Go4miii commited on
Commit
b68d805
1 Parent(s): 3fe5124

Upload 17 files

Browse files
all_results.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2.0,
3
+ "train_loss": 0.5708327819994277,
4
+ "train_runtime": 105327.89,
5
+ "train_samples_per_second": 4.797,
6
+ "train_steps_per_second": 0.019
7
+ }
config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "_name_or_path": "baichuan-inc/Baichuan-13B-Chat",
4
+ "architectures": [
5
+ "BaichuanForCausalLM"
6
+ ],
7
+ "auto_map": {
8
+ "AutoConfig": "configuration_baichuan.BaichuanConfig",
9
+ "AutoModel": "modeling_baichuan.BaichuanForCausalLM",
10
+ "AutoModelForCausalLM": "baichuan-inc/Baichuan-13B-Chat--modeling_baichuan.BaichuanForCausalLM"
11
+ },
12
+ "bos_token_id": 1,
13
+ "eos_token_id": 2,
14
+ "hidden_act": "silu",
15
+ "hidden_size": 5120,
16
+ "initializer_range": 0.02,
17
+ "intermediate_size": 13696,
18
+ "model_max_length": 4096,
19
+ "model_type": "baichuan",
20
+ "num_attention_heads": 40,
21
+ "num_hidden_layers": 40,
22
+ "pad_token_id": 0,
23
+ "rms_norm_eps": 1e-06,
24
+ "tie_word_embeddings": false,
25
+ "torch_dtype": "float16",
26
+ "transformers_version": "4.31.0",
27
+ "use_cache": false,
28
+ "vocab_size": 64000
29
+ }
configuration_baichuan.py ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
2
+
3
+ from transformers.configuration_utils import PretrainedConfig
4
+
5
+ class BaichuanConfig(PretrainedConfig):
6
+ model_type = "baichuan"
7
+ keys_to_ignore_at_inference = ["past_key_values"]
8
+
9
+ def __init__(
10
+ self,
11
+ vocab_size=64000,
12
+ hidden_size=5120,
13
+ intermediate_size=13696,
14
+ num_hidden_layers=40,
15
+ num_attention_heads=40,
16
+ hidden_act="silu",
17
+ model_max_length=4096,
18
+ initializer_range=0.02,
19
+ rms_norm_eps=1e-6,
20
+ use_cache=True,
21
+ pad_token_id=0,
22
+ bos_token_id=1,
23
+ eos_token_id=2,
24
+ tie_word_embeddings=False,
25
+ gradient_checkpointing=False,
26
+ **kwargs,
27
+ ):
28
+ self.vocab_size = vocab_size
29
+ self.model_max_length = model_max_length
30
+ self.hidden_size = hidden_size
31
+ self.intermediate_size = intermediate_size
32
+ self.num_hidden_layers = num_hidden_layers
33
+ self.num_attention_heads = num_attention_heads
34
+ self.hidden_act = hidden_act
35
+ self.initializer_range = initializer_range
36
+ self.rms_norm_eps = rms_norm_eps
37
+ self.use_cache = use_cache
38
+ self.gradient_checkpointing = gradient_checkpointing,
39
+ super().__init__(
40
+ pad_token_id=pad_token_id,
41
+ bos_token_id=bos_token_id,
42
+ eos_token_id=eos_token_id,
43
+ tie_word_embeddings=tie_word_embeddings,
44
+ **kwargs,
45
+ )
46
+
generation_config.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "assistant_token_id": 196,
3
+ "bos_token_id": 1,
4
+ "do_sample": true,
5
+ "eos_token_id": 2,
6
+ "max_new_tokens": 2048,
7
+ "pad_token_id": 0,
8
+ "repetition_penalty": 1.1,
9
+ "temperature": 0.3,
10
+ "top_k": 5,
11
+ "top_p": 0.85,
12
+ "transformers_version": "4.31.0",
13
+ "user_token_id": 195
14
+ }
generation_utils.py ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from typing import List
2
+ from queue import Queue
3
+
4
+ import torch
5
+
6
+
7
+ def build_chat_input(model, tokenizer, messages: List[dict], max_new_tokens: int=0):
8
+ def _parse_messages(messages, split_role="user"):
9
+ system, rounds = "", []
10
+ round = []
11
+ for i, message in enumerate(messages):
12
+ if message["role"] == "system":
13
+ assert i == 0
14
+ system = message["content"]
15
+ continue
16
+ if message["role"] == split_role and round:
17
+ rounds.append(round)
18
+ round = []
19
+ round.append(message)
20
+ if round:
21
+ rounds.append(round)
22
+ return system, rounds
23
+
24
+ max_new_tokens = max_new_tokens or model.generation_config.max_new_tokens
25
+ max_input_tokens = model.config.model_max_length - max_new_tokens
26
+ system, rounds = _parse_messages(messages, split_role="user")
27
+ system_tokens = tokenizer.encode(system)
28
+ max_history_tokens = max_input_tokens - len(system_tokens)
29
+
30
+ history_tokens = []
31
+ for round in rounds[::-1]:
32
+ round_tokens = []
33
+ for message in round:
34
+ if message["role"] == "user":
35
+ round_tokens.append(model.generation_config.user_token_id)
36
+ else:
37
+ round_tokens.append(model.generation_config.assistant_token_id)
38
+ round_tokens.extend(tokenizer.encode(message["content"]))
39
+ if len(history_tokens) == 0 or len(history_tokens) + len(round_tokens) <= max_history_tokens:
40
+ history_tokens = round_tokens + history_tokens # concat left
41
+ if len(history_tokens) < max_history_tokens:
42
+ continue
43
+ break
44
+
45
+ input_tokens = system_tokens + history_tokens
46
+ if messages[-1]["role"] != "assistant":
47
+ input_tokens.append(model.generation_config.assistant_token_id)
48
+ input_tokens = input_tokens[-max_input_tokens:] # truncate left
49
+ return torch.LongTensor([input_tokens]).to(model.device)
50
+
51
+
52
+ class TextIterStreamer:
53
+ def __init__(self, tokenizer, skip_prompt=False, skip_special_tokens=False):
54
+ self.tokenizer = tokenizer
55
+ self.skip_prompt = skip_prompt
56
+ self.skip_special_tokens = skip_special_tokens
57
+ self.tokens = []
58
+ self.text_queue = Queue()
59
+ self.next_tokens_are_prompt = True
60
+
61
+ def put(self, value):
62
+ if self.skip_prompt and self.next_tokens_are_prompt:
63
+ self.next_tokens_are_prompt = False
64
+ else:
65
+ if len(value.shape) > 1:
66
+ value = value[0]
67
+ self.tokens.extend(value.tolist())
68
+ self.text_queue.put(
69
+ self.tokenizer.decode(self.tokens, skip_special_tokens=self.skip_special_tokens))
70
+
71
+ def end(self):
72
+ self.text_queue.put(None)
73
+
74
+ def __iter__(self):
75
+ return self
76
+
77
+ def __next__(self):
78
+ value = self.text_queue.get()
79
+ if value is None:
80
+ raise StopIteration()
81
+ else:
82
+ return value
modeling_baichuan.py ADDED
@@ -0,0 +1,572 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
2
+
3
+ import math
4
+ from threading import Thread
5
+ from typing import List, Optional, Tuple, Union
6
+
7
+ import torch
8
+ import torch.utils.checkpoint
9
+ from torch.nn import CrossEntropyLoss
10
+ from transformers import PreTrainedModel
11
+ from transformers.activations import ACT2FN
12
+ from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
13
+ from transformers.utils import logging
14
+ from transformers.generation.utils import GenerationConfig
15
+
16
+ from .configuration_baichuan import BaichuanConfig
17
+ from .generation_utils import build_chat_input, TextIterStreamer
18
+
19
+ logger = logging.get_logger(__name__)
20
+
21
+
22
+ def _get_interleave(n):
23
+ def _get_interleave_power_of_2(n):
24
+ start = (2 ** (-2 ** -(math.log2(n) - 3)))
25
+ ratio = start
26
+ return [start * ratio ** i for i in range(n)]
27
+
28
+ if math.log2(n).is_integer():
29
+ return _get_interleave_power_of_2(n)
30
+ else:
31
+ closest_power_of_2 = 2 ** math.floor(math.log2(n))
32
+ return _get_interleave_power_of_2(closest_power_of_2) + \
33
+ _get_interleave(2 * closest_power_of_2)[0::2][:n - closest_power_of_2]
34
+
35
+ def _fill_with_neg_inf(t):
36
+ """FP16-compatible function that fills a tensor with -inf."""
37
+ return t.float().fill_(float("-inf")).type_as(t)
38
+
39
+ def _gen_alibi_mask(n_head, max_pos):
40
+ """used in inference only"""
41
+ slopes = torch.Tensor(_get_interleave(n_head))
42
+ alibi = slopes.unsqueeze(1).unsqueeze(1) * torch.arange(max_pos).unsqueeze(0).unsqueeze(0).expand(
43
+ n_head, -1, -1)
44
+ alibi = alibi.view(n_head, 1, max_pos)
45
+ alibi_mask = torch.triu(
46
+ _fill_with_neg_inf(torch.zeros([max_pos, max_pos])), 1
47
+ )
48
+ alibi_mask = alibi_mask.unsqueeze(0) + alibi
49
+ return alibi_mask
50
+
51
+ def _buffered_future_mask(tensor, maxpos, alibi, attn_heads):
52
+ """used in training only"""
53
+ dim = tensor.size(1)
54
+ _future_mask = torch.triu(
55
+ _fill_with_neg_inf(torch.zeros([maxpos, maxpos])), 1
56
+ )
57
+ _future_mask = _future_mask.unsqueeze(0) + alibi
58
+ _future_mask = _future_mask.to(tensor)
59
+ return _future_mask[:tensor.shape[0] * attn_heads, :maxpos, :maxpos]
60
+
61
+
62
+ class RMSNorm(torch.nn.Module):
63
+ def __init__(self, hidden_size, epsilon=1e-6):
64
+ super().__init__()
65
+ self.weight = torch.nn.Parameter(torch.empty(hidden_size))
66
+ self.epsilon = epsilon
67
+
68
+ def forward(self, hidden_states):
69
+ variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
70
+ hidden_states = hidden_states * torch.rsqrt(variance + self.epsilon)
71
+
72
+ # convert into half-precision
73
+ if self.weight.dtype in [torch.float16, torch.bfloat16]:
74
+ hidden_states = hidden_states.to(self.weight.dtype)
75
+
76
+ return self.weight * hidden_states
77
+
78
+
79
+ class MLP(torch.nn.Module):
80
+ def __init__(
81
+ self,
82
+ hidden_size: int,
83
+ intermediate_size: int,
84
+ hidden_act: str,
85
+ ):
86
+ super().__init__()
87
+ self.gate_proj = torch.nn.Linear(hidden_size, intermediate_size, bias=False)
88
+ self.down_proj = torch.nn.Linear(intermediate_size, hidden_size, bias=False)
89
+ self.up_proj = torch.nn.Linear(hidden_size, intermediate_size, bias=False)
90
+ self.act_fn = ACT2FN[hidden_act]
91
+
92
+ def forward(self, x):
93
+ return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
94
+
95
+
96
+ class BaichuanAttention(torch.nn.Module):
97
+ def __init__(self, config: BaichuanConfig):
98
+ super().__init__()
99
+ self.config = config
100
+ self.hidden_size = config.hidden_size
101
+ self.num_heads = config.num_attention_heads
102
+ self.head_dim = self.hidden_size // self.num_heads
103
+ self.max_position_embeddings = config.model_max_length
104
+
105
+ if (self.head_dim * self.num_heads) != self.hidden_size:
106
+ raise ValueError(
107
+ f"hidden_size {self.hidden_size} is not divisible by num_heads {self.num_heads}"
108
+ )
109
+ self.W_pack = torch.nn.Linear(self.hidden_size, 3 * self.hidden_size, bias=False)
110
+ self.o_proj = torch.nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
111
+
112
+ def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
113
+ return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
114
+
115
+ def forward(
116
+ self,
117
+ hidden_states: torch.Tensor,
118
+ attention_mask: Optional[torch.Tensor] = None,
119
+ past_key_value: Optional[Tuple[torch.Tensor]] = None,
120
+ output_attentions: bool = False,
121
+ use_cache: bool = False,
122
+ ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
123
+
124
+ bsz, q_len, _ = hidden_states.size()
125
+
126
+ proj = self.W_pack(hidden_states)
127
+ proj = proj.unflatten(-1, (3, self.hidden_size)).unsqueeze(0).transpose(0, -2).squeeze(-2)
128
+ query_states = proj[0].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
129
+ key_states = proj[1].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
130
+ value_states = proj[2].view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
131
+
132
+ kv_seq_len = key_states.shape[-2]
133
+ if past_key_value is not None:
134
+ kv_seq_len += past_key_value[0].shape[-2]
135
+
136
+ if past_key_value is not None:
137
+ # reuse k, v, self_attention
138
+ key_states = torch.cat([past_key_value[0], key_states], dim=2)
139
+ value_states = torch.cat([past_key_value[1], value_states], dim=2)
140
+
141
+ past_key_value = (key_states, value_states) if use_cache else None
142
+
143
+ attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
144
+
145
+ if attention_mask is not None:
146
+ if q_len == 1: # inference with cache
147
+ if len(attention_mask.size()) == 4:
148
+ attention_mask = attention_mask[:, :, -1:, :]
149
+ else:
150
+ attention_mask = attention_mask[:, -1:, :]
151
+ attn_weights = attn_weights + attention_mask
152
+ attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min))
153
+
154
+ attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1)
155
+
156
+ attn_output = torch.matmul(attn_weights, value_states)
157
+
158
+ attn_output = attn_output.transpose(1, 2)
159
+ attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
160
+ attn_output = self.o_proj(attn_output)
161
+
162
+ if not output_attentions:
163
+ attn_weights = None
164
+
165
+ return attn_output, attn_weights, past_key_value
166
+
167
+
168
+ class BaichuanLayer(torch.nn.Module):
169
+ def __init__(self, config: BaichuanConfig):
170
+ super().__init__()
171
+ self.hidden_size = config.hidden_size
172
+ self.self_attn = BaichuanAttention(config=config)
173
+ self.mlp = MLP(
174
+ hidden_size=self.hidden_size,
175
+ intermediate_size=config.intermediate_size,
176
+ hidden_act=config.hidden_act,
177
+ )
178
+ self.input_layernorm = RMSNorm(config.hidden_size, epsilon=config.rms_norm_eps)
179
+ self.post_attention_layernorm = RMSNorm(config.hidden_size, epsilon=config.rms_norm_eps)
180
+
181
+ def forward(
182
+ self,
183
+ hidden_states: torch.Tensor,
184
+ attention_mask: Optional[torch.Tensor] = None,
185
+ past_key_value: Optional[Tuple[torch.Tensor]] = None,
186
+ output_attentions: Optional[bool] = False,
187
+ use_cache: Optional[bool] = False,
188
+ ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
189
+
190
+ residual = hidden_states
191
+
192
+ hidden_states = self.input_layernorm(hidden_states)
193
+
194
+ # Self Attention
195
+ hidden_states, self_attn_weights, present_key_value = self.self_attn(
196
+ hidden_states=hidden_states,
197
+ attention_mask=attention_mask,
198
+ past_key_value=past_key_value,
199
+ output_attentions=output_attentions,
200
+ use_cache=use_cache,
201
+ )
202
+ hidden_states = residual + hidden_states
203
+
204
+ # Fully Connected
205
+ residual = hidden_states
206
+ hidden_states = self.post_attention_layernorm(hidden_states)
207
+ hidden_states = self.mlp(hidden_states)
208
+ hidden_states = residual + hidden_states
209
+
210
+ outputs = (hidden_states,)
211
+
212
+ if use_cache:
213
+ outputs += (present_key_value,)
214
+
215
+ return outputs
216
+
217
+
218
+ class BaichuanPreTrainedModel(PreTrainedModel):
219
+ config_class = BaichuanConfig
220
+ base_model_prefix = "model"
221
+ supports_gradient_checkpointing = True
222
+ _no_split_modules = ["BaichuanLayer"]
223
+ _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]
224
+
225
+ def _init_weights(self, module):
226
+ std = self.config.initializer_range
227
+ if isinstance(module, torch.nn.Linear):
228
+ module.weight.data.normal_(mean=0.0, std=std)
229
+ if module.bias is not None:
230
+ module.bias.data.zero_()
231
+ elif isinstance(module, torch.nn.Embedding):
232
+ module.weight.data.normal_(mean=0.0, std=std)
233
+ if module.padding_idx is not None:
234
+ module.weight.data[module.padding_idx].zero_()
235
+
236
+ def _set_gradient_checkpointing(self, module, value=False):
237
+ if isinstance(module, BaichuanModel):
238
+ module.gradient_checkpointing = value
239
+
240
+
241
+ class BaichuanModel(BaichuanPreTrainedModel):
242
+ def __init__(self, config: BaichuanConfig):
243
+ super().__init__(config)
244
+ self.padding_idx = config.pad_token_id
245
+ self.vocab_size = config.vocab_size
246
+ self.n_head = config.num_attention_heads
247
+ self.embed_tokens = torch.nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
248
+ self.layers = torch.nn.ModuleList([BaichuanLayer(config) for _ in range(config.num_hidden_layers)])
249
+ self.norm = RMSNorm(config.hidden_size, epsilon=config.rms_norm_eps)
250
+
251
+ self.gradient_checkpointing = config.gradient_checkpointing
252
+ self.post_init()
253
+ self.max_cache_pos = config.model_max_length
254
+ self.first_run = True
255
+ self.alibi_mask = None
256
+
257
+ def get_input_embeddings(self):
258
+ return self.embed_tokens
259
+
260
+ def set_input_embeddings(self, value):
261
+ self.embed_tokens = value
262
+
263
+ def get_alibi_mask(self, tensor, seq_length_with_past):
264
+ if self.training:
265
+ slopes = torch.Tensor(_get_interleave(self.n_head))
266
+ alibi = slopes.unsqueeze(1).unsqueeze(1) * torch.arange(seq_length_with_past).unsqueeze(0).unsqueeze(0).expand(
267
+ self.n_head,
268
+ -1, -1)
269
+ alibi = alibi.view(self.n_head, 1, seq_length_with_past)
270
+ mask = _buffered_future_mask(tensor, seq_length_with_past, alibi, self.n_head)
271
+ else:
272
+ if self.first_run:
273
+ self.first_run = False
274
+ self.register_buffer("future_mask", _gen_alibi_mask(self.n_head, self.max_cache_pos).to(tensor), persistent=False)
275
+ if seq_length_with_past > self.max_cache_pos:
276
+ self.max_cache_pos = seq_length_with_past
277
+ self.register_buffer("future_mask", _gen_alibi_mask(self.n_head, self.max_cache_pos).to(tensor), persistent=False)
278
+ mask = self.future_mask[:self.n_head, :seq_length_with_past, :seq_length_with_past]
279
+ return mask
280
+
281
+ def forward(
282
+ self,
283
+ input_ids: torch.LongTensor = None,
284
+ attention_mask: Optional[torch.Tensor] = None,
285
+ past_key_values: Optional[List[torch.FloatTensor]] = None,
286
+ inputs_embeds: Optional[torch.FloatTensor] = None,
287
+ use_cache: Optional[bool] = False,
288
+ output_attentions: Optional[bool] = False,
289
+ output_hidden_states: Optional[bool] = False,
290
+ return_dict: Optional[bool] = True,
291
+ ) -> Union[Tuple, BaseModelOutputWithPast]:
292
+
293
+ if input_ids is not None and inputs_embeds is not None:
294
+ raise ValueError("You cannot provide both input_ids and inputs_embeds simultaneously")
295
+ elif input_ids is not None:
296
+ batch_size, seq_length = input_ids.shape
297
+ elif inputs_embeds is not None:
298
+ batch_size, seq_length, _ = inputs_embeds.shape
299
+ else:
300
+ raise ValueError("You need to provide input_ids or inputs_embeds")
301
+
302
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
303
+
304
+ seq_length_with_past = seq_length
305
+
306
+ if past_key_values is not None:
307
+ past_key_values_length = past_key_values[0][0].shape[2]
308
+ seq_length_with_past = seq_length_with_past + past_key_values_length
309
+
310
+ if inputs_embeds is None:
311
+ inputs_embeds = self.embed_tokens(input_ids)
312
+
313
+ if self.training:
314
+ if self.alibi_mask is None or self.alibi_mask.shape[-1] != seq_length_with_past:
315
+ self.alibi_mask = self.get_alibi_mask(inputs_embeds, seq_length_with_past)
316
+ alibi_mask = self.alibi_mask
317
+ else:
318
+ alibi_mask = self.get_alibi_mask(inputs_embeds, seq_length_with_past)
319
+
320
+ if attention_mask is not None:
321
+ if len(attention_mask.shape) == 2:
322
+ expanded_mask = attention_mask.to(alibi_mask.dtype)
323
+ expanded_mask = torch.tril(torch.gt(expanded_mask[:, :, None] * expanded_mask[:, None, :], 0)
324
+ ) * torch.eq(expanded_mask[:, :, None] - expanded_mask[:, None, :], 0)
325
+ else:
326
+ expanded_mask = attention_mask
327
+ bsz = inputs_embeds.size(0)
328
+ src_len, tgt_len = alibi_mask.size()[-2:]
329
+ expanded_mask = expanded_mask.unsqueeze(1).expand(bsz, 1, src_len, tgt_len).to(alibi_mask.dtype)
330
+ inverted_mask = 1.0 - expanded_mask
331
+ inverted_mask = inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(alibi_mask.dtype).min)
332
+ attention_mask = inverted_mask + alibi_mask.unsqueeze(0)
333
+ else:
334
+ attention_mask = alibi_mask
335
+
336
+ hidden_states = inputs_embeds
337
+
338
+ if self.gradient_checkpointing and self.training:
339
+ if use_cache:
340
+ logger.warning_once(
341
+ "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
342
+ )
343
+ use_cache = False
344
+
345
+ # decoder layers
346
+ all_hidden_states = () if output_hidden_states else None
347
+ all_self_attns = () if output_attentions else None
348
+ next_decoder_cache = () if use_cache else None
349
+
350
+ for idx, decoder_layer in enumerate(self.layers):
351
+ if output_hidden_states:
352
+ all_hidden_states += (hidden_states,)
353
+
354
+ past_key_value = past_key_values[idx] if past_key_values is not None else None
