Hugging Face's logo

Datasets:
MallikaWy
/
project-backup

Dataset card Files
xet
 Community
MallikaWy commited on
Commit
b915bbb
·
verified ·
1 Parent(s): ef74ff1

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/README.md +202 -0
  2. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/adapter_config.json +29 -0
  3. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/added_tokens.json +24 -0
  4. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/latest +1 -0
  5. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/merges.txt +0 -0
  6. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/special_tokens_map.json +31 -0
  7. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/tokenizer_config.json +210 -0
  8. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/trainer_state.json +1433 -0
  9. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/vocab.json +0 -0
  10. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/zero_to_fp32.py +760 -0
  11. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/README.md +202 -0
  12. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/adapter_config.json +29 -0
  13. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/added_tokens.json +24 -0
  14. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/latest +1 -0
  15. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/merges.txt +0 -0
  16. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/special_tokens_map.json +31 -0
  17. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/tokenizer_config.json +210 -0
  18. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/trainer_state.json +2133 -0
  19. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/vocab.json +0 -0
  20. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/zero_to_fp32.py +760 -0
  21. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/README.md +202 -0
  22. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/adapter_config.json +29 -0
  23. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/added_tokens.json +24 -0
  24. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/config.json +61 -0
  25. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/latest +1 -0
  26. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/merges.txt +0 -0
  27. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/preprocessor_config.json +19 -0
  28. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/special_tokens_map.json +31 -0
  29. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/tokenizer_config.json +207 -0
  30. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/trainer_state.json +2833 -0
  31. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/vocab.json +0 -0
  32. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/zero_to_fp32.py +760 -0
  33. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch1-mlp/checkpoint-100/README.md +202 -0
  34. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch1-mlp/checkpoint-100/special_tokens_map.json +31 -0
  35. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch1-mlp/checkpoint-100/zero_to_fp32.py +760 -0
  36. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/README.md +202 -0
  37. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/adapter_config.json +111 -0
  38. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/added_tokens.json +24 -0
  39. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/chat_template.json +3 -0
  40. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/README.md +202 -0
  41. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/adapter_config.json +111 -0
  42. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/added_tokens.json +24 -0
  43. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/latest +1 -0
  44. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/merges.txt +0 -0
  45. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/special_tokens_map.json +31 -0
  46. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/tokenizer_config.json +210 -0
  47. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/trainer_state.json +733 -0
  48. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/vocab.json +0 -0
  49. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/zero_to_fp32.py +760 -0
  50. InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-200/README.md +202 -0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/adapter_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 16,
14
+ "lora_dropout": 0.05,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": null,
18
+ "peft_type": "LORA",
19
+ "r": 8,
20
+ "rank_pattern": {},
21
+ "revision": null,
22
+ "target_modules": [
23
+ "mlp.2",
24
+ "mlp.0"
25
+ ],
26
+ "task_type": "CAUSAL_LM",
27
+ "use_dora": false,
28
+ "use_rslora": false
29
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/added_tokens.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</tool_call>": 151658,
3
+ "<tool_call>": 151657,
4
+ "<|box_end|>": 151649,
5
+ "<|box_start|>": 151648,
6
+ "<|endoftext|>": 151643,
7
+ "<|file_sep|>": 151664,
8
+ "<|fim_middle|>": 151660,
9
+ "<|fim_pad|>": 151662,
10
+ "<|fim_prefix|>": 151659,
11
+ "<|fim_suffix|>": 151661,
12
+ "<|im_end|>": 151645,
13
+ "<|im_start|>": 151644,
14
+ "<|image_pad|>": 151655,
15
+ "<|object_ref_end|>": 151647,
16
+ "<|object_ref_start|>": 151646,
17
+ "<|quad_end|>": 151651,
18
+ "<|quad_start|>": 151650,
19
+ "<|repo_name|>": 151663,
20
+ "<|video_pad|>": 151656,
21
+ "<|vision_end|>": 151653,
22
+ "<|vision_pad|>": 151654,
23
+ "<|vision_start|>": 151652
24
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step200
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/tokenizer_config.json ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ }
181
+ },
182
+ "additional_special_tokens": [
183
+ "<|im_start|>",
184
+ "<|im_end|>",
185
+ "<|object_ref_start|>",
186
+ "<|object_ref_end|>",
187
+ "<|box_start|>",
188
+ "<|box_end|>",
189
+ "<|quad_start|>",
190
+ "<|quad_end|>",
191
+ "<|vision_start|>",
192
+ "<|vision_end|>",
193
+ "<|vision_pad|>",
194
+ "<|image_pad|>",
195
+ "<|video_pad|>"
196
+ ],
197
+ "bos_token": null,
198
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
199
+ "clean_up_tokenization_spaces": false,
200
+ "eos_token": "<|im_end|>",
201
+ "errors": "replace",
202
+ "extra_special_tokens": {},
203
+ "model_max_length": 8192,
204
+ "pad_token": "<|endoftext|>",
205
+ "padding_side": "right",
206
+ "split_special_tokens": false,
207
+ "tokenizer_class": "Qwen2Tokenizer",
208
+ "truncation": true,
209
+ "unk_token": null
210
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/trainer_state.json ADDED
@@ -0,0 +1,1433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.21511158913686476,
5
+ "eval_steps": 500,
6
+ "global_step": 200,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0010755579456843238,
13
+ "grad_norm": 0.08796706795692444,
14
+ "learning_rate": 0.0,
15
+ "loss": 3.92,
16
+ "step": 1
17
+ },
18
+ {
19
+ "epoch": 0.0021511158913686476,
20
+ "grad_norm": 0.03759637102484703,
21
+ "learning_rate": 5.252990700743872e-06,
22
+ "loss": 2.2035,
23
+ "step": 2
24
+ },
25
+ {
26
+ "epoch": 0.0032266738370529714,
27
+ "grad_norm": 0.08365879952907562,
28
+ "learning_rate": 8.325793277315987e-06,
29
+ "loss": 4.949,
30
+ "step": 3
31
+ },
32
+ {
33
+ "epoch": 0.004302231782737295,
34
+ "grad_norm": 0.08474036306142807,
35
+ "learning_rate": 1.0505981401487744e-05,
36
+ "loss": 4.1587,
37
+ "step": 4
38
+ },
39
+ {
40
+ "epoch": 0.005377789728421619,
41
+ "grad_norm": 0.04265577346086502,
42
+ "learning_rate": 1.2197066690239248e-05,
43
+ "loss": 2.5989,
44
+ "step": 5
45
+ },
46
+ {
47
+ "epoch": 0.006453347674105943,
48
+ "grad_norm": 0.12465030699968338,
49
+ "learning_rate": 1.3578783978059857e-05,
50
+ "loss": 3.7898,
51
+ "step": 6
52
+ },
53
+ {
54
+ "epoch": 0.007528905619790266,
55
+ "grad_norm": 0.07569626718759537,
56
+ "learning_rate": 1.4747009299256131e-05,
57
+ "loss": 4.4848,
58
+ "step": 7
59
+ },
60
+ {
61
+ "epoch": 0.00860446356547459,
62
+ "grad_norm": 0.047666411846876144,
63
+ "learning_rate": 1.5758972102231614e-05,
64
+ "loss": 4.0286,
65
+ "step": 8
66
+ },
67
+ {
68
+ "epoch": 0.009680021511158914,
69
+ "grad_norm": 0.06569754332304001,
70
+ "learning_rate": 1.6651586554631973e-05,
71
+ "loss": 4.8475,
72
+ "step": 9
73
+ },
74
+ {
75
+ "epoch": 0.010755579456843238,
76
+ "grad_norm": 0.08362340182065964,
77
+ "learning_rate": 1.745005739098312e-05,
78
+ "loss": 4.1507,
79
