hrnph
commited on
Commit
•
f66a816
1
Parent(s):
663c947
Add files!
Browse files- checkpoint-300/config.json +37 -0
- checkpoint-300/configuration_gpt2_mq.py +201 -0
- checkpoint-300/modeling_gpt2_mq.py +346 -0
- checkpoint-300/optimizer.pt +3 -0
- checkpoint-300/pytorch_model.bin +3 -0
- checkpoint-300/rng_state.pth +3 -0
- checkpoint-300/scaler.pt +3 -0
- checkpoint-300/scheduler.pt +3 -0
- checkpoint-300/trainer_state.json +2056 -0
- checkpoint-300/training_args.bin +3 -0
checkpoint-300/config.json
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "bigcode/santacoder",
|
3 |
+
"activation_function": "gelu_fast",
|
4 |
+
"architectures": [
|
5 |
+
"GPT2LMHeadCustomModel"
|
6 |
+
],
|
7 |
+
"attention_head_type": "multiquery",
|
8 |
+
"attn_pdrop": 0.1,
|
9 |
+
"auto_map": {
|
10 |
+
"AutoConfig": "configuration_gpt2_mq.GPT2CustomConfig",
|
11 |
+
"AutoModelForCausalLM": "modeling_gpt2_mq.GPT2LMHeadCustomModel"
|
12 |
+
},
|
13 |
+
"bos_token_id": 49152,
|
14 |
+
"embd_pdrop": 0.1,
|
15 |
+
"eos_token_id": 49152,
|
16 |
+
"initializer_range": 0.02,
|
17 |
+
"layer_norm_epsilon": 1e-05,
|
18 |
+
"model_type": "gpt2",
|
19 |
+
"n_embd": 2048,
|
20 |
+
"n_head": 16,
|
21 |
+
"n_inner": 8192,
|
22 |
+
"n_layer": 24,
|
23 |
+
"n_positions": 2048,
|
24 |
+
"reorder_and_upcast_attn": false,
|
25 |
+
"resid_pdrop": 0.1,
|
26 |
+
"scale_attn_by_inverse_layer_idx": false,
|
27 |
+
"scale_attn_weights": true,
|
28 |
+
"summary_activation": null,
|
29 |
+
"summary_first_dropout": 0.1,
|
30 |
+
"summary_proj_to_labels": true,
|
31 |
+
"summary_type": "cls_index",
|
32 |
+
"summary_use_proj": true,
|
33 |
+
"torch_dtype": "float32",
|
34 |
+
"transformers_version": "4.26.0.dev0",
|
35 |
+
"use_cache": false,
|
36 |
+
"vocab_size": 49280
|
37 |
+
}
|
checkpoint-300/configuration_gpt2_mq.py
ADDED
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2018 The OpenAI Team Authors and Hugging Face Inc. team.
|
3 |
+
# Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
|
4 |
+
#
|
5 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
6 |
+
# you may not use this file except in compliance with the License.
|
7 |
+
# You may obtain a copy of the License at
|
8 |
+
#
|
9 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
10 |
+
#
|
11 |
+
# Unless required by applicable law or agreed to in writing, software
|
12 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
13 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
14 |
+
# See the License for the specific language governing permissions and
|
15 |
+
# limitations under the License.
|
16 |
+
""" Custom GPT-2 configuration"""
|
17 |
+
from collections import OrderedDict
|
18 |
+
from typing import Any, List, Mapping, Optional
|
19 |
+
from enum import Enum
|
20 |
+
|
21 |
+
from transformers import PreTrainedTokenizer, TensorType, is_torch_available
|
22 |
+
|
23 |
+
from transformers.configuration_utils import PretrainedConfig
|
24 |
+
from transformers.onnx import OnnxConfigWithPast, PatchingSpec
|
25 |
+
from transformers.utils import logging
|
26 |
+
|
27 |
+
|
28 |
+
logger = logging.get_logger(__name__)
|
29 |
+
|
30 |
+
GPT2_PRETRAINED_CONFIG_ARCHIVE_MAP = {
|
31 |
+
"gpt2": "https://huggingface.co/gpt2/resolve/main/config.json",
|
32 |
+
"gpt2-medium": "https://huggingface.co/gpt2-medium/resolve/main/config.json",
|
33 |
+
"gpt2-large": "https://huggingface.co/gpt2-large/resolve/main/config.json",
|
34 |
+
"gpt2-xl": "https://huggingface.co/gpt2-xl/resolve/main/config.json",
|
35 |
+
"distilgpt2": "https://huggingface.co/distilgpt2/resolve/main/config.json",
|
36 |
+
}
|
37 |
+
|
38 |
+
MULTI_HEAD = "multihead"
|
39 |
+
MULTI_QUERY = "multiquery"
|
40 |
+
|
41 |
+
|
42 |
+
class GPT2CustomConfig(PretrainedConfig):
|
43 |
+
"""
|
44 |
+
This is the configuration class to store the configuration of a [`GPT2Model`] or a [`TFGPT2Model`]. It is used to
|
45 |
+
instantiate a GPT-2 model according to the specified arguments, defining the model architecture. Instantiating a
|
46 |
+
configuration with the defaults will yield a similar configuration to that of the GPT-2
|
47 |
+
[gpt2](https://huggingface.co/gpt2) architecture.
|
48 |
+
|
49 |
+
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
|
50 |
+
documentation from [`PretrainedConfig`] for more information.
|
51 |
+
|
52 |
+
|
53 |
+
Args:
|
54 |
+
vocab_size (`int`, *optional*, defaults to 50257):
|
55 |
+
Vocabulary size of the GPT-2 model. Defines the number of different tokens that can be represented by the
|
56 |
+
`inputs_ids` passed when calling [`GPT2Model`] or [`TFGPT2Model`].
|
57 |
+
n_positions (`int`, *optional*, defaults to 1024):
|
58 |
+
The maximum sequence length that this model might ever be used with. Typically set this to something large
|
59 |
+
just in case (e.g., 512 or 1024 or 2048).
|
60 |
+
n_embd (`int`, *optional*, defaults to 768):
|
61 |
+
Dimensionality of the embeddings and hidden states.
|
62 |
+
n_layer (`int`, *optional*, defaults to 12):
|
63 |
+
Number of hidden layers in the Transformer encoder.
|
64 |
+
n_head (`int`, *optional*, defaults to 12):
|
65 |
+
Number of attention heads for each attention layer in the Transformer encoder.
|
66 |
+
n_inner (`int`, *optional*, defaults to None):
|
67 |
+
Dimensionality of the inner feed-forward layers. `None` will set it to 4 times n_embd
|
68 |
+
activation_function (`str`, *optional*, defaults to `"gelu"`):
|
69 |
+
Activation function, to be selected in the list `["relu", "silu", "gelu", "tanh", "gelu_new"]`.
|
70 |
+
resid_pdrop (`float`, *optional*, defaults to 0.1):
|
71 |
+
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
|
72 |
+
embd_pdrop (`int`, *optional*, defaults to 0.1):
|
73 |
+
The dropout ratio for the embeddings.
|
74 |
+
attn_pdrop (`float`, *optional*, defaults to 0.1):
|
75 |
+
The dropout ratio for the attention.
|
76 |
+
layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
|
77 |
+
The epsilon to use in the layer normalization layers.
|
78 |
+
initializer_range (`float`, *optional*, defaults to 0.02):
|
79 |
+
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
|
80 |
+
summary_type (`string`, *optional*, defaults to `"cls_index"`):
|
81 |
+
Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
|
82 |
+
[`TFGPT2DoubleHeadsModel`].
|
83 |
+
|
84 |
+
Has to be one of the following options:
|
85 |
+
|
86 |
+
- `"last"`: Take the last token hidden state (like XLNet).
|
87 |
+
- `"first"`: Take the first token hidden state (like BERT).
|
88 |
+
- `"mean"`: Take the mean of all tokens hidden states.
|
89 |
+
- `"cls_index"`: Supply a Tensor of classification token position (like GPT/GPT-2).
|
90 |
+
- `"attn"`: Not implemented now, use multi-head attention.
|
91 |
+
summary_use_proj (`bool`, *optional*, defaults to `True`):
|
92 |
+
Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
|
93 |
+
[`TFGPT2DoubleHeadsModel`].
|
94 |
+
|
95 |
+
Whether or not to add a projection after the vector extraction.
|
96 |
+
summary_activation (`str`, *optional*):
|
97 |
+
Argument used when doing sequence summary. Used in for the multiple choice head in
|
98 |
+
[`GPT2DoubleHeadsModel`].
|
99 |
+
|
100 |
+
Pass `"tanh"` for a tanh activation to the output, any other value will result in no activation.
|
101 |
+
summary_proj_to_labels (`bool`, *optional*, defaults to `True`):
|
102 |
+
Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
|
103 |
+
[`TFGPT2DoubleHeadsModel`].
|
104 |
+
|
105 |
+
Whether the projection outputs should have `config.num_labels` or `config.hidden_size` classes.
|
106 |
+
summary_first_dropout (`float`, *optional*, defaults to 0.1):
|
107 |
+
Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
|
108 |
+
[`TFGPT2DoubleHeadsModel`].
|
109 |
+
|
110 |
+
The dropout ratio to be used after the projection and activation.
|
111 |
+
scale_attn_weights (`bool`, *optional*, defaults to `True`):
|
112 |
+
Scale attention weights by dividing by sqrt(head_dim)..
|
113 |
+
use_cache (`bool`, *optional*, defaults to `True`):
|
114 |
+
Whether or not the model should return the last key/values attentions (not used by all models).
|
115 |
+
scale_attn_by_inverse_layer_idx (`bool`, *optional*, defaults to `False`):
|
116 |
+
Whether to additionally scale attention weights by `1 / layer_idx + 1`.
|
117 |
+
reorder_and_upcast_attn (`bool`, *optional*, defaults to `False`):
|
118 |
+
Whether to scale keys (K) prior to computing attention (dot-product) and upcast attention
|
119 |
+
dot-product/softmax to float() when training with mixed precision.
|
120 |
+
|
121 |
+
Example:
|
122 |
+
|
123 |
+
```python
|
124 |
+
>>> from transformers import GPT2Config, GPT2Model
|
125 |
+
|
126 |
+
>>> # Initializing a GPT2 configuration
|
127 |
+
>>> configuration = GPT2Config()
|
128 |
+
|
129 |
+
>>> # Initializing a model (with random weights) from the configuration
|
130 |
+
>>> model = GPT2Model(configuration)
|
131 |
+
|
132 |
+
>>> # Accessing the model configuration
|
133 |
+
>>> configuration = model.config
|
134 |
+
```"""
|
135 |
+
|
136 |
+
model_type = "gpt2"
|
137 |
+
keys_to_ignore_at_inference = ["past_key_values"]
|
138 |
+
attribute_map = {
|
139 |
+
"hidden_size": "n_embd",
|
140 |
+
"max_position_embeddings": "n_positions",
|
141 |
+
"num_attention_heads": "n_head",
|
142 |
+
"num_hidden_layers": "n_layer",
|
143 |
+
}
|
144 |
+
|
145 |
+
def __init__(
|
146 |
+
self,
|
147 |
+
vocab_size=50257,
|
148 |
+
n_positions=1024,
|
149 |
+
n_embd=768,
|
150 |
+
n_layer=12,
|
151 |
+
n_head=12,
|
152 |
+
n_inner=None,
|
153 |
+
activation_function="gelu_new",
|
154 |
+
resid_pdrop=0.1,
|
155 |
+
embd_pdrop=0.1,
|
156 |
+
attn_pdrop=0.1,
|
157 |
+
layer_norm_epsilon=1e-5,
|
158 |
+
initializer_range=0.02,
|
159 |
+
summary_type="cls_index",
|
160 |
+
summary_use_proj=True,
|
161 |
+
summary_activation=None,
|
162 |
+
summary_proj_to_labels=True,
|
163 |
+
summary_first_dropout=0.1,
|
164 |
+
scale_attn_weights=True,
|
165 |
+
use_cache=True,
|
166 |
+
bos_token_id=50256,
|
167 |
+
eos_token_id=50256,
|
168 |
+
scale_attn_by_inverse_layer_idx=False,
|
169 |
+
reorder_and_upcast_attn=False,
|
170 |
+
attention_head_type=MULTI_HEAD,
|
171 |
+
**kwargs,
|
172 |
+
):
|
173 |
+
self.vocab_size = vocab_size
|
174 |
+
self.n_positions = n_positions
|
175 |
+
self.n_embd = n_embd
|
176 |
+
self.n_layer = n_layer
|
177 |
+
self.n_head = n_head
|
178 |
+
self.n_inner = n_inner
|
179 |
+
self.activation_function = activation_function
|
180 |
+
self.resid_pdrop = resid_pdrop
|
181 |
+
self.embd_pdrop = embd_pdrop
|
182 |
+
self.attn_pdrop = attn_pdrop
|
183 |
+
self.layer_norm_epsilon = layer_norm_epsilon
|
184 |
+
self.initializer_range = initializer_range
|
185 |
+
self.summary_type = summary_type
|
186 |
+
self.summary_use_proj = summary_use_proj
|
187 |
+
self.summary_activation = summary_activation
|
188 |
+
self.summary_first_dropout = summary_first_dropout
|
189 |
+
self.summary_proj_to_labels = summary_proj_to_labels
|
190 |
+
self.scale_attn_weights = scale_attn_weights
|
191 |
+
self.use_cache = use_cache
|
192 |
+
self.scale_attn_by_inverse_layer_idx = scale_attn_by_inverse_layer_idx
|
193 |
+
self.reorder_and_upcast_attn = reorder_and_upcast_attn
|
194 |
+
self.attention_head_type = attention_head_type
|
195 |
+
# assert attention_head_type in [AttentionType.MULTI_HEAD, AttentionType.MULTI_QUERY]
|
196 |
+
assert attention_head_type in [MULTI_HEAD, MULTI_QUERY]
|
197 |
+
|
198 |
+
self.bos_token_id = bos_token_id
|
199 |
+
self.eos_token_id = eos_token_id
|
200 |
+
|
201 |
+
super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
|
checkpoint-300/modeling_gpt2_mq.py
ADDED
@@ -0,0 +1,346 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""PyTorch OpenAI GPT-2 model modified with MultiQuery attention"""
|
2 |
+
|
3 |
+
|
4 |
+
import math
|
5 |
+
import os
|
6 |
+
from dataclasses import dataclass
|
7 |
+
from typing import Optional, Tuple, Union
|
8 |
+
|
9 |
+
import torch
|
10 |
+
import torch.utils.checkpoint
|
11 |
+
from torch import nn
|
12 |
+
from torch.cuda.amp import autocast
|
13 |
+
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
|
14 |
+
|
15 |
+
from transformers.activations import ACT2FN
|
16 |
+
from transformers.modeling_outputs import (
|
17 |
+
BaseModelOutputWithPastAndCrossAttentions,
|
18 |
+
CausalLMOutputWithCrossAttentions,
|
19 |
+
SequenceClassifierOutputWithPast,
|
20 |
+
TokenClassifierOutput,
|
21 |
+
)
|
22 |
+
from transformers.modeling_utils import PreTrainedModel, SequenceSummary
|
23 |
+
from transformers.pytorch_utils import Conv1D, find_pruneable_heads_and_indices, prune_conv1d_layer
|
24 |
+
|
25 |
+
from transformers.utils import (
|
26 |
+
ModelOutput,
|
27 |
+
add_code_sample_docstrings,
|
28 |
+
add_start_docstrings,
|
29 |
+
add_start_docstrings_to_model_forward,
|
30 |
+
logging,
|
31 |
+
replace_return_docstrings,
|
32 |
+
)
|
33 |
+
from transformers.utils.model_parallel_utils import assert_device_map, get_device_map
|
34 |
+
from transformers.models.gpt2.modeling_gpt2 import GPT2Model, GPT2Block, GPT2PreTrainedModel, GPT2LMHeadModel
|
35 |
+
from .configuration_gpt2_mq import GPT2CustomConfig, MULTI_QUERY, MULTI_HEAD
|
36 |
+
|
37 |
+
|
38 |
+
|
39 |
+
class GPT2MQAttention(nn.Module):
|
40 |
+
def __init__(self, config, is_cross_attention=False, layer_idx=None):
|
41 |
+
super().__init__()
|
42 |
+
assert config.attention_head_type == MULTI_QUERY
|
43 |
+
|
44 |
+
max_positions = config.max_position_embeddings
|
45 |
+
self.register_buffer(
|
46 |
+
"bias",
|
47 |
+
torch.tril(torch.ones((max_positions, max_positions), dtype=torch.uint8)).view(
|
48 |
+
1, 1, max_positions, max_positions
|
49 |
+
),
|
50 |
+
)
|
51 |
+
self.register_buffer("masked_bias", torch.tensor(-1e4))
|
52 |
+
|
53 |
+
self.embed_dim = config.hidden_size
|
54 |
+
self.num_heads = config.num_attention_heads
|
55 |
+
self.head_dim = self.embed_dim // self.num_heads
|
56 |
+
self.split_size = self.embed_dim
|
57 |
+
if self.head_dim * self.num_heads != self.embed_dim:
|
58 |
+
raise ValueError(
|
59 |
+
f"`embed_dim` must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
|
60 |
+
f" {self.num_heads})."