355
+
356
+ if self.gradient_checkpointing and self.training:
357
+
358
+ def create_custom_forward(module):
359
+ def custom_forward(*inputs):
360
+ # None for past_key_value
361
+ return module(*inputs, output_attentions, None)
362
+
363
+ return custom_forward
364
+
365
+ layer_outputs = torch.utils.checkpoint.checkpoint(
366
+ create_custom_forward(decoder_layer),
367
+ hidden_states,
368
+ attention_mask,
369
+ None,
370
+ )
371
+ else:
372
+ layer_outputs = decoder_layer(
373
+ hidden_states,
374
+ attention_mask=attention_mask,
375
+ past_key_value=past_key_value,
376
+ output_attentions=output_attentions,
377
+ use_cache=use_cache,
378
+ )
379
+
380
+ hidden_states = layer_outputs[0]
381
+
382
+ if use_cache:
383
+ next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
384
+
385
+ if output_attentions:
386
+ all_self_attns += (layer_outputs[1],)
387
+
388
+ hidden_states = self.norm(hidden_states)
389
+
390
+ # add hidden states from the last decoder layer
391
+ if output_hidden_states:
392
+ all_hidden_states += (hidden_states,)
393
+
394
+ next_cache = next_decoder_cache if use_cache else None
395
+ if not return_dict:
396
+ return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
397
+ return BaseModelOutputWithPast(
398
+ last_hidden_state=hidden_states,
399
+ past_key_values=next_cache,
400
+ hidden_states=all_hidden_states,
401
+ attentions=all_self_attns,
402
+ )
403
+
404
+
405
+ class BaichuanForCausalLM(BaichuanPreTrainedModel):
406
+ def __init__(self, config):
407
+ super().__init__(config)
408
+ self.model = BaichuanModel(config)
409
+ self.lm_head = torch.nn.Linear(config.hidden_size, config.vocab_size, bias=False)
410
+
411
+ # Initialize weights and apply final processing
412
+ self.post_init()
413
+
414
+ def get_input_embeddings(self):
415
+ return self.model.embed_tokens
416
+
417
+ def set_input_embeddings(self, value):
418
+ self.model.embed_tokens = value
419
+
420
+ def get_output_embeddings(self):
421
+ return self.lm_head
422
+
423
+ def set_output_embeddings(self, new_embeddings):
424
+ self.lm_head = new_embeddings
425
+
426
+ def set_decoder(self, decoder):
427
+ self.model = decoder
428
+
429
+ def get_decoder(self):
430
+ return self.model
431
+
432
+ def forward(
433
+ self,
434
+ input_ids: torch.LongTensor = None,
435
+ attention_mask: Optional[torch.Tensor] = None,
436
+ past_key_values: Optional[List[torch.FloatTensor]] = None,
437
+ inputs_embeds: Optional[torch.FloatTensor] = None,
438
+ labels: Optional[torch.LongTensor] = None,
439
+ use_cache: Optional[bool] = None,
440
+ output_attentions: Optional[bool] = False,
441
+ output_hidden_states: Optional[bool] = False,
442
+ return_dict: Optional[bool] = True,
443
+ **kwargs
444
+ ) -> Union[Tuple, CausalLMOutputWithPast]:
445
+
446
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
447
+
448
+ # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
449
+ outputs = self.model(
450
+ input_ids=input_ids,
451
+ attention_mask=attention_mask,
452
+ past_key_values=past_key_values,
453
+ inputs_embeds=inputs_embeds,
454
+ use_cache=use_cache,
455
+ output_attentions=output_attentions,
456
+ output_hidden_states=output_hidden_states,
457
+ return_dict=return_dict,
458
+ )
459
+
460
+ hidden_states = outputs[0]
461
+ logits = self.lm_head(hidden_states)
462
+
463
+ loss = None
464
+ if labels is not None:
465
+ # Shift so that tokens < n predict n
466
+ shift_logits = logits[..., :-1, :].contiguous()
467
+ shift_labels = labels[..., 1:].contiguous()
468
+ # Flatten the tokens
469
+ loss_fct = CrossEntropyLoss()
470
+ shift_logits = shift_logits.view(-1, self.config.vocab_size)
471
+ shift_labels = shift_labels.view(-1)
472
+ # Enable model parallelism
473
+ shift_labels = shift_labels.to(shift_logits.device)
474
+ loss = loss_fct(shift_logits, shift_labels)
475
+
476
+ if not return_dict:
477
+ output = (logits,) + outputs[1:]
478
+ return (loss,) + output if loss is not None else output
479
+
480
+ return CausalLMOutputWithPast(
481
+ loss=loss,
482
+ logits=logits,
483
+ past_key_values=outputs.past_key_values,
484
+ hidden_states=outputs.hidden_states,
485
+ attentions=outputs.attentions,
486
+ )
487
+
488
+ def prepare_inputs_for_generation(
489
+ self,
490
+ input_ids: torch.LongTensor,
491
+ past_key_values: Optional[torch.Tensor] = None,
492
+ attention_mask: Optional[torch.Tensor] = None,
493
+ inputs_embeds: Optional[torch.Tensor] = None,
494
+ **kwargs
495
+ ):
496
+ if past_key_values:
497
+ input_ids = input_ids[:, -1:]
498
+
499
+ # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
500
+ if inputs_embeds is not None and past_key_values is None:
501
+ model_inputs = {"inputs_embeds": inputs_embeds}
502
+ else:
503
+ model_inputs = {"input_ids": input_ids}
504
+
505
+ model_inputs.update(
506
+ {
507
+ "past_key_values": past_key_values,
508
+ "use_cache": kwargs.get("use_cache"),
509
+ "attention_mask": attention_mask
510
+ }
511
+ )
512
+ return model_inputs
513
+
514
+ @staticmethod
515
+ def _reorder_cache(past_key_values, beam_idx):
516
+ return tuple(
517
+ tuple(past_state.index_select(0, beam_idx) for past_state in layer_past)
518
+ for layer_past in past_key_values
519
+ )
520
+
521
+ def quantize(self, bits: int):
522
+ try:
523
+ from .quantizer import QLinear
524
+ except ImportError:
525
+ raise ImportError(
526
+ f"Needs QLinear to run quantize."
527
+ )
528
+
529
+ for layer in self.model.layers:
530
+ layer.self_attn.W_pack = QLinear(
531
+ bits=bits,
532
+ weight=layer.self_attn.W_pack.weight,
533
+ bias = None,
534
+ )
535
+ layer.self_attn.o_proj = QLinear(
536
+ bits=bits,
537
+ weight=layer.self_attn.o_proj.weight,
538
+ bias = None,
539
+ )
540
+ layer.mlp.gate_proj = QLinear(
541
+ bits=bits,
542
+ weight=layer.mlp.gate_proj.weight,
543
+ bias = None,
544
+ )
545
+ layer.mlp.down_proj = QLinear(
546
+ bits=bits,
547
+ weight=layer.mlp.down_proj.weight,
548
+ bias = None,
549
+ )
550
+ layer.mlp.up_proj = QLinear(
551
+ bits=bits,
552
+ weight=layer.mlp.up_proj.weight,
553
+ bias = None,
554
+ )
555
+ return self
556
+
557
+ @torch.no_grad()
558
+ def chat(self, tokenizer, messages: List[dict], stream=False,
559
+ generation_config: Optional[GenerationConfig]=None):
560
+ generation_config = generation_config or self.generation_config
561
+ input_ids = build_chat_input(self, tokenizer, messages, generation_config.max_new_tokens)
562
+ if stream:
563
+ streamer = TextIterStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
564
+ Thread(target=self.generate, kwargs=dict(
565
+ inputs=input_ids, streamer=streamer,
566
+ generation_config=generation_config,
567
+ )).start()
568
+ return streamer
569
+ else:
570
+ outputs = self.generate(input_ids, generation_config=generation_config)
571
+ response = tokenizer.decode(outputs[0][len(input_ids[0]):], skip_special_tokens=True)
572
+ return response
pytorch_model.bin.index.json ADDED
@@ -0,0 +1,290 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 26529802240
4
+ },
5
+ "weight_map": {
6
+ "lm_head.weight": "pytorch_model-00003-of-00003.bin",
7
+ "model.embed_tokens.weight": "pytorch_model-00001-of-00003.bin",
8
+ "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
9
+ "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
10
+ "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
11
+ "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
12
+ "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
13
+ "model.layers.0.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
14
+ "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
15
+ "model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
16
+ "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
17
+ "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
18
+ "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
19
+ "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
20
+ "model.layers.1.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
21
+ "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
22
+ "model.layers.10.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
23
+ "model.layers.10.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
24
+ "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
25
+ "model.layers.10.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
26
+ "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
27
+ "model.layers.10.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
28
+ "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
29
+ "model.layers.11.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
30
+ "model.layers.11.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
31
+ "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
32
+ "model.layers.11.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
33
+ "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
34
+ "model.layers.11.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
35
+ "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
36
+ "model.layers.12.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
37
+ "model.layers.12.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
38
+ "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
39
+ "model.layers.12.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
40
+ "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
41
+ "model.layers.12.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
42
+ "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
43
+ "model.layers.13.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
44
+ "model.layers.13.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
45
+ "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
46
+ "model.layers.13.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
47
+ "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
48
+ "model.layers.13.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
49
+ "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
50
+ "model.layers.14.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
51
+ "model.layers.14.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
52
+ "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
53
+ "model.layers.14.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
54
+ "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
55
+ "model.layers.14.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
56
+ "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
57
+ "model.layers.15.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
58
+ "model.layers.15.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
59
+ "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
60
+ "model.layers.15.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
61
+ "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
62
+ "model.layers.15.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
63
+ "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
64
+ "model.layers.16.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
65
+ "model.layers.16.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
66
+ "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
67
+ "model.layers.16.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
68
+ "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
69
+ "model.layers.16.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
70
+ "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
71
+ "model.layers.17.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
72
+ "model.layers.17.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
73
+ "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
74
+ "model.layers.17.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
75
+ "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
76
+ "model.layers.17.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
77
+ "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
78
+ "model.layers.18.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
79
+ "model.layers.18.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
80
+ "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
81
+ "model.layers.18.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
82
+ "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
83
+ "model.layers.18.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
84
+ "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
85
+ "model.layers.19.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
86
+ "model.layers.19.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
87
+ "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
88
+ "model.layers.19.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
89
+ "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
90
+ "model.layers.19.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
91
+ "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
92
+ "model.layers.2.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
93
+ "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
94
+ "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
95
+ "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
96
+ "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
97
+ "model.layers.2.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
98
+ "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
99
+ "model.layers.20.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
100
+ "model.layers.20.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
101
+ "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
102
+ "model.layers.20.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
103
+ "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
104
+ "model.layers.20.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
105
+ "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
106
+ "model.layers.21.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
107
+ "model.layers.21.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
108
+ "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
109
+ "model.layers.21.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
110
+ "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
111
+ "model.layers.21.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
112
+ "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
113
+ "model.layers.22.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
114
+ "model.layers.22.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
115
+ "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
116
+ "model.layers.22.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
117
+ "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
118
+ "model.layers.22.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
119
+ "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
120
+ "model.layers.23.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
121
+ "model.layers.23.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
122
+ "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
123
+ "model.layers.23.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
124
+ "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
125
+ "model.layers.23.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
126
+ "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
127
+ "model.layers.24.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
128
+ "model.layers.24.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
129
+ "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
130
+ "model.layers.24.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
131
+ "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
132
+ "model.layers.24.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
133
+ "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
134
+ "model.layers.25.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
135
+ "model.layers.25.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
136
+ "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
137
+ "model.layers.25.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
138
+ "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
139
+ "model.layers.25.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
140
+ "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
141
+ "model.layers.26.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
142
+ "model.layers.26.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
143
+ "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
144
+ "model.layers.26.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
145
+ "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
146
+ "model.layers.26.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
147
+ "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
148
+ "model.layers.27.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
149
+ "model.layers.27.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
150
+ "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
151
+ "model.layers.27.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
152
+ "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
153
+ "model.layers.27.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
154
+ "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
155
+ "model.layers.28.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
156
+ "model.layers.28.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
157
+ "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
158
+ "model.layers.28.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
159
+ "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
160
+ "model.layers.28.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
161
+ "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
162
+ "model.layers.29.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
163
+ "model.layers.29.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
164
+ "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
165
+ "model.layers.29.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
166
+ "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
167
+ "model.layers.29.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
168
+ "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
169
+ "model.layers.3.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
170
+ "model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
171
+ "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
172
+ "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
173
+ "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
174
+ "model.layers.3.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
175
+ "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
176
+ "model.layers.30.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
177
+ "model.layers.30.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
178
+ "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
179
+ "model.layers.30.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
180
+ "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
181
+ "model.layers.30.self_attn.W_pack.weight": "pytorch_model-00002-of-00003.bin",
182
+ "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
183
+ "model.layers.31.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
184
+ "model.layers.31.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
185
+ "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
186
+ "model.layers.31.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
187
+ "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
188
+ "model.layers.31.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
189
+ "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
190