+ "step": 10
80
+ },
81
+ {
82
+ "epoch": 0.011831137402527562,
83
+ "grad_norm": 0.0652509331703186,
84
+ "learning_rate": 1.8172362122561044e-05,
85
+ "loss": 3.1503,
86
+ "step": 11
87
+ },
88
+ {
89
+ "epoch": 0.012906695348211886,
90
+ "grad_norm": 0.07024404406547546,
91
+ "learning_rate": 1.883177467880373e-05,
92
+ "loss": 4.5793,
93
+ "step": 12
94
+ },
95
+ {
96
+ "epoch": 0.01398225329389621,
97
+ "grad_norm": 0.08304284512996674,
98
+ "learning_rate": 1.9438375428058432e-05,
99
+ "loss": 2.2916,
100
+ "step": 13
101
+ },
102
+ {
103
+ "epoch": 0.015057811239580532,
104
+ "grad_norm": 0.09187567979097366,
105
+ "learning_rate": 2e-05,
106
+ "loss": 4.3406,
107
+ "step": 14
108
+ },
109
+ {
110
+ "epoch": 0.016133369185264857,
111
+ "grad_norm": 0.09281238913536072,
112
+ "learning_rate": 2e-05,
113
+ "loss": 3.9966,
114
+ "step": 15
115
+ },
116
+ {
117
+ "epoch": 0.01720892713094918,
118
+ "grad_norm": 0.06343083083629608,
119
+ "learning_rate": 2e-05,
120
+ "loss": 3.1267,
121
+ "step": 16
122
+ },
123
+ {
124
+ "epoch": 0.018284485076633505,
125
+ "grad_norm": 0.05553126335144043,
126
+ "learning_rate": 2e-05,
127
+ "loss": 2.6709,
128
+ "step": 17
129
+ },
130
+ {
131
+ "epoch": 0.01936004302231783,
132
+ "grad_norm": 0.07925406098365784,
133
+ "learning_rate": 2e-05,
134
+ "loss": 3.0051,
135
+ "step": 18
136
+ },
137
+ {
138
+ "epoch": 0.020435600968002152,
139
+ "grad_norm": 0.07598422467708588,
140
+ "learning_rate": 2e-05,
141
+ "loss": 2.5423,
142
+ "step": 19
143
+ },
144
+ {
145
+ "epoch": 0.021511158913686476,
146
+ "grad_norm": 0.11351823806762695,
147
+ "learning_rate": 2e-05,
148
+ "loss": 4.2694,
149
+ "step": 20
150
+ },
151
+ {
152
+ "epoch": 0.0225867168593708,
153
+ "grad_norm": 0.07363948971033096,
154
+ "learning_rate": 2e-05,
155
+ "loss": 3.2626,
156
+ "step": 21
157
+ },
158
+ {
159
+ "epoch": 0.023662274805055124,
160
+ "grad_norm": 0.06679286062717438,
161
+ "learning_rate": 2e-05,
162
+ "loss": 3.476,
163
+ "step": 22
164
+ },
165
+ {
166
+ "epoch": 0.024737832750739448,
167
+ "grad_norm": 0.0646878108382225,
168
+ "learning_rate": 2e-05,
169
+ "loss": 2.6229,
170
+ "step": 23
171
+ },
172
+ {
173
+ "epoch": 0.02581339069642377,
174
+ "grad_norm": 0.14322379231452942,
175
+ "learning_rate": 2e-05,
176
+ "loss": 4.1941,
177
+ "step": 24
178
+ },
179
+ {
180
+ "epoch": 0.026888948642108095,
181
+ "grad_norm": 0.11759044229984283,
182
+ "learning_rate": 2e-05,
183
+ "loss": 4.4217,
184
+ "step": 25
185
+ },
186
+ {
187
+ "epoch": 0.02796450658779242,
188
+ "grad_norm": 0.11527741700410843,
189
+ "learning_rate": 2e-05,
190
+ "loss": 4.2243,
191
+ "step": 26
192
+ },
193
+ {
194
+ "epoch": 0.02904006453347674,
195
+ "grad_norm": 0.06782741844654083,
196
+ "learning_rate": 2e-05,
197
+ "loss": 2.8814,
198
+ "step": 27
199
+ },
200
+ {
201
+ "epoch": 0.030115622479161063,
202
+ "grad_norm": 0.1451217532157898,
203
+ "learning_rate": 2e-05,
204
+ "loss": 3.0934,
205
+ "step": 28
206
+ },
207
+ {
208
+ "epoch": 0.031191180424845387,
209
+ "grad_norm": 0.11510222405195236,
210
+ "learning_rate": 2e-05,
211
+ "loss": 2.9115,
212
+ "step": 29
213
+ },
214
+ {
215
+ "epoch": 0.032266738370529714,
216
+ "grad_norm": 0.14234976470470428,
217
+ "learning_rate": 2e-05,
218
+ "loss": 4.2921,
219
+ "step": 30
220
+ },
221
+ {
222
+ "epoch": 0.033342296316214035,
223
+ "grad_norm": 0.12387730926275253,
224
+ "learning_rate": 2e-05,
225
+ "loss": 3.8533,
226
+ "step": 31
227
+ },
228
+ {
229
+ "epoch": 0.03441785426189836,
230
+ "grad_norm": 0.08505982905626297,
231
+ "learning_rate": 2e-05,
232
+ "loss": 2.293,
233
+ "step": 32
234
+ },
235
+ {
236
+ "epoch": 0.03549341220758268,
237
+ "grad_norm": 0.1067776158452034,
238
+ "learning_rate": 2e-05,
239
+ "loss": 3.5162,
240
+ "step": 33
241
+ },
242
+ {
243
+ "epoch": 0.03656897015326701,
244
+ "grad_norm": 0.1792977899312973,
245
+ "learning_rate": 2e-05,
246
+ "loss": 5.5227,
247
+ "step": 34
248
+ },
249
+ {
250
+ "epoch": 0.03764452809895133,
251
+ "grad_norm": 0.1272369921207428,
252
+ "learning_rate": 2e-05,
253
+ "loss": 3.1357,
254
+ "step": 35
255
+ },
256
+ {
257
+ "epoch": 0.03872008604463566,
258
+ "grad_norm": 0.19525307416915894,
259
+ "learning_rate": 2e-05,
260
+ "loss": 3.4933,
261
+ "step": 36
262
+ },
263
+ {
264
+ "epoch": 0.03979564399031998,
265
+ "grad_norm": 0.15390488505363464,
266
+ "learning_rate": 2e-05,
267
+ "loss": 3.9449,
268
+ "step": 37
269
+ },
270
+ {
271
+ "epoch": 0.040871201936004305,
272
+ "grad_norm": 0.1725727617740631,
273
+ "learning_rate": 2e-05,
274
+ "loss": 4.6611,
275
+ "step": 38
276
+ },
277
+ {
278
+ "epoch": 0.041946759881688625,
279
+ "grad_norm": 0.14692844450473785,
280
+ "learning_rate": 2e-05,
281
+ "loss": 3.9717,
282
+ "step": 39
283
+ },
284
+ {
285
+ "epoch": 0.04302231782737295,
286
+ "grad_norm": 0.21558256447315216,
287
+ "learning_rate": 2e-05,
288
+ "loss": 4.4963,
289
+ "step": 40
290
+ },
291
+ {
292
+ "epoch": 0.04409787577305727,
293
+ "grad_norm": 0.16119123995304108,
294
+ "learning_rate": 2e-05,
295
+ "loss": 3.7143,
296
+ "step": 41
297
+ },
298
+ {
299
+ "epoch": 0.0451734337187416,
300
+ "grad_norm": 0.17820394039154053,
301
+ "learning_rate": 2e-05,
302
+ "loss": 3.6309,
303
+ "step": 42
304
+ },
305
+ {
306
+ "epoch": 0.04624899166442592,
307
+ "grad_norm": 0.2944539785385132,
308
+ "learning_rate": 2e-05,
309
+ "loss": 4.682,
310
+ "step": 43
311
+ },
312
+ {
313
+ "epoch": 0.04732454961011025,
314
+ "grad_norm": 0.24868269264698029,
315
+ "learning_rate": 2e-05,
316
+ "loss": 4.4821,
317
+ "step": 44
318
+ },
319
+ {
320
+ "epoch": 0.04840010755579457,
321
+ "grad_norm": 0.22005093097686768,
322
+ "learning_rate": 2e-05,
323
+ "loss": 3.8853,
324
+ "step": 45
325
+ },
326
+ {
327
+ "epoch": 0.049475665501478895,
328
+ "grad_norm": 0.219339519739151,
329
+ "learning_rate": 2e-05,
330
+ "loss": 3.1609,
331
+ "step": 46
332
+ },
333
+ {
334
+ "epoch": 0.050551223447163215,
335
+ "grad_norm": 0.18265879154205322,
336
+ "learning_rate": 2e-05,
337
+ "loss": 3.5927,
338
+ "step": 47
339
+ },
340
+ {
341
+ "epoch": 0.05162678139284754,
342
+ "grad_norm": 0.14955416321754456,
343
+ "learning_rate": 2e-05,
344
+ "loss": 3.2562,
345
+ "step": 48
346
+ },
347
+ {
348
+ "epoch": 0.05270233933853186,
349
+ "grad_norm": 0.15133006870746613,
350
+ "learning_rate": 2e-05,
351
+ "loss": 3.3774,
352
+ "step": 49
353
+ },
354
+ {
355
+ "epoch": 0.05377789728421619,
356
+ "grad_norm": 0.29029661417007446,
357
+ "learning_rate": 2e-05,
358
+ "loss": 4.0942,
359
+ "step": 50
360
+ },
361
+ {
362
+ "epoch": 0.05485345522990051,
363
+ "grad_norm": 0.22209279239177704,
364
+ "learning_rate": 2e-05,
365
+ "loss": 3.3751,
366
+ "step": 51
367
+ },
368
+ {
369
+ "epoch": 0.05592901317558484,
370
+ "grad_norm": 0.3011445701122284,
371
+ "learning_rate": 2e-05,
372
+ "loss": 3.2212,
373
+ "step": 52
374
+ },
375
+ {
376
+ "epoch": 0.05700457112126916,
377
+ "grad_norm": 0.16566476225852966,
378
+ "learning_rate": 2e-05,
379
+ "loss": 3.3951,
380
+ "step": 53
381
+ },
382
+ {
383
+ "epoch": 0.05808012906695348,
384
+ "grad_norm": 0.21326138079166412,
385
+ "learning_rate": 2e-05,
386
+ "loss": 2.4738,
387
+ "step": 54
388
+ },
389
+ {
390
+ "epoch": 0.059155687012637806,
391
+ "grad_norm": 0.37225082516670227,
392
+ "learning_rate": 2e-05,
393
+ "loss": 3.9215,
394
+ "step": 55
395
+ },
396
+ {
397
+ "epoch": 0.060231244958322126,
398
+ "grad_norm": 0.3678470849990845,
399
+ "learning_rate": 2e-05,
400
+ "loss": 5.1287,
401
+ "step": 56
402
+ },
403
+ {
404
+ "epoch": 0.061306802904006454,
405
+ "grad_norm": 0.3504056930541992,
406
+ "learning_rate": 2e-05,
407
+ "loss": 4.5698,
408
+ "step": 57
409
+ },
410
+ {
411
+ "epoch": 0.062382360849690774,
412
+ "grad_norm": 0.3031858503818512,
413
+ "learning_rate": 2e-05,
414
+ "loss": 3.0631,
415
+ "step": 58
416
+ },
417
+ {
418
+ "epoch": 0.0634579187953751,
419
+ "grad_norm": 0.19123384356498718,
420
+ "learning_rate": 2e-05,
421
+ "loss": 3.2144,
422
+ "step": 59
423
+ },
424
+ {
425
+ "epoch": 0.06453347674105943,
426
+ "grad_norm": 0.4370661973953247,
427
+ "learning_rate": 2e-05,
428
+ "loss": 4.4572,
429
+ "step": 60
430
+ },
431
+ {
432
+ "epoch": 0.06560903468674374,
433
+ "grad_norm": 0.22062987089157104,
434
+ "learning_rate": 2e-05,
435
+ "loss": 3.6537,
436
+ "step": 61
437
+ },
438
+ {
439
+ "epoch": 0.06668459263242807,
440
+ "grad_norm": 0.3883632719516754,
441
+ "learning_rate": 2e-05,
442
+ "loss": 4.5733,
443
+ "step": 62
444
+ },
445
+ {
446
