "step": 4400
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 8.214574898785425e-06,
+      "loss": 1.1149,
+      "step": 4410
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 8.210526315789475e-06,
+      "loss": 1.0624,
+      "step": 4420
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 8.206477732793523e-06,
+      "loss": 1.1197,
+      "step": 4430
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 8.20242914979757e-06,
+      "loss": 1.0567,
+      "step": 4440
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 1.0876251459121704,
+      "eval_runtime": 10.6357,
+      "eval_samples_per_second": 103.425,
+      "eval_steps_per_second": 12.975,
+      "step": 4447
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 8.19838056680162e-06,
+      "loss": 1.0768,
+      "step": 4450
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 8.194331983805668e-06,
+      "loss": 1.0687,
+      "step": 4460
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 8.190283400809716e-06,
+      "loss": 1.1029,
+      "step": 4470
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 8.186234817813766e-06,
+      "loss": 1.0569,
+      "step": 4480
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 8.182186234817814e-06,
+      "loss": 1.0151,
+      "step": 4490
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 8.178137651821862e-06,
+      "loss": 1.0405,
+      "step": 4500
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 8.174089068825911e-06,
+      "loss": 1.07,
+      "step": 4510
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 8.17004048582996e-06,
+      "loss": 1.0643,
+      "step": 4520
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 8.165991902834009e-06,
+      "loss": 1.0859,
+      "step": 4530
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 8.161943319838057e-06,
+      "loss": 1.0384,
+      "step": 4540
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 8.157894736842105e-06,
+      "loss": 1.0873,
+      "step": 4550
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 8.153846153846154e-06,
+      "loss": 1.0847,
+      "step": 4560
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 8.149797570850202e-06,
+      "loss": 1.0105,
+      "step": 4570
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 8.145748987854252e-06,
+      "loss": 1.0407,
+      "step": 4580
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 8.1417004048583e-06,
+      "loss": 1.0437,
+      "step": 4590
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 8.13765182186235e-06,
+      "loss": 1.0641,
+      "step": 4600
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 8.133603238866397e-06,
+      "loss": 1.0662,
+      "step": 4610
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 8.129554655870445e-06,
+      "loss": 1.0608,
+      "step": 4620
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 8.125506072874495e-06,
+      "loss": 1.0728,
+      "step": 4630
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 8.121457489878543e-06,
+      "loss": 1.039,
+      "step": 4640
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 8.117408906882591e-06,
+      "loss": 1.0898,
+      "step": 4650
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 8.11336032388664e-06,
+      "loss": 1.107,
+      "step": 4660
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 8.109311740890688e-06,
+      "loss": 1.1001,
+      "step": 4670
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 8.105263157894738e-06,
+      "loss": 1.0371,
+      "step": 4680
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 8.101214574898786e-06,
+      "loss": 1.0945,
+      "step": 4690
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 8.097165991902834e-06,
+      "loss": 1.0509,
+      "step": 4700
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 8.093117408906884e-06,
+      "loss": 1.0694,
+      "step": 4710
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 8.089068825910931e-06,
+      "loss": 1.0153,
+      "step": 4720
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 8.08502024291498e-06,
+      "loss": 1.0373,
+      "step": 4730
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 8.080971659919029e-06,
+      "loss": 1.0583,
+      "step": 4740
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 8.076923076923077e-06,
+      "loss": 1.0213,
+      "step": 4750
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 8.072874493927125e-06,
+      "loss": 1.0707,
+      "step": 4760
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 8.068825910931175e-06,
+      "loss": 1.0801,
+      "step": 4770
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 8.064777327935222e-06,
+      "loss": 1.0336,
+      "step": 4780
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 8.060728744939272e-06,
+      "loss": 1.0829,
+      "step": 4790
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 8.05668016194332e-06,
+      "loss": 1.0853,
+      "step": 4800
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 8.05263157894737e-06,
+      "loss": 1.074,
+      "step": 4810
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 8.048582995951418e-06,
+      "loss": 1.0805,
+      "step": 4820
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 8.044534412955467e-06,
+      "loss": 1.0644,
+      "step": 4830
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 8.040485829959514e-06,
+      "loss": 1.039,
+      "step": 4840
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 8.036437246963563e-06,
+      "loss": 1.0326,
+      "step": 4850
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 8.032388663967611e-06,
+      "loss": 1.0776,
+      "step": 4860
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 8.02834008097166e-06,
+      "loss": 1.043,
+      "step": 4870
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 8.024291497975709e-06,
+      "loss": 1.0784,
+      "step": 4880
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 8.020242914979758e-06,
+      "loss": 1.0783,
+      "step": 4890
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 8.016194331983806e-06,
+      "loss": 1.0987,
+      "step": 4900
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 8.012145748987854e-06,
+      "loss": 1.0515,
+      "step": 4910