+ "model.layers.32.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
191
+ "model.layers.32.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
192
+ "model.layers.32.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
193
+ "model.layers.32.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
194
+ "model.layers.32.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
195
+ "model.layers.32.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
196
+ "model.layers.32.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
197
+ "model.layers.33.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
198
+ "model.layers.33.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
199
+ "model.layers.33.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
200
+ "model.layers.33.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
201
+ "model.layers.33.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
202
+ "model.layers.33.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
203
+ "model.layers.33.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
204
+ "model.layers.34.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
205
+ "model.layers.34.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
206
+ "model.layers.34.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
207
+ "model.layers.34.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
208
+ "model.layers.34.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
209
+ "model.layers.34.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
210
+ "model.layers.34.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
211
+ "model.layers.35.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
212
+ "model.layers.35.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
213
+ "model.layers.35.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
214
+ "model.layers.35.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
215
+ "model.layers.35.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
216
+ "model.layers.35.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
217
+ "model.layers.35.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
218
+ "model.layers.36.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
219
+ "model.layers.36.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
220
+ "model.layers.36.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
221
+ "model.layers.36.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
222
+ "model.layers.36.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
223
+ "model.layers.36.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
224
+ "model.layers.36.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
225
+ "model.layers.37.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
226
+ "model.layers.37.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
227
+ "model.layers.37.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
228
+ "model.layers.37.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
229
+ "model.layers.37.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
230
+ "model.layers.37.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
231
+ "model.layers.37.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
232
+ "model.layers.38.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
233
+ "model.layers.38.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
234
+ "model.layers.38.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
235
+ "model.layers.38.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
236
+ "model.layers.38.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
237
+ "model.layers.38.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
238
+ "model.layers.38.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
239
+ "model.layers.39.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
240
+ "model.layers.39.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
241
+ "model.layers.39.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
242
+ "model.layers.39.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
243
+ "model.layers.39.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
244
+ "model.layers.39.self_attn.W_pack.weight": "pytorch_model-00003-of-00003.bin",
245
+ "model.layers.39.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
246
+ "model.layers.4.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
247
+ "model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
248
+ "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
249
+ "model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
250
+ "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
251
+ "model.layers.4.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
252
+ "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
253
+ "model.layers.5.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
254
+ "model.layers.5.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
255
+ "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
256
+ "model.layers.5.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
257
+ "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
258
+ "model.layers.5.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
259
+ "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
260
+ "model.layers.6.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
261
+ "model.layers.6.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
262
+ "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
263
+ "model.layers.6.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
264
+ "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
265
+ "model.layers.6.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
266
+ "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
267
+ "model.layers.7.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
268
+ "model.layers.7.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
269
+ "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
270
+ "model.layers.7.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
271
+ "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
272
+ "model.layers.7.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
273
+ "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
274
+ "model.layers.8.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
275
+ "model.layers.8.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
276
+ "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
277
+ "model.layers.8.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
278
+ "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
279
+ "model.layers.8.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
280
+ "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
281
+ "model.layers.9.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
282
+ "model.layers.9.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
283
+ "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
284
+ "model.layers.9.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
285
+ "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
286
+ "model.layers.9.self_attn.W_pack.weight": "pytorch_model-00001-of-00003.bin",
287
+ "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
288
+ "model.norm.weight": "pytorch_model-00003-of-00003.bin"
289
+ }
290
+ }
quantizer.py ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
2
+
3
+ import torch
4
+ from typing import List
5
+ import bz2
6
+ import base64
7
+ import ctypes
8
+ from transformers.utils import logging
9
+ logger = logging.get_logger(__name__)
10
+
11
+ try:
12
+ from cpm_kernels.kernels.base import LazyKernelCModule, KernelFunction, round_up
13
+
14
+ class Kernel:
15
+ def __init__(self, code: bytes, function_names: List[str]):
16
+ self.code = code
17
+ self._function_names = function_names
18
+ self._cmodule = LazyKernelCModule(self.code)
19
+
20
+ for name in self._function_names:
21
+ setattr(self, name, KernelFunction(self._cmodule, name))
22
+ quantization_code = "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"
23
+ kernels = Kernel(
24
+ bz2.decompress(base64.b64decode(quantization_code)),
25
+ [
26
+ "int4_to_fp16",
27
+ "fp16_to_int4",
28
+ "int8_to_fp16",
29
+ "fp16_to_int8",
30
+ "int4_to_bf16",
31
+ "bf16_to_int4",
32
+ "int8_to_bf16",
33
+ "bf16_to_int8",
34
+ ],
35
+ )
36
+ except Exception as exception:
37
+ kernels = None
38
+ logger.warning("Failed to load kernels:" + str(exception))
39
+
40
+ def quant4(weight: torch.Tensor, scale: torch.Tensor):
41
+ stream = torch.cuda.current_stream()
42
+ num_row = weight.size(0)
43
+ num_chan_fp16 = weight.size(1)
44
+ # 4bit
45
+ num_chan_int = num_chan_fp16 // 8
46
+ qweight = torch.zeros((num_row, num_chan_int), dtype=torch.int32, device=weight.device)
47
+ intweight = torch.empty(num_row, num_chan_fp16, dtype = torch.int32)
48
+ intweight = torch.clip(torch.round(weight.to(scale.dtype) / scale[:, None]),-16, 15).to(dtype=torch.int32)
49
+
50
+ for j in range(num_chan_int):
51
+ qweight[:, j] = ((intweight[:, j*8+7] & 0x0f) << 28) \
52
+ | ((intweight[:, j*8+6] & 0x0f) << 24) \
53
+ | ((intweight[:, j*8+5] & 0x0f) << 20) \
54
+ | ((intweight[:, j*8+4] & 0x0f) << 16) \
55
+ | ((intweight[:, j*8+3] & 0x0f) << 12) \
56
+ | ((intweight[:, j*8+2] & 0x0f) << 8) \
57
+ | ((intweight[:, j*8+1] & 0x0f) << 4) \
58
+ | ((intweight[:, j*8] & 0x0f))
59
+ return qweight
60
+
61
+ def dequant4(qweight: torch.Tensor, scale: torch.Tensor, input: torch.Tensor):
62
+ stream = torch.cuda.current_stream()
63
+ num_row = qweight.size(0)
64
+ num_chan_int = qweight.size(1)
65
+ # 4bit
66
+ num_chan_fp16 = num_chan_int * 8
67
+
68
+ out = torch.empty((num_row, num_chan_fp16), dtype=input.dtype, device=qweight.device)
69
+
70
+ blockDim = (128, 1, 1)
71
+ gridDim = ((num_chan_int + blockDim[0] - 1) // blockDim[0], num_row, 1)
72
+ if input.dtype == torch.bfloat16:
73
+ kernels.int4_to_bf16(
74
+ gridDim,
75
+ blockDim,
76
+ 0,
77
+ stream,
78
+ [ctypes.c_void_p(out.data_ptr()), ctypes.c_void_p(qweight.data_ptr()),
79
+ ctypes.c_void_p(scale.data_ptr()), ctypes.c_int32(num_row), ctypes.c_int32(num_chan_int), ctypes.c_int32(num_chan_fp16)],
80
+ )
81
+ elif input.dtype == torch.float16:
82
+ kernels.int4_to_fp16(
83
+ gridDim,
84
+ blockDim,
85
+ 0,
86
+ stream,
87
+ [ctypes.c_void_p(out.data_ptr()), ctypes.c_void_p(qweight.data_ptr()),
88
+ ctypes.c_void_p(scale.data_ptr()), ctypes.c_int32(num_row), ctypes.c_int32(num_chan_int), ctypes.c_int32(num_chan_fp16)],
89
+ )
90
+ return out
91
+
92
+ class QLinear(torch.nn.Module):
93
+ def __init__(self, bits: int, weight: torch.Tensor, bias=None):
94
+ super().__init__()
95
+ self.quant_bits = bits
96
+ self.scale = weight.abs().max(dim=-1).values / ((2 ** (bits - 1)) - 1)
97
+ self.scale = self.scale.to(torch.float32)
98
+ if self.quant_bits == 4:
99
+ self.weight = quant4(weight, self.scale)
100
+ elif self.quant_bits == 8:
101
+ self.weight = torch.round(weight.to(self.scale.dtype) / self.scale[:, None]).to(torch.int8)
102
+ if self.quant_bits == 8:
103
+ self.weight = self.weight.T
104
+ self.bias = None
105
+
106
+ def forward(self, input):
107
+ if self.quant_bits == 4:
108
+ assert(input.dtype == torch.bfloat16 or input.dtype == torch.float16)
109
+
110
+ if self.weight.device != input.device:
111
+ self.weight = self.weight.to(input.device)
112
+ self.scale = self.scale.to(input.device)
113
+
114
+ if self.quant_bits == 4:
115
+ self.scale = self.scale.to(input.dtype)
116
+ rweight = dequant4(self.weight, self.scale, input).T
117
+ output = torch.matmul(input, rweight)
118
+ elif self.quant_bits == 8:
119
+ rweight = self.weight.to(input.dtype) * self.scale.to(input.dtype)
120
+ output = torch.matmul(input, rweight)
121
+ if self.bias is not None:
122
+ output = output + self.bias
123
+ return output
special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": true,
6
+ "rstrip": false,
7
+ "single_word": true
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": true,
13
+ "rstrip": false,
14
+ "single_word": true
15
+ },
16
+ "pad_token": {
17
+ "content": "<unk>",
18
+ "lstrip": false,
19
+ "normalized": true,
20
+ "rstrip": false,
21
+ "single_word": true
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": true,
27
+ "rstrip": false,
28
+ "single_word": true
29
+ }
30
+ }
tokenization_baichuan.py ADDED
@@ -0,0 +1,232 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright (c) 2023, Baichuan Intelligent Technology. All rights reserved.
2
+
3
+ import os
4
+ from shutil import copyfile
5
+ from typing import Any, Dict, List, Optional, Tuple
6
+
7
+ import sentencepiece as spm
8
+ from transformers.tokenization_utils import AddedToken, PreTrainedTokenizer
9
+ from transformers.utils import logging
10
+
11
+
12
+ logger = logging.get_logger(__name__)
13
+
14
+ VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.model"}
15
+
16
+ PRETRAINED_VOCAB_FILES_MAP = {
17
+ "vocab_file": {},
18
+ "tokenizer_file": {},
19
+ }
20
+ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {}
21
+
22
+
23
+ class BaichuanTokenizer(PreTrainedTokenizer):
24
+ """
25
+ Construct a Baichuan tokenizer. Based on byte-level Byte-Pair-Encoding.
26
+
27
+ Args:
28
+ vocab_file (`str`):
29
+ Path to the vocabulary file.
30
+ """
31
+
32
+ vocab_files_names = VOCAB_FILES_NAMES
33
+ pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
34
+ max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
35
+ model_input_names = ["input_ids", "attention_mask"]
36
+
37
+ def __init__(
38
+ self,
39
+ vocab_file,
40
+ unk_token="<unk>",
41
+ bos_token="<s>",
42
+ eos_token="</s>",
43
+ pad_token=None,
44
+ sp_model_kwargs: Optional[Dict[str, Any]] = None,
45
+ add_bos_token=True,
46
+ add_eos_token=False,
47
+ clean_up_tokenization_spaces=False,
48
+ **kwargs,
49
+ ):
50
+ self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
51
+ bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
52
+ eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
53
+ unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
54
+ pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token
55
+ super().__init__(
56
+ bos_token=bos_token,
57
+ eos_token=eos_token,
58
+ unk_token=unk_token,
59
+ pad_token=pad_token,
60
+ add_bos_token=add_bos_token,
61
+ add_eos_token=add_eos_token,
62
+ sp_model_kwargs=self.sp_model_kwargs,
63
+ clean_up_tokenization_spaces=clean_up_tokenization_spaces,
64
+ **kwargs,
65
+ )
66
+ self.vocab_file = vocab_file
67
+ self.add_bos_token = add_bos_token
68
+ self.add_eos_token = add_eos_token
69
+ self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
70
+ self.sp_model.Load(vocab_file)
71
+
72
+ def __getstate__(self):
73
+ state = self.__dict__.copy()
74
+ state["sp_model"] = None
75
+ return state
76
+
77
+ def __setstate__(self, d):
78
+ self.__dict__ = d
79
+ self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
80
+ self.sp_model.Load(self.vocab_file)
81
+
82
+ @property
83
+ def vocab_size(self):
84
+ """Returns vocab size"""
85
+ return self.sp_model.get_piece_size()
86
+
87
+ def get_vocab(self):
88
+ """Returns vocab as a dict"""
89
+ vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
90
+ vocab.update(self.added_tokens_encoder)
91
+ return vocab
92
+
93
+ def _tokenize(self, text):
94
+ """Returns a tokenized string."""
95
+ return self.sp_model.encode(text, out_type=str)
96
+
97
+ def _convert_token_to_id(self, token):
98
+ """Converts a token (str) in an id using the vocab."""
99
+ return self.sp_model.piece_to_id(token)
100
+
101
+ def _convert_id_to_token(self, index):
102
+ """Converts an index (integer) in a token (str) using the vocab."""
103
+ token = self.sp_model.IdToPiece(index)
104
+ return token
105
+
106
+ def convert_tokens_to_string(self, tokens):
107
+ """Converts a sequence of tokens (string) in a single string."""
108
+ current_sub_tokens = []
109
+ out_string = ""
110
+ prev_is_special = False
111
+ for i, token in enumerate(tokens):
112
+ # make sure that special tokens are not decoded using sentencepiece model
113
+ if token in self.all_special_tokens:
114
+ if not prev_is_special and i != 0:
115
+ out_string += " "
116
+ out_string += self.sp_model.decode(current_sub_tokens) + token
117
+ prev_is_special = True
118
+ current_sub_tokens = []
119
+ else:
120
+ current_sub_tokens.append(token)
121
+ prev_is_special = False
122
+ out_string += self.sp_model.decode(current_sub_tokens)
123
+ return out_string
124
+
125
+ def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None) -> Tuple[str]:
126
+ """
127
+ Save the vocabulary and special tokens file to a directory.
128
+
129
+ Args:
130
+ save_directory (`str`):
131
+ The directory in which to save the vocabulary.
132
+
133
+ Returns:
134
+ `Tuple(str)`: Paths to the files saved.
135
+ """
136
+ if not os.path.isdir(save_directory):
137
+ logger.error(f"Vocabulary path ({save_directory}) should be a directory")
138
+ return
139
+ out_vocab_file = os.path.join(
140
+ save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
141
+ )
142
+
143
+ if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
144
+ copyfile(self.vocab_file, out_vocab_file)
145
+ elif not os.path.isfile(self.vocab_file):
146
+ with open(out_vocab_file, "wb") as fi:
147
+ content_spiece_model = self.sp_model.serialized_model_proto()
148
+ fi.write(content_spiece_model)
149
+
150
+ return (out_vocab_file,)
151
+
152
+ def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
153
+ bos_token_id = [self.bos_token_id] if self.add_bos_token else []
154
+ eos_token_id = [self.eos_token_id] if self.add_eos_token else []
155
+
156
+ output = bos_token_id + token_ids_0 + eos_token_id
157
+
158
+ if token_ids_1 is not None:
159
+ output = output + bos_token_id + token_ids_1 + eos_token_id
160
+
161
+ return output
162
+
163
+ def get_special_tokens_mask(
164
+ self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
165
+ ) -> List[int]:
166
+ """
167
+ Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
168
+ special tokens using the tokenizer `prepare_for_model` method.
169
+
170
+ Args:
171
+ token_ids_0 (`List[int]`):
172
+ List of IDs.
173
+ token_ids_1 (`List[int]`, *optional*):
174
+ Optional second list of IDs for sequence pairs.
175
+ already_has_special_tokens (`bool`, *optional*, defaults to `False`):
176
+ Whether or not the token list is already formatted with special tokens for the model.
177
+
178
+ Returns:
179
+ `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
180
+ """
181
+ if already_has_special_tokens:
182
+ return super().get_special_tokens_mask(
183
+ token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
184
+ )
185
+
186
+ bos_token_id = [1] if self.add_bos_token else []
187
+ eos_token_id = [1] if self.add_eos_token else []
188
+
189
+ if token_ids_1 is None:
190
+ return bos_token_id + ([0] * len(token_ids_0)) + eos_token_id
191
+ return (
192
+ bos_token_id
193
+ + ([0] * len(token_ids_0))
194
+ + eos_token_id
195
+ + bos_token_id
196
+ + ([0] * len(token_ids_1))
197
+ + eos_token_id
198
+ )
199
+
200
+ def create_token_type_ids_from_sequences(
201
+ self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
202
+ ) -> List[int]:
203
+ """
204
+ Creates a mask from the two sequences passed to be used in a sequence-pair classification task. An ALBERT
205
+ sequence pair mask has the following format:
206
+
207
+ ```
208
+ 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
209
+ | first sequence | second sequence |
210
+ ```
211
+
212
+ if token_ids_1 is None, only returns the first portion of the mask (0s).
213
+
214
+ Args:
215
+ token_ids_0 (`List[int]`):
216
+ List of ids.
217
+ token_ids_1 (`List[int]`, *optional*):
218
+ Optional second list of IDs for sequence pairs.