+ "epoch": 0.0677601505781124,
447
+ "grad_norm": 0.5261430144309998,
448
+ "learning_rate": 2e-05,
449
+ "loss": 4.7935,
450
+ "step": 63
451
+ },
452
+ {
453
+ "epoch": 0.06883570852379672,
454
+ "grad_norm": 0.3224443197250366,
455
+ "learning_rate": 2e-05,
456
+ "loss": 4.2075,
457
+ "step": 64
458
+ },
459
+ {
460
+ "epoch": 0.06991126646948104,
461
+ "grad_norm": 0.31252115964889526,
462
+ "learning_rate": 2e-05,
463
+ "loss": 3.8278,
464
+ "step": 65
465
+ },
466
+ {
467
+ "epoch": 0.07098682441516536,
468
+ "grad_norm": 0.3743115961551666,
469
+ "learning_rate": 2e-05,
470
+ "loss": 4.0193,
471
+ "step": 66
472
+ },
473
+ {
474
+ "epoch": 0.07206238236084969,
475
+ "grad_norm": 0.08219049125909805,
476
+ "learning_rate": 2e-05,
477
+ "loss": 2.1053,
478
+ "step": 67
479
+ },
480
+ {
481
+ "epoch": 0.07313794030653402,
482
+ "grad_norm": 0.32555970549583435,
483
+ "learning_rate": 2e-05,
484
+ "loss": 3.6152,
485
+ "step": 68
486
+ },
487
+ {
488
+ "epoch": 0.07421349825221833,
489
+ "grad_norm": 0.5192983746528625,
490
+ "learning_rate": 2e-05,
491
+ "loss": 2.7418,
492
+ "step": 69
493
+ },
494
+ {
495
+ "epoch": 0.07528905619790266,
496
+ "grad_norm": 0.40597131848335266,
497
+ "learning_rate": 2e-05,
498
+ "loss": 3.8889,
499
+ "step": 70
500
+ },
501
+ {
502
+ "epoch": 0.07636461414358699,
503
+ "grad_norm": 0.2159370630979538,
504
+ "learning_rate": 2e-05,
505
+ "loss": 2.2449,
506
+ "step": 71
507
+ },
508
+ {
509
+ "epoch": 0.07744017208927131,
510
+ "grad_norm": 0.30155783891677856,
511
+ "learning_rate": 2e-05,
512
+ "loss": 3.261,
513
+ "step": 72
514
+ },
515
+ {
516
+ "epoch": 0.07851573003495563,
517
+ "grad_norm": 0.8526483774185181,
518
+ "learning_rate": 2e-05,
519
+ "loss": 4.2702,
520
+ "step": 73
521
+ },
522
+ {
523
+ "epoch": 0.07959128798063995,
524
+ "grad_norm": 0.20504102110862732,
525
+ "learning_rate": 2e-05,
526
+ "loss": 2.1707,
527
+ "step": 74
528
+ },
529
+ {
530
+ "epoch": 0.08066684592632428,
531
+ "grad_norm": 0.4856877028942108,
532
+ "learning_rate": 2e-05,
533
+ "loss": 2.8522,
534
+ "step": 75
535
+ },
536
+ {
537
+ "epoch": 0.08174240387200861,
538
+ "grad_norm": 0.5779041647911072,
539
+ "learning_rate": 2e-05,
540
+ "loss": 3.46,
541
+ "step": 76
542
+ },
543
+ {
544
+ "epoch": 0.08281796181769292,
545
+ "grad_norm": 0.440947949886322,
546
+ "learning_rate": 2e-05,
547
+ "loss": 3.5463,
548
+ "step": 77
549
+ },
550
+ {
551
+ "epoch": 0.08389351976337725,
552
+ "grad_norm": 0.20853711664676666,
553
+ "learning_rate": 2e-05,
554
+ "loss": 2.2373,
555
+ "step": 78
556
+ },
557
+ {
558
+ "epoch": 0.08496907770906158,
559
+ "grad_norm": 0.23959238827228546,
560
+ "learning_rate": 2e-05,
561
+ "loss": 2.2488,
562
+ "step": 79
563
+ },
564
+ {
565
+ "epoch": 0.0860446356547459,
566
+ "grad_norm": 0.5024415850639343,
567
+ "learning_rate": 2e-05,
568
+ "loss": 4.206,
569
+ "step": 80
570
+ },
571
+ {
572
+ "epoch": 0.08712019360043022,
573
+ "grad_norm": 0.559302806854248,
574
+ "learning_rate": 2e-05,
575
+ "loss": 3.4094,
576
+ "step": 81
577
+ },
578
+ {
579
+ "epoch": 0.08819575154611455,
580
+ "grad_norm": 0.6160392165184021,
581
+ "learning_rate": 2e-05,
582
+ "loss": 3.4671,
583
+ "step": 82
584
+ },
585
+ {
586
+ "epoch": 0.08927130949179887,
587
+ "grad_norm": 0.38725361227989197,
588
+ "learning_rate": 2e-05,
589
+ "loss": 2.444,
590
+ "step": 83
591
+ },
592
+ {
593
+ "epoch": 0.0903468674374832,
594
+ "grad_norm": 0.8837510943412781,
595
+ "learning_rate": 2e-05,
596
+ "loss": 2.8208,
597
+ "step": 84
598
+ },
599
+ {
600
+ "epoch": 0.09142242538316751,
601
+ "grad_norm": 0.7302711606025696,
602
+ "learning_rate": 2e-05,
603
+ "loss": 4.2269,
604
+ "step": 85
605
+ },
606
+ {
607
+ "epoch": 0.09249798332885184,
608
+ "grad_norm": 0.5721362233161926,
609
+ "learning_rate": 2e-05,
610
+ "loss": 3.7523,
611
+ "step": 86
612
+ },
613
+ {
614
+ "epoch": 0.09357354127453617,
615
+ "grad_norm": 0.5297729969024658,
616
+ "learning_rate": 2e-05,
617
+ "loss": 3.9402,
618
+ "step": 87
619
+ },
620
+ {
621
+ "epoch": 0.0946490992202205,
622
+ "grad_norm": 0.4848865270614624,
623
+ "learning_rate": 2e-05,
624
+ "loss": 4.0042,
625
+ "step": 88
626
+ },
627
+ {
628
+ "epoch": 0.09572465716590481,
629
+ "grad_norm": 0.7766829133033752,
630
+ "learning_rate": 2e-05,
631
+ "loss": 4.9007,
632
+ "step": 89
633
+ },
634
+ {
635
+ "epoch": 0.09680021511158914,
636
+ "grad_norm": 0.7517933249473572,
637
+ "learning_rate": 2e-05,
638
+ "loss": 3.4699,
639
+ "step": 90
640
+ },
641
+ {
642
+ "epoch": 0.09787577305727346,
643
+ "grad_norm": 0.8348935842514038,
644
+ "learning_rate": 2e-05,
645
+ "loss": 3.3264,
646
+ "step": 91
647
+ },
648
+ {
649
+ "epoch": 0.09895133100295779,
650
+ "grad_norm": 0.6233230233192444,
651
+ "learning_rate": 2e-05,
652
+ "loss": 3.0732,
653
+ "step": 92
654
+ },
655
+ {
656
+ "epoch": 0.1000268889486421,
657
+ "grad_norm": 0.7525995373725891,
658
+ "learning_rate": 2e-05,
659
+ "loss": 4.2947,
660
+ "step": 93
661
+ },
662
+ {
663
+ "epoch": 0.10110244689432643,
664
+ "grad_norm": 1.0709550380706787,
665
+ "learning_rate": 2e-05,
666
+ "loss": 3.9258,
667
+ "step": 94
668
+ },
669
+ {
670
+ "epoch": 0.10217800484001076,
671
+ "grad_norm": 0.5882276892662048,
672
+ "learning_rate": 2e-05,
673
+ "loss": 3.2486,
674
+ "step": 95
675
+ },
676
+ {
677
+ "epoch": 0.10325356278569509,
678
+ "grad_norm": 1.3848390579223633,
679
+ "learning_rate": 2e-05,
680
+ "loss": 3.1059,
681
+ "step": 96
682
+ },
683
+ {
684
+ "epoch": 0.1043291207313794,
685
+ "grad_norm": 1.0833474397659302,
686
+ "learning_rate": 2e-05,
687
+ "loss": 3.176,
688
+ "step": 97
689
+ },
690
+ {
691
+ "epoch": 0.10540467867706373,
692
+ "grad_norm": 0.6375811100006104,
693
+ "learning_rate": 2e-05,
694
+ "loss": 3.2412,
695
+ "step": 98
696
+ },
697
+ {
698
+ "epoch": 0.10648023662274805,
699
+ "grad_norm": 1.6618685722351074,
700
+ "learning_rate": 2e-05,
701
+ "loss": 3.2567,
702
+ "step": 99
703
+ },
704
+ {
705
+ "epoch": 0.10755579456843238,
706
+ "grad_norm": 0.978987991809845,
707
+ "learning_rate": 2e-05,
708
+ "loss": 4.2846,
709
+ "step": 100
710
+ },
711
+ {
712
+ "epoch": 0.1086313525141167,
713
+ "grad_norm": 0.857375979423523,
714
+ "learning_rate": 2e-05,
715
+ "loss": 3.5841,
716
+ "step": 101
717
+ },
718
+ {
719
+ "epoch": 0.10970691045980102,
720
+ "grad_norm": 1.7573585510253906,
721
+ "learning_rate": 2e-05,
722
+ "loss": 4.3449,
723
+ "step": 102
724
+ },
725
+ {
726
+ "epoch": 0.11078246840548535,
727
+ "grad_norm": 0.925139307975769,
728
+ "learning_rate": 2e-05,
729
+ "loss": 2.9236,
730
+ "step": 103
731
+ },
732
+ {
733
+ "epoch": 0.11185802635116968,
734
+ "grad_norm": 1.0588639974594116,
735
+ "learning_rate": 2e-05,
736
+ "loss": 3.5894,
737
+ "step": 104
738
+ },
739
+ {
740
+ "epoch": 0.11293358429685399,
741
+ "grad_norm": 2.0492420196533203,
742
+ "learning_rate": 2e-05,
743
+ "loss": 3.3241,
744
+ "step": 105
745
+ },
746
+ {
747
+ "epoch": 0.11400914224253832,
748
+ "grad_norm": 1.2681959867477417,
749
+ "learning_rate": 2e-05,
750
+ "loss": 2.9072,
751
+ "step": 106
752
+ },
753
+ {
754
+ "epoch": 0.11508470018822264,
755
+ "grad_norm": 0.8113515377044678,
756
+ "learning_rate": 2e-05,
757
+ "loss": 3.7567,
758
+ "step": 107
759
+ },
760
+ {
761
+ "epoch": 0.11616025813390696,
762
+ "grad_norm": 0.8880893588066101,
763
+ "learning_rate": 2e-05,
764
+ "loss": 2.9194,
765
+ "step": 108
766
+ },
767
+ {
768
+ "epoch": 0.11723581607959128,
769
+ "grad_norm": 0.3360120356082916,
770
+ "learning_rate": 2e-05,
771
+ "loss": 2.0488,
772
+ "step": 109
773
+ },
774
+ {
775
+ "epoch": 0.11831137402527561,
776
+ "grad_norm": 0.9331343173980713,
777
+ "learning_rate": 2e-05,
778
+ "loss": 3.0238,
779
+ "step": 110
780
+ },
781
+ {
782
+ "epoch": 0.11938693197095994,
783
+ "grad_norm": 1.8001103401184082,
784
+ "learning_rate": 2e-05,
785
+ "loss": 4.6799,
786
+ "step": 111
787
+ },
788
+ {
789
+ "epoch": 0.12046248991664425,
790
+ "grad_norm": 0.6512516140937805,
791
+ "learning_rate": 2e-05,
792
+ "loss": 2.1307,
793
+ "step": 112
794
+ },
795
+ {
796
+ "epoch": 0.12153804786232858,
797
+ "grad_norm": 1.2599865198135376,
798
+ "learning_rate": 2e-05,
799
+ "loss": 2.6622,
800
+ "step": 113
801
+ },
802
+ {
803
+ "epoch": 0.12261360580801291,
804
+ "grad_norm": 0.7627917528152466,
805
+ "learning_rate": 2e-05,
806
+ "loss": 3.5136,
807
+ "step": 114
808
+ },
809
+ {
810
+ "epoch": 0.12368916375369723,
811
+ "grad_norm": 0.6828168630599976,
812
+ "learning_rate": 2e-05,
813