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 8.008097165991904e-06,
+      "loss": 1.0667,
+      "step": 4920
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 8.004048582995952e-06,
+      "loss": 1.1095,
+      "step": 4930
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 8e-06,
+      "loss": 1.0245,
+      "step": 4940
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 7.99595141700405e-06,
+      "loss": 1.0753,
+      "step": 4950
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 7.991902834008097e-06,
+      "loss": 1.0522,
+      "step": 4960
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 7.987854251012147e-06,
+      "loss": 1.0106,
+      "step": 4970
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 7.983805668016195e-06,
+      "loss": 1.0948,
+      "step": 4980
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 7.979757085020243e-06,
+      "loss": 1.0649,
+      "step": 4990
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 7.975708502024292e-06,
+      "loss": 1.0096,
+      "step": 5000
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 7.97165991902834e-06,
+      "loss": 1.0076,
+      "step": 5010
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 7.967611336032388e-06,
+      "loss": 1.0605,
+      "step": 5020
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 7.963562753036438e-06,
+      "loss": 1.0544,
+      "step": 5030
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 7.959514170040486e-06,
+      "loss": 1.0638,
+      "step": 5040
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 7.955465587044534e-06,
+      "loss": 1.021,
+      "step": 5050
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 7.951417004048583e-06,
+      "loss": 1.0622,
+      "step": 5060
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 7.947368421052631e-06,
+      "loss": 1.0452,
+      "step": 5070
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 7.943319838056681e-06,
+      "loss": 1.0678,
+      "step": 5080
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 7.939271255060729e-06,
+      "loss": 0.9841,
+      "step": 5090
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 7.935222672064778e-06,
+      "loss": 1.0684,
+      "step": 5100
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 7.931174089068826e-06,
+      "loss": 1.0393,
+      "step": 5110
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 7.927125506072876e-06,
+      "loss": 1.0816,
+      "step": 5120
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 7.923076923076922e-06,
+      "loss": 1.0744,
+      "step": 5130
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 7.919028340080972e-06,
+      "loss": 1.0575,
+      "step": 5140
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 7.914979757085022e-06,
+      "loss": 1.119,
+      "step": 5150
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 7.91093117408907e-06,
+      "loss": 1.0652,
+      "step": 5160
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 7.906882591093117e-06,
+      "loss": 0.9888,
+      "step": 5170
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 7.902834008097167e-06,
+      "loss": 1.0415,
+      "step": 5180
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 7.898785425101215e-06,
+      "loss": 1.0266,
+      "step": 5190
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 7.894736842105263e-06,
+      "loss": 1.0204,
+      "step": 5200
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 7.890688259109313e-06,
+      "loss": 1.0709,
+      "step": 5210
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 7.88663967611336e-06,
+      "loss": 1.0342,
+      "step": 5220
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 7.882591093117408e-06,
+      "loss": 1.0492,
+      "step": 5230
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 7.878542510121458e-06,
+      "loss": 1.0375,
+      "step": 5240
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 7.874493927125506e-06,
+      "loss": 1.0188,
+      "step": 5250
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 7.870445344129556e-06,
+      "loss": 1.0951,
+      "step": 5260
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 7.866396761133604e-06,
+      "loss": 1.0133,
+      "step": 5270
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 7.862348178137651e-06,
+      "loss": 1.0414,
+      "step": 5280
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 7.858299595141701e-06,
+      "loss": 1.0863,
+      "step": 5290
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 7.854251012145749e-06,
+      "loss": 1.0998,
+      "step": 5300
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 7.850202429149797e-06,
+      "loss": 1.0854,
+      "step": 5310
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 7.846153846153847e-06,
+      "loss": 1.0594,
+      "step": 5320
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 7.842105263157896e-06,
+      "loss": 1.0225,
+      "step": 5330
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 7.838056680161942e-06,
+      "loss": 1.0728,
+      "step": 5340
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 7.834008097165992e-06,
+      "loss": 1.0417,
+      "step": 5350
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 7.82995951417004e-06,
+      "loss": 1.0226,
+      "step": 5360
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 7.82591093117409e-06,
+      "loss": 1.0953,
+      "step": 5370
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 7.821862348178138e-06,
+      "loss": 1.053,
+      "step": 5380
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 7.817813765182187e-06,
+      "loss": 1.0523,
+      "step": 5390
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 7.813765182186235e-06,
+      "loss": 1.1017,
+      "step": 5400
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 7.809716599190285e-06,
+      "loss": 1.0032,
+      "step": 5410
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 7.805668016194333e-06,
+      "loss": 1.0331,
+      "step": 5420
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 7.80161943319838e-06,
+      "loss": 1.0483,
+      "step": 5430
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 7.79757085020243e-06,
+      "loss": 1.0291,
+      "step": 5440