219
+
220
+ Returns:
221
+ `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
222
+ """
223
+ bos_token_id = [self.bos_token_id] if self.add_bos_token else []
224
+ eos_token_id = [self.eos_token_id] if self.add_eos_token else []
225
+
226
+ output = [0] * len(bos_token_id + token_ids_0 + eos_token_id)
227
+
228
+ if token_ids_1 is not None:
229
+ output += [1] * len(bos_token_id + token_ids_1 + eos_token_id)
230
+
231
+ return output
232
+
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f7d1ab69d25c74644af5c5e4dcd1cc6e96d33783dbd257b6bdea55b643c72813
3
+ size 1136765
tokenizer_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_eos_token": false,
4
+ "auto_map": {
5
+ "AutoTokenizer": [
6
+ "tokenization_baichuan.BaichuanTokenizer",
7
+ null
8
+ ]
9
+ },
10
+ "bos_token": {
11
+ "__type": "AddedToken",
12
+ "content": "<s>",
13
+ "lstrip": false,
14
+ "normalized": true,
15
+ "rstrip": false,
16
+ "single_word": true
17
+ },
18
+ "clean_up_tokenization_spaces": false,
19
+ "eos_token": {
20
+ "__type": "AddedToken",
21
+ "content": "</s>",
22
+ "lstrip": false,
23
+ "normalized": true,
24
+ "rstrip": false,
25
+ "single_word": true
26
+ },
27
+ "model_max_length": 4096,
28
+ "pad_token": {
29
+ "__type": "AddedToken",
30
+ "content": "<unk>",
31
+ "lstrip": false,
32
+ "normalized": true,
33
+ "rstrip": false,
34
+ "single_word": true
35
+ },
36
+ "padding_side": "right",
37
+ "sp_model_kwargs": {},
38
+ "split_special_tokens": false,
39
+ "tokenizer_class": "BaichuanTokenizer",
40
+ "unk_token": {
41
+ "__type": "AddedToken",
42
+ "content": "<unk>",
43
+ "lstrip": false,
44
+ "normalized": true,
45
+ "rstrip": false,
46
+ "single_word": true
47
+ }
48
+ }
train_results.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 2.0,
3
+ "train_loss": 0.5708327819994277,
4
+ "train_runtime": 105327.89,
5
+ "train_samples_per_second": 4.797,
6
+ "train_steps_per_second": 0.019
7
+ }
trainer_log.jsonl ADDED
@@ -0,0 +1,198 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"current_steps": 10, "total_steps": 1974, "loss": 0.9175, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.9996834033646177e-05, "epoch": 0.01, "percentage": 0.51, "elapsed_time": "0:08:53", "remaining_time": "1 day, 5:06:00"}
2
+ {"current_steps": 20, "total_steps": 1974, "loss": 0.7595, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.998733693645213e-05, "epoch": 0.02, "percentage": 1.01, "elapsed_time": "0:17:16", "remaining_time": "1 day, 4:07:31"}
3
+ {"current_steps": 30, "total_steps": 1974, "loss": 0.7375, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.997151111381707e-05, "epoch": 0.03, "percentage": 1.52, "elapsed_time": "0:25:37", "remaining_time": "1 day, 3:40:03"}
4
+ {"current_steps": 40, "total_steps": 1974, "loss": 0.7227, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.9949360574062774e-05, "epoch": 0.04, "percentage": 2.03, "elapsed_time": "0:33:53", "remaining_time": "1 day, 3:18:52"}
5
+ {"current_steps": 50, "total_steps": 1974, "loss": 0.7147, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.9920890927418316e-05, "epoch": 0.05, "percentage": 2.53, "elapsed_time": "0:42:08", "remaining_time": "1 day, 3:01:22"}
6
+ {"current_steps": 60, "total_steps": 1974, "loss": 0.7102, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.988610938459917e-05, "epoch": 0.06, "percentage": 3.04, "elapsed_time": "0:50:24", "remaining_time": "1 day, 2:47:46"}
7
+ {"current_steps": 70, "total_steps": 1974, "loss": 0.7056, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.9845024754980876e-05, "epoch": 0.07, "percentage": 3.55, "elapsed_time": "0:58:37", "remaining_time": "1 day, 2:34:35"}
8
+ {"current_steps": 80, "total_steps": 1974, "loss": 0.7128, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.979764744436784e-05, "epoch": 0.08, "percentage": 4.05, "elapsed_time": "1:06:53", "remaining_time": "1 day, 2:23:46"}
9
+ {"current_steps": 90, "total_steps": 1974, "loss": 0.6982, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.9743989452357756e-05, "epoch": 0.09, "percentage": 4.56, "elapsed_time": "1:15:09", "remaining_time": "1 day, 2:13:11"}
10
+ {"current_steps": 100, "total_steps": 1974, "loss": 0.7258, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.968406436930243e-05, "epoch": 0.1, "percentage": 5.07, "elapsed_time": "1:23:27", "remaining_time": "1 day, 2:04:01"}
11
+ {"current_steps": 110, "total_steps": 1974, "loss": 0.7095, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.961788737286559e-05, "epoch": 0.11, "percentage": 5.57, "elapsed_time": "1:31:45", "remaining_time": "1 day, 1:54:50"}
12
+ {"current_steps": 120, "total_steps": 1974, "loss": 0.7048, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.954547522417877e-05, "epoch": 0.12, "percentage": 6.08, "elapsed_time": "1:40:03", "remaining_time": "1 day, 1:45:50"}
13
+ {"current_steps": 130, "total_steps": 1974, "loss": 0.6805, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.946684626359607e-05, "epoch": 0.13, "percentage": 6.59, "elapsed_time": "1:48:21", "remaining_time": "1 day, 1:36:59"}
14
+ {"current_steps": 140, "total_steps": 1974, "loss": 0.6798, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.938202040604898e-05, "epoch": 0.14, "percentage": 7.09, "elapsed_time": "1:56:36", "remaining_time": "1 day, 1:27:37"}
15
+ {"current_steps": 150, "total_steps": 1974, "loss": 0.7134, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.929101913600238e-05, "epoch": 0.15, "percentage": 7.6, "elapsed_time": "2:04:53", "remaining_time": "1 day, 1:18:36"}
16
+ {"current_steps": 160, "total_steps": 1974, "loss": 0.6895, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.919386550201299e-05, "epoch": 0.16, "percentage": 8.11, "elapsed_time": "2:13:08", "remaining_time": "1 day, 1:09:32"}
17
+ {"current_steps": 170, "total_steps": 1974, "loss": 0.705, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.909058411089174e-05, "epoch": 0.17, "percentage": 8.61, "elapsed_time": "2:21:25", "remaining_time": "1 day, 1:00:44"}
18
+ {"current_steps": 180, "total_steps": 1974, "loss": 0.6712, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.8981201121471356e-05, "epoch": 0.18, "percentage": 9.12, "elapsed_time": "2:29:41", "remaining_time": "1 day, 0:51:58"}
19
+ {"current_steps": 190, "total_steps": 1974, "loss": 0.6744, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.886574423798097e-05, "epoch": 0.19, "percentage": 9.63, "elapsed_time": "2:37:59", "remaining_time": "1 day, 0:43:26"}
20
+ {"current_steps": 200, "total_steps": 1974, "loss": 0.6675, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.874424270302927e-05, "epoch": 0.2, "percentage": 10.13, "elapsed_time": "2:46:13", "remaining_time": "1 day, 0:34:29"}
21
+ {"current_steps": 210, "total_steps": 1974, "loss": 0.6726, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.861672729019797e-05, "epoch": 0.21, "percentage": 10.64, "elapsed_time": "3:06:14", "remaining_time": "1 day, 2:04:28"}
22
+ {"current_steps": 220, "total_steps": 1974, "loss": 0.6821, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.848323029624761e-05, "epoch": 0.22, "percentage": 11.14, "elapsed_time": "3:14:32", "remaining_time": "1 day, 1:50:57"}
23
+ {"current_steps": 230, "total_steps": 1974, "loss": 0.7133, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.834378553293748e-05, "epoch": 0.23, "percentage": 11.65, "elapsed_time": "3:22:48", "remaining_time": "1 day, 1:37:50"}
24
+ {"current_steps": 240, "total_steps": 1974, "loss": 0.6681, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.81984283184619e-05, "epoch": 0.24, "percentage": 12.16, "elapsed_time": "3:31:01", "remaining_time": "1 day, 1:24:41"}
25
+ {"current_steps": 250, "total_steps": 1974, "loss": 0.682, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.804719546850487e-05, "epoch": 0.25, "percentage": 12.66, "elapsed_time": "3:39:18", "remaining_time": "1 day, 1:12:19"}
26
+ {"current_steps": 260, "total_steps": 1974, "loss": 0.6755, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.789012528691558e-05, "epoch": 0.26, "percentage": 13.17, "elapsed_time": "3:47:36", "remaining_time": "1 day, 1:00:28"}
27
+ {"current_steps": 270, "total_steps": 1974, "loss": 0.68, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.772725755600682e-05, "epoch": 0.27, "percentage": 13.68, "elapsed_time": "3:55:51", "remaining_time": "1 day, 0:48:32"}
28
+ {"current_steps": 280, "total_steps": 1974, "loss": 0.6663, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.755863352647909e-05, "epoch": 0.28, "percentage": 14.18, "elapsed_time": "4:04:07", "remaining_time": "1 day, 0:36:58"}
29
+ {"current_steps": 290, "total_steps": 1974, "loss": 0.6672, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.738429590697271e-05, "epoch": 0.29, "percentage": 14.69, "elapsed_time": "4:12:21", "remaining_time": "1 day, 0:25:24"}
30
+ {"current_steps": 300, "total_steps": 1974, "loss": 0.6466, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.720428885325069e-05, "epoch": 0.3, "percentage": 15.2, "elapsed_time": "4:20:38", "remaining_time": "1 day, 0:14:20"}
31
+ {"current_steps": 310, "total_steps": 1974, "loss": 0.6668, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.701865795701505e-05, "epoch": 0.31, "percentage": 15.7, "elapsed_time": "4:28:57", "remaining_time": "1 day, 0:03:40"}
32
+ {"current_steps": 320, "total_steps": 1974, "loss": 0.6591, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.684682059461469e-05, "epoch": 0.32, "percentage": 16.21, "elapsed_time": "4:37:10", "remaining_time": "23:52:39"}
33
+ {"current_steps": 330, "total_steps": 1974, "loss": 0.661, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.665063509461097e-05, "epoch": 0.33, "percentage": 16.72, "elapsed_time": "4:45:27", "remaining_time": "23:42:06"}
34
+ {"current_steps": 340, "total_steps": 1974, "loss": 0.6749, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.644896598002736e-05, "epoch": 0.34, "percentage": 17.22, "elapsed_time": "4:53:43", "remaining_time": "23:31:35"}
35
+ {"current_steps": 350, "total_steps": 1974, "loss": 0.6654, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.624186432907437e-05, "epoch": 0.35, "percentage": 17.73, "elapsed_time": "5:01:59", "remaining_time": "23:21:16"}
36
+ {"current_steps": 360, "total_steps": 1974, "loss": 0.6716, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.602938259590072e-05, "epoch": 0.36, "percentage": 18.24, "elapsed_time": "5:10:16", "remaining_time": "23:11:05"}
37
+ {"current_steps": 370, "total_steps": 1974, "loss": 0.6796, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.581157459730783e-05, "epoch": 0.37, "percentage": 18.74, "elapsed_time": "5:18:35", "remaining_time": "23:01:09"}
38
+ {"current_steps": 380, "total_steps": 1974, "loss": 0.6794, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.558849549911931e-05, "epoch": 0.39, "percentage": 19.25, "elapsed_time": "5:26:55", "remaining_time": "22:51:20"}
39
+ {"current_steps": 390, "total_steps": 1974, "loss": 0.6651, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.536020180220871e-05, "epoch": 0.4, "percentage": 19.76, "elapsed_time": "5:35:12", "remaining_time": "22:41:26"}
40
+ {"current_steps": 400, "total_steps": 1974, "loss": 0.6823, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.515032675559024e-05, "epoch": 0.41, "percentage": 20.26, "elapsed_time": "5:43:27", "remaining_time": "22:31:30"}
41
+ {"current_steps": 410, "total_steps": 1974, "loss": 0.6677, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.4912285699446786e-05, "epoch": 0.42, "percentage": 20.77, "elapsed_time": "6:04:58", "remaining_time": "23:12:15"}
42
+ {"current_steps": 420, "total_steps": 1974, "loss": 0.6704, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.4669201313179155e-05, "epoch": 0.43, "percentage": 21.28, "elapsed_time": "6:13:18", "remaining_time": "23:01:12"}
43
+ {"current_steps": 430, "total_steps": 1974, "loss": 0.6481, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.442113516454638e-05, "epoch": 0.44, "percentage": 21.78, "elapsed_time": "6:21:37", "remaining_time": "22:50:18"}
44
+ {"current_steps": 440, "total_steps": 1974, "loss": 0.665, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.416815008307488e-05, "epoch": 0.45, "percentage": 22.29, "elapsed_time": "6:29:58", "remaining_time": "22:39:35"}
45
+ {"current_steps": 450, "total_steps": 1974, "loss": 0.6658, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.391031014414514e-05, "epoch": 0.46, "percentage": 22.8, "elapsed_time": "6:38:16", "remaining_time": "22:28:49"}
46
+ {"current_steps": 460, "total_steps": 1974, "loss": 0.6699, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.364768065276284e-05, "epoch": 0.47, "percentage": 23.3, "elapsed_time": "6:46:39", "remaining_time": "22:18:26"}
47
+ {"current_steps": 470, "total_steps": 1974, "loss": 0.6664, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.338032812701867e-05, "epoch": 0.48, "percentage": 23.81, "elapsed_time": "6:55:02", "remaining_time": "22:08:08"}
48
+ {"current_steps": 480, "total_steps": 1974, "loss": 0.6817, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.310832028124069e-05, "epoch": 0.49, "percentage": 24.32, "elapsed_time": "7:03:24", "remaining_time": "21:57:52"}
49
+ {"current_steps": 490, "total_steps": 1974, "loss": 0.6791, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.283172600884393e-05, "epoch": 0.5, "percentage": 24.82, "elapsed_time": "7:11:47", "remaining_time": "21:47:43"}
50
+ {"current_steps": 500, "total_steps": 1974, "loss": 0.6679, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.2550615364881194e-05, "epoch": 0.51, "percentage": 25.33, "elapsed_time": "7:20:10", "remaining_time": "21:37:38"}
51
+ {"current_steps": 510, "total_steps": 1974, "loss": 0.6476, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.226505954829973e-05, "epoch": 0.52, "percentage": 25.84, "elapsed_time": "7:28:32", "remaining_time": "21:27:33"}
52
+ {"current_steps": 520, "total_steps": 1974, "loss": 0.6721, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.197513088390813e-05, "epoch": 0.53, "percentage": 26.34, "elapsed_time": "7:36:52", "remaining_time": "21:17:30"}
53
+ {"current_steps": 530, "total_steps": 1974, "loss": 0.6563, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.1680902804058095e-05, "epoch": 0.54, "percentage": 26.85, "elapsed_time": "7:45:10", "remaining_time": "21:07:22"}
54
+ {"current_steps": 540, "total_steps": 1974, "loss": 0.6391, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.138244983004574e-05, "epoch": 0.55, "percentage": 27.36, "elapsed_time": "7:53:26", "remaining_time": "20:57:14"}
55
+ {"current_steps": 550, "total_steps": 1974, "loss": 0.6672, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.107984755323697e-05, "epoch": 0.56, "percentage": 27.86, "elapsed_time": "8:01:42", "remaining_time": "20:47:11"}