+ "loss": 2.471,
814
+ "step": 115
815
+ },
816
+ {
817
+ "epoch": 0.12476472169938155,
818
+ "grad_norm": 2.2847108840942383,
819
+ "learning_rate": 2e-05,
820
+ "loss": 5.1293,
821
+ "step": 116
822
+ },
823
+ {
824
+ "epoch": 0.1258402796450659,
825
+ "grad_norm": 1.5811688899993896,
826
+ "learning_rate": 2e-05,
827
+ "loss": 3.4057,
828
+ "step": 117
829
+ },
830
+ {
831
+ "epoch": 0.1269158375907502,
832
+ "grad_norm": 0.6133345365524292,
833
+ "learning_rate": 2e-05,
834
+ "loss": 2.1214,
835
+ "step": 118
836
+ },
837
+ {
838
+ "epoch": 0.12799139553643452,
839
+ "grad_norm": 1.0904330015182495,
840
+ "learning_rate": 2e-05,
841
+ "loss": 3.5764,
842
+ "step": 119
843
+ },
844
+ {
845
+ "epoch": 0.12906695348211886,
846
+ "grad_norm": 3.196547269821167,
847
+ "learning_rate": 2e-05,
848
+ "loss": 4.751,
849
+ "step": 120
850
+ },
851
+ {
852
+ "epoch": 0.13014251142780317,
853
+ "grad_norm": 2.6566030979156494,
854
+ "learning_rate": 2e-05,
855
+ "loss": 4.1988,
856
+ "step": 121
857
+ },
858
+ {
859
+ "epoch": 0.13121806937348748,
860
+ "grad_norm": 2.3683555126190186,
861
+ "learning_rate": 2e-05,
862
+ "loss": 4.2292,
863
+ "step": 122
864
+ },
865
+ {
866
+ "epoch": 0.13229362731917182,
867
+ "grad_norm": 2.0530343055725098,
868
+ "learning_rate": 2e-05,
869
+ "loss": 3.3207,
870
+ "step": 123
871
+ },
872
+ {
873
+ "epoch": 0.13336918526485614,
874
+ "grad_norm": 1.4857006072998047,
875
+ "learning_rate": 2e-05,
876
+ "loss": 3.1832,
877
+ "step": 124
878
+ },
879
+ {
880
+ "epoch": 0.13444474321054048,
881
+ "grad_norm": 1.1598551273345947,
882
+ "learning_rate": 2e-05,
883
+ "loss": 3.7383,
884
+ "step": 125
885
+ },
886
+ {
887
+ "epoch": 0.1355203011562248,
888
+ "grad_norm": 1.7023690938949585,
889
+ "learning_rate": 2e-05,
890
+ "loss": 4.4133,
891
+ "step": 126
892
+ },
893
+ {
894
+ "epoch": 0.1365958591019091,
895
+ "grad_norm": 2.516563653945923,
896
+ "learning_rate": 2e-05,
897
+ "loss": 4.4345,
898
+ "step": 127
899
+ },
900
+ {
901
+ "epoch": 0.13767141704759345,
902
+ "grad_norm": 1.2092735767364502,
903
+ "learning_rate": 2e-05,
904
+ "loss": 2.0301,
905
+ "step": 128
906
+ },
907
+ {
908
+ "epoch": 0.13874697499327776,
909
+ "grad_norm": 3.6382813453674316,
910
+ "learning_rate": 2e-05,
911
+ "loss": 3.1083,
912
+ "step": 129
913
+ },
914
+ {
915
+ "epoch": 0.13982253293896207,
916
+ "grad_norm": 1.815604567527771,
917
+ "learning_rate": 2e-05,
918
+ "loss": 3.6486,
919
+ "step": 130
920
+ },
921
+ {
922
+ "epoch": 0.14089809088464642,
923
+ "grad_norm": 2.0142934322357178,
924
+ "learning_rate": 2e-05,
925
+ "loss": 3.137,
926
+ "step": 131
927
+ },
928
+ {
929
+ "epoch": 0.14197364883033073,
930
+ "grad_norm": 1.4500945806503296,
931
+ "learning_rate": 2e-05,
932
+ "loss": 2.9764,
933
+ "step": 132
934
+ },
935
+ {
936
+ "epoch": 0.14304920677601507,
937
+ "grad_norm": 3.060540199279785,
938
+ "learning_rate": 2e-05,
939
+ "loss": 3.8098,
940
+ "step": 133
941
+ },
942
+ {
943
+ "epoch": 0.14412476472169938,
944
+ "grad_norm": 1.1867421865463257,
945
+ "learning_rate": 2e-05,
946
+ "loss": 2.3144,
947
+ "step": 134
948
+ },
949
+ {
950
+ "epoch": 0.1452003226673837,
951
+ "grad_norm": 2.2631044387817383,
952
+ "learning_rate": 2e-05,
953
+ "loss": 3.6758,
954
+ "step": 135
955
+ },
956
+ {
957
+ "epoch": 0.14627588061306804,
958
+ "grad_norm": 1.878165364265442,
959
+ "learning_rate": 2e-05,
960
+ "loss": 2.954,
961
+ "step": 136
962
+ },
963
+ {
964
+ "epoch": 0.14735143855875235,
965
+ "grad_norm": 2.5874857902526855,
966
+ "learning_rate": 2e-05,
967
+ "loss": 2.9404,
968
+ "step": 137
969
+ },
970
+ {
971
+ "epoch": 0.14842699650443666,
972
+ "grad_norm": 1.9505641460418701,
973
+ "learning_rate": 2e-05,
974
+ "loss": 3.8377,
975
+ "step": 138
976
+ },
977
+ {
978
+ "epoch": 0.149502554450121,
979
+ "grad_norm": 3.316041946411133,
980
+ "learning_rate": 2e-05,
981
+ "loss": 3.3591,
982
+ "step": 139
983
+ },
984
+ {
985
+ "epoch": 0.15057811239580532,
986
+ "grad_norm": 1.034047245979309,
987
+ "learning_rate": 2e-05,
988
+ "loss": 2.7709,
989
+ "step": 140
990
+ },
991
+ {
992
+ "epoch": 0.15165367034148966,
993
+ "grad_norm": 3.9413716793060303,
994
+ "learning_rate": 2e-05,
995
+ "loss": 4.6814,
996
+ "step": 141
997
+ },
998
+ {
999
+ "epoch": 0.15272922828717397,
1000
+ "grad_norm": 1.6777610778808594,
1001
+ "learning_rate": 2e-05,
1002
+ "loss": 3.0112,
1003
+ "step": 142
1004
+ },
1005
+ {
1006
+ "epoch": 0.1538047862328583,
1007
+ "grad_norm": 2.4318885803222656,
1008
+ "learning_rate": 2e-05,
1009
+ "loss": 3.0604,
1010
+ "step": 143
1011
+ },
1012
+ {
1013
+ "epoch": 0.15488034417854263,
1014
+ "grad_norm": 1.0361096858978271,
1015
+ "learning_rate": 2e-05,
1016
+ "loss": 2.4662,
1017
+ "step": 144
1018
+ },
1019
+ {
1020
+ "epoch": 0.15595590212422694,
1021
+ "grad_norm": 1.5604453086853027,
1022
+ "learning_rate": 2e-05,
1023
+ "loss": 2.6922,
1024
+ "step": 145
1025
+ },
1026
+ {
1027
+ "epoch": 0.15703146006991126,
1028
+ "grad_norm": 2.758985996246338,
1029
+ "learning_rate": 2e-05,
1030
+ "loss": 4.1986,
1031
+ "step": 146
1032
+ },
1033
+ {
1034
+ "epoch": 0.1581070180155956,
1035
+ "grad_norm": 3.1454544067382812,
1036
+ "learning_rate": 2e-05,
1037
+ "loss": 3.3037,
1038
+ "step": 147
1039
+ },
1040
+ {
1041
+ "epoch": 0.1591825759612799,
1042
+ "grad_norm": 5.285416126251221,
1043
+ "learning_rate": 2e-05,
1044
+ "loss": 4.41,
1045
+ "step": 148
1046
+ },
1047
+ {
1048
+ "epoch": 0.16025813390696425,
1049
+ "grad_norm": 4.258360385894775,
1050
+ "learning_rate": 2e-05,
1051
+ "loss": 3.552,
1052
+ "step": 149
1053
+ },
1054
+ {
1055
+ "epoch": 0.16133369185264856,
1056
+ "grad_norm": 3.8160877227783203,
1057
+ "learning_rate": 2e-05,
1058
+ "loss": 2.3886,
1059
+ "step": 150
1060
+ },
1061
+ {
1062
+ "epoch": 0.16240924979833288,
1063
+ "grad_norm": 2.338573694229126,
1064
+ "learning_rate": 2e-05,
1065
+ "loss": 2.7356,
1066
+ "step": 151
1067
+ },
1068
+ {
1069
+ "epoch": 0.16348480774401722,
1070
+ "grad_norm": 4.280124664306641,
1071
+ "learning_rate": 2e-05,
1072
+ "loss": 3.6049,
1073
+ "step": 152
1074
+ },
1075
+ {
1076
+ "epoch": 0.16456036568970153,
1077
+ "grad_norm": 2.4771525859832764,
1078
+ "learning_rate": 2e-05,
1079
+ "loss": 4.0302,
1080
+ "step": 153
1081
+ },
1082
+ {
1083
+ "epoch": 0.16563592363538585,
1084
+ "grad_norm": 1.8418432474136353,
1085
+ "learning_rate": 2e-05,
1086
+ "loss": 3.6569,
1087
+ "step": 154
1088
+ },
1089
+ {
1090
+ "epoch": 0.1667114815810702,
1091
+ "grad_norm": 2.0972683429718018,
1092
+ "learning_rate": 2e-05,
1093
+ "loss": 3.0681,
1094
+ "step": 155
1095
+ },
1096
+ {
1097
+ "epoch": 0.1677870395267545,
1098
+ "grad_norm": 1.6193856000900269,
1099
+ "learning_rate": 2e-05,
1100
+ "loss": 2.8379,
1101
+ "step": 156
1102
+ },
1103
+ {
1104
+ "epoch": 0.16886259747243884,
1105
+ "grad_norm": 4.455961227416992,
1106
+ "learning_rate": 2e-05,
1107
+ "loss": 2.7098,
1108
+ "step": 157
1109
+ },
1110
+ {
1111
+ "epoch": 0.16993815541812315,
1112
+ "grad_norm": 1.417285680770874,
1113
+ "learning_rate": 2e-05,
1114
+ "loss": 1.7622,
1115
+ "step": 158
1116
+ },
1117
+ {
1118
+ "epoch": 0.17101371336380747,
1119
+ "grad_norm": 4.029820442199707,
1120
+ "learning_rate": 2e-05,
1121
+ "loss": 4.3837,
1122
+ "step": 159
1123
+ },
1124
+ {
1125
+ "epoch": 0.1720892713094918,
1126
+ "grad_norm": 1.492792010307312,
1127
+ "learning_rate": 2e-05,
1128
+ "loss": 2.9025,
1129
+ "step": 160
1130
+ },
1131
+ {
1132
+ "epoch": 0.17316482925517612,
1133
+ "grad_norm": 4.077206134796143,
1134
+ "learning_rate": 2e-05,
1135
+ "loss": 3.2485,
1136
+ "step": 161
1137
+ },
1138
+ {
1139
+ "epoch": 0.17424038720086044,
1140
+ "grad_norm": 3.3235838413238525,
1141
+ "learning_rate": 2e-05,
1142
+ "loss": 2.7973,
1143
+ "step": 162
1144
+ },
1145
+ {
1146
+ "epoch": 0.17531594514654478,
1147
+ "grad_norm": 3.8488590717315674,
1148
+ "learning_rate": 2e-05,
1149
+ "loss": 3.8877,
1150
+ "step": 163
1151
+ },
1152
+ {
1153
+ "epoch": 0.1763915030922291,
1154
+ "grad_norm": 2.2555062770843506,
1155
+ "learning_rate": 2e-05,
1156
+ "loss": 3.0205,
1157
+ "step": 164
1158
+ },
1159
+ {
1160
+ "epoch": 0.1774670610379134,
1161
+ "grad_norm": 1.7995167970657349,
1162
+ "learning_rate": 2e-05,
1163
+ "loss": 2.2315,
1164
+ "step": 165
1165
+ },
1166
+ {
1167
+ "epoch": 0.17854261898359775,
1168
+ "grad_norm": 1.2209360599517822,
1169
+ "learning_rate": 2e-05,
1170
+ "loss": 2.4366,
1171
+ "step": 166
1172
+ },
1173
+ {
1174
+ "epoch": 0.17961817692928206,