|
61 |
+
)
|
62 |
+
|
63 |
+
self.scale_attn_weights = config.scale_attn_weights
|
64 |
+
if is_cross_attention:
|
65 |
+
raise NotImplementedError("Cross-attention not implemented for MQA")
|
66 |
+
self.is_cross_attention = is_cross_attention
|
67 |
+
|
68 |
+
# Layer-wise attention scaling, reordering, and upcasting
|
69 |
+
self.scale_attn_by_inverse_layer_idx = config.scale_attn_by_inverse_layer_idx
|
70 |
+
self.layer_idx = layer_idx
|
71 |
+
self.reorder_and_upcast_attn = config.reorder_and_upcast_attn
|
72 |
+
|
73 |
+
if self.is_cross_attention:
|
74 |
+
self.c_attn = Conv1D(2 * self.embed_dim, self.embed_dim)
|
75 |
+
self.q_attn = Conv1D(self.embed_dim, self.embed_dim)
|
76 |
+
else:
|
77 |
+
# self.c_attn = Conv1D(3 * self.embed_dim, self.embed_dim)
|
78 |
+
self.q_attn = Conv1D(self.embed_dim, self.embed_dim)
|
79 |
+
# Keys and values are shared across heads
|
80 |
+
self.kv_attn = Conv1D(2 * self.head_dim, self.embed_dim)
|
81 |
+
self.c_proj = Conv1D(self.embed_dim, self.embed_dim)
|
82 |
+
|
83 |
+
self.attn_dropout = nn.Dropout(config.attn_pdrop)
|
84 |
+
self.resid_dropout = nn.Dropout(config.resid_pdrop)
|
85 |
+
|
86 |
+
self.pruned_heads = set()
|
87 |
+
|
88 |
+
def prune_heads(self, heads):
|
89 |
+
if len(heads) == 0:
|
90 |
+
return
|
91 |
+
heads, index = find_pruneable_heads_and_indices(heads, self.num_heads, self.head_dim, self.pruned_heads)
|
92 |
+
index_attn = torch.cat([index, index + self.split_size, index + (2 * self.split_size)])
|
93 |
+
|
94 |
+
# Prune conv1d layers
|
95 |
+
self.c_attn = prune_conv1d_layer(self.c_attn, index_attn, dim=1)
|
96 |
+
self.c_proj = prune_conv1d_layer(self.c_proj, index, dim=0)
|
97 |
+
|
98 |
+
# Update hyper params
|
99 |
+
self.split_size = (self.split_size // self.num_heads) * (self.num_heads - len(heads))
|
100 |
+
self.num_heads = self.num_heads - len(heads)
|
101 |
+
self.pruned_heads = self.pruned_heads.union(heads)
|
102 |
+
|
103 |
+
def _attn(self, query, key, value, attention_mask=None, head_mask=None):
|
104 |
+
# query: (b, num_heads * sq, head_dim)
|
105 |
+
# key: (b, head_dim, sk)
|
106 |
+
# value: (b, sk, head_dim)
|
107 |
+
batch_size = query.size(0)
|
108 |
+
query_length = query.size(1) // self.num_heads
|
109 |
+
key_length = key.size(2)
|
110 |
+
# (b, num_heads * sq, head_dim) x (b, head_dim, sk) -> (b, num_heads * sq, sk)
|
111 |
+
attn_weights = torch.bmm(query, key)
|
112 |
+
# -> (b, num_heads, sq, sk)
|
113 |
+
attn_weights = attn_weights.view(batch_size, self.num_heads, query_length, key_length)
|
114 |
+
|
115 |
+
if self.scale_attn_weights:
|
116 |
+
attn_weights = attn_weights / torch.tensor(
|
117 |
+
value.size(-1) ** 0.5, dtype=attn_weights.dtype, device=attn_weights.device
|
118 |
+
)
|
119 |
+
|
120 |
+
# Layer-wise attention scaling
|
121 |
+
if self.scale_attn_by_inverse_layer_idx:
|
122 |
+
attn_weights = attn_weights / float(self.layer_idx + 1)
|
123 |
+
|
124 |
+
if not self.is_cross_attention:
|
125 |
+
# if only "normal" attention layer implements causal mask
|
126 |
+
causal_mask = self.bias[:, :, key_length - query_length : key_length, :key_length].to(torch.bool)
|
127 |
+
mask_value = torch.finfo(attn_weights.dtype).min
|
128 |
+
# Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
|
129 |
+
# Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
|
130 |
+
mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(attn_weights.device)
|
131 |
+
attn_weights = torch.where(causal_mask, attn_weights, mask_value)
|
132 |
+
|
133 |
+
if attention_mask is not None:
|
134 |
+
# Apply the attention mask
|
135 |
+
attn_weights = attn_weights + attention_mask
|
136 |
+
|
137 |
+
attn_weights = nn.functional.softmax(attn_weights, dim=-1)
|
138 |
+
|
139 |
+
# Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op otherwise
|
140 |
+
attn_weights = attn_weights.type(value.dtype)
|
141 |
+
attn_weights = self.attn_dropout(attn_weights)
|
142 |
+
|
143 |
+
# Mask heads if we want to
|
144 |
+
if head_mask is not None:
|
145 |
+
attn_weights = attn_weights * head_mask
|
146 |
+
|
147 |
+
# (b, num_heads, sq, sk) -> (b, num_heads * sq, sk)
|
148 |
+
_attn_weights = attn_weights.view(batch_size, self.num_heads * query_length, key_length)
|
149 |
+
# (b, num_heads * sq, sk) x (b, sk, head_dim) -> (b, num_heads * sq, head_dim)
|
150 |
+
attn_output = torch.bmm(_attn_weights, value)
|
151 |
+
attn_output = attn_output.view(batch_size, self.num_heads, query_length, self.head_dim)
|
152 |
+
|
153 |
+
return attn_output, attn_weights
|
154 |
+
|
155 |
+
def _upcast_and_reordered_attn(self, query, key, value, attention_mask=None, head_mask=None):
|
156 |
+
# Use `torch.baddbmm` (a bit more efficient w/ alpha param for scaling -- from Megatron-LM)
|
157 |
+
bsz, num_heads, q_seq_len, dk = query.size()
|
158 |
+
_, _, k_seq_len, _ = key.size()
|
159 |
+
|
160 |
+
# Preallocate attn_weights for `baddbmm`
|
161 |
+
attn_weights = torch.empty(bsz * num_heads, q_seq_len, k_seq_len, dtype=torch.float32, device=query.device)
|
162 |
+
|
163 |
+
# Compute Scale Factor
|
164 |
+
scale_factor = 1.0
|
165 |
+
if self.scale_attn_weights:
|
166 |
+
scale_factor /= float(value.size(-1)) ** 0.5
|
167 |
+
|
168 |
+
if self.scale_attn_by_inverse_layer_idx:
|
169 |
+
scale_factor /= float(self.layer_idx + 1)
|
170 |
+
|
171 |
+
# Upcast (turn off autocast) and reorder (Scale K by 1 / root(dk))
|
172 |
+
with autocast(enabled=False):
|
173 |
+
q, k = query.reshape(-1, q_seq_len, dk), key.transpose(-1, -2).reshape(-1, dk, k_seq_len)
|
174 |
+
attn_weights = torch.baddbmm(attn_weights, q.float(), k.float(), beta=0, alpha=scale_factor)
|
175 |
+
attn_weights = attn_weights.reshape(bsz, num_heads, q_seq_len, k_seq_len)
|
176 |
+
|
177 |
+
if not self.is_cross_attention:
|
178 |
+
# if only "normal" attention layer implements causal mask
|
179 |
+
query_length, key_length = query.size(-2), key.size(-2)
|
180 |
+
causal_mask = self.bias[:, :, key_length - query_length : key_length, :key_length].bool()
|
181 |
+
mask_value = torch.finfo(attn_weights.dtype).min
|
182 |
+
# Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
|
183 |
+
# Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
|
184 |
+
mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(attn_weights.device)
|
185 |
+
attn_weights = torch.where(causal_mask, attn_weights, mask_value)
|
186 |
+
|
187 |
+
if attention_mask is not None:
|
188 |
+
# Apply the attention mask
|
189 |
+
attn_weights = attn_weights + attention_mask
|
190 |
+
|
191 |
+
attn_weights = nn.functional.softmax(attn_weights, dim=-1)
|
192 |
+
|
193 |
+
# Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op if otherwise
|
194 |
+
if attn_weights.dtype != torch.float32:
|
195 |
+
raise RuntimeError("Error with upcasting, attn_weights does not have dtype torch.float32")
|
196 |
+
attn_weights = attn_weights.type(value.dtype)
|
197 |
+
attn_weights = self.attn_dropout(attn_weights)
|
198 |
+
|
199 |
+
# Mask heads if we want to
|
200 |
+
if head_mask is not None:
|
201 |
+
attn_weights = attn_weights * head_mask
|
202 |
+
|
203 |
+
attn_output = torch.matmul(attn_weights, value)
|
204 |
+
|
205 |
+
return attn_output, attn_weights
|
206 |
+
|
207 |
+
def _split_heads(self, tensor, num_heads, attn_head_size):
|
208 |
+
"""
|
209 |
+
Splits hidden_size dim into attn_head_size and num_heads
|
210 |
+
"""
|
211 |
+
new_shape = tensor.size()[:-1] + (num_heads, attn_head_size)
|
212 |
+
tensor = tensor.view(new_shape)
|
213 |
+
return tensor.permute(0, 2, 1, 3) # (batch, head, seq_length, head_features)
|
214 |
+
|
215 |
+
def _merge_heads(self, tensor, num_heads, attn_head_size):
|
216 |
+
"""
|
217 |
+
Merges attn_head_size dim and num_attn_heads dim into hidden_size
|
218 |
+
"""
|
219 |
+
tensor = tensor.permute(0, 2, 1, 3).contiguous()
|
220 |
+
new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,)
|
221 |
+
return tensor.view(new_shape)
|
222 |
+
|
223 |
+
def forward(
|
224 |
+
self,
|
225 |
+
hidden_states: Optional[Tuple[torch.FloatTensor]],
|
226 |
+
layer_past: Optional[Tuple[torch.Tensor]] = None,
|
227 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
228 |
+
head_mask: Optional[torch.FloatTensor] = None,
|
229 |
+
encoder_hidden_states: Optional[torch.Tensor] = None,
|
230 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
231 |
+
use_cache: Optional[bool] = False,
|
232 |
+
output_attentions: Optional[bool] = False,
|
233 |
+
) -> Tuple[Union[torch.Tensor, Tuple[torch.Tensor]], ...]:
|
234 |
+
if encoder_hidden_states is not None:
|
235 |
+
raise NotImplementedError("Cross-attention not implemented for MQA")
|
236 |
+
if not hasattr(self, "q_attn"):
|
237 |
+
raise ValueError(
|
238 |
+
"If class is used as cross attention, the weights `q_attn` have to be defined. "
|
239 |
+
"Please make sure to instantiate class with `GPT2Attention(..., is_cross_attention=True)`."