56
+ {"current_steps": 560, "total_steps": 1974, "loss": 0.6497, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.077317261592194e-05, "epoch": 0.57, "percentage": 28.37, "elapsed_time": "8:09:57", "remaining_time": "20:37:08"}
57
+ {"current_steps": 570, "total_steps": 1974, "loss": 0.668, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.04625026919033e-05, "epoch": 0.58, "percentage": 28.88, "elapsed_time": "8:18:10", "remaining_time": "20:27:04"}
58
+ {"current_steps": 580, "total_steps": 1974, "loss": 0.6682, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.0147916466823174e-05, "epoch": 0.59, "percentage": 29.38, "elapsed_time": "8:26:24", "remaining_time": "20:17:07"}
59
+ {"current_steps": 590, "total_steps": 1974, "loss": 0.6352, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.982949361823388e-05, "epoch": 0.6, "percentage": 29.89, "elapsed_time": "8:34:38", "remaining_time": "20:07:14"}
60
+ {"current_steps": 600, "total_steps": 1974, "loss": 0.6698, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.950731479541743e-05, "epoch": 0.61, "percentage": 30.4, "elapsed_time": "8:42:55", "remaining_time": "19:57:30"}
61
+ {"current_steps": 610, "total_steps": 1974, "loss": 0.6549, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.918146159895882e-05, "epoch": 0.62, "percentage": 30.9, "elapsed_time": "9:01:46", "remaining_time": "20:11:25"}
62
+ {"current_steps": 620, "total_steps": 1974, "loss": 0.6516, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.8852016560078605e-05, "epoch": 0.63, "percentage": 31.41, "elapsed_time": "9:10:12", "remaining_time": "20:01:35"}
63
+ {"current_steps": 630, "total_steps": 1974, "loss": 0.6629, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.851906311972943e-05, "epoch": 0.64, "percentage": 31.91, "elapsed_time": "9:18:37", "remaining_time": "19:51:44"}
64
+ {"current_steps": 640, "total_steps": 1974, "loss": 0.6764, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.821647502051616e-05, "epoch": 0.65, "percentage": 32.42, "elapsed_time": "9:26:58", "remaining_time": "19:41:46"}
65
+ {"current_steps": 650, "total_steps": 1974, "loss": 0.6415, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.787708866250794e-05, "epoch": 0.66, "percentage": 32.93, "elapsed_time": "9:35:20", "remaining_time": "19:31:54"}
66
+ {"current_steps": 660, "total_steps": 1974, "loss": 0.6463, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.7534440830144466e-05, "epoch": 0.67, "percentage": 33.43, "elapsed_time": "9:43:42", "remaining_time": "19:22:06"}
67
+ {"current_steps": 670, "total_steps": 1974, "loss": 0.6508, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.71886183083464e-05, "epoch": 0.68, "percentage": 33.94, "elapsed_time": "9:52:04", "remaining_time": "19:12:19"}
68
+ {"current_steps": 680, "total_steps": 1974, "loss": 0.6411, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.683970868611123e-05, "epoch": 0.69, "percentage": 34.45, "elapsed_time": "10:00:24", "remaining_time": "19:02:31"}
69
+ {"current_steps": 690, "total_steps": 1974, "loss": 0.6266, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.648780033432891e-05, "epoch": 0.7, "percentage": 34.95, "elapsed_time": "10:08:43", "remaining_time": "18:52:44"}
70
+ {"current_steps": 700, "total_steps": 1974, "loss": 0.6409, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.613298238339955e-05, "epoch": 0.71, "percentage": 35.46, "elapsed_time": "10:17:01", "remaining_time": "18:42:58"}
71
+ {"current_steps": 710, "total_steps": 1974, "loss": 0.6594, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.5775344700658705e-05, "epoch": 0.72, "percentage": 35.97, "elapsed_time": "10:25:18", "remaining_time": "18:33:13"}
72
+ {"current_steps": 720, "total_steps": 1974, "loss": 0.6427, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.5414977867616006e-05, "epoch": 0.73, "percentage": 36.47, "elapsed_time": "10:33:40", "remaining_time": "18:23:38"}
73
+ {"current_steps": 730, "total_steps": 1974, "loss": 0.6462, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.505197315701292e-05, "epoch": 0.74, "percentage": 36.98, "elapsed_time": "10:41:59", "remaining_time": "18:14:01"}
74
+ {"current_steps": 740, "total_steps": 1974, "loss": 0.6277, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.468642250970547e-05, "epoch": 0.75, "percentage": 37.49, "elapsed_time": "10:50:18", "remaining_time": "18:04:25"}
75
+ {"current_steps": 750, "total_steps": 1974, "loss": 0.6551, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.431841851137764e-05, "epoch": 0.76, "percentage": 37.99, "elapsed_time": "10:58:35", "remaining_time": "17:54:48"}
76
+ {"current_steps": 760, "total_steps": 1974, "loss": 0.6402, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.394805436909157e-05, "epoch": 0.77, "percentage": 38.5, "elapsed_time": "11:06:52", "remaining_time": "17:45:15"}
77
+ {"current_steps": 770, "total_steps": 1974, "loss": 0.6515, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.357542388768033e-05, "epoch": 0.78, "percentage": 39.01, "elapsed_time": "11:15:12", "remaining_time": "17:35:47"}
78
+ {"current_steps": 780, "total_steps": 1974, "loss": 0.6489, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.3200621445989226e-05, "epoch": 0.79, "percentage": 39.51, "elapsed_time": "11:23:30", "remaining_time": "17:26:18"}
79
+ {"current_steps": 790, "total_steps": 1974, "loss": 0.6568, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.282374197297185e-05, "epoch": 0.8, "percentage": 40.02, "elapsed_time": "11:31:50", "remaining_time": "17:16:53"}
80
+ {"current_steps": 800, "total_steps": 1974, "loss": 0.615, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.2444880923646674e-05, "epoch": 0.81, "percentage": 40.53, "elapsed_time": "11:40:07", "remaining_time": "17:07:26"}
81
+ {"current_steps": 810, "total_steps": 1974, "loss": 0.6447, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.20641342549205e-05, "epoch": 0.82, "percentage": 41.03, "elapsed_time": "12:02:31", "remaining_time": "17:18:17"}
82
+ {"current_steps": 820, "total_steps": 1974, "loss": 0.6159, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.168159840128472e-05, "epoch": 0.83, "percentage": 41.54, "elapsed_time": "12:10:54", "remaining_time": "17:08:37"}
83
+ {"current_steps": 830, "total_steps": 1974, "loss": 0.6347, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.129737025039068e-05, "epoch": 0.84, "percentage": 42.05, "elapsed_time": "12:19:18", "remaining_time": "16:58:59"}
84
+ {"current_steps": 840, "total_steps": 1974, "loss": 0.6361, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.091154711851022e-05, "epoch": 0.85, "percentage": 42.55, "elapsed_time": "12:27:42", "remaining_time": "16:49:24"}
85
+ {"current_steps": 850, "total_steps": 1974, "loss": 0.6504, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.052422672588765e-05, "epoch": 0.86, "percentage": 43.06, "elapsed_time": "12:36:05", "remaining_time": "16:39:48"}
86
+ {"current_steps": 860, "total_steps": 1974, "loss": 0.6467, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.013550717198948e-05, "epoch": 0.87, "percentage": 43.57, "elapsed_time": "12:44:27", "remaining_time": "16:30:14"}
87
+ {"current_steps": 870, "total_steps": 1974, "loss": 0.6364, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.9745486910657993e-05, "epoch": 0.88, "percentage": 44.07, "elapsed_time": "12:52:51", "remaining_time": "16:20:44"}
88
+ {"current_steps": 880, "total_steps": 1974, "loss": 0.6361, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.9354264725175185e-05, "epoch": 0.89, "percentage": 44.58, "elapsed_time": "13:01:16", "remaining_time": "16:11:15"}
89
+ {"current_steps": 890, "total_steps": 1974, "loss": 0.6441, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.8961939703243122e-05, "epoch": 0.9, "percentage": 45.09, "elapsed_time": "13:09:39", "remaining_time": "16:01:47"}
90
+ {"current_steps": 900, "total_steps": 1974, "loss": 0.6404, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.856861121188735e-05, "epoch": 0.91, "percentage": 45.59, "elapsed_time": "13:18:01", "remaining_time": "15:52:19"}
91
+ {"current_steps": 910, "total_steps": 1974, "loss": 0.6307, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.8174378872289446e-05, "epoch": 0.92, "percentage": 46.1, "elapsed_time": "13:26:21", "remaining_time": "15:42:49"}
92
+ {"current_steps": 920, "total_steps": 1974, "loss": 0.6484, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.777934253455522e-05, "epoch": 0.93, "percentage": 46.61, "elapsed_time": "13:34:42", "remaining_time": "15:33:21"}
93
+ {"current_steps": 930, "total_steps": 1974, "loss": 0.6237, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.7383602252424985e-05, "epoch": 0.94, "percentage": 47.11, "elapsed_time": "13:43:00", "remaining_time": "15:23:53"}
94
+ {"current_steps": 940, "total_steps": 1974, "loss": 0.6161, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.6987258257932175e-05, "epoch": 0.95, "percentage": 47.62, "elapsed_time": "13:51:20", "remaining_time": "15:14:29"}
95
+ {"current_steps": 950, "total_steps": 1974, "loss": 0.6381, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.6590410936016895e-05, "epoch": 0.96, "percentage": 48.13, "elapsed_time": "13:59:40", "remaining_time": "15:05:05"}
96
+ {"current_steps": 960, "total_steps": 1974, "loss": 0.6366, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.619316079910063e-05, "epoch": 0.97, "percentage": 48.63, "elapsed_time": "14:07:59", "remaining_time": "14:55:41"}
97
+ {"current_steps": 970, "total_steps": 1974, "loss": 0.6202, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.5795608461628802e-05, "epoch": 0.98, "percentage": 49.14, "elapsed_time": "14:16:20", "remaining_time": "14:46:21"}
98
+ {"current_steps": 980, "total_steps": 1974, "loss": 0.6334, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.5397854614587334e-05, "epoch": 0.99, "percentage": 49.65, "elapsed_time": "14:24:38", "remaining_time": "14:36:59"}
99
+ {"current_steps": 990, "total_steps": 1974, "loss": 0.5954, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.5e-05, "epoch": 1.0, "percentage": 50.15, "elapsed_time": "14:32:58", "remaining_time": "14:27:41"}
100
+ {"current_steps": 1000, "total_steps": 1974, "loss": 0.4963, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.460214538541267e-05, "epoch": 1.01, "percentage": 50.66, "elapsed_time": "14:41:17", "remaining_time": "14:18:23"}
101
+ {"current_steps": 1010, "total_steps": 1974, "loss": 0.487, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.4204391538371207e-05, "epoch": 1.02, "percentage": 51.17, "elapsed_time": "15:03:28", "remaining_time": "14:22:20"}
102
+ {"current_steps": 1020, "total_steps": 1974, "loss": 0.489, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.3806839200899377e-05, "epoch": 1.03, "percentage": 51.67, "elapsed_time": "15:11:48", "remaining_time": "14:12:48"}
103
+ {"current_steps": 1030, "total_steps": 1974, "loss": 0.4805, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.3409589063983117e-05, "epoch": 1.04, "percentage": 52.18, "elapsed_time": "15:20:06", "remaining_time": "14:03:17"}
104
+ {"current_steps": 1040, "total_steps": 1974, "loss": 0.4907, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.3012741742067838e-05, "epoch": 1.05, "percentage": 52.68, "elapsed_time": "15:28:27", "remaining_time": "13:53:49"}
105
+ {"current_steps": 1050, "total_steps": 1974, "loss": 0.4719, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.261639774757503e-05, "epoch": 1.06, "percentage": 53.19, "elapsed_time": "15:36:47", "remaining_time": "13:44:22"}
106
+ {"current_steps": 1060, "total_steps": 1974, "loss": 0.4914, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.2220657465444782e-05, "epoch": 1.07, "percentage": 53.7, "elapsed_time": "15:45:07", "remaining_time": "13:34:56"}
107
+ {"current_steps": 1070, "total_steps": 1974, "loss": 0.4775, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.182562112771056e-05, "epoch": 1.08, "percentage": 54.2, "elapsed_time": "15:53:27", "remaining_time": "13:25:32"}
108
+ {"current_steps": 1080, "total_steps": 1974, "loss": 0.4935, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.143138878811265e-05, "epoch": 1.09, "percentage": 54.71, "elapsed_time": "16:01:47", "remaining_time": "13:16:09"}
109
+ {"current_steps": 1090, "total_steps": 1974, "loss": 0.5082, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.1038060296756883e-05, "epoch": 1.1, "percentage": 55.22, "elapsed_time": "16:10:09", "remaining_time": "13:06:48"}
110
+ {"current_steps": 1100, "total_steps": 1974, "loss": 0.4868, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.064573527482482e-05, "epoch": 1.11, "percentage": 55.72, "elapsed_time": "16:18:28", "remaining_time": "12:57:26"}
111
+ {"current_steps": 1110, "total_steps": 1974, "loss": 0.4988, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.025451308934201e-05, "epoch": 1.12, "percentage": 56.23, "elapsed_time": "16:26:44", "remaining_time": "12:48:03"}
112
+ {"current_steps": 1120, "total_steps": 1974, "loss": 0.4653, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.9864492828010526e-05, "epoch": 1.13, "percentage": 56.74, "elapsed_time": "16:35:02", "remaining_time": "12:38:43"}
113
+ {"current_steps": 1130, "total_steps": 1974, "loss": 0.4915, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.9475773274112354e-05, "epoch": 1.14, "percentage": 57.24, "elapsed_time": "16:43:23", "remaining_time": "12:29:25"}
114
+ {"current_steps": 1140, "total_steps": 1974, "loss": 0.4763, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.9088452881489787e-05, "epoch": 1.16, "percentage": 57.75, "elapsed_time": "16:51:43", "remaining_time": "12:20:09"}
115
+ {"current_steps": 1150, "total_steps": 1974, "loss": 0.4807, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.8702629749609324e-05, "epoch": 1.17, "percentage": 58.26, "elapsed_time": "17:00:05", "remaining_time": "12:10:54"}
116
+ {"current_steps": 1160, "total_steps": 1974, "loss": 0.4653, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.8318401598715284e-05, "epoch": 1.18, "percentage": 58.76, "elapsed_time": "17:08:24", "remaining_time": "12:01:39"}
117
+ {"current_steps": 1170, "total_steps": 1974, "loss": 0.4778, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.793586574507951e-05, "epoch": 1.19, "percentage": 59.27, "elapsed_time": "17:16:43", "remaining_time": "11:52:24"}
118
+ {"current_steps": 1180, "total_steps": 1974, "loss": 0.4839, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.7555119076353338e-05, "epoch": 1.2, "percentage": 59.78, "elapsed_time": "17:25:04", "remaining_time": "11:43:12"}
119