1175
+ "grad_norm": 1.4641145467758179,
1176
+ "learning_rate": 2e-05,
1177
+ "loss": 1.9656,
1178
+ "step": 167
1179
+ },
1180
+ {
1181
+ "epoch": 0.1806937348749664,
1182
+ "grad_norm": 2.6921463012695312,
1183
+ "learning_rate": 2e-05,
1184
+ "loss": 4.2631,
1185
+ "step": 168
1186
+ },
1187
+ {
1188
+ "epoch": 0.1817692928206507,
1189
+ "grad_norm": 4.518846035003662,
1190
+ "learning_rate": 2e-05,
1191
+ "loss": 3.7601,
1192
+ "step": 169
1193
+ },
1194
+ {
1195
+ "epoch": 0.18284485076633503,
1196
+ "grad_norm": 1.3827426433563232,
1197
+ "learning_rate": 2e-05,
1198
+ "loss": 2.3307,
1199
+ "step": 170
1200
+ },
1201
+ {
1202
+ "epoch": 0.18392040871201937,
1203
+ "grad_norm": 2.110506057739258,
1204
+ "learning_rate": 2e-05,
1205
+ "loss": 2.8058,
1206
+ "step": 171
1207
+ },
1208
+ {
1209
+ "epoch": 0.18499596665770368,
1210
+ "grad_norm": 1.1899346113204956,
1211
+ "learning_rate": 2e-05,
1212
+ "loss": 1.9843,
1213
+ "step": 172
1214
+ },
1215
+ {
1216
+ "epoch": 0.186071524603388,
1217
+ "grad_norm": 3.0731053352355957,
1218
+ "learning_rate": 2e-05,
1219
+ "loss": 2.3099,
1220
+ "step": 173
1221
+ },
1222
+ {
1223
+ "epoch": 0.18714708254907234,
1224
+ "grad_norm": 3.0657575130462646,
1225
+ "learning_rate": 2e-05,
1226
+ "loss": 2.8818,
1227
+ "step": 174
1228
+ },
1229
+ {
1230
+ "epoch": 0.18822264049475665,
1231
+ "grad_norm": 2.161705493927002,
1232
+ "learning_rate": 2e-05,
1233
+ "loss": 2.4349,
1234
+ "step": 175
1235
+ },
1236
+ {
1237
+ "epoch": 0.189298198440441,
1238
+ "grad_norm": 3.0202341079711914,
1239
+ "learning_rate": 2e-05,
1240
+ "loss": 2.1977,
1241
+ "step": 176
1242
+ },
1243
+ {
1244
+ "epoch": 0.1903737563861253,
1245
+ "grad_norm": 2.346747398376465,
1246
+ "learning_rate": 2e-05,
1247
+ "loss": 2.399,
1248
+ "step": 177
1249
+ },
1250
+ {
1251
+ "epoch": 0.19144931433180962,
1252
+ "grad_norm": 1.5756202936172485,
1253
+ "learning_rate": 2e-05,
1254
+ "loss": 2.2689,
1255
+ "step": 178
1256
+ },
1257
+ {
1258
+ "epoch": 0.19252487227749396,
1259
+ "grad_norm": 2.398148775100708,
1260
+ "learning_rate": 2e-05,
1261
+ "loss": 3.8229,
1262
+ "step": 179
1263
+ },
1264
+ {
1265
+ "epoch": 0.19360043022317827,
1266
+ "grad_norm": 3.2819738388061523,
1267
+ "learning_rate": 2e-05,
1268
+ "loss": 2.6033,
1269
+ "step": 180
1270
+ },
1271
+ {
1272
+ "epoch": 0.19467598816886258,
1273
+ "grad_norm": 5.887080192565918,
1274
+ "learning_rate": 2e-05,
1275
+ "loss": 2.4956,
1276
+ "step": 181
1277
+ },
1278
+ {
1279
+ "epoch": 0.19575154611454693,
1280
+ "grad_norm": 2.208775520324707,
1281
+ "learning_rate": 2e-05,
1282
+ "loss": 2.1286,
1283
+ "step": 182
1284
+ },
1285
+ {
1286
+ "epoch": 0.19682710406023124,
1287
+ "grad_norm": 1.817888617515564,
1288
+ "learning_rate": 2e-05,
1289
+ "loss": 2.4608,
1290
+ "step": 183
1291
+ },
1292
+ {
1293
+ "epoch": 0.19790266200591558,
1294
+ "grad_norm": 2.150890588760376,
1295
+ "learning_rate": 2e-05,
1296
+ "loss": 3.2177,
1297
+ "step": 184
1298
+ },
1299
+ {
1300
+ "epoch": 0.1989782199515999,
1301
+ "grad_norm": 6.783865928649902,
1302
+ "learning_rate": 2e-05,
1303
+ "loss": 3.6292,
1304
+ "step": 185
1305
+ },
1306
+ {
1307
+ "epoch": 0.2000537778972842,
1308
+ "grad_norm": 3.8008947372436523,
1309
+ "learning_rate": 2e-05,
1310
+ "loss": 2.9182,
1311
+ "step": 186
1312
+ },
1313
+ {
1314
+ "epoch": 0.20112933584296855,
1315
+ "grad_norm": 4.612886905670166,
1316
+ "learning_rate": 2e-05,
1317
+ "loss": 3.1909,
1318
+ "step": 187
1319
+ },
1320
+ {
1321
+ "epoch": 0.20220489378865286,
1322
+ "grad_norm": 5.594083786010742,
1323
+ "learning_rate": 2e-05,
1324
+ "loss": 3.0986,
1325
+ "step": 188
1326
+ },
1327
+ {
1328
+ "epoch": 0.20328045173433718,
1329
+ "grad_norm": 3.358548402786255,
1330
+ "learning_rate": 2e-05,
1331
+ "loss": 2.2792,
1332
+ "step": 189
1333
+ },
1334
+ {
1335
+ "epoch": 0.20435600968002152,
1336
+ "grad_norm": 2.6911849975585938,
1337
+ "learning_rate": 2e-05,
1338
+ "loss": 3.4548,
1339
+ "step": 190
1340
+ },
1341
+ {
1342
+ "epoch": 0.20543156762570583,
1343
+ "grad_norm": 2.4795117378234863,
1344
+ "learning_rate": 2e-05,
1345
+ "loss": 2.3414,
1346
+ "step": 191
1347
+ },
1348
+ {
1349
+ "epoch": 0.20650712557139017,
1350
+ "grad_norm": 1.7064529657363892,
1351
+ "learning_rate": 2e-05,
1352
+ "loss": 1.8158,
1353
+ "step": 192
1354
+ },
1355
+ {
1356
+ "epoch": 0.20758268351707448,
1357
+ "grad_norm": 7.239320755004883,
1358
+ "learning_rate": 2e-05,
1359
+ "loss": 2.6676,
1360
+ "step": 193
1361
+ },
1362
+ {
1363
+ "epoch": 0.2086582414627588,
1364
+ "grad_norm": 4.747775077819824,
1365
+ "learning_rate": 2e-05,
1366
+ "loss": 3.2929,
1367
+ "step": 194
1368
+ },
1369
+ {
1370
+ "epoch": 0.20973379940844314,
1371
+ "grad_norm": 1.850651502609253,
1372
+ "learning_rate": 2e-05,
1373
+ "loss": 2.3005,
1374
+ "step": 195
1375
+ },
1376
+ {
1377
+ "epoch": 0.21080935735412745,
1378
+ "grad_norm": 2.4029853343963623,
1379
+ "learning_rate": 2e-05,
1380
+ "loss": 3.7213,
1381
+ "step": 196
1382
+ },
1383
+ {
1384
+ "epoch": 0.21188491529981177,
1385
+ "grad_norm": 2.5401084423065186,
1386
+ "learning_rate": 2e-05,
1387
+ "loss": 2.7665,
1388
+ "step": 197
1389
+ },
1390
+ {
1391
+ "epoch": 0.2129604732454961,
1392
+ "grad_norm": 4.080130577087402,
1393
+ "learning_rate": 2e-05,
1394
+ "loss": 1.8886,
1395
+ "step": 198
1396
+ },
1397
+ {
1398
+ "epoch": 0.21403603119118042,
1399
+ "grad_norm": 1.7884845733642578,
1400
+ "learning_rate": 2e-05,
1401
+ "loss": 2.4168,
1402
+ "step": 199
1403
+ },
1404
+ {
1405
+ "epoch": 0.21511158913686476,
1406
+ "grad_norm": 3.1292009353637695,
1407
+ "learning_rate": 2e-05,
1408
+ "loss": 2.4442,
1409
+ "step": 200
1410
+ }
1411
+ ],
1412
+ "logging_steps": 1.0,
1413
+ "max_steps": 465,
1414
+ "num_input_tokens_seen": 0,
1415
+ "num_train_epochs": 1,
1416
+ "save_steps": 100,
1417
+ "stateful_callbacks": {
1418
+ "TrainerControl": {
1419
+ "args": {
1420
+ "should_epoch_stop": false,
1421
+ "should_evaluate": false,
1422
+ "should_log": false,
1423
+ "should_save": true,
1424
+ "should_training_stop": false
1425
+ },
1426
+ "attributes": {}
1427
+ }
1428
+ },
1429
+ "total_flos": 1909776771072.0,
1430
+ "train_batch_size": 4,
1431
+ "trial_name": null,
1432
+ "trial_params": null
1433
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-200/zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/adapter_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 16,
14
+ "lora_dropout": 0.05,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": null,
18
+ "peft_type": "LORA",
19
+ "r": 8,
20
+ "rank_pattern": {},
21
+ "revision": null,
22
+ "target_modules": [
23
+ "mlp.2",
24
+ "mlp.0"
25
+ ],
26
+ "task_type": "CAUSAL_LM",
27
+ "use_dora": false,
28
+ "use_rslora": false
29
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/added_tokens.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</tool_call>": 151658,
3
+ "<tool_call>": 151657,
4
+ "<|box_end|>": 151649,
5
+ "<|box_start|>": 151648,
6
+ "<|endoftext|>": 151643,
7
+ "<|file_sep|>": 151664,
8
+ "<|fim_middle|>": 151660,
9
+ "<|fim_pad|>": 151662,
10
+ "<|fim_prefix|>": 151659,
11
+ "<|fim_suffix|>": 151661,
12
+ "<|im_end|>": 151645,
13
+ "<|im_start|>": 151644,
14
+ "<|image_pad|>": 151655,
15
+ "<|object_ref_end|>": 151647,
16
+ "<|object_ref_start|>": 151646,
17
+ "<|quad_end|>": 151651,
18
+ "<|quad_start|>": 151650,
19
+ "<|repo_name|>": 151663,
20
+ "<|video_pad|>": 151656,
21
+ "<|vision_end|>": 151653,
22
+ "<|vision_pad|>": 151654,
23
+ "<|vision_start|>": 151652
24
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step300
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/tokenizer_config.json ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ }
181
+ },
182
+ "additional_special_tokens": [
183
+ "<|im_start|>",
184
+ "<|im_end|>",
185
+ "<|object_ref_start|>",
186
+ "<|object_ref_end|>",
187
+ "<|box_start|>",
188
+ "<|box_end|>",
189
+ "<|quad_start|>",
190
+ "<|quad_end|>",
191
+ "<|vision_start|>",
192
+ "<|vision_end|>",
193
+ "<|vision_pad|>",
194
+ "<|image_pad|>",
195
+ "<|video_pad|>"
196
+ ],
197
+ "bos_token": null,
198
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
199
+ "clean_up_tokenization_spaces": false,
200
+ "eos_token": "<|im_end|>",
201
+ "errors": "replace",
202
+ "extra_special_tokens": {},
203
+ "model_max_length": 8192,
204
+ "pad_token": "<|endoftext|>",
205
+ "padding_side": "right",
206
+ "split_special_tokens": false,
207
+ "tokenizer_class": "Qwen2Tokenizer",
208