|
240 |
+
)
|
241 |
+
|
242 |
+
query = self.q_attn(hidden_states)
|
243 |
+
key, value = self.c_attn(encoder_hidden_states).split(self.split_size, dim=2)
|
244 |
+
attention_mask = encoder_attention_mask
|
245 |
+
else:
|
246 |
+
query = self.q_attn(hidden_states)
|
247 |
+
key, value = self.kv_attn(hidden_states).split(self.head_dim, dim=2)
|
248 |
+
|
249 |
+
|
250 |
+
batch_size, seq_length = query.shape[:2]
|
251 |
+
# (query_length, batch, num_heads, head_dim)
|
252 |
+
# (batch, num_heads * query_length, head_dim)\
|
253 |
+
|
254 |
+
# (batch, query_length, hidden_size) -> (batch, num_heads, query_length, head_dim)
|
255 |
+
query = query.view(batch_size, seq_length, self.num_heads, self.head_dim).permute([0, 2, 1, 3])
|
256 |
+
# -> (batch, num_heads * query_length, head_dim)
|
257 |
+
query = query.reshape(batch_size, self.num_heads * seq_length, self.head_dim)
|
258 |
+
|
259 |
+
# (batch, query_length, hidden_size) -> (batch, query_length * num_heads, head_dim)
|
260 |
+
# query = query.view(
|
261 |
+
# batch_size, seq_length, self.num_heads, self.head_dim,
|
262 |
+
# ).reshape(
|
263 |
+
# batch_size, seq_length * self.num_heads, self.head_dim
|
264 |
+
# )
|
265 |
+
key = key.permute(0, 2, 1) # (batch_size, head_dim, seq_length)
|
266 |
+
# value (batch_size, seq_length, head_dim)
|
267 |
+
|
268 |
+
if layer_past is not None:
|
269 |
+
past_key, past_value = layer_past
|
270 |
+
# Concatenate on sequence dimension
|
271 |
+
key = torch.cat((past_key, key), dim=-1)
|
272 |
+
value = torch.cat((past_value, value), dim=-2)
|
273 |
+
|
274 |
+
if use_cache is True:
|
275 |
+
present = (key, value)
|
276 |
+
else:
|
277 |
+
present = None
|
278 |
+
|
279 |
+
if self.reorder_and_upcast_attn:
|
280 |
+
raise NotImplementedError("Reorder and upcast attention not implemented for MQA")
|
281 |
+
attn_output, attn_weights = self._upcast_and_reordered_attn(query, key, value, attention_mask, head_mask)
|
282 |
+
else:
|
283 |
+
attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
|
284 |
+
|
285 |
+
attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim)
|
286 |
+
attn_output = self.c_proj(attn_output)
|
287 |
+
attn_output = self.resid_dropout(attn_output)
|
288 |
+
|
289 |
+
outputs = (attn_output, present)
|
290 |
+
if output_attentions:
|
291 |
+
outputs += (attn_weights,)
|
292 |
+
|
293 |
+
return outputs # a, present, (attentions)
|
294 |
+
|
295 |
+
|
296 |
+
# inherit from gpt_modeling.py, and override `attn` module
|
297 |
+
class GPT2CustomBlock(GPT2Block):
|
298 |
+
|
299 |
+
def __init__(self, config: GPT2CustomConfig, layer_idx=None):
|
300 |
+
super().__init__(config, layer_idx)
|
301 |
+
# Override attention module if using multiquery
|
302 |
+
if config.attention_head_type == MULTI_QUERY:
|
303 |
+
self.attn = GPT2MQAttention(config, layer_idx=layer_idx)
|
304 |
+
if config.add_cross_attention:
|
305 |
+
raise NotImplementedError("Cross-attention not implemented for MQA")
|
306 |
+
|
307 |
+
|
308 |
+
# inherit from gpt_modeling.py and override `__init__` method
|
309 |
+
class GPT2CustomModel(GPT2Model):
|
310 |
+
config_class = GPT2CustomConfig
|
311 |
+
|
312 |
+
def __init__(self, config):
|
313 |
+
GPT2PreTrainedModel.__init__(self, config)
|
314 |
+
|
315 |
+
self.embed_dim = config.hidden_size
|
316 |
+
|
317 |
+
self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
|
318 |
+
self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim)
|
319 |
+
|
320 |
+
self.drop = nn.Dropout(config.embd_pdrop)
|
321 |
+
self.h = nn.ModuleList([GPT2CustomBlock(config, layer_idx=i) for i in range(config.num_hidden_layers)])
|
322 |
+
self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
|
323 |
+
|
324 |
+
# Model parallel
|
325 |
+
self.model_parallel = False
|
326 |
+
self.device_map = None
|
327 |
+
self.gradient_checkpointing = False
|
328 |
+
|
329 |
+
# Initialize weights and apply final processing
|
330 |
+
self.post_init()
|
331 |
+
|
332 |
+
|
333 |
+
class GPT2LMHeadCustomModel(GPT2LMHeadModel):
|
334 |
+
config_class = GPT2CustomConfig
|
335 |
+
|
336 |
+
def __init__(self, config):
|
337 |
+
GPT2PreTrainedModel.__init__(self, config)
|
338 |
+
self.transformer = GPT2CustomModel(config)
|
339 |
+
self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
|
340 |
+
|
341 |
+
# Model parallel
|
342 |
+
self.model_parallel = False
|
343 |
+
self.device_map = None
|
344 |
+
|
345 |
+
# Initialize weights and apply final processing
|
346 |
+
self.post_init()
|
checkpoint-300/optimizer.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:84d14294d941b0df2fd04f37973b38a81296ea1f41f9bf9e16b871c0ab8bd14b
|
3 |
+
size 1459
|
checkpoint-300/pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6928345a3054ec03006d7e1b519f551c79196893f038cb6cabc68a9a4f246e42
|
3 |
+
size 4600336581
|
checkpoint-300/rng_state.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c7744e589658518cdc45ed0562339611dea49d59dab5f337b421d0fee1ff7d2e
|
3 |
+
size 14575
|
checkpoint-300/scaler.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:7fa181fa360d46feed4180ea17c8b6a4a879a9b4231c2e91aff2be20be9076cc
|
3 |
+
size 557
|
checkpoint-300/scheduler.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:11c20fa9a427ab3fa3faf10cbf9ec355ead8f7d713d92d58a7afc36da47f1a0b
|
3 |
+
size 627
|
checkpoint-300/trainer_state.json
ADDED
@@ -0,0 +1,2056 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 0.3,
|
5 |
+
"global_step": 300,
|
6 |
+
"is_hyper_param_search": false,
|
7 |
+
"is_local_process_zero": true,
|
8 |
+
"is_world_process_zero": true,
|
9 |
+
"log_history": [
|
10 |
+
{
|
11 |
+
"epoch": 0.0,
|
12 |
+
"learning_rate": 0.0,
|
13 |
+
"loss": 1.4642,
|
14 |
+
"step": 1
|
15 |
+
},
|
16 |
+
{
|
17 |
+
"epoch": 0.0,
|
18 |
+
"learning_rate": 0.0,
|
19 |
+
"loss": 1.4922,
|
20 |
+
"step": 2
|
21 |
+
},
|
22 |
+
{
|
23 |
+
"epoch": 0.0,
|
24 |
+
"learning_rate": 0.0,
|
25 |
+
"loss": 1.4662,
|
26 |
+
"step": 3
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"epoch": 0.0,
|
30 |
+
"learning_rate": 0.0,
|
31 |
+
"loss": 1.4925,
|
32 |
+
"step": 4
|
33 |
+
},
|
34 |
+
{
|
35 |
+
"epoch": 0.01,
|
36 |
+
"learning_rate": 0.0,
|
37 |
+
"loss": 1.4821,
|
38 |
+
"step": 5
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"epoch": 0.01,
|
42 |
+
"learning_rate": 0.0,
|
43 |
+
"loss": 1.4877,
|
44 |
+
"step": 6
|
45 |
+
},
|
46 |
+
{
|
47 |
+
"epoch": 0.01,
|
48 |
+
"learning_rate": 0.0,
|
49 |
+
"loss": 1.4703,
|
50 |
+
"step": 7
|
51 |
+
},
|
52 |
+
{
|
53 |
+
"epoch": 0.01,
|
54 |
+
"learning_rate": 0.0,
|
55 |
+
"loss": 1.4768,
|
56 |
+
"step": 8
|
57 |
+
},
|
58 |
+
{
|
59 |
+
"epoch": 0.01,
|
60 |
+
"learning_rate": 0.0,
|
61 |
+
"loss": 1.479,
|
62 |
+
"step": 9
|
63 |
+
},
|
64 |
+
{
|
65 |
+
"epoch": 0.01,
|
66 |
+
"learning_rate": 0.0,
|
67 |
+
"loss": 1.4581,
|
68 |
+
"step": 10
|
69 |
+
},
|
70 |
+
{
|
71 |
+
"epoch": 0.01,
|
72 |
+
"eval_loss": 1.497441053390503,
|
73 |
+
"eval_runtime": 2.4584,
|
74 |
+
"eval_samples_per_second": 34.575,
|
75 |
+
"eval_steps_per_second": 2.034,
|
76 |
+
"step": 10
|
77 |
+
},
|
78 |
+
{
|
79 |
+
"epoch": 0.01,
|
80 |
+
"learning_rate": 0.0,
|
81 |
+
"loss": 1.5086,
|
82 |
+
"step": 11
|
83 |
+
},
|
84 |
+
{
|
85 |
+
"epoch": 0.01,
|
86 |
+
"learning_rate": 0.0,
|
87 |
+
"loss": 1.5105,
|
88 |
+
"step": 12
|
89 |
+
},
|
90 |
+
{
|
91 |
+
"epoch": 0.01,
|
92 |
+
"learning_rate": 0.0,
|
93 |
+
"loss": 1.4347,
|
94 |
+
"step": 13
|
95 |
+
},
|
96 |
+
{
|
97 |
+
"epoch": 0.01,
|
98 |
+
"learning_rate": 0.0,
|
99 |
+
"loss": 1.4871,
|
100 |
+
"step": 14
|
101 |
+
},
|
102 |
+
{
|
103 |
+
"epoch": 0.01,
|
104 |
+
"learning_rate": 0.0,
|
105 |
+
"loss": 1.504,
|
106 |
+
"step": 15
|
107 |
+
},
|
108 |
+
{
|
109 |
+
"epoch": 0.02,
|
110 |
+
"learning_rate": 0.0,
|
111 |
+
"loss": 1.4708,
|
112 |
+
"step": 16
|
113 |
+
},
|
114 |
+
{
|
115 |
+
"epoch": 0.02,
|
116 |
+
"learning_rate": 0.0,
|
117 |
+
"loss": 1.5296,
|
118 |
+
"step": 17
|
119 |
+
},
|
120 |
+
{
|
121 |
+
"epoch": 0.02,
|
122 |
+
"learning_rate": 0.0,
|
123 |
+
"loss": 1.4056,
|
124 |
+
"step": 18
|
125 |
+
},
|
126 |
+
{
|
127 |
+
"epoch": 0.02,
|
128 |
+
"learning_rate": 0.0,
|
129 |
+
"loss": 1.4928,
|
130 |
+
"step": 19
|
131 |
+
},
|
132 |
+
{
|
133 |
+
"epoch": 0.02,
|
134 |
+
"learning_rate": 0.0,
|
135 |
+
"loss": 1.4764,
|
136 |
+
"step": 20
|
137 |
+
},
|
138 |
+
{
|
139 |
+
"epoch": 0.02,
|
140 |
+
"eval_loss": 1.497441053390503,
|
141 |
+
"eval_runtime": 2.4788,
|
142 |
+
"eval_samples_per_second": 34.291,
|
143 |
+
"eval_steps_per_second": 2.017,
|
144 |
+
"step": 20
|
145 |
+
},
|
146 |
+
{
|
147 |
+
"epoch": 0.02,
|
148 |
+
"learning_rate": 0.0,
|
149 |
+
"loss": 1.4898,
|
150 |
+
"step": 21
|
151 |
+
},
|
152 |
+
{
|
153 |
+
"epoch": 0.02,
|
154 |
+
"learning_rate": 0.0,
|
155 |
+
"loss": 1.4475,
|
156 |
+
"step": 22
|
157 |
+
},
|
158 |
+
{
|
159 |
+
"epoch": 0.02,
|
160 |
+
"learning_rate": 0.0,
|
161 |
+
"loss": 1.4926,
|
162 |
+
"step": 23
|
163 |
+
},
|
164 |
+
{
|
165 |
+
"epoch": 0.02,
|
166 |
+
"learning_rate": 0.0,
|
167 |
+
"loss": 1.4854,
|
168 |
+
"step": 24
|
169 |
+
},
|
170 |
+
{
|
171 |
+
"epoch": 0.03,
|
172 |
+
"learning_rate": 0.0,
|
173 |
+
"loss": 1.4763,
|
174 |
+
"step": 25