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 7.793522267206478e-06,
+      "loss": 1.0113,
+      "step": 5450
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 7.789473684210526e-06,
+      "loss": 1.0407,
+      "step": 5460
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 7.785425101214576e-06,
+      "loss": 1.077,
+      "step": 5470
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 7.781376518218624e-06,
+      "loss": 1.1063,
+      "step": 5480
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 7.777327935222672e-06,
+      "loss": 1.0377,
+      "step": 5490
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 7.773279352226721e-06,
+      "loss": 1.0841,
+      "step": 5500
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 7.76923076923077e-06,
+      "loss": 1.0638,
+      "step": 5510
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 7.765182186234819e-06,
+      "loss": 1.0525,
+      "step": 5520
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 7.761133603238867e-06,
+      "loss": 1.0638,
+      "step": 5530
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 7.757085020242915e-06,
+      "loss": 1.0478,
+      "step": 5540
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 7.753036437246964e-06,
+      "loss": 1.0429,
+      "step": 5550
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 7.748987854251014e-06,
+      "loss": 1.0615,
+      "step": 5560
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 7.74493927125506e-06,
+      "loss": 1.0653,
+      "step": 5570
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 7.74089068825911e-06,
+      "loss": 1.0755,
+      "step": 5580
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 7.736842105263158e-06,
+      "loss": 1.0447,
+      "step": 5590
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 7.732793522267207e-06,
+      "loss": 1.0372,
+      "step": 5600
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 7.728744939271255e-06,
+      "loss": 1.1158,
+      "step": 5610
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 7.724696356275305e-06,
+      "loss": 1.1195,
+      "step": 5620
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 7.720647773279351e-06,
+      "loss": 1.0246,
+      "step": 5630
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 7.716599190283401e-06,
+      "loss": 1.0339,
+      "step": 5640
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 7.712550607287449e-06,
+      "loss": 1.0578,
+      "step": 5650
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 7.708502024291498e-06,
+      "loss": 1.0747,
+      "step": 5660
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 7.704453441295546e-06,
+      "loss": 1.0301,
+      "step": 5670
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 7.700404858299596e-06,
+      "loss": 0.9873,
+      "step": 5680
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 7.696356275303644e-06,
+      "loss": 1.0217,
+      "step": 5690
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 1.0941,
+      "step": 5700
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 7.688259109311742e-06,
+      "loss": 1.0842,
+      "step": 5710
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 7.68421052631579e-06,
+      "loss": 1.0163,
+      "step": 5720
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 7.680161943319839e-06,
+      "loss": 1.0387,
+      "step": 5730
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 7.676113360323887e-06,
+      "loss": 1.0719,
+      "step": 5740
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 7.672064777327935e-06,
+      "loss": 1.0495,
+      "step": 5750
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 7.668016194331985e-06,
+      "loss": 1.0669,
+      "step": 5760
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 7.663967611336033e-06,
+      "loss": 1.1226,
+      "step": 5770
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 7.65991902834008e-06,
+      "loss": 1.0147,
+      "step": 5780
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 7.65587044534413e-06,
+      "loss": 1.0244,
+      "step": 5790
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 7.651821862348178e-06,
+      "loss": 1.0134,
+      "step": 5800
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 7.647773279352228e-06,
+      "loss": 1.0215,
+      "step": 5810
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 7.643724696356276e-06,
+      "loss": 1.1177,
+      "step": 5820
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 7.639676113360325e-06,
+      "loss": 1.0472,
+      "step": 5830
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 7.635627530364373e-06,
+      "loss": 1.0335,
+      "step": 5840
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 7.631578947368423e-06,
+      "loss": 1.0346,
+      "step": 5850
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 7.627530364372468e-06,
+      "loss": 1.0354,
+      "step": 5860
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 7.623481781376519e-06,
+      "loss": 1.0663,
+      "step": 5870
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 7.619433198380567e-06,
+      "loss": 1.0602,
+      "step": 5880
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 7.615384615384615e-06,
+      "loss": 1.0458,
+      "step": 5890
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 7.611336032388663e-06,
+      "loss": 1.079,
+      "step": 5900
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 7.607287449392713e-06,
+      "loss": 1.0888,
+      "step": 5910
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 7.60323886639676e-06,
+      "loss": 1.0395,
+      "step": 5920
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 7.5991902834008105e-06,
+      "loss": 1.1178,
+      "step": 5930
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 1.0923843383789062,
+      "eval_runtime": 10.6309,
+      "eval_samples_per_second": 103.472,
+      "eval_steps_per_second": 12.981,
+      "step": 5930
+    }
+  ],
+  "max_steps": 22230,
+  "num_train_epochs": 15,
+  "total_flos": 6.084121621168128e+18,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b1cdc00ed543342a31cca3e2c908e7445062959a1a372014b76c542fbe5d37d
+size 5307

zero_to_fp32.py

@@ -0,0 +1,578 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)