+ "truncation": true,
209
+ "unk_token": null
210
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/trainer_state.json ADDED
@@ -0,0 +1,2133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.32266738370529713,
5
+ "eval_steps": 500,
6
+ "global_step": 300,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0010755579456843238,
13
+ "grad_norm": 0.08796706795692444,
14
+ "learning_rate": 0.0,
15
+ "loss": 3.92,
16
+ "step": 1
17
+ },
18
+ {
19
+ "epoch": 0.0021511158913686476,
20
+ "grad_norm": 0.03759637102484703,
21
+ "learning_rate": 5.252990700743872e-06,
22
+ "loss": 2.2035,
23
+ "step": 2
24
+ },
25
+ {
26
+ "epoch": 0.0032266738370529714,
27
+ "grad_norm": 0.08365879952907562,
28
+ "learning_rate": 8.325793277315987e-06,
29
+ "loss": 4.949,
30
+ "step": 3
31
+ },
32
+ {
33
+ "epoch": 0.004302231782737295,
34
+ "grad_norm": 0.08474036306142807,
35
+ "learning_rate": 1.0505981401487744e-05,
36
+ "loss": 4.1587,
37
+ "step": 4
38
+ },
39
+ {
40
+ "epoch": 0.005377789728421619,
41
+ "grad_norm": 0.04265577346086502,
42
+ "learning_rate": 1.2197066690239248e-05,
43
+ "loss": 2.5989,
44
+ "step": 5
45
+ },
46
+ {
47
+ "epoch": 0.006453347674105943,
48
+ "grad_norm": 0.12465030699968338,
49
+ "learning_rate": 1.3578783978059857e-05,
50
+ "loss": 3.7898,
51
+ "step": 6
52
+ },
53
+ {
54
+ "epoch": 0.007528905619790266,
55
+ "grad_norm": 0.07569626718759537,
56
+ "learning_rate": 1.4747009299256131e-05,
57
+ "loss": 4.4848,
58
+ "step": 7
59
+ },
60
+ {
61
+ "epoch": 0.00860446356547459,
62
+ "grad_norm": 0.047666411846876144,
63
+ "learning_rate": 1.5758972102231614e-05,
64
+ "loss": 4.0286,
65
+ "step": 8
66
+ },
67
+ {
68
+ "epoch": 0.009680021511158914,
69
+ "grad_norm": 0.06569754332304001,
70
+ "learning_rate": 1.6651586554631973e-05,
71
+ "loss": 4.8475,
72
+ "step": 9
73
+ },
74
+ {
75
+ "epoch": 0.010755579456843238,
76
+ "grad_norm": 0.08362340182065964,
77
+ "learning_rate": 1.745005739098312e-05,
78
+ "loss": 4.1507,
79
+ "step": 10
80
+ },
81
+ {
82
+ "epoch": 0.011831137402527562,
83
+ "grad_norm": 0.0652509331703186,
84
+ "learning_rate": 1.8172362122561044e-05,
85
+ "loss": 3.1503,
86
+ "step": 11
87
+ },
88
+ {
89
+ "epoch": 0.012906695348211886,
90
+ "grad_norm": 0.07024404406547546,
91
+ "learning_rate": 1.883177467880373e-05,
92
+ "loss": 4.5793,
93
+ "step": 12
94
+ },
95
+ {
96
+ "epoch": 0.01398225329389621,
97
+ "grad_norm": 0.08304284512996674,
98
+ "learning_rate": 1.9438375428058432e-05,
99
+ "loss": 2.2916,
100
+ "step": 13
101
+ },
102
+ {
103
+ "epoch": 0.015057811239580532,
104
+ "grad_norm": 0.09187567979097366,
105
+ "learning_rate": 2e-05,
106
+ "loss": 4.3406,
107
+ "step": 14
108
+ },
109
+ {
110
+ "epoch": 0.016133369185264857,
111
+ "grad_norm": 0.09281238913536072,
112
+ "learning_rate": 2e-05,
113
+ "loss": 3.9966,
114
+ "step": 15
115
+ },
116
+ {
117
+ "epoch": 0.01720892713094918,
118
+ "grad_norm": 0.06343083083629608,
119
+ "learning_rate": 2e-05,
120
+ "loss": 3.1267,
121
+ "step": 16
122
+ },
123
+ {
124
+ "epoch": 0.018284485076633505,
125
+ "grad_norm": 0.05553126335144043,
126
+ "learning_rate": 2e-05,
127
+ "loss": 2.6709,
128
+ "step": 17
129
+ },
130
+ {
131
+ "epoch": 0.01936004302231783,
132
+ "grad_norm": 0.07925406098365784,
133
+ "learning_rate": 2e-05,
134
+ "loss": 3.0051,
135
+ "step": 18
136
+ },
137
+ {
138
+ "epoch": 0.020435600968002152,
139
+ "grad_norm": 0.07598422467708588,
140
+ "learning_rate": 2e-05,
141
+ "loss": 2.5423,
142
+ "step": 19
143
+ },
144
+ {
145
+ "epoch": 0.021511158913686476,
146
+ "grad_norm": 0.11351823806762695,
147
+ "learning_rate": 2e-05,
148
+ "loss": 4.2694,
149
+ "step": 20
150
+ },
151
+ {
152
+ "epoch": 0.0225867168593708,
153
+ "grad_norm": 0.07363948971033096,
154
+ "learning_rate": 2e-05,
155
+ "loss": 3.2626,
156
+ "step": 21
157
+ },
158
+ {
159
+ "epoch": 0.023662274805055124,
160
+ "grad_norm": 0.06679286062717438,
161
+ "learning_rate": 2e-05,
162
+ "loss": 3.476,
163
+ "step": 22
164
+ },
165
+ {
166
+ "epoch": 0.024737832750739448,
167
+ "grad_norm": 0.0646878108382225,
168
+ "learning_rate": 2e-05,
169
+ "loss": 2.6229,
170
+ "step": 23
171
+ },
172
+ {
173
+ "epoch": 0.02581339069642377,
174
+ "grad_norm": 0.14322379231452942,
175
+ "learning_rate": 2e-05,
176
+ "loss": 4.1941,
177
+ "step": 24
178
+ },
179
+ {
180
+ "epoch": 0.026888948642108095,
181
+ "grad_norm": 0.11759044229984283,
182
+ "learning_rate": 2e-05,
183
+ "loss": 4.4217,
184
+ "step": 25
185
+ },
186
+ {
187
+ "epoch": 0.02796450658779242,
188
+ "grad_norm": 0.11527741700410843,
189
+ "learning_rate": 2e-05,
190
+ "loss": 4.2243,
191
+ "step": 26
192
+ },
193
+ {
194
+ "epoch": 0.02904006453347674,
195
+ "grad_norm": 0.06782741844654083,
196
+ "learning_rate": 2e-05,
197
+ "loss": 2.8814,
198
+ "step": 27
199
+ },
200
+ {
201
+ "epoch": 0.030115622479161063,
202
+ "grad_norm": 0.1451217532157898,
203
+ "learning_rate": 2e-05,
204
+ "loss": 3.0934,
205
+ "step": 28
206
+ },
207
+ {
208
+ "epoch": 0.031191180424845387,
209
+ "grad_norm": 0.11510222405195236,
210
+ "learning_rate": 2e-05,
211
+ "loss": 2.9115,
212
+ "step": 29
213
+ },
214
+ {
215
+ "epoch": 0.032266738370529714,
216
+ "grad_norm": 0.14234976470470428,
217
+ "learning_rate": 2e-05,
218
+ "loss": 4.2921,
219
+ "step": 30
220
+ },
221
+ {
222
+ "epoch": 0.033342296316214035,
223
+ "grad_norm": 0.12387730926275253,
224
+ "learning_rate": 2e-05,
225
+ "loss": 3.8533,
226
+ "step": 31
227
+ },
228
+ {
229
+ "epoch": 0.03441785426189836,
230
+ "grad_norm": 0.08505982905626297,
231
+ "learning_rate": 2e-05,
232
+ "loss": 2.293,
233
+ "step": 32
234
+ },
235
+ {
236
+ "epoch": 0.03549341220758268,
237
+ "grad_norm": 0.1067776158452034,
238
+ "learning_rate": 2e-05,
239
+ "loss": 3.5162,
240
+ "step": 33
241
+ },
242
+ {
243
+ "epoch": 0.03656897015326701,
244
+ "grad_norm": 0.1792977899312973,
245
+ "learning_rate": 2e-05,
246
+ "loss": 5.5227,
247
+ "step": 34
248
+ },
249
+ {
250
+ "epoch": 0.03764452809895133,
251
+ "grad_norm": 0.1272369921207428,
252
+ "learning_rate": 2e-05,
253
+ "loss": 3.1357,
254
+ "step": 35
255
+ },
256
+ {
257
+ "epoch": 0.03872008604463566,
258
+ "grad_norm": 0.19525307416915894,
259
+ "learning_rate": 2e-05,
260
+ "loss": 3.4933,
261
+ "step": 36
262
+ },
263
+ {
264
+ "epoch": 0.03979564399031998,
265
+ "grad_norm": 0.15390488505363464,
266
+ "learning_rate": 2e-05,
267
+ "loss": 3.9449,
268
+ "step": 37
269
+ },
270
+ {
271
+ "epoch": 0.040871201936004305,
272
+ "grad_norm": 0.1725727617740631,
273
+ "learning_rate": 2e-05,
274
+ "loss": 4.6611,
275
+ "step": 38
276
+ },
277
+ {
278
+ "epoch": 0.041946759881688625,
279
+ "grad_norm": 0.14692844450473785,
280
+ "learning_rate": 2e-05,
281
+ "loss": 3.9717,
282
+ "step": 39
283
+ },
284
+ {
285
+ "epoch": 0.04302231782737295,
286
+ "grad_norm": 0.21558256447315216,
287
+ "learning_rate": 2e-05,
288
+ "loss": 4.4963,
289
+ "step": 40
290
+ },
291
+ {
292
+ "epoch": 0.04409787577305727,
293
+ "grad_norm": 0.16119123995304108,
294
+ "learning_rate": 2e-05,
295
+ "loss": 3.7143,
296
+ "step": 41
297
+ },
298
+ {
299
+ "epoch": 0.0451734337187416,
300
+ "grad_norm": 0.17820394039154053,
301
+ "learning_rate": 2e-05,
302
+ "loss": 3.6309,
303
+ "step": 42
304
+ },
305
+ {
306
+ "epoch": 0.04624899166442592,
307
+ "grad_norm": 0.2944539785385132,
308
+ "learning_rate": 2e-05,
309
+ "loss": 4.682,
310
+ "step": 43
311
+ },
312
+ {
313
+ "epoch": 0.04732454961011025,
314
+ "grad_norm": 0.24868269264698029,
315
+ "learning_rate": 2e-05,
316
+ "loss": 4.4821,
317
+ "step": 44
318
+ },
319
+ {
320
+ "epoch": 0.04840010755579457,
321
+ "grad_norm": 0.22005093097686768,
322
+ "learning_rate": 2e-05,
323
+ "loss": 3.8853,
324
+ "step": 45
325
+ },
326
+ {
327
+ "epoch": 0.049475665501478895,
328
+ "grad_norm": 0.219339519739151,
329
+ "learning_rate": 2e-05,
330
+ "loss": 3.1609,
331
+ "step": 46
332
+ },
333
+ {
334
+ "epoch": 0.050551223447163215,
335
+ "grad_norm": 0.18265879154205322,
336
+ "learning_rate": 2e-05,
337
+ "loss": 3.5927,
338
+ "step": 47
339
+ },
340
+ {
341
+ "epoch": 0.05162678139284754,
342
+ "grad_norm": 0.14955416321754456,
343
+ "learning_rate": 2e-05,
344
+ "loss": 3.2562,
345
+ "step": 48
346
+ },
347
+ {
348
+ "epoch": 0.05270233933853186,
349
+ "grad_norm": 0.15133006870746613,
350
+ "learning_rate": 2e-05,
351
+ "loss": 3.3774,
352
+ "step": 49
353
+ },
354
+ {
355
+ "epoch": 0.05377789728421619,
356
+ "grad_norm": 0.29029661417007446,
357
+ "learning_rate": 2e-05,
358
+ "loss": 4.0942,
359