|
175 |
+
},
|
176 |
+
{
|
177 |
+
"epoch": 0.03,
|
178 |
+
"learning_rate": 0.0,
|
179 |
+
"loss": 1.4496,
|
180 |
+
"step": 26
|
181 |
+
},
|
182 |
+
{
|
183 |
+
"epoch": 0.03,
|
184 |
+
"learning_rate": 0.0,
|
185 |
+
"loss": 1.4878,
|
186 |
+
"step": 27
|
187 |
+
},
|
188 |
+
{
|
189 |
+
"epoch": 0.03,
|
190 |
+
"learning_rate": 0.0,
|
191 |
+
"loss": 1.4838,
|
192 |
+
"step": 28
|
193 |
+
},
|
194 |
+
{
|
195 |
+
"epoch": 0.03,
|
196 |
+
"learning_rate": 0.0,
|
197 |
+
"loss": 1.5233,
|
198 |
+
"step": 29
|
199 |
+
},
|
200 |
+
{
|
201 |
+
"epoch": 0.03,
|
202 |
+
"learning_rate": 0.0,
|
203 |
+
"loss": 1.4398,
|
204 |
+
"step": 30
|
205 |
+
},
|
206 |
+
{
|
207 |
+
"epoch": 0.03,
|
208 |
+
"eval_loss": 1.4974406957626343,
|
209 |
+
"eval_runtime": 2.4674,
|
210 |
+
"eval_samples_per_second": 34.449,
|
211 |
+
"eval_steps_per_second": 2.026,
|
212 |
+
"step": 30
|
213 |
+
},
|
214 |
+
{
|
215 |
+
"epoch": 0.03,
|
216 |
+
"learning_rate": 0.0,
|
217 |
+
"loss": 1.4769,
|
218 |
+
"step": 31
|
219 |
+
},
|
220 |
+
{
|
221 |
+
"epoch": 0.03,
|
222 |
+
"learning_rate": 0.0,
|
223 |
+
"loss": 1.5235,
|
224 |
+
"step": 32
|
225 |
+
},
|
226 |
+
{
|
227 |
+
"epoch": 0.03,
|
228 |
+
"learning_rate": 0.0,
|
229 |
+
"loss": 1.4976,
|
230 |
+
"step": 33
|
231 |
+
},
|
232 |
+
{
|
233 |
+
"epoch": 0.03,
|
234 |
+
"learning_rate": 0.0,
|
235 |
+
"loss": 1.5019,
|
236 |
+
"step": 34
|
237 |
+
},
|
238 |
+
{
|
239 |
+
"epoch": 0.04,
|
240 |
+
"learning_rate": 0.0,
|
241 |
+
"loss": 1.4437,
|
242 |
+
"step": 35
|
243 |
+
},
|
244 |
+
{
|
245 |
+
"epoch": 0.04,
|
246 |
+
"learning_rate": 0.0,
|
247 |
+
"loss": 1.4414,
|
248 |
+
"step": 36
|
249 |
+
},
|
250 |
+
{
|
251 |
+
"epoch": 0.04,
|
252 |
+
"learning_rate": 0.0,
|
253 |
+
"loss": 1.527,
|
254 |
+
"step": 37
|
255 |
+
},
|
256 |
+
{
|
257 |
+
"epoch": 0.04,
|
258 |
+
"learning_rate": 0.0,
|
259 |
+
"loss": 1.4977,
|
260 |
+
"step": 38
|
261 |
+
},
|
262 |
+
{
|
263 |
+
"epoch": 0.04,
|
264 |
+
"learning_rate": 0.0,
|
265 |
+
"loss": 1.4703,
|
266 |
+
"step": 39
|
267 |
+
},
|
268 |
+
{
|
269 |
+
"epoch": 0.04,
|
270 |
+
"learning_rate": 0.0,
|
271 |
+
"loss": 1.4633,
|
272 |
+
"step": 40
|
273 |
+
},
|
274 |
+
{
|
275 |
+
"epoch": 0.04,
|
276 |
+
"eval_loss": 1.497441053390503,
|
277 |
+
"eval_runtime": 2.4589,
|
278 |
+
"eval_samples_per_second": 34.568,
|
279 |
+
"eval_steps_per_second": 2.033,
|
280 |
+
"step": 40
|
281 |
+
},
|
282 |
+
{
|
283 |
+
"epoch": 0.04,
|
284 |
+
"learning_rate": 0.0,
|
285 |
+
"loss": 1.5169,
|
286 |
+
"step": 41
|
287 |
+
},
|
288 |
+
{
|
289 |
+
"epoch": 0.04,
|
290 |
+
"learning_rate": 0.0,
|
291 |
+
"loss": 1.5016,
|
292 |
+
"step": 42
|
293 |
+
},
|
294 |
+
{
|
295 |
+
"epoch": 0.04,
|
296 |
+
"learning_rate": 0.0,
|
297 |
+
"loss": 1.4505,
|
298 |
+
"step": 43
|
299 |
+
},
|
300 |
+
{
|
301 |
+
"epoch": 0.04,
|
302 |
+
"learning_rate": 0.0,
|
303 |
+
"loss": 1.4539,
|
304 |
+
"step": 44
|
305 |
+
},
|
306 |
+
{
|
307 |
+
"epoch": 0.04,
|
308 |
+
"learning_rate": 0.0,
|
309 |
+
"loss": 1.5177,
|
310 |
+
"step": 45
|
311 |
+
},
|
312 |
+
{
|
313 |
+
"epoch": 0.05,
|
314 |
+
"learning_rate": 0.0,
|
315 |
+
"loss": 1.4662,
|
316 |
+
"step": 46
|
317 |
+
},
|
318 |
+
{
|
319 |
+
"epoch": 0.05,
|
320 |
+
"learning_rate": 0.0,
|
321 |
+
"loss": 1.4824,
|
322 |
+
"step": 47
|
323 |
+
},
|
324 |
+
{
|
325 |
+
"epoch": 0.05,
|
326 |
+
"learning_rate": 0.0,
|
327 |
+
"loss": 1.4901,
|
328 |
+
"step": 48
|
329 |
+
},
|
330 |
+
{
|
331 |
+
"epoch": 0.05,
|
332 |
+
"learning_rate": 0.0,
|
333 |
+
"loss": 1.4714,
|
334 |
+
"step": 49
|
335 |
+
},
|
336 |
+
{
|
337 |
+
"epoch": 0.05,
|
338 |
+
"learning_rate": 0.0,
|
339 |
+
"loss": 1.4663,
|
340 |
+
"step": 50
|
341 |
+
},
|
342 |
+
{
|
343 |
+
"epoch": 0.05,
|
344 |
+
"eval_loss": 1.4974409341812134,
|
345 |
+
"eval_runtime": 2.4708,
|
346 |
+
"eval_samples_per_second": 34.401,
|
347 |
+
"eval_steps_per_second": 2.024,
|
348 |
+
"step": 50
|
349 |
+
},
|
350 |
+
{
|
351 |
+
"epoch": 0.05,
|
352 |
+
"learning_rate": 0.0,
|
353 |
+
"loss": 1.4819,
|
354 |
+
"step": 51
|
355 |
+
},
|
356 |
+
{
|
357 |
+
"epoch": 0.05,
|
358 |
+
"learning_rate": 0.0,
|
359 |
+
"loss": 1.495,
|
360 |
+
"step": 52
|
361 |
+
},
|
362 |
+
{
|
363 |
+
"epoch": 0.05,
|
364 |
+
"learning_rate": 0.0,
|
365 |
+
"loss": 1.4785,
|
366 |
+
"step": 53
|
367 |
+
},
|
368 |
+
{
|
369 |
+
"epoch": 0.05,
|
370 |
+
"learning_rate": 0.0,
|
371 |
+
"loss": 1.4907,
|
372 |
+
"step": 54
|
373 |
+
},
|
374 |
+
{
|
375 |
+
"epoch": 0.06,
|
376 |
+
"learning_rate": 0.0,
|
377 |
+
"loss": 1.4766,
|
378 |
+
"step": 55
|
379 |
+
},
|
380 |
+
{
|
381 |
+
"epoch": 0.06,
|
382 |
+
"learning_rate": 0.0,
|
383 |
+
"loss": 1.4638,
|
384 |
+
"step": 56
|
385 |
+
},
|
386 |
+
{
|
387 |
+
"epoch": 0.06,
|
388 |
+
"learning_rate": 0.0,
|
389 |
+
"loss": 1.4695,
|
390 |
+
"step": 57
|
391 |
+
},
|
392 |
+
{
|
393 |
+
"epoch": 0.06,
|
394 |
+
"learning_rate": 0.0,
|
395 |
+
"loss": 1.4272,
|
396 |
+
"step": 58
|
397 |
+
},
|
398 |
+
{
|
399 |
+
"epoch": 0.06,
|
400 |
+
"learning_rate": 0.0,
|
401 |
+
"loss": 1.5211,
|
402 |
+
"step": 59
|
403 |
+
},
|
404 |
+
{
|
405 |
+
"epoch": 0.06,
|
406 |
+
"learning_rate": 0.0,
|
407 |
+
"loss": 1.5044,
|
408 |
+
"step": 60
|
409 |
+
},
|
410 |
+
{
|
411 |
+
"epoch": 0.06,
|
412 |
+
"eval_loss": 1.4974411725997925,
|
413 |
+
"eval_runtime": 2.4841,
|
414 |
+
"eval_samples_per_second": 34.217,
|
415 |
+
"eval_steps_per_second": 2.013,
|
416 |
+
"step": 60
|
417 |
+
},
|
418 |
+
{
|
419 |
+
"epoch": 0.06,
|
420 |
+
"learning_rate": 0.0,
|
421 |
+
"loss": 1.5065,
|
422 |
+
"step": 61
|
423 |
+
},
|
424 |
+
{
|
425 |
+
"epoch": 0.06,
|
426 |
+
"learning_rate": 0.0,
|
427 |
+
"loss": 1.4428,
|
428 |
+
"step": 62
|
429 |
+
},
|
430 |
+
{
|
431 |
+
"epoch": 0.06,
|
432 |
+
"learning_rate": 0.0,
|
433 |
+
"loss": 1.4665,
|
434 |
+
"step": 63
|
435 |
+
},
|
436 |
+
{
|
437 |
+
"epoch": 0.06,
|
438 |
+
"learning_rate": 0.0,
|
439 |
+
"loss": 1.4986,
|
440 |
+
"step": 64
|
441 |
+
},
|
442 |
+
{
|
443 |
+
"epoch": 0.07,
|
444 |
+
"learning_rate": 0.0,
|
445 |
+
"loss": 1.4946,
|
446 |
+
"step": 65
|
447 |
+
},
|
448 |
+
{
|
449 |
+
"epoch": 0.07,
|
450 |
+
"learning_rate": 0.0,
|
451 |
+
"loss": 1.4675,
|
452 |
+
"step": 66
|
453 |
+
},
|
454 |
+
{
|
455 |
+
"epoch": 0.07,
|
456 |
+
"learning_rate": 0.0,
|
457 |
+
"loss": 1.4636,
|
458 |
+
"step": 67
|
459 |
+
},
|
460 |
+
{
|
461 |
+
"epoch": 0.07,
|
462 |
+
"learning_rate": 0.0,
|
463 |
+
"loss": 1.5105,
|
464 |
+
"step": 68
|
465 |
+
},
|
466 |
+
{
|
467 |
+
"epoch": 0.07,
|
468 |
+
"learning_rate": 0.0,
|
469 |
+
"loss": 1.4805,
|
470 |
+
"step": 69
|
471 |
+
},
|
472 |
+
{
|
473 |
+
"epoch": 0.07,
|
474 |
+
"learning_rate": 0.0,
|
475 |
+
"loss": 1.4444,
|
476 |
+
"step": 70
|
477 |
+
},
|
478 |
+
{
|
479 |
+
"epoch": 0.07,
|
480 |
+
"eval_loss": 1.497441053390503,
|
481 |
+
"eval_runtime": 2.7704,
|
482 |
+
"eval_samples_per_second": 30.682,
|
483 |
+
"eval_steps_per_second": 1.805,
|
484 |
+
"step": 70
|
485 |
+
},
|
486 |
+
{
|
487 |
+
"epoch": 0.07,
|
488 |
+
"learning_rate": 0.0,
|
489 |
+
"loss": 1.5231,
|
490 |
+
"step": 71
|
491 |
+
},
|
492 |
+
{
|
493 |
+
"epoch": 0.07,
|
494 |
+
"learning_rate": 0.0,
|
495 |
+
"loss": 1.4438,
|
496 |
+
"step": 72
|
497 |
+
},
|
498 |
+
{
|
499 |
+
"epoch": 0.07,
|
500 |
+
"learning_rate": 0.0,
|
501 |
+
"loss": 1.4733,
|
502 |
+
"step": 73
|
503 |
+
},
|
504 |
+
{
|
505 |
+
"epoch": 0.07,
|
506 |
+
"learning_rate": 0.0,
|
507 |
+
"loss": 1.4863,
|
508 |
+
"step": 74
|
509 |
+
},
|
510 |
+
{
|
511 |
+
"epoch": 0.07,
|
512 |
+
"learning_rate": 0.0,
|
513 |
+
"loss": 1.5116,
|
514 |
+
"step": 75
|
515 |
+
},
|
516 |
+
{
|
517 |
+
"epoch": 0.08,
|
518 |
+
"learning_rate": 0.0,
|
519 |
+
"loss": 1.4434,
|
520 |
+
"step": 76
|
521 |
+
},
|
522 |
+
{
|
523 |
+
"epoch": 0.08,
|
524 |
+
"learning_rate": 0.0,
|
525 |
+
"loss": 1.4413,
|
526 |
+
"step": 77
|
527 |
+
},
|
528 |
+
{
|
529 |
+
"epoch": 0.08,
|
530 |
+
"learning_rate": 0.0,
|
531 |
+
"loss": 1.4878,
|
532 |
+
"step": 78
|
533 |
+
},
|
534 |
+
{
|
535 |
+
"epoch": 0.08,
|
536 |
+
"learning_rate": 0.0,
|
537 |
+
"loss": 1.4866,
|
538 |
+
"step": 79
|
539 |
+
},
|
540 |
+
{
|
541 |
+
"epoch": 0.08,
|
542 |
+
"learning_rate": 0.0,
|
543 |
+
"loss": 1.4683,
|
544 |
+
"step": 80
|
545 |
+
},
|
546 |
+
{
|
547 |
+
"epoch": 0.08,
|
548 |
+
"eval_loss": 1.4974406957626343,
|
549 |
+
"eval_runtime": 2.4599,
|
550 |
+
"eval_samples_per_second": 34.555,
|
551 |
+
"eval_steps_per_second": 2.033,
|
552 |
+
"step": 80
|
553 |
+
},
|
554 |
+
{
|
555 |
+
"epoch": 0.08,
|
556 |
+
"learning_rate": 0.0,
|
557 |
+
"loss": 1.4787,
|
558 |
+
"step": 81
|
559 |
+
},
|
560 |
+
{
|
561 |
+
"epoch": 0.08,
|
562 |
+
"learning_rate": 0.0,
|
563 |
+
"loss": 1.4832,
|
564 |
+
"step": 82
|
565 |
+
},
|
566 |
+
{
|
567 |
+
"epoch": 0.08,
|
568 |
+
"learning_rate": 0.0,
|
569 |
+
"loss": 1.4494,
|
570 |
+
"step": 83
|
571 |
+
},
|
572 |
+
{
|
573 |
+
"epoch": 0.08,
|
574 |
+
"learning_rate": 0.0,
|
575 |
+
"loss": 1.4606,
|
576 |
+
"step": 84
|
577 |
+
},
|
578 |
+
{
|
579 |
+
"epoch": 0.09,
|
580 |
+
"learning_rate": 0.0,
|
581 |
+
"loss": 1.4981,
|
582 |
+
"step": 85
|
583 |
+
},
|
584 |
+
{
|
585 |
+