+ {"current_steps": 1190, "total_steps": 1974, "loss": 0.4718, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.7176258027028152e-05, "epoch": 1.21, "percentage": 60.28, "elapsed_time": "17:33:23", "remaining_time": "11:33:59"}
120
+ {"current_steps": 1200, "total_steps": 1974, "loss": 0.4793, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.6799378554010773e-05, "epoch": 1.22, "percentage": 60.79, "elapsed_time": "17:41:44", "remaining_time": "11:24:49"}
121
+ {"current_steps": 1210, "total_steps": 1974, "loss": 0.4825, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.6424576112319672e-05, "epoch": 1.23, "percentage": 61.3, "elapsed_time": "18:04:33", "remaining_time": "11:24:47"}
122
+ {"current_steps": 1220, "total_steps": 1974, "loss": 0.4857, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.6051945630908426e-05, "epoch": 1.24, "percentage": 61.8, "elapsed_time": "18:12:53", "remaining_time": "11:15:26"}
123
+ {"current_steps": 1230, "total_steps": 1974, "loss": 0.4802, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.5681581488622367e-05, "epoch": 1.25, "percentage": 62.31, "elapsed_time": "18:21:13", "remaining_time": "11:06:06"}
124
+ {"current_steps": 1240, "total_steps": 1974, "loss": 0.4812, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.5313577490294538e-05, "epoch": 1.26, "percentage": 62.82, "elapsed_time": "18:29:36", "remaining_time": "10:56:49"}
125
+ {"current_steps": 1250, "total_steps": 1974, "loss": 0.4682, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.4948026842987084e-05, "epoch": 1.27, "percentage": 63.32, "elapsed_time": "18:38:01", "remaining_time": "10:47:33"}
126
+ {"current_steps": 1260, "total_steps": 1974, "loss": 0.4974, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.4585022132384008e-05, "epoch": 1.28, "percentage": 63.83, "elapsed_time": "18:46:25", "remaining_time": "10:38:18"}
127
+ {"current_steps": 1270, "total_steps": 1974, "loss": 0.4737, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.4224655299341304e-05, "epoch": 1.29, "percentage": 64.34, "elapsed_time": "18:54:44", "remaining_time": "10:29:01"}
128
+ {"current_steps": 1280, "total_steps": 1974, "loss": 0.4877, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.3867017616600456e-05, "epoch": 1.3, "percentage": 64.84, "elapsed_time": "19:03:02", "remaining_time": "10:19:44"}
129
+ {"current_steps": 1290, "total_steps": 1974, "loss": 0.4753, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.3512199665671094e-05, "epoch": 1.31, "percentage": 65.35, "elapsed_time": "19:11:23", "remaining_time": "10:10:30"}
130
+ {"current_steps": 1300, "total_steps": 1974, "loss": 0.4638, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.316029131388878e-05, "epoch": 1.32, "percentage": 65.86, "elapsed_time": "19:19:45", "remaining_time": "10:01:17"}
131
+ {"current_steps": 1310, "total_steps": 1974, "loss": 0.4626, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.2811381691653607e-05, "epoch": 1.33, "percentage": 66.36, "elapsed_time": "19:28:04", "remaining_time": "9:52:03"}
132
+ {"current_steps": 1320, "total_steps": 1974, "loss": 0.4786, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.2465559169855535e-05, "epoch": 1.34, "percentage": 66.87, "elapsed_time": "19:36:26", "remaining_time": "9:42:52"}
133
+ {"current_steps": 1330, "total_steps": 1974, "loss": 0.4717, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.212291133749206e-05, "epoch": 1.35, "percentage": 67.38, "elapsed_time": "19:44:49", "remaining_time": "9:33:42"}
134
+ {"current_steps": 1340, "total_steps": 1974, "loss": 0.4803, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.178352497948384e-05, "epoch": 1.36, "percentage": 67.88, "elapsed_time": "19:53:08", "remaining_time": "9:24:31"}
135
+ {"current_steps": 1350, "total_steps": 1974, "loss": 0.4803, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.1447486054694112e-05, "epoch": 1.37, "percentage": 68.39, "elapsed_time": "20:01:28", "remaining_time": "9:15:20"}
136
+ {"current_steps": 1360, "total_steps": 1974, "loss": 0.4739, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.1114879674157233e-05, "epoch": 1.38, "percentage": 68.9, "elapsed_time": "20:09:47", "remaining_time": "9:06:11"}
137
+ {"current_steps": 1370, "total_steps": 1974, "loss": 0.471, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.0785790079522001e-05, "epoch": 1.39, "percentage": 69.4, "elapsed_time": "20:18:06", "remaining_time": "8:57:01"}
138
+ {"current_steps": 1380, "total_steps": 1974, "loss": 0.4799, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.046030062171512e-05, "epoch": 1.4, "percentage": 69.91, "elapsed_time": "20:26:25", "remaining_time": "8:47:53"}
139
+ {"current_steps": 1390, "total_steps": 1974, "loss": 0.4689, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.0138493739830352e-05, "epoch": 1.41, "percentage": 70.42, "elapsed_time": "20:34:44", "remaining_time": "8:38:46"}
140
+ {"current_steps": 1400, "total_steps": 1974, "loss": 0.4599, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 9.820450940248544e-06, "epoch": 1.42, "percentage": 70.92, "elapsed_time": "20:43:03", "remaining_time": "8:29:39"}
141
+ {"current_steps": 1410, "total_steps": 1974, "loss": 0.5019, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 9.506252775993882e-06, "epoch": 1.43, "percentage": 71.43, "elapsed_time": "21:01:37", "remaining_time": "8:24:39"}
142
+ {"current_steps": 1420, "total_steps": 1974, "loss": 0.4764, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 9.195978826331697e-06, "epoch": 1.44, "percentage": 71.94, "elapsed_time": "21:10:01", "remaining_time": "8:15:29"}
143
+ {"current_steps": 1430, "total_steps": 1974, "loss": 0.4579, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 8.889707676612791e-06, "epoch": 1.45, "percentage": 72.44, "elapsed_time": "21:18:24", "remaining_time": "8:06:19"}
144
+ {"current_steps": 1440, "total_steps": 1974, "loss": 0.4592, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 8.587516898369589e-06, "epoch": 1.46, "percentage": 72.95, "elapsed_time": "21:26:47", "remaining_time": "7:57:11"}
145
+ {"current_steps": 1450, "total_steps": 1974, "loss": 0.4861, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 8.289483029668972e-06, "epoch": 1.47, "percentage": 73.45, "elapsed_time": "21:35:09", "remaining_time": "7:48:02"}
146
+ {"current_steps": 1460, "total_steps": 1974, "loss": 0.4753, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 7.99568155572701e-06, "epoch": 1.48, "percentage": 73.96, "elapsed_time": "21:43:33", "remaining_time": "7:38:55"}
147
+ {"current_steps": 1470, "total_steps": 1974, "loss": 0.4929, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 7.706186889790209e-06, "epoch": 1.49, "percentage": 74.47, "elapsed_time": "21:51:52", "remaining_time": "7:29:47"}
148
+ {"current_steps": 1480, "total_steps": 1974, "loss": 0.4594, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 7.421072354288302e-06, "epoch": 1.5, "percentage": 74.97, "elapsed_time": "22:00:14", "remaining_time": "7:20:40"}
149
+ {"current_steps": 1490, "total_steps": 1974, "loss": 0.4912, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 7.140410162263414e-06, "epoch": 1.51, "percentage": 75.48, "elapsed_time": "22:08:35", "remaining_time": "7:11:34"}
150
+ {"current_steps": 1500, "total_steps": 1974, "loss": 0.4681, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 6.86427139908008e-06, "epoch": 1.52, "percentage": 75.99, "elapsed_time": "22:16:55", "remaining_time": "7:02:28"}
151
+ {"current_steps": 1510, "total_steps": 1974, "loss": 0.4816, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 6.5927260044209655e-06, "epoch": 1.53, "percentage": 76.49, "elapsed_time": "22:25:16", "remaining_time": "6:53:22"}
152
+ {"current_steps": 1520, "total_steps": 1974, "loss": 0.4723, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 6.3258427545727e-06, "epoch": 1.54, "percentage": 77.0, "elapsed_time": "22:33:37", "remaining_time": "6:44:18"}
153
+ {"current_steps": 1530, "total_steps": 1974, "loss": 0.4856, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 6.063689245006443e-06, "epoch": 1.55, "percentage": 77.51, "elapsed_time": "22:41:56", "remaining_time": "6:35:13"}
154
+ {"current_steps": 1540, "total_steps": 1974, "loss": 0.4829, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 5.806331873257462e-06, "epoch": 1.56, "percentage": 78.01, "elapsed_time": "22:50:16", "remaining_time": "6:26:10"}
155
+ {"current_steps": 1550, "total_steps": 1974, "loss": 0.4741, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 5.553835822108152e-06, "epoch": 1.57, "percentage": 78.52, "elapsed_time": "22:58:38", "remaining_time": "6:17:07"}
156
+ {"current_steps": 1560, "total_steps": 1974, "loss": 0.4654, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 5.306265043078693e-06, "epoch": 1.58, "percentage": 79.03, "elapsed_time": "23:06:58", "remaining_time": "6:08:04"}
157
+ {"current_steps": 1570, "total_steps": 1974, "loss": 0.4668, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 5.0636822402296165e-06, "epoch": 1.59, "percentage": 79.53, "elapsed_time": "23:15:15", "remaining_time": "5:59:02"}
158
+ {"current_steps": 1580, "total_steps": 1974, "loss": 0.4723, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.826148854280277e-06, "epoch": 1.6, "percentage": 80.04, "elapsed_time": "23:23:36", "remaining_time": "5:50:00"}
159
+ {"current_steps": 1590, "total_steps": 1974, "loss": 0.4639, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.593725047047293e-06, "epoch": 1.61, "percentage": 80.55, "elapsed_time": "23:31:55", "remaining_time": "5:40:59"}
160
+ {"current_steps": 1600, "total_steps": 1974, "loss": 0.4777, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.3664696862069505e-06, "epoch": 1.62, "percentage": 81.05, "elapsed_time": "23:40:12", "remaining_time": "5:31:58"}
161
+ {"current_steps": 1610, "total_steps": 1974, "loss": 0.4546, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.144440330385347e-06, "epoch": 1.63, "percentage": 81.56, "elapsed_time": "1 day, 0:03:58", "remaining_time": "5:26:27"}
162
+ {"current_steps": 1620, "total_steps": 1974, "loss": 0.4543, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.927693214580075e-06, "epoch": 1.64, "percentage": 82.07, "elapsed_time": "1 day, 0:12:13", "remaining_time": "5:17:20"}
163
+ {"current_steps": 1630, "total_steps": 1974, "loss": 0.4543, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.71628323591722e-06, "epoch": 1.65, "percentage": 82.57, "elapsed_time": "1 day, 0:20:27", "remaining_time": "5:08:13"}
164
+ {"current_steps": 1640, "total_steps": 1974, "loss": 0.4659, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.5102639397471214e-06, "epoch": 1.66, "percentage": 83.08, "elapsed_time": "1 day, 0:28:43", "remaining_time": "4:59:07"}
165
+ {"current_steps": 1650, "total_steps": 1974, "loss": 0.485, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.3096875060825845e-06, "epoch": 1.67, "percentage": 83.59, "elapsed_time": "1 day, 0:36:59", "remaining_time": "4:50:01"}
166
+ {"current_steps": 1660, "total_steps": 1974, "loss": 0.4768, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.11460473638282e-06, "epoch": 1.68, "percentage": 84.09, "elapsed_time": "1 day, 0:45:17", "remaining_time": "4:40:57"}
167
+ {"current_steps": 1670, "total_steps": 1974, "loss": 0.4635, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.925065040686642e-06, "epoch": 1.69, "percentage": 84.6, "elapsed_time": "1 day, 0:53:32", "remaining_time": "4:31:52"}
168
+ {"current_steps": 1680, "total_steps": 1974, "loss": 0.4681, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.741116425097995e-06, "epoch": 1.7, "percentage": 85.11, "elapsed_time": "1 day, 1:01:47", "remaining_time": "4:22:48"}
169
+ {"current_steps": 1690, "total_steps": 1974, "loss": 0.4492, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.5628054796271063e-06, "epoch": 1.71, "percentage": 85.61, "elapsed_time": "1 day, 1:10:02", "remaining_time": "4:13:45"}
170
+ {"current_steps": 1700, "total_steps": 1974, "loss": 0.4664, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.390177366390273e-06, "epoch": 1.72, "percentage": 86.12, "elapsed_time": "1 day, 1:18:18", "remaining_time": "4:04:42"}
171
+ {"current_steps": 1710, "total_steps": 1974, "loss": 0.48, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.22327580817136e-06, "epoch": 1.73, "percentage": 86.63, "elapsed_time": "1 day, 1:26:34", "remaining_time": "3:55:40"}
172
+ {"current_steps": 1720, "total_steps": 1974, "loss": 0.4616, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.0621430773477947e-06, "epoch": 1.74, "percentage": 87.13, "elapsed_time": "1 day, 1:34:47", "remaining_time": "3:46:38"}
173
+ {"current_steps": 1730, "total_steps": 1974, "loss": 0.4854, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.906819985183908e-06, "epoch": 1.75, "percentage": 87.64, "elapsed_time": "1 day, 1:43:03", "remaining_time": "3:37:38"}
174
+ {"current_steps": 1740, "total_steps": 1974, "loss": 0.4846, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.7573458714944063e-06, "epoch": 1.76, "percentage": 88.15, "elapsed_time": "1 day, 1:51:18", "remaining_time": "3:28:37"}
175
+ {"current_steps": 1750, "total_steps": 1974, "loss": 0.4552, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.6137585946804674e-06, "epoch": 1.77, "percentage": 88.65, "elapsed_time": "1 day, 1:59:35", "remaining_time": "3:19:37"}
176
+ {"current_steps": 1760, "total_steps": 1974, "loss": 0.4615, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.4760945221410638e-06, "epoch": 1.78, "percentage": 89.16, "elapsed_time": "1 day, 2:07:51", "remaining_time": "3:10:38"}
177
+ {"current_steps": 1770, "total_steps": 1974, "loss": 0.4735, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.3443885210619428e-06, "epoch": 1.79, "percentage": 89.67, "elapsed_time": "1 day, 2:16:10", "remaining_time": "3:01:39"}
178
+ {"current_steps": 1780, "total_steps": 1974, "loss": 0.4705, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.2186739495845477e-06, "epoch": 1.8, "percentage": 90.17, "elapsed_time": "1 day, 2:24:26", "remaining_time": "2:52:41"}
179
+ {"current_steps": 1790, "total_steps": 1974, "loss": 0.4653, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.0989826483571552e-06, "epoch": 1.81, "percentage": 90.68, "elapsed_time": "1 day, 2:32:40", "remaining_time": "2:43:42"}
180
+ {"current_steps": 1800, "total_steps": 1974, "loss": 0.4632, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 9.85344932470364e-07, "epoch": 1.82, "percentage": 91.19, "elapsed_time": "1 day, 2:40:51", "remaining_time": "2:34:44"}
181
+ {"current_steps": 1810, "total_steps": 1974, "loss": 0.481, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 8.77789583778979e-07, "epoch": 1.83, "percentage": 91.69, "elapsed_time": "1 day, 2:59:37", "remaining_time": "2:26:45"}