+ "step": 50
360
+ },
361
+ {
362
+ "epoch": 0.05485345522990051,
363
+ "grad_norm": 0.22209279239177704,
364
+ "learning_rate": 2e-05,
365
+ "loss": 3.3751,
366
+ "step": 51
367
+ },
368
+ {
369
+ "epoch": 0.05592901317558484,
370
+ "grad_norm": 0.3011445701122284,
371
+ "learning_rate": 2e-05,
372
+ "loss": 3.2212,
373
+ "step": 52
374
+ },
375
+ {
376
+ "epoch": 0.05700457112126916,
377
+ "grad_norm": 0.16566476225852966,
378
+ "learning_rate": 2e-05,
379
+ "loss": 3.3951,
380
+ "step": 53
381
+ },
382
+ {
383
+ "epoch": 0.05808012906695348,
384
+ "grad_norm": 0.21326138079166412,
385
+ "learning_rate": 2e-05,
386
+ "loss": 2.4738,
387
+ "step": 54
388
+ },
389
+ {
390
+ "epoch": 0.059155687012637806,
391
+ "grad_norm": 0.37225082516670227,
392
+ "learning_rate": 2e-05,
393
+ "loss": 3.9215,
394
+ "step": 55
395
+ },
396
+ {
397
+ "epoch": 0.060231244958322126,
398
+ "grad_norm": 0.3678470849990845,
399
+ "learning_rate": 2e-05,
400
+ "loss": 5.1287,
401
+ "step": 56
402
+ },
403
+ {
404
+ "epoch": 0.061306802904006454,
405
+ "grad_norm": 0.3504056930541992,
406
+ "learning_rate": 2e-05,
407
+ "loss": 4.5698,
408
+ "step": 57
409
+ },
410
+ {
411
+ "epoch": 0.062382360849690774,
412
+ "grad_norm": 0.3031858503818512,
413
+ "learning_rate": 2e-05,
414
+ "loss": 3.0631,
415
+ "step": 58
416
+ },
417
+ {
418
+ "epoch": 0.0634579187953751,
419
+ "grad_norm": 0.19123384356498718,
420
+ "learning_rate": 2e-05,
421
+ "loss": 3.2144,
422
+ "step": 59
423
+ },
424
+ {
425
+ "epoch": 0.06453347674105943,
426
+ "grad_norm": 0.4370661973953247,
427
+ "learning_rate": 2e-05,
428
+ "loss": 4.4572,
429
+ "step": 60
430
+ },
431
+ {
432
+ "epoch": 0.06560903468674374,
433
+ "grad_norm": 0.22062987089157104,
434
+ "learning_rate": 2e-05,
435
+ "loss": 3.6537,
436
+ "step": 61
437
+ },
438
+ {
439
+ "epoch": 0.06668459263242807,
440
+ "grad_norm": 0.3883632719516754,
441
+ "learning_rate": 2e-05,
442
+ "loss": 4.5733,
443
+ "step": 62
444
+ },
445
+ {
446
+ "epoch": 0.0677601505781124,
447
+ "grad_norm": 0.5261430144309998,
448
+ "learning_rate": 2e-05,
449
+ "loss": 4.7935,
450
+ "step": 63
451
+ },
452
+ {
453
+ "epoch": 0.06883570852379672,
454
+ "grad_norm": 0.3224443197250366,
455
+ "learning_rate": 2e-05,
456
+ "loss": 4.2075,
457
+ "step": 64
458
+ },
459
+ {
460
+ "epoch": 0.06991126646948104,
461
+ "grad_norm": 0.31252115964889526,
462
+ "learning_rate": 2e-05,
463
+ "loss": 3.8278,
464
+ "step": 65
465
+ },
466
+ {
467
+ "epoch": 0.07098682441516536,
468
+ "grad_norm": 0.3743115961551666,
469
+ "learning_rate": 2e-05,
470
+ "loss": 4.0193,
471
+ "step": 66
472
+ },
473
+ {
474
+ "epoch": 0.07206238236084969,
475
+ "grad_norm": 0.08219049125909805,
476
+ "learning_rate": 2e-05,
477
+ "loss": 2.1053,
478
+ "step": 67
479
+ },
480
+ {
481
+ "epoch": 0.07313794030653402,
482
+ "grad_norm": 0.32555970549583435,
483
+ "learning_rate": 2e-05,
484
+ "loss": 3.6152,
485
+ "step": 68
486
+ },
487
+ {
488
+ "epoch": 0.07421349825221833,
489
+ "grad_norm": 0.5192983746528625,
490
+ "learning_rate": 2e-05,
491
+ "loss": 2.7418,
492
+ "step": 69
493
+ },
494
+ {
495
+ "epoch": 0.07528905619790266,
496
+ "grad_norm": 0.40597131848335266,
497
+ "learning_rate": 2e-05,
498
+ "loss": 3.8889,
499
+ "step": 70
500
+ },
501
+ {
502
+ "epoch": 0.07636461414358699,
503
+ "grad_norm": 0.2159370630979538,
504
+ "learning_rate": 2e-05,
505
+ "loss": 2.2449,
506
+ "step": 71
507
+ },
508
+ {
509
+ "epoch": 0.07744017208927131,
510
+ "grad_norm": 0.30155783891677856,
511
+ "learning_rate": 2e-05,
512
+ "loss": 3.261,
513
+ "step": 72
514
+ },
515
+ {
516
+ "epoch": 0.07851573003495563,
517
+ "grad_norm": 0.8526483774185181,
518
+ "learning_rate": 2e-05,
519
+ "loss": 4.2702,
520
+ "step": 73
521
+ },
522
+ {
523
+ "epoch": 0.07959128798063995,
524
+ "grad_norm": 0.20504102110862732,
525
+ "learning_rate": 2e-05,
526
+ "loss": 2.1707,
527
+ "step": 74
528
+ },
529
+ {
530
+ "epoch": 0.08066684592632428,
531
+ "grad_norm": 0.4856877028942108,
532
+ "learning_rate": 2e-05,
533
+ "loss": 2.8522,
534
+ "step": 75
535
+ },
536
+ {
537
+ "epoch": 0.08174240387200861,
538
+ "grad_norm": 0.5779041647911072,
539
+ "learning_rate": 2e-05,
540
+ "loss": 3.46,
541
+ "step": 76
542
+ },
543
+ {
544
+ "epoch": 0.08281796181769292,
545
+ "grad_norm": 0.440947949886322,
546
+ "learning_rate": 2e-05,
547
+ "loss": 3.5463,
548
+ "step": 77
549
+ },
550
+ {
551
+ "epoch": 0.08389351976337725,
552
+ "grad_norm": 0.20853711664676666,
553
+ "learning_rate": 2e-05,
554
+ "loss": 2.2373,
555
+ "step": 78
556
+ },
557
+ {
558
+ "epoch": 0.08496907770906158,
559
+ "grad_norm": 0.23959238827228546,
560
+ "learning_rate": 2e-05,
561
+ "loss": 2.2488,
562
+ "step": 79
563
+ },
564
+ {
565
+ "epoch": 0.0860446356547459,
566
+ "grad_norm": 0.5024415850639343,
567
+ "learning_rate": 2e-05,
568
+ "loss": 4.206,
569
+ "step": 80
570
+ },
571
+ {
572
+ "epoch": 0.08712019360043022,
573
+ "grad_norm": 0.559302806854248,
574
+ "learning_rate": 2e-05,
575
+ "loss": 3.4094,
576
+ "step": 81
577
+ },
578
+ {
579
+ "epoch": 0.08819575154611455,
580
+ "grad_norm": 0.6160392165184021,
581
+ "learning_rate": 2e-05,
582
+ "loss": 3.4671,
583
+ "step": 82
584
+ },
585
+ {
586
+ "epoch": 0.08927130949179887,
587
+ "grad_norm": 0.38725361227989197,
588
+ "learning_rate": 2e-05,
589
+ "loss": 2.444,
590
+ "step": 83
591
+ },
592
+ {
593
+ "epoch": 0.0903468674374832,
594
+ "grad_norm": 0.8837510943412781,
595
+ "learning_rate": 2e-05,
596
+ "loss": 2.8208,
597
+ "step": 84
598
+ },
599
+ {
600
+ "epoch": 0.09142242538316751,
601
+ "grad_norm": 0.7302711606025696,
602
+ "learning_rate": 2e-05,
603
+ "loss": 4.2269,
604
+ "step": 85
605
+ },
606
+ {
607
+ "epoch": 0.09249798332885184,
608
+ "grad_norm": 0.5721362233161926,
609
+ "learning_rate": 2e-05,
610
+ "loss": 3.7523,
611
+ "step": 86
612
+ },
613
+ {
614
+ "epoch": 0.09357354127453617,
615
+ "grad_norm": 0.5297729969024658,
616
+ "learning_rate": 2e-05,
617
+ "loss": 3.9402,
618
+ "step": 87
619
+ },
620
+ {
621
+ "epoch": 0.0946490992202205,
622
+ "grad_norm": 0.4848865270614624,
623
+ "learning_rate": 2e-05,
624
+ "loss": 4.0042,
625
+ "step": 88
626
+ },
627
+ {
628
+ "epoch": 0.09572465716590481,
629
+ "grad_norm": 0.7766829133033752,
630
+ "learning_rate": 2e-05,
631
+ "loss": 4.9007,
632
+ "step": 89
633
+ },
634
+ {
635
+ "epoch": 0.09680021511158914,
636
+ "grad_norm": 0.7517933249473572,
637
+ "learning_rate": 2e-05,
638
+ "loss": 3.4699,
639
+ "step": 90
640
+ },
641
+ {
642
+ "epoch": 0.09787577305727346,
643
+ "grad_norm": 0.8348935842514038,
644
+ "learning_rate": 2e-05,
645
+ "loss": 3.3264,
646
+ "step": 91
647
+ },
648
+ {
649
+ "epoch": 0.09895133100295779,
650
+ "grad_norm": 0.6233230233192444,
651
+ "learning_rate": 2e-05,
652
+ "loss": 3.0732,
653
+ "step": 92
654
+ },
655
+ {
656
+ "epoch": 0.1000268889486421,
657
+ "grad_norm": 0.7525995373725891,
658
+ "learning_rate": 2e-05,
659
+ "loss": 4.2947,
660
+ "step": 93
661
+ },
662
+ {
663
+ "epoch": 0.10110244689432643,
664
+ "grad_norm": 1.0709550380706787,
665
+ "learning_rate": 2e-05,
666
+ "loss": 3.9258,
667
+ "step": 94
668
+ },
669
+ {
670
+ "epoch": 0.10217800484001076,
671
+ "grad_norm": 0.5882276892662048,
672
+ "learning_rate": 2e-05,
673
+ "loss": 3.2486,
674
+ "step": 95
675
+ },
676
+ {
677
+ "epoch": 0.10325356278569509,
678
+ "grad_norm": 1.3848390579223633,
679
+ "learning_rate": 2e-05,
680
+ "loss": 3.1059,
681
+ "step": 96
682
+ },
683
+ {
684
+ "epoch": 0.1043291207313794,
685
+ "grad_norm": 1.0833474397659302,
686
+ "learning_rate": 2e-05,
687
+ "loss": 3.176,
688
+ "step": 97
689
+ },
690
+ {
691
+ "epoch": 0.10540467867706373,
692
+ "grad_norm": 0.6375811100006104,
693
+ "learning_rate": 2e-05,
694
+ "loss": 3.2412,
695
+ "step": 98
696
+ },
697
+ {
698
+ "epoch": 0.10648023662274805,
699
+ "grad_norm": 1.6618685722351074,
700
+ "learning_rate": 2e-05,
701
+ "loss": 3.2567,
702
+ "step": 99
703
+ },
704
+ {
705
+ "epoch": 0.10755579456843238,
706
+ "grad_norm": 0.978987991809845,
707
+ "learning_rate": 2e-05,
708
+ "loss": 4.2846,
709
+ "step": 100
710
+ },
711
+ {
712
+ "epoch": 0.1086313525141167,
713
+ "grad_norm": 0.857375979423523,
714
+ "learning_rate": 2e-05,
715
+ "loss": 3.5841,
716
+ "step": 101
717
+ },
718
+ {
719
+ "epoch": 0.10970691045980102,
720
+ "grad_norm": 1.7573585510253906,
721
+ "learning_rate": 2e-05,
722
+ "loss": 4.3449,
723
+ "step": 102
724
+ },
725
+ {
726
+ "epoch": 0.11078246840548535,
727
+ "grad_norm": 0.925139307975769,
728