"epoch": 0.09,
|
586 |
+
"learning_rate": 0.0,
|
587 |
+
"loss": 1.5046,
|
588 |
+
"step": 86
|
589 |
+
},
|
590 |
+
{
|
591 |
+
"epoch": 0.09,
|
592 |
+
"learning_rate": 0.0,
|
593 |
+
"loss": 1.4937,
|
594 |
+
"step": 87
|
595 |
+
},
|
596 |
+
{
|
597 |
+
"epoch": 0.09,
|
598 |
+
"learning_rate": 0.0,
|
599 |
+
"loss": 1.4954,
|
600 |
+
"step": 88
|
601 |
+
},
|
602 |
+
{
|
603 |
+
"epoch": 0.09,
|
604 |
+
"learning_rate": 0.0,
|
605 |
+
"loss": 1.4731,
|
606 |
+
"step": 89
|
607 |
+
},
|
608 |
+
{
|
609 |
+
"epoch": 0.09,
|
610 |
+
"learning_rate": 0.0,
|
611 |
+
"loss": 1.5075,
|
612 |
+
"step": 90
|
613 |
+
},
|
614 |
+
{
|
615 |
+
"epoch": 0.09,
|
616 |
+
"eval_loss": 1.4974411725997925,
|
617 |
+
"eval_runtime": 2.4585,
|
618 |
+
"eval_samples_per_second": 34.575,
|
619 |
+
"eval_steps_per_second": 2.034,
|
620 |
+
"step": 90
|
621 |
+
},
|
622 |
+
{
|
623 |
+
"epoch": 0.09,
|
624 |
+
"learning_rate": 0.0,
|
625 |
+
"loss": 1.4989,
|
626 |
+
"step": 91
|
627 |
+
},
|
628 |
+
{
|
629 |
+
"epoch": 0.09,
|
630 |
+
"learning_rate": 0.0,
|
631 |
+
"loss": 1.4831,
|
632 |
+
"step": 92
|
633 |
+
},
|
634 |
+
{
|
635 |
+
"epoch": 0.09,
|
636 |
+
"learning_rate": 0.0,
|
637 |
+
"loss": 1.4624,
|
638 |
+
"step": 93
|
639 |
+
},
|
640 |
+
{
|
641 |
+
"epoch": 0.09,
|
642 |
+
"learning_rate": 0.0,
|
643 |
+
"loss": 1.4936,
|
644 |
+
"step": 94
|
645 |
+
},
|
646 |
+
{
|
647 |
+
"epoch": 0.1,
|
648 |
+
"learning_rate": 0.0,
|
649 |
+
"loss": 1.4631,
|
650 |
+
"step": 95
|
651 |
+
},
|
652 |
+
{
|
653 |
+
"epoch": 0.1,
|
654 |
+
"learning_rate": 0.0,
|
655 |
+
"loss": 1.4991,
|
656 |
+
"step": 96
|
657 |
+
},
|
658 |
+
{
|
659 |
+
"epoch": 0.1,
|
660 |
+
"learning_rate": 0.0,
|
661 |
+
"loss": 1.4646,
|
662 |
+
"step": 97
|
663 |
+
},
|
664 |
+
{
|
665 |
+
"epoch": 0.1,
|
666 |
+
"learning_rate": 0.0,
|
667 |
+
"loss": 1.4986,
|
668 |
+
"step": 98
|
669 |
+
},
|
670 |
+
{
|
671 |
+
"epoch": 0.1,
|
672 |
+
"learning_rate": 0.0,
|
673 |
+
"loss": 1.4815,
|
674 |
+
"step": 99
|
675 |
+
},
|
676 |
+
{
|
677 |
+
"epoch": 0.1,
|
678 |
+
"learning_rate": 0.0,
|
679 |
+
"loss": 1.4876,
|
680 |
+
"step": 100
|
681 |
+
},
|
682 |
+
{
|
683 |
+
"epoch": 0.1,
|
684 |
+
"eval_loss": 1.497441053390503,
|
685 |
+
"eval_runtime": 2.485,
|
686 |
+
"eval_samples_per_second": 34.205,
|
687 |
+
"eval_steps_per_second": 2.012,
|
688 |
+
"step": 100
|
689 |
+
},
|
690 |
+
{
|
691 |
+
"epoch": 0.1,
|
692 |
+
"learning_rate": 0.0,
|
693 |
+
"loss": 1.4663,
|
694 |
+
"step": 101
|
695 |
+
},
|
696 |
+
{
|
697 |
+
"epoch": 0.1,
|
698 |
+
"learning_rate": 0.0,
|
699 |
+
"loss": 1.4616,
|
700 |
+
"step": 102
|
701 |
+
},
|
702 |
+
{
|
703 |
+
"epoch": 0.1,
|
704 |
+
"learning_rate": 0.0,
|
705 |
+
"loss": 1.4779,
|
706 |
+
"step": 103
|
707 |
+
},
|
708 |
+
{
|
709 |
+
"epoch": 0.1,
|
710 |
+
"learning_rate": 0.0,
|
711 |
+
"loss": 1.5175,
|
712 |
+
"step": 104
|
713 |
+
},
|
714 |
+
{
|
715 |
+
"epoch": 0.1,
|
716 |
+
"learning_rate": 0.0,
|
717 |
+
"loss": 1.48,
|
718 |
+
"step": 105
|
719 |
+
},
|
720 |
+
{
|
721 |
+
"epoch": 0.11,
|
722 |
+
"learning_rate": 0.0,
|
723 |
+
"loss": 1.4722,
|
724 |
+
"step": 106
|
725 |
+
},
|
726 |
+
{
|
727 |
+
"epoch": 0.11,
|
728 |
+
"learning_rate": 0.0,
|
729 |
+
"loss": 1.4856,
|
730 |
+
"step": 107
|
731 |
+
},
|
732 |
+
{
|
733 |
+
"epoch": 0.11,
|
734 |
+
"learning_rate": 0.0,
|
735 |
+
"loss": 1.4342,
|
736 |
+
"step": 108
|
737 |
+
},
|
738 |
+
{
|
739 |
+
"epoch": 0.11,
|
740 |
+
"learning_rate": 0.0,
|
741 |
+
"loss": 1.4481,
|
742 |
+
"step": 109
|
743 |
+
},
|
744 |
+
{
|
745 |
+
"epoch": 0.11,
|
746 |
+
"learning_rate": 0.0,
|
747 |
+
"loss": 1.4997,
|
748 |
+
"step": 110
|
749 |
+
},
|
750 |
+
{
|
751 |
+
"epoch": 0.11,
|
752 |
+
"eval_loss": 1.497441053390503,
|
753 |
+
"eval_runtime": 2.4899,
|
754 |
+
"eval_samples_per_second": 34.137,
|
755 |
+
"eval_steps_per_second": 2.008,
|
756 |
+
"step": 110
|
757 |
+
},
|
758 |
+
{
|
759 |
+
"epoch": 0.11,
|
760 |
+
"learning_rate": 0.0,
|
761 |
+
"loss": 1.4932,
|
762 |
+
"step": 111
|
763 |
+
},
|
764 |
+
{
|
765 |
+
"epoch": 0.11,
|
766 |
+
"learning_rate": 0.0,
|
767 |
+
"loss": 1.5032,
|
768 |
+
"step": 112
|
769 |
+
},
|
770 |
+
{
|
771 |
+
"epoch": 0.11,
|
772 |
+
"learning_rate": 0.0,
|
773 |
+
"loss": 1.493,
|
774 |
+
"step": 113
|
775 |
+
},
|
776 |
+
{
|
777 |
+
"epoch": 0.11,
|
778 |
+
"learning_rate": 0.0,
|
779 |
+
"loss": 1.4659,
|
780 |
+
"step": 114
|
781 |
+
},
|
782 |
+
{
|
783 |
+
"epoch": 0.12,
|
784 |
+
"learning_rate": 0.0,
|
785 |
+
"loss": 1.4918,
|
786 |
+
"step": 115
|
787 |
+
},
|
788 |
+
{
|
789 |
+
"epoch": 0.12,
|
790 |
+
"learning_rate": 0.0,
|
791 |
+
"loss": 1.5082,
|
792 |
+
"step": 116
|
793 |
+
},
|
794 |
+
{
|
795 |
+
"epoch": 0.12,
|
796 |
+
"learning_rate": 0.0,
|
797 |
+
"loss": 1.4699,
|
798 |
+
"step": 117
|
799 |
+
},
|
800 |
+
{
|
801 |
+
"epoch": 0.12,
|
802 |
+
"learning_rate": 0.0,
|
803 |
+
"loss": 1.464,
|
804 |
+
"step": 118
|
805 |
+
},
|
806 |
+
{
|
807 |
+
"epoch": 0.12,
|
808 |
+
"learning_rate": 0.0,
|
809 |
+
"loss": 1.4729,
|
810 |
+
"step": 119
|
811 |
+
},
|
812 |
+
{
|
813 |
+
"epoch": 0.12,
|
814 |
+
"learning_rate": 0.0,
|
815 |
+
"loss": 1.4768,
|
816 |
+
"step": 120
|
817 |
+
},
|
818 |
+
{
|
819 |
+
"epoch": 0.12,
|
820 |
+
"eval_loss": 1.497441291809082,
|
821 |
+
"eval_runtime": 2.4613,
|
822 |
+
"eval_samples_per_second": 34.535,
|
823 |
+
"eval_steps_per_second": 2.031,
|
824 |
+
"step": 120
|
825 |
+
},
|
826 |
+
{
|
827 |
+
"epoch": 0.12,
|
828 |
+
"learning_rate": 0.0,
|
829 |
+
"loss": 1.473,
|
830 |
+
"step": 121
|
831 |
+
},
|
832 |
+
{
|
833 |
+
"epoch": 0.12,
|
834 |
+
"learning_rate": 0.0,
|
835 |
+
"loss": 1.4733,
|
836 |
+
"step": 122
|
837 |
+
},
|
838 |
+
{
|
839 |
+
"epoch": 0.12,
|
840 |
+
"learning_rate": 0.0,
|
841 |
+
"loss": 1.4845,
|
842 |
+
"step": 123
|
843 |
+
},
|
844 |
+
{
|
845 |
+
"epoch": 0.12,
|
846 |
+
"learning_rate": 0.0,
|
847 |
+
"loss": 1.4629,
|
848 |
+
"step": 124
|
849 |
+
},
|
850 |
+
{
|
851 |
+
"epoch": 0.12,
|
852 |
+
"learning_rate": 0.0,
|
853 |
+
"loss": 1.4659,
|
854 |
+
"step": 125
|
855 |
+
},
|
856 |
+
{
|
857 |
+
"epoch": 0.13,
|
858 |
+
"learning_rate": 0.0,
|
859 |
+
"loss": 1.4863,
|
860 |
+
"step": 126
|
861 |
+
},
|
862 |
+
{
|
863 |
+
"epoch": 0.13,
|
864 |
+
"learning_rate": 0.0,
|
865 |
+
"loss": 1.4926,
|
866 |
+
"step": 127
|
867 |
+
},
|
868 |
+
{
|
869 |
+
"epoch": 0.13,
|
870 |
+
"learning_rate": 0.0,
|
871 |
+
"loss": 1.4879,
|
872 |
+
"step": 128
|
873 |
+
},
|
874 |
+
{
|
875 |
+
"epoch": 0.13,
|
876 |
+
"learning_rate": 0.0,
|
877 |
+
"loss": 1.4636,
|
878 |
+
"step": 129
|
879 |
+
},
|
880 |
+
{
|
881 |
+
"epoch": 0.13,
|
882 |
+
"learning_rate": 0.0,
|
883 |
+
"loss": 1.501,
|
884 |
+
"step": 130
|
885 |
+
},
|
886 |
+
{
|
887 |
+
"epoch": 0.13,
|
888 |
+
"eval_loss": 1.4974408149719238,
|
889 |
+
"eval_runtime": 2.4972,
|
890 |
+
"eval_samples_per_second": 34.037,
|
891 |
+
"eval_steps_per_second": 2.002,
|
892 |
+
"step": 130
|
893 |
+
},
|
894 |
+
{
|
895 |
+
"epoch": 0.13,
|
896 |
+
"learning_rate": 0.0,
|
897 |
+
"loss": 1.4754,
|
898 |
+
"step": 131
|
899 |
+
},
|
900 |
+
{
|
901 |
+
"epoch": 0.13,
|
902 |
+
"learning_rate": 0.0,
|
903 |
+
"loss": 1.4732,
|
904 |
+
"step": 132
|
905 |
+
},
|
906 |
+
{
|
907 |
+
"epoch": 0.13,
|
908 |
+
"learning_rate": 0.0,
|
909 |
+
"loss": 1.4862,
|
910 |
+
"step": 133
|
911 |
+
},
|
912 |
+
{
|
913 |
+
"epoch": 0.13,
|
914 |
+
"learning_rate": 0.0,
|
915 |
+
"loss": 1.4766,
|
916 |
+
"step": 134
|
917 |
+
},
|
918 |
+
{
|
919 |
+
"epoch": 0.14,
|
920 |
+
"learning_rate": 0.0,
|
921 |
+
"loss": 1.4898,
|
922 |
+
"step": 135
|
923 |
+
},
|
924 |
+
{
|
925 |
+
"epoch": 0.14,
|
926 |
+
"learning_rate": 0.0,
|
927 |
+
"loss": 1.4533,
|
928 |
+
"step": 136
|
929 |
+
},
|
930 |
+
{
|
931 |
+
"epoch": 0.14,
|
932 |
+
"learning_rate": 0.0,
|
933 |
+
"loss": 1.491,
|
934 |
+
"step": 137
|
935 |
+
},
|
936 |
+
{
|
937 |
+
"epoch": 0.14,
|
938 |
+
"learning_rate": 0.0,
|
939 |
+
"loss": 1.4539,
|
940 |
+
"step": 138
|
941 |
+
},
|
942 |
+
{
|
943 |
+
"epoch": 0.14,
|
944 |
+
"learning_rate": 0.0,
|
945 |
+
"loss": 1.4875,
|
946 |
+
"step": 139
|
947 |
+
},
|
948 |
+
{
|
949 |
+
"epoch": 0.14,
|
950 |
+
"learning_rate": 0.0,
|
951 |
+
"loss": 1.5224,
|
952 |
+
"step": 140
|
953 |
+
},
|
954 |
+
{
|
955 |
+
"epoch": 0.14,
|
956 |
+
"eval_loss": 1.4974406957626343,
|
957 |
+
"eval_runtime": 2.4604,
|
958 |
+
"eval_samples_per_second": 34.547,
|
959 |
+
"eval_steps_per_second": 2.032,
|
960 |
+
"step": 140
|
961 |
+
},
|
962 |
+
{
|
963 |
+
"epoch": 0.14,
|
964 |
+
"learning_rate": 0.0,
|
965 |
+
"loss": 1.4881,
|
966 |
+
"step": 141
|
967 |
+
},
|
968 |
+
{
|
969 |
+
"epoch": 0.14,
|
970 |
+
"learning_rate": 0.0,
|
971 |
+
"loss": 1.4815,
|
972 |
+
"step": 142
|
973 |
+
},
|
974 |
+
{
|
975 |
+
"epoch": 0.14,
|
976 |
+
"learning_rate": 0.0,
|
977 |
+
"loss": 1.474,
|
978 |
+
"step": 143
|
979 |
+
},
|
980 |
+
{
|
981 |
+
"epoch": 0.14,
|
982 |
+
"learning_rate": 0.0,
|
983 |
+
"loss": 1.4913,
|
984 |
+
"step": 144
|
985 |
+
},
|
986 |
+
{
|
987 |
+
"epoch": 0.14,
|
988 |
+
"learning_rate": 0.0,
|
989 |
+
"loss": 1.4527,
|
990 |
+
"step": 145
|
991 |
+
},
|
992 |
+
{
|
993 |
+
"epoch": 0.15,
|