182
+ {"current_steps": 1820, "total_steps": 1974, "loss": 0.4773, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 7.763438436122122e-07, "epoch": 1.84, "percentage": 92.2, "elapsed_time": "1 day, 3:07:54", "remaining_time": "2:17:44"}
183
+ {"current_steps": 1830, "total_steps": 1974, "loss": 0.4791, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 6.810334058740736e-07, "epoch": 1.85, "percentage": 92.71, "elapsed_time": "1 day, 3:16:15", "remaining_time": "2:08:45"}
184
+ {"current_steps": 1840, "total_steps": 1974, "loss": 0.4793, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 5.918824105356797e-07, "epoch": 1.86, "percentage": 93.21, "elapsed_time": "1 day, 3:24:34", "remaining_time": "1:59:46"}
185
+ {"current_steps": 1850, "total_steps": 1974, "loss": 0.4647, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 5.08913437521169e-07, "epoch": 1.87, "percentage": 93.72, "elapsed_time": "1 day, 3:32:49", "remaining_time": "1:50:47"}
186
+ {"current_steps": 1860, "total_steps": 1974, "loss": 0.4765, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.3214750098869995e-07, "epoch": 1.88, "percentage": 94.22, "elapsed_time": "1 day, 3:41:06", "remaining_time": "1:41:48"}
187
+ {"current_steps": 1870, "total_steps": 1974, "loss": 0.4787, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 3.616040440080432e-07, "epoch": 1.89, "percentage": 94.73, "elapsed_time": "1 day, 3:49:26", "remaining_time": "1:32:50"}
188
+ {"current_steps": 1880, "total_steps": 1974, "loss": 0.4723, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.973009336361021e-07, "epoch": 1.9, "percentage": 95.24, "elapsed_time": "1 day, 3:57:44", "remaining_time": "1:23:53"}
189
+ {"current_steps": 1890, "total_steps": 1974, "loss": 0.453, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.392544563915883e-07, "epoch": 1.91, "percentage": 95.74, "elapsed_time": "1 day, 4:06:01", "remaining_time": "1:14:56"}
190
+ {"current_steps": 1900, "total_steps": 1974, "loss": 0.4561, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.8747931413001795e-07, "epoch": 1.93, "percentage": 96.25, "elapsed_time": "1 day, 4:14:18", "remaining_time": "1:05:59"}
191
+ {"current_steps": 1910, "total_steps": 1974, "loss": 0.4612, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.4198862032005488e-07, "epoch": 1.94, "percentage": 96.76, "elapsed_time": "1 day, 4:22:35", "remaining_time": "0:57:03"}
192
+ {"current_steps": 1920, "total_steps": 1974, "loss": 0.4774, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.0279389672218365e-07, "epoch": 1.95, "percentage": 97.26, "elapsed_time": "1 day, 4:30:53", "remaining_time": "0:48:07"}
193
+ {"current_steps": 1930, "total_steps": 1974, "loss": 0.4635, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 6.990507047049676e-08, "epoch": 1.96, "percentage": 97.77, "elapsed_time": "1 day, 4:39:09", "remaining_time": "0:39:11"}
194
+ {"current_steps": 1940, "total_steps": 1974, "loss": 0.4761, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 4.3330471558378213e-08, "epoch": 1.97, "percentage": 98.28, "elapsed_time": "1 day, 4:47:22", "remaining_time": "0:30:16"}
195
+ {"current_steps": 1950, "total_steps": 1974, "loss": 0.4593, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 2.3076830728713252e-08, "epoch": 1.98, "percentage": 98.78, "elapsed_time": "1 day, 4:55:39", "remaining_time": "0:21:21"}
196
+ {"current_steps": 1960, "total_steps": 1974, "loss": 0.4736, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 9.149277769132658e-09, "epoch": 1.99, "percentage": 99.29, "elapsed_time": "1 day, 5:03:51", "remaining_time": "0:12:27"}
197
+ {"current_steps": 1970, "total_steps": 1974, "loss": 0.4486, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": 1.551340212760377e-09, "epoch": 2.0, "percentage": 99.8, "elapsed_time": "1 day, 5:12:08", "remaining_time": "0:03:33"}
198
+ {"current_steps": 1974, "total_steps": 1974, "loss": null, "eval_loss": null, "predict_loss": null, "reward": null, "learning_rate": null, "epoch": 2.0, "percentage": 100.0, "elapsed_time": "1 day, 5:15:27", "remaining_time": "0:00:00"}
trainer_state.json ADDED
@@ -0,0 +1,1207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.0,
5
+ "global_step": 1974,
6
+ "is_hyper_param_search": false,
7
+ "is_local_process_zero": true,
8
+ "is_world_process_zero": true,
9
+ "log_history": [
10
+ {
11
+ "epoch": 0.01,
12
+ "learning_rate": 4.9996834033646177e-05,
13
+ "loss": 0.9175,
14
+ "step": 10
15
+ },
16
+ {
17
+ "epoch": 0.02,
18
+ "learning_rate": 4.998733693645213e-05,
19
+ "loss": 0.7595,
20
+ "step": 20
21
+ },
22
+ {
23
+ "epoch": 0.03,
24
+ "learning_rate": 4.997151111381707e-05,
25
+ "loss": 0.7375,
26
+ "step": 30
27
+ },
28
+ {
29
+ "epoch": 0.04,
30
+ "learning_rate": 4.9949360574062774e-05,
31
+ "loss": 0.7227,
32
+ "step": 40
33
+ },
34
+ {
35
+ "epoch": 0.05,
36
+ "learning_rate": 4.9920890927418316e-05,
37
+ "loss": 0.7147,
38
+ "step": 50
39
+ },
40
+ {
41
+ "epoch": 0.06,
42
+ "learning_rate": 4.988610938459917e-05,
43
+ "loss": 0.7102,
44
+ "step": 60
45
+ },
46
+ {
47
+ "epoch": 0.07,
48
+ "learning_rate": 4.9845024754980876e-05,
49
+ "loss": 0.7056,
50
+ "step": 70
51
+ },
52
+ {
53
+ "epoch": 0.08,
54
+ "learning_rate": 4.979764744436784e-05,
55
+ "loss": 0.7128,
56
+ "step": 80
57
+ },
58
+ {
59
+ "epoch": 0.09,
60
+ "learning_rate": 4.9743989452357756e-05,
61
+ "loss": 0.6982,
62
+ "step": 90
63
+ },
64
+ {
65
+ "epoch": 0.1,
66
+ "learning_rate": 4.968406436930243e-05,
67
+ "loss": 0.7258,
68
+ "step": 100
69
+ },
70
+ {
71
+ "epoch": 0.11,
72
+ "learning_rate": 4.961788737286559e-05,
73
+ "loss": 0.7095,
74
+ "step": 110
75
+ },
76
+ {
77
+ "epoch": 0.12,
78
+ "learning_rate": 4.954547522417877e-05,
79
+ "loss": 0.7048,
80
+ "step": 120
81
+ },
82
+ {
83
+ "epoch": 0.13,
84
+ "learning_rate": 4.946684626359607e-05,
85
+ "loss": 0.6805,
86
+ "step": 130
87
+ },
88
+ {
89
+ "epoch": 0.14,
90
+ "learning_rate": 4.938202040604898e-05,
91
+ "loss": 0.6798,
92
+ "step": 140
93
+ },
94
+ {
95
+ "epoch": 0.15,
96
+ "learning_rate": 4.929101913600238e-05,
97
+ "loss": 0.7134,
98
+ "step": 150
99
+ },
100
+ {
101
+ "epoch": 0.16,
102
+ "learning_rate": 4.919386550201299e-05,
103
+ "loss": 0.6895,
104
+ "step": 160
105
+ },
106
+ {
107
+ "epoch": 0.17,
108
+ "learning_rate": 4.909058411089174e-05,
109
+ "loss": 0.705,
110
+ "step": 170
111
+ },
112
+ {
113
+ "epoch": 0.18,
114
+ "learning_rate": 4.8981201121471356e-05,
115
+ "loss": 0.6712,
116
+ "step": 180
117
+ },
118
+ {
119
+ "epoch": 0.19,
120
+ "learning_rate": 4.886574423798097e-05,
121
+ "loss": 0.6744,
122
+ "step": 190
123
+ },
124
+ {
125
+ "epoch": 0.2,
126
+ "learning_rate": 4.874424270302927e-05,
127
+ "loss": 0.6675,
128
+ "step": 200
129
+ },
130
+ {
131
+ "epoch": 0.21,
132
+ "learning_rate": 4.861672729019797e-05,
133
+ "loss": 0.6726,
134
+ "step": 210
135
+ },
136
+ {
137
+ "epoch": 0.22,
138
+ "learning_rate": 4.848323029624761e-05,
139
+ "loss": 0.6821,
140
+ "step": 220
141
+ },
142
+ {
143
+ "epoch": 0.23,
144
+ "learning_rate": 4.834378553293748e-05,
145
+ "loss": 0.7133,
146
+ "step": 230
147
+ },
148
+ {
149
+ "epoch": 0.24,
150
+ "learning_rate": 4.81984283184619e-05,
151
+ "loss": 0.6681,
152
+ "step": 240
153
+ },
154
+ {
155
+ "epoch": 0.25,
156
+ "learning_rate": 4.804719546850487e-05,
157
+ "loss": 0.682,
158
+ "step": 250
159
+ },
160
+ {
161
+ "epoch": 0.26,
162
+ "learning_rate": 4.789012528691558e-05,
163
+ "loss": 0.6755,
164
+ "step": 260
165
+ },
166
+ {
167
+ "epoch": 0.27,
168
+ "learning_rate": 4.772725755600682e-05,
169
+ "loss": 0.68,
170
+ "step": 270
171
+ },
172
+ {
173
+ "epoch": 0.28,
174
+ "learning_rate": 4.755863352647909e-05,
175
+ "loss": 0.6663,
176
+ "step": 280
177
+ },
178
+ {
179
+ "epoch": 0.29,
180
+ "learning_rate": 4.738429590697271e-05,
181
+ "loss": 0.6672,
182
+ "step": 290
183
+ },
184
+ {
185
+ "epoch": 0.3,
186
+ "learning_rate": 4.720428885325069e-05,
187
+ "loss": 0.6466,
188
+ "step": 300
189
+ },
190
+ {
191
+ "epoch": 0.31,
192
+ "learning_rate": 4.701865795701505e-05,
193
+ "loss": 0.6668,
194
+ "step": 310
195
+ },
196
+ {
197
+ "epoch": 0.32,
198
+ "learning_rate": 4.684682059461469e-05,
199
+ "loss": 0.6591,
200
+ "step": 320
201
+ },
202
+ {
203
+ "epoch": 0.33,
204
+ "learning_rate": 4.665063509461097e-05,
205
+ "loss": 0.661,
206
+ "step": 330
207
+ },
208
+ {
209
+ "epoch": 0.34,
210
+ "learning_rate": 4.644896598002736e-05,
211
+ "loss": 0.6749,
212
+ "step": 340
213
+ },
214
+ {
215
+ "epoch": 0.35,
216
+ "learning_rate": 4.624186432907437e-05,
217
+ "loss": 0.6654,
218
+ "step": 350
219
+ },
220
+ {
221
+ "epoch": 0.36,
222
+ "learning_rate": 4.602938259590072e-05,
223
+ "loss": 0.6716,
224
+ "step": 360
225
+ },
226
+ {
227
+ "epoch": 0.37,
228
+ "learning_rate": 4.581157459730783e-05,
229
+ "loss": 0.6796,
230
+ "step": 370
231
+ },
232
+ {
233
+ "epoch": 0.39,
234
+ "learning_rate": 4.558849549911931e-05,
235
+ "loss": 0.6794,
236
+ "step": 380
237
+ },
238
+ {
239
+ "epoch": 0.4,
240
+ "learning_rate": 4.536020180220871e-05,
241
+ "loss": 0.6651,
242
+ "step": 390
243
+ },
244
+ {
245
+ "epoch": 0.41,
246
+ "learning_rate": 4.515032675559024e-05,
247
+ "loss": 0.6823,
248
+ "step": 400
249
+ },
250
+ {
251
+ "epoch": 0.42,
252
+ "learning_rate": 4.4912285699446786e-05,
253
+ "loss": 0.6677,
254
+ "step": 410
255
+ },
256
+ {
257
+ "epoch": 0.43,
258
+ "learning_rate": 4.4669201313179155e-05,
259
+ "loss": 0.6704,
260
+ "step": 420
261
+ },
262
+ {
263
+ "epoch": 0.44,
264
+ "learning_rate": 4.442113516454638e-05,
265
+ "loss": 0.6481,
266
+ "step": 430
267
+ },
268
+ {
269
+ "epoch": 0.45,
270
+ "learning_rate": 4.416815008307488e-05,
271
+ "loss": 0.665,
272
+ "step": 440
273
+ },
274
+ {
275
+ "epoch": 0.46,
276
+ "learning_rate": 4.391031014414514e-05,
277
+ "loss": 0.6658,
278
+ "step": 450
279
+ },
280
+ {
281
+ "epoch": 0.47,
282
+ "learning_rate": 4.364768065276284e-05,
283
+ "loss": 0.6699,
284
+ "step": 460
285
+ },
286
+ {
287
+ "epoch": 0.48,
288
+ "learning_rate": 4.338032812701867e-05,
289
+ "loss": 0.6664,
290
+ "step": 470
291
+ },
292
+ {
293
+ "epoch": 0.49,
294
+ "learning_rate": 4.310832028124069e-05,
295
+ "loss": 0.6817,
296
+ "step": 480
297
+ },
298
+ {
299
+ "epoch": 0.5,
300
+ "learning_rate": 4.283172600884393e-05,
301
+ "loss": 0.6791,
302
+ "step": 490
303
+ },
304
+ {
305
+ "epoch": 0.51,
306
+ "learning_rate": 4.2550615364881194e-05,
307
+ "loss": 0.6679,
308
+ "step": 500
309
+ },
310
+ {
311
+ "epoch": 0.52,
312
+ "learning_rate": 4.226505954829973e-05,
313
+ "loss": 0.6476,
314
+ "step": 510
315
+ },
316
+ {
317
+ "epoch": 0.53,
318
+ "learning_rate": 4.197513088390813e-05,
319
+ "loss": 0.6721,
320
+ "step": 520
321
+ },
322
+ {
323
+ "epoch": 0.54,
324
+ "learning_rate": 4.1680902804058095e-05,
325
+ "loss": 0.6563,
326
+ "step": 530
327
+ },
328
+ {
329
+ "epoch": 0.55,
330
+ "learning_rate": 4.138244983004574e-05,
331
+ "loss": 0.6391,
332
+ "step": 540
333
+ },
334
+ {
335
+ "epoch": 0.56,
336
+ "learning_rate": 4.107984755323697e-05,
337
+ "loss": 0.6672,
338
+ "step": 550
339
+ },
340
+ {
341
+ "epoch": 0.57,
342
+ "learning_rate": 4.077317261592194e-05,
343
+ "loss": 0.6497,
344
+ "step": 560
345
+ },
346
+ {
347
+ "epoch": 0.58,
348
+ "learning_rate": 4.04625026919033e-05,
349
+ "loss": 0.668,
350
+ "step": 570
351
+ },
352
+ {
353
+ "epoch": 0.59,
354
+ "learning_rate": 4.0147916466823174e-05,
355
+ "loss": 0.6682,
356
+ "step": 580
357
+ },
358
+ {
359
+ "epoch": 0.6,
360
+ "learning_rate": 3.982949361823388e-05,
361
+ "loss": 0.6352,
362
+ "step": 590
363
+ },
364
+ {
365
+ "epoch": 0.61,
366
+ "learning_rate": 3.950731479541743e-05,
367
+ "loss": 0.6698,
368
+ "step": 600
369
+ },
370
+ {
371
+ "epoch": 0.62,
372
+ "learning_rate": 3.918146159895882e-05,
373
+ "loss": 0.6549,
374
+ "step": 610
375
+ },
376
+ {
377
+ "epoch": 0.63,
378
+ "learning_rate": 3.8852016560078605e-05,
379
+ "loss": 0.6516,
380
+ "step": 620
381
+ },
382
+ {
383
+ "epoch": 0.64,
384
+ "learning_rate": 3.851906311972943e-05,
385
+ "loss": 0.6629,
386
+ "step": 630
387
+ },
388
+ {
389
+ "epoch": 0.65,
390
+ "learning_rate": 3.821647502051616e-05,
391
+ "loss": 0.6764,
392
+ "step": 640
393
+ },
394
+ {
395
+ "epoch": 0.66,
396
+ "learning_rate": 3.787708866250794e-05,
397
+ "loss": 0.6415,
398
+ "step": 650
399
+ },
400
+ {
401
+ "epoch": 0.67,
402
+ "learning_rate": 3.7534440830144466e-05,
403
+ "loss": 0.6463,
404
+ "step": 660
405
+ },
406
+ {
407
+ "epoch": 0.68,
408
+ "learning_rate": 3.71886183083464e-05,
409
+ "loss": 0.6508,
410
+ "step": 670
411
+ },
412
+ {
413
+ "epoch": 0.69,
414
+ "learning_rate": 3.683970868611123e-05,
415
+ "loss": 0.6411,
416
+ "step": 680
417
+ },
418
+ {
419
+ "epoch": 0.7,
420
+ "learning_rate": 3.648780033432891e-05,
421
+ "loss": 0.6266,
422
+ "step": 690
423
+ },
424
+ {
425
+ "epoch": 0.71,
426
+ "learning_rate": 3.613298238339955e-05,
427
+ "loss": 0.6409,
428
+ "step": 700
429
+ },
430
+ {
431
+ "epoch": 0.72,
432
+ "learning_rate": 3.5775344700658705e-05,
433
+ "loss": 0.6594,
434
+ "step": 710
435
+ },
436
+ {
437
+ "epoch": 0.73,
438
+ "learning_rate": 3.5414977867616006e-05,
439
+ "loss": 0.6427,
440
+ "step": 720
441
+ },
442
+ {
443
+ "epoch": 0.74,
444
+ "learning_rate": 3.505197315701292e-05,
445
+ "loss": 0.6462,
446
+ "step": 730
447
+ },
448
+ {
449
+ "epoch": 0.75,
450
+ "learning_rate": 3.468642250970547e-05,
451
+ "loss": 0.6277,
452
+ "step": 740
453
+ },
454
+ {
455
+ "epoch": 0.76,
456
+ "learning_rate": 3.431841851137764e-05,
457
+ "loss": 0.6551,
458
+ "step": 750
459
+ },
460
+ {
461
+ "epoch": 0.77,
462
+ "learning_rate": 3.394805436909157e-05,
463
+ "loss": 0.6402,
464
+ "step": 760
465
+ },
466
+ {
467
+ "epoch": 0.78,
468
+ "learning_rate": 3.357542388768033e-05,
469
+ "loss": 0.6515,
470
+ "step": 770
471
+ },
472
+ {
473
+ "epoch": 0.79,
474
+ "learning_rate": 3.3200621445989226e-05,
475
+ "loss": 0.6489,
476
+ "step": 780
477
+ },
478
+ {
479
+ "epoch": 0.8,
480
+ "learning_rate": 3.282374197297185e-05,
481
+ "loss": 0.6568,
482
+ "step": 790
483
+ },
484
+ {
485
+ "epoch": 0.81,
486
+ "learning_rate": 3.2444880923646674e-05,
487
+ "loss": 0.615,
488
+ "step": 800
489
+ },
490
+ {
491
+ "epoch": 0.82,
492
+ "learning_rate": 3.20641342549205e-05,
493
+ "loss": 0.6447,
494
+ "step": 810
495
+ },
496
+ {
497
+ "epoch": 0.83,
498
+ "learning_rate": 3.168159840128472e-05,
499
+ "loss": 0.6159,
500
+ "step": 820
501
+ },
502
+ {
503
+ "epoch": 0.84,
504
+ "learning_rate": 3.129737025039068e-05,
505
+ "loss": 0.6347,
506
+ "step": 830
507
+ },
508
+ {
509
+ "epoch": 0.85,
510
+ "learning_rate": 3.091154711851022e-05,
511
+ "loss": 0.6361,
512
+ "step": 840
513
+ },
514
+ {
515
+ "epoch": 0.86,
516
+ "learning_rate": 3.052422672588765e-05,
517