994 |
+
"learning_rate": 0.0,
|
995 |
+
"loss": 1.4874,
|
996 |
+
"step": 146
|
997 |
+
},
|
998 |
+
{
|
999 |
+
"epoch": 0.15,
|
1000 |
+
"learning_rate": 0.0,
|
1001 |
+
"loss": 1.4907,
|
1002 |
+
"step": 147
|
1003 |
+
},
|
1004 |
+
{
|
1005 |
+
"epoch": 0.15,
|
1006 |
+
"learning_rate": 0.0,
|
1007 |
+
"loss": 1.4855,
|
1008 |
+
"step": 148
|
1009 |
+
},
|
1010 |
+
{
|
1011 |
+
"epoch": 0.15,
|
1012 |
+
"learning_rate": 0.0,
|
1013 |
+
"loss": 1.4746,
|
1014 |
+
"step": 149
|
1015 |
+
},
|
1016 |
+
{
|
1017 |
+
"epoch": 0.15,
|
1018 |
+
"learning_rate": 0.0,
|
1019 |
+
"loss": 1.4988,
|
1020 |
+
"step": 150
|
1021 |
+
},
|
1022 |
+
{
|
1023 |
+
"epoch": 0.15,
|
1024 |
+
"eval_loss": 1.4974409341812134,
|
1025 |
+
"eval_runtime": 2.4749,
|
1026 |
+
"eval_samples_per_second": 34.345,
|
1027 |
+
"eval_steps_per_second": 2.02,
|
1028 |
+
"step": 150
|
1029 |
+
},
|
1030 |
+
{
|
1031 |
+
"epoch": 0.15,
|
1032 |
+
"learning_rate": 0.0,
|
1033 |
+
"loss": 1.5209,
|
1034 |
+
"step": 151
|
1035 |
+
},
|
1036 |
+
{
|
1037 |
+
"epoch": 0.15,
|
1038 |
+
"learning_rate": 0.0,
|
1039 |
+
"loss": 1.4406,
|
1040 |
+
"step": 152
|
1041 |
+
},
|
1042 |
+
{
|
1043 |
+
"epoch": 0.15,
|
1044 |
+
"learning_rate": 0.0,
|
1045 |
+
"loss": 1.501,
|
1046 |
+
"step": 153
|
1047 |
+
},
|
1048 |
+
{
|
1049 |
+
"epoch": 0.15,
|
1050 |
+
"learning_rate": 0.0,
|
1051 |
+
"loss": 1.4227,
|
1052 |
+
"step": 154
|
1053 |
+
},
|
1054 |
+
{
|
1055 |
+
"epoch": 0.15,
|
1056 |
+
"learning_rate": 0.0,
|
1057 |
+
"loss": 1.474,
|
1058 |
+
"step": 155
|
1059 |
+
},
|
1060 |
+
{
|
1061 |
+
"epoch": 0.16,
|
1062 |
+
"learning_rate": 0.0,
|
1063 |
+
"loss": 1.493,
|
1064 |
+
"step": 156
|
1065 |
+
},
|
1066 |
+
{
|
1067 |
+
"epoch": 0.16,
|
1068 |
+
"learning_rate": 0.0,
|
1069 |
+
"loss": 1.4953,
|
1070 |
+
"step": 157
|
1071 |
+
},
|
1072 |
+
{
|
1073 |
+
"epoch": 0.16,
|
1074 |
+
"learning_rate": 0.0,
|
1075 |
+
"loss": 1.4475,
|
1076 |
+
"step": 158
|
1077 |
+
},
|
1078 |
+
{
|
1079 |
+
"epoch": 0.16,
|
1080 |
+
"learning_rate": 0.0,
|
1081 |
+
"loss": 1.5084,
|
1082 |
+
"step": 159
|
1083 |
+
},
|
1084 |
+
{
|
1085 |
+
"epoch": 0.16,
|
1086 |
+
"learning_rate": 0.0,
|
1087 |
+
"loss": 1.4582,
|
1088 |
+
"step": 160
|
1089 |
+
},
|
1090 |
+
{
|
1091 |
+
"epoch": 0.16,
|
1092 |
+
"eval_loss": 1.497441053390503,
|
1093 |
+
"eval_runtime": 2.4861,
|
1094 |
+
"eval_samples_per_second": 34.19,
|
1095 |
+
"eval_steps_per_second": 2.011,
|
1096 |
+
"step": 160
|
1097 |
+
},
|
1098 |
+
{
|
1099 |
+
"epoch": 0.16,
|
1100 |
+
"learning_rate": 0.0,
|
1101 |
+
"loss": 1.4891,
|
1102 |
+
"step": 161
|
1103 |
+
},
|
1104 |
+
{
|
1105 |
+
"epoch": 0.16,
|
1106 |
+
"learning_rate": 0.0,
|
1107 |
+
"loss": 1.5041,
|
1108 |
+
"step": 162
|
1109 |
+
},
|
1110 |
+
{
|
1111 |
+
"epoch": 0.16,
|
1112 |
+
"learning_rate": 0.0,
|
1113 |
+
"loss": 1.4514,
|
1114 |
+
"step": 163
|
1115 |
+
},
|
1116 |
+
{
|
1117 |
+
"epoch": 0.16,
|
1118 |
+
"learning_rate": 0.0,
|
1119 |
+
"loss": 1.4876,
|
1120 |
+
"step": 164
|
1121 |
+
},
|
1122 |
+
{
|
1123 |
+
"epoch": 0.17,
|
1124 |
+
"learning_rate": 0.0,
|
1125 |
+
"loss": 1.4778,
|
1126 |
+
"step": 165
|
1127 |
+
},
|
1128 |
+
{
|
1129 |
+
"epoch": 0.17,
|
1130 |
+
"learning_rate": 0.0,
|
1131 |
+
"loss": 1.4555,
|
1132 |
+
"step": 166
|
1133 |
+
},
|
1134 |
+
{
|
1135 |
+
"epoch": 0.17,
|
1136 |
+
"learning_rate": 5e-06,
|
1137 |
+
"loss": 1.5126,
|
1138 |
+
"step": 167
|
1139 |
+
},
|
1140 |
+
{
|
1141 |
+
"epoch": 0.17,
|
1142 |
+
"learning_rate": 1e-05,
|
1143 |
+
"loss": 1.4723,
|
1144 |
+
"step": 168
|
1145 |
+
},
|
1146 |
+
{
|
1147 |
+
"epoch": 0.17,
|
1148 |
+
"learning_rate": 1.5e-05,
|
1149 |
+
"loss": 1.4596,
|
1150 |
+
"step": 169
|
1151 |
+
},
|
1152 |
+
{
|
1153 |
+
"epoch": 0.17,
|
1154 |
+
"learning_rate": 2e-05,
|
1155 |
+
"loss": 1.5069,
|
1156 |
+
"step": 170
|
1157 |
+
},
|
1158 |
+
{
|
1159 |
+
"epoch": 0.17,
|
1160 |
+
"eval_loss": 1.497441291809082,
|
1161 |
+
"eval_runtime": 2.4879,
|
1162 |
+
"eval_samples_per_second": 34.165,
|
1163 |
+
"eval_steps_per_second": 2.01,
|
1164 |
+
"step": 170
|
1165 |
+
},
|
1166 |
+
{
|
1167 |
+
"epoch": 0.17,
|
1168 |
+
"learning_rate": 2.5e-05,
|
1169 |
+
"loss": 1.4832,
|
1170 |
+
"step": 171
|
1171 |
+
},
|
1172 |
+
{
|
1173 |
+
"epoch": 0.17,
|
1174 |
+
"learning_rate": 3e-05,
|
1175 |
+
"loss": 1.4793,
|
1176 |
+
"step": 172
|
1177 |
+
},
|
1178 |
+
{
|
1179 |
+
"epoch": 0.17,
|
1180 |
+
"learning_rate": 3.5e-05,
|
1181 |
+
"loss": 1.5123,
|
1182 |
+
"step": 173
|
1183 |
+
},
|
1184 |
+
{
|
1185 |
+
"epoch": 0.17,
|
1186 |
+
"learning_rate": 4e-05,
|
1187 |
+
"loss": 1.4773,
|
1188 |
+
"step": 174
|
1189 |
+
},
|
1190 |
+
{
|
1191 |
+
"epoch": 0.17,
|
1192 |
+
"learning_rate": 4.5e-05,
|
1193 |
+
"loss": 1.4608,
|
1194 |
+
"step": 175
|
1195 |
+
},
|
1196 |
+
{
|
1197 |
+
"epoch": 0.18,
|
1198 |
+
"learning_rate": 5e-05,
|
1199 |
+
"loss": 1.4544,
|
1200 |
+
"step": 176
|
1201 |
+
},
|
1202 |
+
{
|
1203 |
+
"epoch": 0.18,
|
1204 |
+
"learning_rate": 4.999987412513878e-05,
|
1205 |
+
"loss": 1.4933,
|
1206 |
+
"step": 177
|
1207 |
+
},
|
1208 |
+
{
|
1209 |
+
"epoch": 0.18,
|
1210 |
+
"learning_rate": 4.999949650182266e-05,
|
1211 |
+
"loss": 1.4753,
|
1212 |
+
"step": 178
|
1213 |
+
},
|
1214 |
+
{
|
1215 |
+
"epoch": 0.18,
|
1216 |
+
"learning_rate": 4.999886713385432e-05,
|
1217 |
+
"loss": 1.4745,
|
1218 |
+
"step": 179
|
1219 |
+
},
|
1220 |
+
{
|
1221 |
+
"epoch": 0.18,
|
1222 |
+
"learning_rate": 4.9997986027571485e-05,
|
1223 |
+
"loss": 1.5125,
|
1224 |
+
"step": 180
|
1225 |
+
},
|
1226 |
+
{
|
1227 |
+
"epoch": 0.18,
|
1228 |
+
"eval_loss": 1.4974409341812134,
|
1229 |
+
"eval_runtime": 2.4894,
|
1230 |
+
"eval_samples_per_second": 34.145,
|
1231 |
+
"eval_steps_per_second": 2.009,
|
1232 |
+
"step": 180
|
1233 |
+
},
|
1234 |
+
{
|
1235 |
+
"epoch": 0.18,
|
1236 |
+
"learning_rate": 4.9996853191846885e-05,
|
1237 |
+
"loss": 1.4941,
|
1238 |
+
"step": 181
|
1239 |
+
},
|
1240 |
+
{
|
1241 |
+
"epoch": 0.18,
|
1242 |
+
"learning_rate": 4.999546863808815e-05,
|
1243 |
+
"loss": 1.4606,
|
1244 |
+
"step": 182
|
1245 |
+
},
|
1246 |
+
{
|
1247 |
+
"epoch": 0.18,
|
1248 |
+
"learning_rate": 4.9993832380237735e-05,
|
1249 |
+
"loss": 1.4429,
|
1250 |
+
"step": 183
|
1251 |
+
},
|
1252 |
+
{
|
1253 |
+
"epoch": 0.18,
|
1254 |
+
"learning_rate": 4.9991944434772734e-05,
|
1255 |
+
"loss": 1.4993,
|
1256 |
+
"step": 184
|
1257 |
+
},
|
1258 |
+
{
|
1259 |
+
"epoch": 0.18,
|
1260 |
+
"learning_rate": 4.9989804820704735e-05,
|
1261 |
+
"loss": 1.4545,
|
1262 |
+
"step": 185
|
1263 |
+
},
|
1264 |
+
{
|
1265 |
+
"epoch": 0.19,
|
1266 |
+
"learning_rate": 4.9987413559579636e-05,
|
1267 |
+
"loss": 1.5229,
|
1268 |
+
"step": 186
|
1269 |
+
},
|
1270 |
+
{
|
1271 |
+
"epoch": 0.19,
|
1272 |
+
"learning_rate": 4.99847706754774e-05,
|
1273 |
+
"loss": 1.4328,
|
1274 |
+
"step": 187
|
1275 |
+
},
|
1276 |
+
{
|
1277 |
+
"epoch": 0.19,
|
1278 |
+
"learning_rate": 4.9981876195011844e-05,
|
1279 |
+
"loss": 1.5037,
|
1280 |
+
"step": 188
|
1281 |
+
},
|
1282 |
+
{
|
1283 |
+
"epoch": 0.19,
|
1284 |
+
"learning_rate": 4.9978730147330355e-05,
|
1285 |
+
"loss": 1.4765,
|
1286 |
+
"step": 189
|
1287 |
+
},
|
1288 |
+
{
|
1289 |
+
"epoch": 0.19,
|
1290 |
+
"learning_rate": 4.99753325641136e-05,
|
1291 |
+
"loss": 1.4936,
|
1292 |
+
"step": 190
|
1293 |
+
},
|
1294 |
+
{
|
1295 |
+
"epoch": 0.19,
|
1296 |
+
"eval_loss": 1.497441053390503,
|
1297 |
+
"eval_runtime": 2.4637,
|
1298 |
+
"eval_samples_per_second": 34.5,
|
1299 |
+
"eval_steps_per_second": 2.029,
|
1300 |
+
"step": 190
|
1301 |
+
},
|
1302 |
+
{
|
1303 |
+
"epoch": 0.19,
|
1304 |
+
"learning_rate": 4.99716834795752e-05,
|
1305 |
+
"loss": 1.5076,
|
1306 |
+
"step": 191
|
1307 |
+
},
|
1308 |
+
{
|
1309 |
+
"epoch": 0.19,
|
1310 |
+
"learning_rate": 4.996778293046141e-05,
|
1311 |
+
"loss": 1.441,
|
1312 |
+
"step": 192
|
1313 |
+
},
|
1314 |
+
{
|
1315 |
+
"epoch": 0.19,
|
1316 |
+
"learning_rate": 4.996363095605069e-05,
|
1317 |
+
"loss": 1.4712,
|
1318 |
+
"step": 193
|
1319 |
+
},
|
1320 |
+
{
|
1321 |
+
"epoch": 0.19,
|
1322 |
+
"learning_rate": 4.995922759815339e-05,
|
1323 |
+
"loss": 1.4352,
|
1324 |
+
"step": 194
|
1325 |
+
},
|
1326 |
+
{
|
1327 |
+
"epoch": 0.2,
|
1328 |
+
"learning_rate": 4.9954572901111286e-05,
|
1329 |
+
"loss": 1.5377,
|
1330 |
+
"step": 195
|
1331 |
+
},
|
1332 |
+
{
|
1333 |
+
"epoch": 0.2,
|
1334 |
+
"learning_rate": 4.994966691179711e-05,
|
1335 |
+
"loss": 1.4753,
|
1336 |
+
"step": 196
|
1337 |
+
},
|
1338 |
+
{
|
1339 |
+
"epoch": 0.2,
|
1340 |
+
"learning_rate": 4.994450967961413e-05,
|
1341 |
+
"loss": 1.474,
|
1342 |
+
"step": 197
|
1343 |
+
},
|
1344 |
+
{
|
1345 |
+
"epoch": 0.2,
|
1346 |
+
"learning_rate": 4.993910125649561e-05,
|
1347 |
+
"loss": 1.4867,
|
1348 |
+
"step": 198
|
1349 |
+
},
|
1350 |
+
{
|
1351 |
+
"epoch": 0.2,
|
1352 |
+
"learning_rate": 4.993344169690431e-05,
|
1353 |
+
"loss": 1.4881,
|
1354 |
+
"step": 199
|
1355 |
+
},
|
1356 |
+
{
|
1357 |
+
"epoch": 0.2,
|
1358 |
+
"learning_rate": 4.992753105783194e-05,
|
1359 |
+
"loss": 1.4942,
|
1360 |
+
"step": 200
|
1361 |
+
},
|
1362 |
+
{
|
1363 |
+
"epoch": 0.2,
|
1364 |
+
"eval_loss": 1.497441053390503,
|
1365 |
+
"eval_runtime": 2.4763,
|
1366 |
+
"eval_samples_per_second": 34.325,
|
1367 |
+
"eval_steps_per_second": 2.019,
|
1368 |
+
"step": 200
|
1369 |
+
},
|
1370 |
+
{
|
1371 |
+
"epoch": 0.2,
|
1372 |
+
"learning_rate": 4.992136939879856e-05,
|
1373 |
+
"loss": 1.4813,
|
1374 |
+
"step": 201
|
1375 |
+
},
|
1376 |
+
{
|
1377 |
+
"epoch": 0.2,
|
1378 |
+
"learning_rate": 4.991495678185202e-05,
|
1379 |
+
"loss": 1.4725,
|
1380 |
+
"step": 202
|
1381 |
+
},
|
1382 |
+
{
|
1383 |
+
"epoch": 0.2,
|
1384 |
+
"learning_rate": 4.9908293271567286e-05,
|
1385 |
+
"loss": 1.4791,
|
1386 |
+
"step": 203
|
1387 |
+
},
|
1388 |
+
{
|
1389 |
+
"epoch": 0.2,
|
1390 |
+
"learning_rate": 4.990137893504585e-05,
|
1391 |
+
"loss": 1.4917,
|
1392 |
+
"step": 204
|
1393 |
+
},
|
1394 |
+
{
|
1395 |
+
"epoch": 0.2,
|
1396 |
+
"learning_rate": 4.989421384191499e-05,
|
1397 |
+
"loss": 1.5229,
|
1398 |
+
"step": 205
|
1399 |
+
},
|
1400 |
+
{
|
1401 |
+
"epoch": 0.21,
|
1402 |
+
"learning_rate": 4.988679806432712e-05,
|
1403 |
+
"loss": 1.4668,
|
1404 |
+
"step": 206
|
1405 |
+
},
|
1406 |
+
{
|
1407 |
+
"epoch": 0.21,
|
1408 |
+
"learning_rate": 4.987913167695904e-05,
|
1409 |
+
"loss": 1.4752,
|
1410 |
+
"step": 207
|
1411 |
+
},
|
1412 |
+
{
|
1413 |
+
"epoch": 0.21,
|
1414 |
+
"learning_rate": 4.9871214757011176e-05,
|
1415 |
+
"loss": 1.505,
|
1416 |
+
"step": 208
|
1417 |
+
},
|
1418 |
+
{
|
1419 |
+
"epoch": 0.21,
|
1420 |
+
"learning_rate": 4.9863047384206835e-05,
|
1421 |
+
"loss": 1.4523,
|
1422 |
+
"step": 209
|
1423 |
+
},
|
1424 |
+
{
|
1425 |
+
"epoch": 0.21,
|
1426 |
+
"learning_rate": 4.985462964079137e-05,
|
1427 |
+
"loss": 1.4324,
|
1428 |
+
"step": 210
|
1429 |
+
},
|
1430 |
+
{
|
1431 |
+
"epoch": 0.21,
|
1432 |
+
"eval_loss": 1.4974406957626343,
|
1433 |
+
"eval_runtime": 2.4644,
|
1434 |
+
"eval_samples_per_second": 34.492,
|
1435 |
+
"eval_steps_per_second": 2.029,
|
1436 |
+
"step": 210
|
1437 |
+
},
|
1438 |
+
{
|
1439 |
+
"epoch": 0.21,
|
1440 |
+
"learning_rate": 4.984596161153136e-05,
|
1441 |
+
"loss": 1.4717,
|
1442 |
+
"step": 211
|
1443 |
+
},
|
1444 |
+
{
|
1445 |
+
"epoch": 0.21,
|
1446 |
+
"learning_rate": 4.9837043383713753e-05,
|
1447 |
+
"loss": 1.493,
|
1448 |
+
"step": 212
|
1449 |
+
},
|
1450 |
+
{
|
1451 |
+
"epoch": 0.21,
|
1452 |
+
"learning_rate": 4.982787504714503e-05,
|
1453 |
+
"loss": 1.4947,
|
1454 |
+
"step": 213
|
1455 |
+
},
|
1456 |
+
{
|
1457 |
+
"epoch": 0.21,
|
1458 |
+
"learning_rate": 4.981845669415022e-05,
|
1459 |
+
"loss": 1.5107,
|
1460 |
+
"step": 214
|
1461 |
+
},
|
1462 |
+
{
|
1463 |
+
"epoch": 0.21,
|
1464 |
+
"learning_rate": 4.980878841957203e-05,
|
1465 |
+
"loss": 1.4815,
|
1466 |
+
"step": 215
|
1467 |
+
},
|
1468 |
+
{
|
1469 |
+
"epoch": 0.22,
|
1470 |
+
"learning_rate": 4.9798870320769886e-05,
|
1471 |
+
"loss": 1.4603,
|
1472 |
+
"step": 216
|
1473 |
+
},
|
1474 |
+
{
|
1475 |
+
"epoch": 0.22,
|
1476 |
+
"learning_rate": 4.978870249761893e-05,
|
1477 |
+
"loss": 1.4934,
|
1478 |
+
"step": 217
|
1479 |
+
},
|
1480 |
+
{
|
1481 |
+
"epoch": 0.22,
|
1482 |
+
"learning_rate": 4.977828505250903e-05,
|
1483 |
+
"loss": 1.4776,
|
1484 |
+
"step": 218
|
1485 |
+
},
|
1486 |
+
{
|
1487 |
+
"epoch": 0.22,
|
1488 |
+
"learning_rate": 4.9767618090343745e-05,
|
1489 |
+
"loss": 1.4958,
|
1490 |
+
"step": 219
|
1491 |
+
},
|
1492 |
+
{
|
1493 |
+
"epoch": 0.22,
|
1494 |
+
"learning_rate": 4.975670171853926e-05,
|
1495 |
+
"loss": 1.5113,
|
1496 |
+
"step": 220
|
1497 |
+
},
|
1498 |
+
{
|
1499 |
+
"epoch": 0.22,
|
1500 |
+
"eval_loss": 1.4974405765533447,
|
1501 |
+
"eval_runtime": 2.464,
|
1502 |
+
"eval_samples_per_second": 34.497,
|
1503 |
+
"eval_steps_per_second": 2.029,
|
1504 |
+
"step": 220
|
1505 |
+
},
|
1506 |
+
{
|
1507 |
+
"epoch": 0.22,
|
1508 |
+
"learning_rate": 4.9745536047023324e-05,
|
1509 |
+
"loss": 1.4685,
|
1510 |
+
"step": 221
|
1511 |
+
},
|
1512 |
+
{
|
1513 |
+
"epoch": 0.22,
|
1514 |
+
"learning_rate": 4.973412118823412e-05,
|
1515 |
+
"loss": 1.4465,
|
1516 |
+
"step": 222
|
1517 |
+
},
|
1518 |
+
{
|
1519 |
+
"epoch": 0.22,
|
1520 |
+
"learning_rate": 4.972245725711914e-05,
|
1521 |
+
"loss": 1.4354,
|
1522 |
+
"step": 223
|
1523 |
+
},
|
1524 |
+
{
|
1525 |
+
"epoch": 0.22,
|
1526 |
+
"learning_rate": 4.971054437113406e-05,
|
1527 |
+
"loss": 1.4926,
|
1528 |
+
"step": 224
|
1529 |
+
},
|
1530 |
+
{
|
1531 |
+
"epoch": 0.23,
|
1532 |
+
"learning_rate": 4.969838265024151e-05,
|
1533 |
+
"loss": 1.4707,
|
1534 |
+
"step": 225
|
1535 |
+
},
|
1536 |
+
{
|
1537 |
+
"epoch": 0.23,
|
1538 |
+
"learning_rate": 4.968597221690986e-05,
|
1539 |
+
"loss": 1.5009,
|
1540 |
+
"step": 226
|
1541 |
+
},
|
1542 |
+
{
|
1543 |
+
"epoch": 0.23,
|
1544 |
+
"learning_rate": 4.967331319611206e-05,
|
1545 |
+
"loss": 1.4871,
|
1546 |
+
"step": 227
|
1547 |
+
},
|
1548 |
+
{
|
1549 |
+
"epoch": 0.23,
|
1550 |
+
"learning_rate": 4.96604057153243e-05,
|
1551 |
+
"loss": 1.5048,
|
1552 |
+
"step": 228
|
1553 |
+
},
|
1554 |
+
{
|
1555 |
+
"epoch": 0.23,
|
1556 |
+
"learning_rate": 4.964724990452476e-05,
|
1557 |
+
"loss": 1.4602,
|
1558 |
+
"step": 229
|
1559 |
+
},
|
1560 |
+
{
|
1561 |
+
"epoch": 0.23,
|
1562 |
+
"learning_rate": 4.963384589619233e-05,
|
1563 |
+
"loss": 1.4779,
|
1564 |
+
"step": 230
|
1565 |
+
},
|
1566 |
+
{
|
1567 |
+
"epoch": 0.23,
|
1568 |
+
"eval_loss": 1.4974411725997925,
|
1569 |
+
"eval_runtime": 2.491,
|
1570 |
+
"eval_samples_per_second": 34.123,
|
1571 |
+
"eval_steps_per_second": 2.007,
|
1572 |
+
"step": 230
|
1573 |
+
},
|
1574 |
+
{
|
1575 |
+
"epoch": 0.23,
|
1576 |
+
"learning_rate": 4.962019382530521e-05,
|
1577 |
+
"loss": 1.501,
|
1578 |
+
"step": 231
|
1579 |
+
},
|
1580 |
+
{
|
1581 |
+
"epoch": 0.23,
|
1582 |
+
"learning_rate": 4.9606293829339595e-05,
|
1583 |
+
"loss": 1.4836,
|
1584 |
+
"step": 232
|
1585 |
+
},
|
1586 |
+
{
|
1587 |
+
"epoch": 0.23,
|
1588 |
+
"learning_rate": 4.959214604826831e-05,
|
1589 |
+
"loss": 1.5055,
|
1590 |
+
"step": 233
|
1591 |
+
},
|
1592 |
+
{
|
1593 |
+
"epoch": 0.23,
|
1594 |
+
"learning_rate": 4.957775062455933e-05,
|
1595 |
+
"loss": 1.4509,
|
1596 |
+
"step": 234
|
1597 |
+
},
|
1598 |
+
{
|
1599 |
+
"epoch": 0.23,
|
1600 |
+
"learning_rate": 4.9563107703174436e-05,
|
1601 |
+
"loss": 1.4911,
|
1602 |
+
"step": 235
|
1603 |
+
},
|
1604 |
+
{
|
1605 |
+
"epoch": 0.24,
|
1606 |
+
"learning_rate": 4.9548217431567665e-05,
|
1607 |
+
"loss": 1.4737,
|
1608 |
+
"step": 236
|
1609 |
+
},
|
1610 |
+
{
|
1611 |
+
"epoch": 0.24,
|
1612 |
+
"learning_rate": 4.95330799596839e-05,
|
1613 |
+
"loss": 1.4336,
|
1614 |
+
"step": 237
|
1615 |
+
},
|
1616 |
+
{
|
1617 |
+
"epoch": 0.24,
|
1618 |
+
"learning_rate": 4.951769543995731e-05,
|
1619 |
+
"loss": 1.5117,
|
1620 |
+
"step": 238
|
1621 |
+
},
|
1622 |
+
{
|
1623 |
+
"epoch": 0.24,
|
1624 |
+
"learning_rate": 4.9502064027309836e-05,
|
1625 |
+
"loss": 1.4661,
|
1626 |
+
"step": 239
|
1627 |
+
},
|
1628 |
+
{
|
1629 |
+
"epoch": 0.24,
|
1630 |
+
"learning_rate": 4.948618587914963e-05,
|
1631 |
+
"loss": 1.4962,
|
1632 |
+
"step": 240
|
1633 |
+
},
|
1634 |
+
{
|
1635 |
+
"epoch": 0.24,
|
1636 |
+
"eval_loss": 1.497441053390503,
|
1637 |
+
"eval_runtime": 2.455,
|
1638 |
+
"eval_samples_per_second": 34.623,
|
1639 |
+
"eval_steps_per_second": 2.037,
|
1640 |
+
"step": 240
|
1641 |
+
},
|
1642 |
+
{
|
1643 |
+
"epoch": 0.24,
|
1644 |
+
"learning_rate": 4.947006115536947e-05,
|
1645 |
+
"loss": 1.4783,
|
1646 |
+
"step": 241
|
1647 |
+
},
|
1648 |
+
{
|
1649 |
+
"epoch": 0.24,
|
1650 |
+
"learning_rate": 4.9453690018345144e-05,
|
1651 |
+
"loss": 1.5097,
|
1652 |
+
"step": 242
|
1653 |
+
},
|
1654 |
+
{
|
1655 |
+
"epoch": 0.24,
|
1656 |
+
"learning_rate": 4.9437072632933814e-05,
|
1657 |
+
"loss": 1.4832,
|
1658 |
+
"step": 243
|
1659 |
+
},
|
1660 |
+
{
|
1661 |
+
"epoch": 0.24,
|
1662 |
+
"learning_rate": 4.942020916647238e-05,
|
1663 |
+
"loss": 1.4817,
|
1664 |
+
"step": 244
|
1665 |
+
},
|
1666 |
+
{
|
1667 |
+
"epoch": 0.24,
|
1668 |
+
"learning_rate": 4.9403099788775754e-05,
|
1669 |
+
"loss": 1.5156,
|
1670 |
+
"step": 245
|
1671 |
+
},
|
1672 |
+
{
|
1673 |
+
"epoch": 0.25,
|
1674 |
+
"learning_rate": 4.938574467213518e-05,
|
1675 |
+
"loss": 1.4654,
|
1676 |
+
"step": 246
|
1677 |
+
},
|
1678 |
+
{
|
1679 |
+
"epoch": 0.25,
|
1680 |
+
"learning_rate": 4.936814399131648e-05,
|
1681 |
+
"loss": 1.4391,
|
1682 |
+
"step": 247
|
1683 |
+
},
|
1684 |
+
{
|
1685 |
+
"epoch": 0.25,
|
1686 |
+
"learning_rate": 4.935029792355834e-05,
|
1687 |
+
"loss": 1.4874,
|
1688 |
+
"step": 248
|
1689 |
+
},
|
1690 |
+
{
|
1691 |
+
"epoch": 0.25,
|
1692 |
+
"learning_rate": 4.933220664857044e-05,
|
1693 |
+
"loss": 1.4714,
|
1694 |
+
"step": 249
|
1695 |
+
},
|
1696 |
+
{
|
1697 |
+
"epoch": 0.25,
|
1698 |
+
"learning_rate": 4.931387034853173e-05,
|
1699 |
+
"loss": 1.4763,
|
1700 |
+
"step": 250
|
1701 |
+
},
|
1702 |
+
{
|
1703 |
+
"epoch": 0.25,
|
1704 |
+
"eval_loss": 1.4974409341812134,
|
1705 |
+
"eval_runtime": 2.4634,
|
1706 |
+
"eval_samples_per_second": 34.505,
|
1707 |
+
"eval_steps_per_second": 2.03,
|
1708 |
+
"step": 250
|
1709 |
+
},
|
1710 |
+
{
|
1711 |
+
"epoch": 0.25,
|
1712 |
+
"learning_rate": 4.929528920808854e-05,
|
1713 |
+
"loss": 1.4579,