+ "loss": 0.6504,
518
+ "step": 850
519
+ },
520
+ {
521
+ "epoch": 0.87,
522
+ "learning_rate": 3.013550717198948e-05,
523
+ "loss": 0.6467,
524
+ "step": 860
525
+ },
526
+ {
527
+ "epoch": 0.88,
528
+ "learning_rate": 2.9745486910657993e-05,
529
+ "loss": 0.6364,
530
+ "step": 870
531
+ },
532
+ {
533
+ "epoch": 0.89,
534
+ "learning_rate": 2.9354264725175185e-05,
535
+ "loss": 0.6361,
536
+ "step": 880
537
+ },
538
+ {
539
+ "epoch": 0.9,
540
+ "learning_rate": 2.8961939703243122e-05,
541
+ "loss": 0.6441,
542
+ "step": 890
543
+ },
544
+ {
545
+ "epoch": 0.91,
546
+ "learning_rate": 2.856861121188735e-05,
547
+ "loss": 0.6404,
548
+ "step": 900
549
+ },
550
+ {
551
+ "epoch": 0.92,
552
+ "learning_rate": 2.8174378872289446e-05,
553
+ "loss": 0.6307,
554
+ "step": 910
555
+ },
556
+ {
557
+ "epoch": 0.93,
558
+ "learning_rate": 2.777934253455522e-05,
559
+ "loss": 0.6484,
560
+ "step": 920
561
+ },
562
+ {
563
+ "epoch": 0.94,
564
+ "learning_rate": 2.7383602252424985e-05,
565
+ "loss": 0.6237,
566
+ "step": 930
567
+ },
568
+ {
569
+ "epoch": 0.95,
570
+ "learning_rate": 2.6987258257932175e-05,
571
+ "loss": 0.6161,
572
+ "step": 940
573
+ },
574
+ {
575
+ "epoch": 0.96,
576
+ "learning_rate": 2.6590410936016895e-05,
577
+ "loss": 0.6381,
578
+ "step": 950
579
+ },
580
+ {
581
+ "epoch": 0.97,
582
+ "learning_rate": 2.619316079910063e-05,
583
+ "loss": 0.6366,
584
+ "step": 960
585
+ },
586
+ {
587
+ "epoch": 0.98,
588
+ "learning_rate": 2.5795608461628802e-05,
589
+ "loss": 0.6202,
590
+ "step": 970
591
+ },
592
+ {
593
+ "epoch": 0.99,
594
+ "learning_rate": 2.5397854614587334e-05,
595
+ "loss": 0.6334,
596
+ "step": 980
597
+ },
598
+ {
599
+ "epoch": 1.0,
600
+ "learning_rate": 2.5e-05,
601
+ "loss": 0.5954,
602
+ "step": 990
603
+ },
604
+ {
605
+ "epoch": 1.01,
606
+ "learning_rate": 2.460214538541267e-05,
607
+ "loss": 0.4963,
608
+ "step": 1000
609
+ },
610
+ {
611
+ "epoch": 1.02,
612
+ "learning_rate": 2.4204391538371207e-05,
613
+ "loss": 0.487,
614
+ "step": 1010
615
+ },
616
+ {
617
+ "epoch": 1.03,
618
+ "learning_rate": 2.3806839200899377e-05,
619
+ "loss": 0.489,
620
+ "step": 1020
621
+ },
622
+ {
623
+ "epoch": 1.04,
624
+ "learning_rate": 2.3409589063983117e-05,
625
+ "loss": 0.4805,
626
+ "step": 1030
627
+ },
628
+ {
629
+ "epoch": 1.05,
630
+ "learning_rate": 2.3012741742067838e-05,
631
+ "loss": 0.4907,
632
+ "step": 1040
633
+ },
634
+ {
635
+ "epoch": 1.06,
636
+ "learning_rate": 2.261639774757503e-05,
637
+ "loss": 0.4719,
638
+ "step": 1050
639
+ },
640
+ {
641
+ "epoch": 1.07,
642
+ "learning_rate": 2.2220657465444782e-05,
643
+ "loss": 0.4914,
644
+ "step": 1060
645
+ },
646
+ {
647
+ "epoch": 1.08,
648
+ "learning_rate": 2.182562112771056e-05,
649
+ "loss": 0.4775,
650
+ "step": 1070
651
+ },
652
+ {
653
+ "epoch": 1.09,
654
+ "learning_rate": 2.143138878811265e-05,
655
+ "loss": 0.4935,
656
+ "step": 1080
657
+ },
658
+ {
659
+ "epoch": 1.1,
660
+ "learning_rate": 2.1038060296756883e-05,
661
+ "loss": 0.5082,
662
+ "step": 1090
663
+ },
664
+ {
665
+ "epoch": 1.11,
666
+ "learning_rate": 2.064573527482482e-05,
667
+ "loss": 0.4868,
668
+ "step": 1100
669
+ },
670
+ {
671
+ "epoch": 1.12,
672
+ "learning_rate": 2.025451308934201e-05,
673
+ "loss": 0.4988,
674
+ "step": 1110
675
+ },
676
+ {
677
+ "epoch": 1.13,
678
+ "learning_rate": 1.9864492828010526e-05,
679
+ "loss": 0.4653,
680
+ "step": 1120
681
+ },
682
+ {
683
+ "epoch": 1.14,
684
+ "learning_rate": 1.9475773274112354e-05,
685
+ "loss": 0.4915,
686
+ "step": 1130
687
+ },
688
+ {
689
+ "epoch": 1.16,
690
+ "learning_rate": 1.9088452881489787e-05,
691
+ "loss": 0.4763,
692
+ "step": 1140
693
+ },
694
+ {
695
+ "epoch": 1.17,
696
+ "learning_rate": 1.8702629749609324e-05,
697
+ "loss": 0.4807,
698
+ "step": 1150
699
+ },
700
+ {
701
+ "epoch": 1.18,
702
+ "learning_rate": 1.8318401598715284e-05,
703
+ "loss": 0.4653,
704
+ "step": 1160
705
+ },
706
+ {
707
+ "epoch": 1.19,
708
+ "learning_rate": 1.793586574507951e-05,
709
+ "loss": 0.4778,
710
+ "step": 1170
711
+ },
712
+ {
713
+ "epoch": 1.2,
714
+ "learning_rate": 1.7555119076353338e-05,
715
+ "loss": 0.4839,
716
+ "step": 1180
717
+ },
718
+ {
719
+ "epoch": 1.21,
720
+ "learning_rate": 1.7176258027028152e-05,
721
+ "loss": 0.4718,
722
+ "step": 1190
723
+ },
724
+ {
725
+ "epoch": 1.22,
726
+ "learning_rate": 1.6799378554010773e-05,
727
+ "loss": 0.4793,
728
+ "step": 1200
729
+ },
730
+ {
731
+ "epoch": 1.23,
732
+ "learning_rate": 1.6424576112319672e-05,
733
+ "loss": 0.4825,
734
+ "step": 1210
735
+ },
736
+ {
737
+ "epoch": 1.24,
738
+ "learning_rate": 1.6051945630908426e-05,
739
+ "loss": 0.4857,
740
+ "step": 1220
741
+ },
742
+ {
743
+ "epoch": 1.25,
744
+ "learning_rate": 1.5681581488622367e-05,
745
+ "loss": 0.4802,
746
+ "step": 1230
747
+ },
748
+ {
749
+ "epoch": 1.26,
750
+ "learning_rate": 1.5313577490294538e-05,
751
+ "loss": 0.4812,
752
+ "step": 1240
753
+ },
754
+ {
755
+ "epoch": 1.27,
756
+ "learning_rate": 1.4948026842987084e-05,
757
+ "loss": 0.4682,
758
+ "step": 1250
759
+ },
760
+ {
761
+ "epoch": 1.28,
762
+ "learning_rate": 1.4585022132384008e-05,
763
+ "loss": 0.4974,
764
+ "step": 1260
765
+ },
766
+ {
767
+ "epoch": 1.29,
768
+ "learning_rate": 1.4224655299341304e-05,
769
+ "loss": 0.4737,
770
+ "step": 1270
771
+ },
772
+ {
773
+ "epoch": 1.3,
774
+ "learning_rate": 1.3867017616600456e-05,
775
+ "loss": 0.4877,
776
+ "step": 1280
777
+ },
778
+ {
779
+ "epoch": 1.31,
780
+ "learning_rate": 1.3512199665671094e-05,
781
+ "loss": 0.4753,
782
+ "step": 1290
783
+ },
784
+ {
785
+ "epoch": 1.32,
786
+ "learning_rate": 1.316029131388878e-05,
787
+ "loss": 0.4638,
788
+ "step": 1300
789
+ },
790
+ {
791
+ "epoch": 1.33,
792
+ "learning_rate": 1.2811381691653607e-05,
793
+ "loss": 0.4626,
794
+ "step": 1310
795
+ },
796
+ {
797
+ "epoch": 1.34,
798
+ "learning_rate": 1.2465559169855535e-05,
799
+ "loss": 0.4786,
800
+ "step": 1320
801
+ },
802
+ {
803
+ "epoch": 1.35,
804
+ "learning_rate": 1.212291133749206e-05,
805
+ "loss": 0.4717,
806
+ "step": 1330
807
+ },
808
+ {
809
+ "epoch": 1.36,
810
+ "learning_rate": 1.178352497948384e-05,
811
+ "loss": 0.4803,
812
+ "step": 1340
813
+ },
814
+ {
815
+ "epoch": 1.37,
816
+ "learning_rate": 1.1447486054694112e-05,
817
+ "loss": 0.4803,
818
+ "step": 1350
819
+ },
820
+ {
821
+ "epoch": 1.38,
822
+ "learning_rate": 1.1114879674157233e-05,
823
+ "loss": 0.4739,
824
+ "step": 1360
825
+ },
826
+ {
827
+ "epoch": 1.39,
828
+ "learning_rate": 1.0785790079522001e-05,
829
+ "loss": 0.471,
830
+ "step": 1370
831
+ },
832
+ {
833
+ "epoch": 1.4,
834
+ "learning_rate": 1.046030062171512e-05,
835
+ "loss": 0.4799,
836
+ "step": 1380
837
+ },
838
+ {
839
+ "epoch": 1.41,
840
+ "learning_rate": 1.0138493739830352e-05,
841
+ "loss": 0.4689,
842
+ "step": 1390
843
+ },
844
+ {
845
+ "epoch": 1.42,
846
+ "learning_rate": 9.820450940248544e-06,
847
+ "loss": 0.4599,
848
+ "step": 1400
849
+ },
850
+ {
851
+ "epoch": 1.43,
852
+ "learning_rate": 9.506252775993882e-06,
853
+ "loss": 0.5019,
854
+ "step": 1410
855
+ },
856
+ {
857
+ "epoch": 1.44,
858
+ "learning_rate": 9.195978826331697e-06,
859
+ "loss": 0.4764,
860
+ "step": 1420
861
+ },
862
+ {
863
+ "epoch": 1.45,
864
+ "learning_rate": 8.889707676612791e-06,
865
+ "loss": 0.4579,
866
+ "step": 1430
867
+ },
868
+ {
869
+ "epoch": 1.46,
870
+ "learning_rate": 8.587516898369589e-06,
871
+ "loss": 0.4592,
872
+ "step": 1440
873
+ },
874
+ {
875
+ "epoch": 1.47,
876
+ "learning_rate": 8.289483029668972e-06,
877
+ "loss": 0.4861,
878
+ "step": 1450
879
+ },
880
+ {
881
+ "epoch": 1.48,
882
+ "learning_rate": 7.99568155572701e-06,
883
+ "loss": 0.4753,
884
+ "step": 1460
885
+ },
886
+ {
887
+ "epoch": 1.49,
888
+ "learning_rate": 7.706186889790209e-06,
889
+ "loss": 0.4929,
890
+ "step": 1470
891
+ },
892
+ {
893
+ "epoch": 1.5,
894
+ "learning_rate": 7.421072354288302e-06,
895
+ "loss": 0.4594,
896
+ "step": 1480
897
+ },
898
+ {
899
+ "epoch": 1.51,
900
+ "learning_rate": 7.140410162263414e-06,
901
+ "loss": 0.4912,
902
+ "step": 1490
903
+ },
904
+ {
905
+ "epoch": 1.52,
906
+ "learning_rate": 6.86427139908008e-06,
907
+ "loss": 0.4681,
908
+ "step": 1500
909
+ },
910
+ {
911
+ "epoch": 1.53,
912
+ "learning_rate": 6.5927260044209655e-06,
913
+ "loss": 0.4816,
914
+ "step": 1510
915
+ },
916
+ {
917
+ "epoch": 1.54,
918
+ "learning_rate": 6.3258427545727e-06,
919
+ "loss": 0.4723,
920
+ "step": 1520
921
+ },
922
+ {
923
+ "epoch": 1.55,
924
+ "learning_rate": 6.063689245006443e-06,
925
+ "loss": 0.4856,
926
+ "step": 1530
927
+ },
928
+ {
929
+ "epoch": 1.56,
930
+ "learning_rate": 5.806331873257462e-06,
931
+ "loss": 0.4829,
932
+ "step": 1540
933
+ },
934
+ {
935
+ "epoch": 1.57,
936
+ "learning_rate": 5.553835822108152e-06,
937
+ "loss": 0.4741,
938
+ "step": 1550
939
+ },
940
+ {
941
+ "epoch": 1.58,
942
+ "learning_rate": 5.306265043078693e-06,
943
+ "loss": 0.4654,
944
+ "step": 1560
945
+ },
946
+ {
947
+ "epoch": 1.59,
948
+ "learning_rate": 5.0636822402296165e-06,
949
+ "loss": 0.4668,
950
+ "step": 1570
951
+ },
952
+ {
953
+ "epoch": 1.6,
954
+ "learning_rate": 4.826148854280277e-06,
955
+ "loss": 0.4723,
956
+ "step": 1580
957
+ },
958
+ {
959
+ "epoch": 1.61,
960
+ "learning_rate": 4.593725047047293e-06,
961
+ "loss": 0.4639,
962
+ "step": 1590
963
+ },
964
+ {
965
+ "epoch": 1.62,
966
+ "learning_rate": 4.3664696862069505e-06,
967
+ "loss": 0.4777,
968
+ "step": 1600
969
+ },
970
+ {
971
+ "epoch": 1.63,
972
+ "learning_rate": 4.144440330385347e-06,
973
+ "loss": 0.4546,
974
+ "step": 1610
975
+ },
976
+ {
977
+ "epoch": 1.64,
978
+ "learning_rate": 3.927693214580075e-06,
979
+ "loss": 0.4543,
980
+ "step": 1620
981
+ },
982
+ {
983
+ "epoch": 1.65,
984
+ "learning_rate": 3.71628323591722e-06,
985
+ "loss": 0.4543,
986
+ "step": 1630
987
+ },
988
+ {
989
+ "epoch": 1.66,
990
+ "learning_rate": 3.5102639397471214e-06,
991
+ "loss": 0.4659,
992
+ "step": 1640
993
+ },
994
+ {
995
+ "epoch": 1.67,
996
+ "learning_rate": 3.3096875060825845e-06,
997
+ "loss": 0.485,
998
+ "step": 1650
999
+ },
1000
+ {
1001
+ "epoch": 1.68,
1002
+ "learning_rate": 3.11460473638282e-06,
1003
+ "loss": 0.4768,
1004
+ "step": 1660
1005
+ },
1006
+ {
1007
+ "epoch": 1.69,
1008
+ "learning_rate": 2.925065040686642e-06,
1009
+ "loss": 0.4635,
1010
+ "step": 1670
1011
+ },
1012
+ {
1013
+ "epoch": 1.7,
1014
+ "learning_rate": 2.741116425097995e-06,
1015
+ "loss": 0.4681,
1016
+ "step": 1680
1017
+ },
1018
+ {
1019
+ "epoch": 1.71,
1020
+ "learning_rate": 2.5628054796271063e-06,
1021
+ "loss": 0.4492,
1022
+ "step": 1690
1023
+ },
1024
+ {
1025
+ "epoch": 1.72,
1026
+ "learning_rate": 2.390177366390273e-06,
1027
+ "loss": 0.4664,
1028
+ "step": 1700
1029
+ },
1030
+ {
1031
+ "epoch": 1.73,
1032
+ "learning_rate": 2.22327580817136e-06,
1033
+ "loss": 0.48,
1034
+ "step": 1710
1035
+ },
1036
+ {
1037
+ "epoch": 1.74,
1038
+ "learning_rate": 2.0621430773477947e-06,
1039
+ "loss": 0.4616,
1040
+ "step": 1720
1041
+ },
1042
+ {
1043
+ "epoch": 1.75,
1044
+ "learning_rate": 1.906819985183908e-06,
1045
+ "loss": 0.4854,
1046
+ "step": 1730
1047
+ },
1048
+ {
1049
+ "epoch": 1.76,
1050
+ "learning_rate": 1.7573458714944063e-06,
1051
+ "loss": 0.4846,
1052
+ "step": 1740
1053
+ },
1054
+ {
1055
+ "epoch": 1.77,
1056
+ "learning_rate": 1.6137585946804674e-06,
1057
+ "loss": 0.4552,
1058
+ "step": 1750
1059
+ },
1060
+ {
1061
+ "epoch": 1.78,
1062
+ "learning_rate": 1.4760945221410638e-06,
1063
+ "loss": 0.4615,
1064
+ "step": 1760
1065
+ },
1066
+ {
1067
+ "epoch": 1.79,
1068
+ "learning_rate": 1.3443885210619428e-06,
1069
+ "loss": 0.4735,
1070
+ "step": 1770
1071
+ },
1072
+ {
1073
+ "epoch": 1.8,
1074
+ "learning_rate": 1.2186739495845477e-06,
1075
+ "loss": 0.4705,
1076
+ "step": 1780
1077
+ },
1078
+ {
1079
+ "epoch": 1.81,
1080
+ "learning_rate": 1.0989826483571552e-06,
1081
+ "loss": 0.4653,
1082
+ "step": 1790
1083
+ },
1084
+ {
1085
+ "epoch": 1.82,
1086
+ "learning_rate": 9.85344932470364e-07,
1087
+ "loss": 0.4632,
1088
+ "step": 1800
1089
+ },
1090
+ {
1091
+ "epoch": 1.83,
1092
+ "learning_rate": 8.77789583778979e-07,
1093
+ "loss": 0.481,
1094
+ "step": 1810
1095
+ },
1096
+ {
1097
+ "epoch": 1.84,
1098
+ "learning_rate": 7.763438436122122e-07,
1099
+ "loss": 0.4773,
1100
+ "step": 1820
1101
+ },
1102
+ {
1103
+ "epoch": 1.85,
1104
+ "learning_rate": 6.810334058740736e-07,
1105
+ "loss": 0.4791,
1106
+ "step": 1830
1107
+ },
1108
+ {
1109
+ "epoch": 1.86,
1110
+ "learning_rate": 5.918824105356797e-07,
1111
+ "loss": 0.4793,
1112
+ "step": 1840
1113
+ },
1114
+ {
1115
+ "epoch": 1.87,
1116
+ "learning_rate": 5.08913437521169e-07,
1117
+ "loss": 0.4647,
1118
+ "step": 1850
1119
+ },
1120
+ {
1121
+ "epoch": 1.88,
1122
+ "learning_rate": 4.3214750098869995e-07,
1123
+ "loss": 0.4765,
1124
+ "step": 1860
1125
+ },
1126
+ {
1127
+ "epoch": 1.89,
1128
+ "learning_rate": 3.616040440080432e-07,
1129
+ "loss": 0.4787,
1130
+ "step": 1870
1131
+ },
1132
+ {
1133
+ "epoch": 1.9,
1134
+ "learning_rate": 2.973009336361021e-07,
1135
+ "loss": 0.4723,
1136
+ "step": 1880
1137
+ },
1138
+ {
1139
+ "epoch": 1.91,
1140
+ "learning_rate": 2.392544563915883e-07,
1141
+ "loss": 0.453,
1142
+ "step": 1890
1143
+ },
1144
+ {
1145
+ "epoch": 1.93,
1146
+ "learning_rate": 1.8747931413001795e-07,
1147
+ "loss": 0.4561,
1148
+ "step": 1900
1149
+ },
1150
+ {
1151
+ "epoch": 1.94,
1152
+ "learning_rate": 1.4198862032005488e-07,
1153
+ "loss": 0.4612,
1154
+ "step": 1910
1155
+ },
1156
+ {
1157
+ "epoch": 1.95,
1158
+ "learning_rate": 1.0279389672218365e-07,
1159
+ "loss": 0.4774,
1160
+ "step": 1920
1161
+ },
1162
+ {
1163
+ "epoch": 1.96,
1164
+ "learning_rate": 6.990507047049676e-08,
1165
+ "loss": 0.4635,
1166
+ "step": 1930
1167
+ },
1168
+ {
1169
+ "epoch": 1.97,
1170
+ "learning_rate": 4.3330471558378213e-08,
1171
+ "loss": 0.4761,
1172
+ "step": 1940
1173
+ },
1174
+ {
1175
+ "epoch": 1.98,
1176
+ "learning_rate": 2.3076830728713252e-08,
1177
+ "loss": 0.4593,
1178
+ "step": 1950
1179
+ },
1180
+ {
1181
+ "epoch": 1.99,
1182
+ "learning_rate": 9.149277769132658e-09,
1183
+ "loss": 0.4736,
1184
+ "step": 1960
1185
+ },
1186
+ {
1187
+ "epoch": 2.0,
1188
+ "learning_rate": 1.551340212760377e-09,
1189
+ "loss": 0.4486,
1190
+ "step": 1970
1191
+ },
1192
+ {
1193
+ "epoch": 2.0,
1194
+ "step": 1974,
1195
+ "total_flos": 3.437359013145084e+19,
1196
+ "train_loss": 0.5708327819994277,
1197
+ "train_runtime": 105327.89,
1198
+ "train_samples_per_second": 4.797,
1199
+ "train_steps_per_second": 0.019
1200
+ }
1201
+ ],
1202
+ "max_steps": 1974,
1203
+ "num_train_epochs": 2,
1204
+ "total_flos": 3.437359013145084e+19,
1205
+ "trial_name": null,
1206
+ "trial_params": null
1207
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe0b7bf81fc699ffb5b38f22a7bd9ad201bc6a51a7928e5d8ae8e14abe781142
3
+ size 5752
training_loss.png ADDED