|
1714 |
+
"step": 251
|
1715 |
+
},
|
1716 |
+
{
|
1717 |
+
"epoch": 0.25,
|
1718 |
+
"learning_rate": 4.9276463414352757e-05,
|
1719 |
+
"loss": 1.4982,
|
1720 |
+
"step": 252
|
1721 |
+
},
|
1722 |
+
{
|
1723 |
+
"epoch": 0.25,
|
1724 |
+
"learning_rate": 4.925739315689991e-05,
|
1725 |
+
"loss": 1.4569,
|
1726 |
+
"step": 253
|
1727 |
+
},
|
1728 |
+
{
|
1729 |
+
"epoch": 0.25,
|
1730 |
+
"learning_rate": 4.923807862776728e-05,
|
1731 |
+
"loss": 1.481,
|
1732 |
+
"step": 254
|
1733 |
+
},
|
1734 |
+
{
|
1735 |
+
"epoch": 0.26,
|
1736 |
+
"learning_rate": 4.921852002145196e-05,
|
1737 |
+
"loss": 1.4755,
|
1738 |
+
"step": 255
|
1739 |
+
},
|
1740 |
+
{
|
1741 |
+
"epoch": 0.26,
|
1742 |
+
"learning_rate": 4.919871753490891e-05,
|
1743 |
+
"loss": 1.4745,
|
1744 |
+
"step": 256
|
1745 |
+
},
|
1746 |
+
{
|
1747 |
+
"epoch": 0.26,
|
1748 |
+
"learning_rate": 4.917867136754893e-05,
|
1749 |
+
"loss": 1.4845,
|
1750 |
+
"step": 257
|
1751 |
+
},
|
1752 |
+
{
|
1753 |
+
"epoch": 0.26,
|
1754 |
+
"learning_rate": 4.915838172123671e-05,
|
1755 |
+
"loss": 1.4617,
|
1756 |
+
"step": 258
|
1757 |
+
},
|
1758 |
+
{
|
1759 |
+
"epoch": 0.26,
|
1760 |
+
"learning_rate": 4.913784880028878e-05,
|
1761 |
+
"loss": 1.4859,
|
1762 |
+
"step": 259
|
1763 |
+
},
|
1764 |
+
{
|
1765 |
+
"epoch": 0.26,
|
1766 |
+
"learning_rate": 4.91170728114714e-05,
|
1767 |
+
"loss": 1.4992,
|
1768 |
+
"step": 260
|
1769 |
+
},
|
1770 |
+
{
|
1771 |
+
"epoch": 0.26,
|
1772 |
+
"eval_loss": 1.4974408149719238,
|
1773 |
+
"eval_runtime": 2.4941,
|
1774 |
+
"eval_samples_per_second": 34.081,
|
1775 |
+
"eval_steps_per_second": 2.005,
|
1776 |
+
"step": 260
|
1777 |
+
},
|
1778 |
+
{
|
1779 |
+
"epoch": 0.26,
|
1780 |
+
"learning_rate": 4.909605396399856e-05,
|
1781 |
+
"loss": 1.4671,
|
1782 |
+
"step": 261
|
1783 |
+
},
|
1784 |
+
{
|
1785 |
+
"epoch": 0.26,
|
1786 |
+
"learning_rate": 4.9074792469529815e-05,
|
1787 |
+
"loss": 1.528,
|
1788 |
+
"step": 262
|
1789 |
+
},
|
1790 |
+
{
|
1791 |
+
"epoch": 0.26,
|
1792 |
+
"learning_rate": 4.9053288542168185e-05,
|
1793 |
+
"loss": 1.4412,
|
1794 |
+
"step": 263
|
1795 |
+
},
|
1796 |
+
{
|
1797 |
+
"epoch": 0.26,
|
1798 |
+
"learning_rate": 4.9031542398457974e-05,
|
1799 |
+
"loss": 1.4627,
|
1800 |
+
"step": 264
|
1801 |
+
},
|
1802 |
+
{
|
1803 |
+
"epoch": 0.27,
|
1804 |
+
"learning_rate": 4.9009554257382616e-05,
|
1805 |
+
"loss": 1.4693,
|
1806 |
+
"step": 265
|
1807 |
+
},
|
1808 |
+
{
|
1809 |
+
"epoch": 0.27,
|
1810 |
+
"learning_rate": 4.898732434036244e-05,
|
1811 |
+
"loss": 1.511,
|
1812 |
+
"step": 266
|
1813 |
+
},
|
1814 |
+
{
|
1815 |
+
"epoch": 0.27,
|
1816 |
+
"learning_rate": 4.896485287125246e-05,
|
1817 |
+
"loss": 1.4986,
|
1818 |
+
"step": 267
|
1819 |
+
},
|
1820 |
+
{
|
1821 |
+
"epoch": 0.27,
|
1822 |
+
"learning_rate": 4.8942140076340135e-05,
|
1823 |
+
"loss": 1.4677,
|
1824 |
+
"step": 268
|
1825 |
+
},
|
1826 |
+
{
|
1827 |
+
"epoch": 0.27,
|
1828 |
+
"learning_rate": 4.8919186184343046e-05,
|
1829 |
+
"loss": 1.4725,
|
1830 |
+
"step": 269
|
1831 |
+
},
|
1832 |
+
{
|
1833 |
+
"epoch": 0.27,
|
1834 |
+
"learning_rate": 4.889599142640663e-05,
|
1835 |
+
"loss": 1.5047,
|
1836 |
+
"step": 270
|
1837 |
+
},
|
1838 |
+
{
|
1839 |
+
"epoch": 0.27,
|
1840 |
+
"eval_loss": 1.4974409341812134,
|
1841 |
+
"eval_runtime": 2.4593,
|
1842 |
+
"eval_samples_per_second": 34.563,
|
1843 |
+
"eval_steps_per_second": 2.033,
|
1844 |
+
"step": 270
|
1845 |
+
},
|
1846 |
+
{
|
1847 |
+
"epoch": 0.27,
|
1848 |
+
"learning_rate": 4.887255603610185e-05,
|
1849 |
+
"loss": 1.4393,
|
1850 |
+
"step": 271
|
1851 |
+
},
|
1852 |
+
{
|
1853 |
+
"epoch": 0.27,
|
1854 |
+
"learning_rate": 4.8848880249422815e-05,
|
1855 |
+
"loss": 1.5039,
|
1856 |
+
"step": 272
|
1857 |
+
},
|
1858 |
+
{
|
1859 |
+
"epoch": 0.27,
|
1860 |
+
"learning_rate": 4.8824964304784446e-05,
|
1861 |
+
"loss": 1.5001,
|
1862 |
+
"step": 273
|
1863 |
+
},
|
1864 |
+
{
|
1865 |
+
"epoch": 0.27,
|
1866 |
+
"learning_rate": 4.880080844302004e-05,
|
1867 |
+
"loss": 1.473,
|
1868 |
+
"step": 274
|
1869 |
+
},
|
1870 |
+
{
|
1871 |
+
"epoch": 0.28,
|
1872 |
+
"learning_rate": 4.877641290737884e-05,
|
1873 |
+
"loss": 1.4691,
|
1874 |
+
"step": 275
|
1875 |
+
},
|
1876 |
+
{
|
1877 |
+
"epoch": 0.28,
|
1878 |
+
"learning_rate": 4.8751777943523634e-05,
|
1879 |
+
"loss": 1.4921,
|
1880 |
+
"step": 276
|
1881 |
+
},
|
1882 |
+
{
|
1883 |
+
"epoch": 0.28,
|
1884 |
+
"learning_rate": 4.8726903799528234e-05,
|
1885 |
+
"loss": 1.4846,
|
1886 |
+
"step": 277
|
1887 |
+
},
|
1888 |
+
{
|
1889 |
+
"epoch": 0.28,
|
1890 |
+
"learning_rate": 4.870179072587499e-05,
|
1891 |
+
"loss": 1.4709,
|
1892 |
+
"step": 278
|
1893 |
+
},
|
1894 |
+
{
|
1895 |
+
"epoch": 0.28,
|
1896 |
+
"learning_rate": 4.8676438975452274e-05,
|
1897 |
+
"loss": 1.4837,
|
1898 |
+
"step": 279
|
1899 |
+
},
|
1900 |
+
{
|
1901 |
+
"epoch": 0.28,
|
1902 |
+
"learning_rate": 4.865084880355193e-05,
|
1903 |
+
"loss": 1.4911,
|
1904 |
+
"step": 280
|
1905 |
+
},
|
1906 |
+
{
|
1907 |
+
"epoch": 0.28,
|
1908 |
+
"eval_loss": 1.4974409341812134,
|
1909 |
+
"eval_runtime": 2.4577,
|
1910 |
+
"eval_samples_per_second": 34.585,
|
1911 |
+
"eval_steps_per_second": 2.034,
|
1912 |
+
"step": 280
|
1913 |
+
},
|
1914 |
+
{
|
1915 |
+
"epoch": 0.28,
|
1916 |
+
"learning_rate": 4.862502046786671e-05,
|
1917 |
+
"loss": 1.4743,
|
1918 |
+
"step": 281
|
1919 |
+
},
|
1920 |
+
{
|
1921 |
+
"epoch": 0.28,
|
1922 |
+
"learning_rate": 4.859895422848767e-05,
|
1923 |
+
"loss": 1.4526,
|
1924 |
+
"step": 282
|
1925 |
+
},
|
1926 |
+
{
|
1927 |
+
"epoch": 0.28,
|
1928 |
+
"learning_rate": 4.8572650347901544e-05,
|
1929 |
+
"loss": 1.5013,
|
1930 |
+
"step": 283
|
1931 |
+
},
|
1932 |
+
{
|
1933 |
+
"epoch": 0.28,
|
1934 |
+
"learning_rate": 4.854610909098812e-05,
|
1935 |
+
"loss": 1.4916,
|
1936 |
+
"step": 284
|
1937 |
+
},
|
1938 |
+
{
|
1939 |
+
"epoch": 0.28,
|
1940 |
+
"learning_rate": 4.851933072501756e-05,
|
1941 |
+
"loss": 1.4802,
|
1942 |
+
"step": 285
|
1943 |
+
},
|
1944 |
+
{
|
1945 |
+
"epoch": 0.29,
|
1946 |
+
"learning_rate": 4.849231551964771e-05,
|
1947 |
+
"loss": 1.4403,
|
1948 |
+
"step": 286
|
1949 |
+
},
|
1950 |
+
{
|
1951 |
+
"epoch": 0.29,
|
1952 |
+
"learning_rate": 4.8465063746921395e-05,
|
1953 |
+
"loss": 1.4976,
|
1954 |
+
"step": 287
|
1955 |
+
},
|
1956 |
+
{
|
1957 |
+
"epoch": 0.29,
|
1958 |
+
"learning_rate": 4.8437575681263656e-05,
|
1959 |
+
"loss": 1.4676,
|
1960 |
+
"step": 288
|
1961 |
+
},
|
1962 |
+
{
|
1963 |
+
"epoch": 0.29,
|
1964 |
+
"learning_rate": 4.8409851599479015e-05,
|
1965 |
+
"loss": 1.4919,
|
1966 |
+
"step": 289
|
1967 |
+
},
|
1968 |
+
{
|
1969 |
+
"epoch": 0.29,
|
1970 |
+
"learning_rate": 4.838189178074867e-05,
|
1971 |
+
"loss": 1.4806,
|
1972 |
+
"step": 290
|
1973 |
+
},
|
1974 |
+
{
|
1975 |
+
"epoch": 0.29,
|
1976 |
+
"eval_loss": 1.4974409341812134,
|
1977 |
+
"eval_runtime": 2.4575,
|
1978 |
+
"eval_samples_per_second": 34.588,
|
1979 |
+
"eval_steps_per_second": 2.035,
|
1980 |
+
"step": 290
|
1981 |
+
},
|
1982 |
+
{
|
1983 |
+
"epoch": 0.29,
|
1984 |
+
"learning_rate": 4.835369650662767e-05,
|
1985 |
+
"loss": 1.4784,
|
1986 |
+
"step": 291
|
1987 |
+
},
|
1988 |
+
{
|
1989 |
+
"epoch": 0.29,
|
1990 |
+
"learning_rate": 4.832526606104213e-05,
|
1991 |
+
"loss": 1.4818,
|
1992 |
+
"step": 292
|
1993 |
+
},
|
1994 |
+
{
|
1995 |
+
"epoch": 0.29,
|
1996 |
+
"learning_rate": 4.829660073028631e-05,
|
1997 |
+
"loss": 1.4512,
|
1998 |
+
"step": 293
|
1999 |
+
},
|
2000 |
+
{
|
2001 |
+
"epoch": 0.29,
|
2002 |
+
"learning_rate": 4.826770080301978e-05,
|
2003 |
+
"loss": 1.4728,
|
2004 |
+
"step": 294
|
2005 |
+
},
|
2006 |
+
{
|
2007 |
+
"epoch": 0.29,
|
2008 |
+
"learning_rate": 4.823856657026448e-05,
|
2009 |
+
"loss": 1.4996,
|
2010 |
+
"step": 295
|
2011 |
+
},
|
2012 |
+
{
|
2013 |
+
"epoch": 0.3,
|
2014 |
+
"learning_rate": 4.8209198325401815e-05,
|
2015 |
+
"loss": 1.4781,
|
2016 |
+
"step": 296
|
2017 |
+
},
|
2018 |
+
{
|
2019 |
+
"epoch": 0.3,
|
2020 |
+
"learning_rate": 4.817959636416969e-05,
|
2021 |
+
"loss": 1.4983,
|
2022 |
+
"step": 297
|
2023 |
+
},
|
2024 |
+
{
|
2025 |
+
"epoch": 0.3,
|
2026 |
+
"learning_rate": 4.8149760984659506e-05,
|
2027 |
+
"loss": 1.4604,
|
2028 |
+
"step": 298
|
2029 |
+
},
|
2030 |
+
{
|
2031 |
+
"epoch": 0.3,
|
2032 |
+
"learning_rate": 4.811969248731323e-05,
|
2033 |
+
"loss": 1.4986,
|
2034 |
+
"step": 299
|
2035 |
+
},
|
2036 |
+
{
|
2037 |
+
"epoch": 0.3,
|
2038 |
+
"learning_rate": 4.8089391174920275e-05,
|
2039 |
+
"loss": 1.4972,
|
2040 |
+
"step": 300
|
2041 |
+
},
|
2042 |
+
{
|
2043 |
+
"epoch": 0.3,
|
2044 |
+
"eval_loss": 1.497441053390503,
|
2045 |
+
"eval_runtime": 2.4854,
|
2046 |
+
"eval_samples_per_second": 34.2,
|
2047 |
+
"eval_steps_per_second": 2.012,
|
2048 |
+
"step": 300
|
2049 |
+
}
|
2050 |
+
],
|
2051 |
+
"max_steps": 1000,
|
2052 |
+
"num_train_epochs": 9223372036854775807,
|
2053 |
+
"total_flos": 3.00741461803008e+17,
|
2054 |
+
"trial_name": null,
|
2055 |
+
"trial_params": null
|
2056 |
+
}
|
checkpoint-300/training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:a56f0a54e3cd2b9c7442d5e94de4b5fa438d7c3e0833995d3351c23b2e7dc832
|
3 |
+
size 3451
|