+ "learning_rate": 2e-05,
729
+ "loss": 2.9236,
730
+ "step": 103
731
+ },
732
+ {
733
+ "epoch": 0.11185802635116968,
734
+ "grad_norm": 1.0588639974594116,
735
+ "learning_rate": 2e-05,
736
+ "loss": 3.5894,
737
+ "step": 104
738
+ },
739
+ {
740
+ "epoch": 0.11293358429685399,
741
+ "grad_norm": 2.0492420196533203,
742
+ "learning_rate": 2e-05,
743
+ "loss": 3.3241,
744
+ "step": 105
745
+ },
746
+ {
747
+ "epoch": 0.11400914224253832,
748
+ "grad_norm": 1.2681959867477417,
749
+ "learning_rate": 2e-05,
750
+ "loss": 2.9072,
751
+ "step": 106
752
+ },
753
+ {
754
+ "epoch": 0.11508470018822264,
755
+ "grad_norm": 0.8113515377044678,
756
+ "learning_rate": 2e-05,
757
+ "loss": 3.7567,
758
+ "step": 107
759
+ },
760
+ {
761
+ "epoch": 0.11616025813390696,
762
+ "grad_norm": 0.8880893588066101,
763
+ "learning_rate": 2e-05,
764
+ "loss": 2.9194,
765
+ "step": 108
766
+ },
767
+ {
768
+ "epoch": 0.11723581607959128,
769
+ "grad_norm": 0.3360120356082916,
770
+ "learning_rate": 2e-05,
771
+ "loss": 2.0488,
772
+ "step": 109
773
+ },
774
+ {
775
+ "epoch": 0.11831137402527561,
776
+ "grad_norm": 0.9331343173980713,
777
+ "learning_rate": 2e-05,
778
+ "loss": 3.0238,
779
+ "step": 110
780
+ },
781
+ {
782
+ "epoch": 0.11938693197095994,
783
+ "grad_norm": 1.8001103401184082,
784
+ "learning_rate": 2e-05,
785
+ "loss": 4.6799,
786
+ "step": 111
787
+ },
788
+ {
789
+ "epoch": 0.12046248991664425,
790
+ "grad_norm": 0.6512516140937805,
791
+ "learning_rate": 2e-05,
792
+ "loss": 2.1307,
793
+ "step": 112
794
+ },
795
+ {
796
+ "epoch": 0.12153804786232858,
797
+ "grad_norm": 1.2599865198135376,
798
+ "learning_rate": 2e-05,
799
+ "loss": 2.6622,
800
+ "step": 113
801
+ },
802
+ {
803
+ "epoch": 0.12261360580801291,
804
+ "grad_norm": 0.7627917528152466,
805
+ "learning_rate": 2e-05,
806
+ "loss": 3.5136,
807
+ "step": 114
808
+ },
809
+ {
810
+ "epoch": 0.12368916375369723,
811
+ "grad_norm": 0.6828168630599976,
812
+ "learning_rate": 2e-05,
813
+ "loss": 2.471,
814
+ "step": 115
815
+ },
816
+ {
817
+ "epoch": 0.12476472169938155,
818
+ "grad_norm": 2.2847108840942383,
819
+ "learning_rate": 2e-05,
820
+ "loss": 5.1293,
821
+ "step": 116
822
+ },
823
+ {
824
+ "epoch": 0.1258402796450659,
825
+ "grad_norm": 1.5811688899993896,
826
+ "learning_rate": 2e-05,
827
+ "loss": 3.4057,
828
+ "step": 117
829
+ },
830
+ {
831
+ "epoch": 0.1269158375907502,
832
+ "grad_norm": 0.6133345365524292,
833
+ "learning_rate": 2e-05,
834
+ "loss": 2.1214,
835
+ "step": 118
836
+ },
837
+ {
838
+ "epoch": 0.12799139553643452,
839
+ "grad_norm": 1.0904330015182495,
840
+ "learning_rate": 2e-05,
841
+ "loss": 3.5764,
842
+ "step": 119
843
+ },
844
+ {
845
+ "epoch": 0.12906695348211886,
846
+ "grad_norm": 3.196547269821167,
847
+ "learning_rate": 2e-05,
848
+ "loss": 4.751,
849
+ "step": 120
850
+ },
851
+ {
852
+ "epoch": 0.13014251142780317,
853
+ "grad_norm": 2.6566030979156494,
854
+ "learning_rate": 2e-05,
855
+ "loss": 4.1988,
856
+ "step": 121
857
+ },
858
+ {
859
+ "epoch": 0.13121806937348748,
860
+ "grad_norm": 2.3683555126190186,
861
+ "learning_rate": 2e-05,
862
+ "loss": 4.2292,
863
+ "step": 122
864
+ },
865
+ {
866
+ "epoch": 0.13229362731917182,
867
+ "grad_norm": 2.0530343055725098,
868
+ "learning_rate": 2e-05,
869
+ "loss": 3.3207,
870
+ "step": 123
871
+ },
872
+ {
873
+ "epoch": 0.13336918526485614,
874
+ "grad_norm": 1.4857006072998047,
875
+ "learning_rate": 2e-05,
876
+ "loss": 3.1832,
877
+ "step": 124
878
+ },
879
+ {
880
+ "epoch": 0.13444474321054048,
881
+ "grad_norm": 1.1598551273345947,
882
+ "learning_rate": 2e-05,
883
+ "loss": 3.7383,
884
+ "step": 125
885
+ },
886
+ {
887
+ "epoch": 0.1355203011562248,
888
+ "grad_norm": 1.7023690938949585,
889
+ "learning_rate": 2e-05,
890
+ "loss": 4.4133,
891
+ "step": 126
892
+ },
893
+ {
894
+ "epoch": 0.1365958591019091,
895
+ "grad_norm": 2.516563653945923,
896
+ "learning_rate": 2e-05,
897
+ "loss": 4.4345,
898
+ "step": 127
899
+ },
900
+ {
901
+ "epoch": 0.13767141704759345,
902
+ "grad_norm": 1.2092735767364502,
903
+ "learning_rate": 2e-05,
904
+ "loss": 2.0301,
905
+ "step": 128
906
+ },
907
+ {
908
+ "epoch": 0.13874697499327776,
909
+ "grad_norm": 3.6382813453674316,
910
+ "learning_rate": 2e-05,
911
+ "loss": 3.1083,
912
+ "step": 129
913
+ },
914
+ {
915
+ "epoch": 0.13982253293896207,
916
+ "grad_norm": 1.815604567527771,
917
+ "learning_rate": 2e-05,
918
+ "loss": 3.6486,
919
+ "step": 130
920
+ },
921
+ {
922
+ "epoch": 0.14089809088464642,
923
+ "grad_norm": 2.0142934322357178,
924
+ "learning_rate": 2e-05,
925
+ "loss": 3.137,
926
+ "step": 131
927
+ },
928
+ {
929
+ "epoch": 0.14197364883033073,
930
+ "grad_norm": 1.4500945806503296,
931
+ "learning_rate": 2e-05,
932
+ "loss": 2.9764,
933
+ "step": 132
934
+ },
935
+ {
936
+ "epoch": 0.14304920677601507,
937
+ "grad_norm": 3.060540199279785,
938
+ "learning_rate": 2e-05,
939
+ "loss": 3.8098,
940
+ "step": 133
941
+ },
942
+ {
943
+ "epoch": 0.14412476472169938,
944
+ "grad_norm": 1.1867421865463257,
945
+ "learning_rate": 2e-05,
946
+ "loss": 2.3144,
947
+ "step": 134
948
+ },
949
+ {
950
+ "epoch": 0.1452003226673837,
951
+ "grad_norm": 2.2631044387817383,
952
+ "learning_rate": 2e-05,
953
+ "loss": 3.6758,
954
+ "step": 135
955
+ },
956
+ {
957
+ "epoch": 0.14627588061306804,
958
+ "grad_norm": 1.878165364265442,
959
+ "learning_rate": 2e-05,
960
+ "loss": 2.954,
961
+ "step": 136
962
+ },
963
+ {
964
+ "epoch": 0.14735143855875235,
965
+ "grad_norm": 2.5874857902526855,
966
+ "learning_rate": 2e-05,
967
+ "loss": 2.9404,
968
+ "step": 137
969
+ },
970
+ {
971
+ "epoch": 0.14842699650443666,
972
+ "grad_norm": 1.9505641460418701,
973
+ "learning_rate": 2e-05,
974
+ "loss": 3.8377,
975
+ "step": 138
976
+ },
977
+ {
978
+ "epoch": 0.149502554450121,
979
+ "grad_norm": 3.316041946411133,
980
+ "learning_rate": 2e-05,
981
+ "loss": 3.3591,
982
+ "step": 139
983
+ },
984
+ {
985
+ "epoch": 0.15057811239580532,
986
+ "grad_norm": 1.034047245979309,
987
+ "learning_rate": 2e-05,
988
+ "loss": 2.7709,
989
+ "step": 140
990
+ },
991
+ {
992
+ "epoch": 0.15165367034148966,
993
+ "grad_norm": 3.9413716793060303,
994
+ "learning_rate": 2e-05,
995
+ "loss": 4.6814,
996
+ "step": 141
997
+ },
998
+ {
999
+ "epoch": 0.15272922828717397,
1000
+ "grad_norm": 1.6777610778808594,
1001
+ "learning_rate": 2e-05,
1002
+ "loss": 3.0112,
1003
+ "step": 142
1004
+ },
1005
+ {
1006
+ "epoch": 0.1538047862328583,
1007
+ "grad_norm": 2.4318885803222656,
1008
+ "learning_rate": 2e-05,
1009
+ "loss": 3.0604,
1010
+ "step": 143
1011
+ },
1012
+ {
1013
+ "epoch": 0.15488034417854263,
1014
+ "grad_norm": 1.0361096858978271,
1015
+ "learning_rate": 2e-05,
1016
+ "loss": 2.4662,
1017
+ "step": 144
1018
+ },
1019
+ {
1020
+ "epoch": 0.15595590212422694,
1021
+ "grad_norm": 1.5604453086853027,
1022
+ "learning_rate": 2e-05,
1023
+ "loss": 2.6922,
1024
+ "step": 145
1025
+ },
1026
+ {
1027
+ "epoch": 0.15703146006991126,
1028
+ "grad_norm": 2.758985996246338,
1029
+ "learning_rate": 2e-05,
1030
+ "loss": 4.1986,
1031
+ "step": 146
1032
+ },
1033
+ {
1034
+ "epoch": 0.1581070180155956,
1035
+ "grad_norm": 3.1454544067382812,
1036
+ "learning_rate": 2e-05,
1037
+ "loss": 3.3037,
1038
+ "step": 147
1039
+ },
1040
+ {
1041
+ "epoch": 0.1591825759612799,
1042
+ "grad_norm": 5.285416126251221,
1043
+ "learning_rate": 2e-05,
1044
+ "loss": 4.41,
1045
+ "step": 148
1046
+ },
1047
+ {
1048
+ "epoch": 0.16025813390696425,
1049
+ "grad_norm": 4.258360385894775,
1050
+ "learning_rate": 2e-05,
1051
+ "loss": 3.552,
1052
+ "step": 149
1053
+ },
1054
+ {
1055
+ "epoch": 0.16133369185264856,
1056
+ "grad_norm": 3.8160877227783203,
1057
+ "learning_rate": 2e-05,
1058
+ "loss": 2.3886,
1059
+ "step": 150
1060
+ },
1061
+ {
1062
+ "epoch": 0.16240924979833288,
1063
+ "grad_norm": 2.338573694229126,
1064
+ "learning_rate": 2e-05,
1065
+ "loss": 2.7356,
1066
+ "step": 151
1067
+ },
1068
+ {
1069
+ "epoch": 0.16348480774401722,
1070
+ "grad_norm": 4.280124664306641,
1071
+ "learning_rate": 2e-05,
1072
+ "loss": 3.6049,
1073
+ "step": 152
1074
+ },
1075
+ {
1076
+ "epoch": 0.16456036568970153,
1077
+ "grad_norm": 2.4771525859832764,
1078
+ "learning_rate": 2e-05,
1079
+ "loss": 4.0302,
1080
+ "step": 153
1081
+ },
1082
+ {
1083
+ "epoch": 0.16563592363538585,
1084
+ "grad_norm": 1.8418432474136353,
1085
+ "learning_rate": 2e-05,
1086
+ "loss": 3.6569,
1087
+ "step": 154
1088
+ },
1089
+ {
1090
+ "epoch": 0.1667114815810702,
1091
+ "grad_norm": 2.0972683429718018,
1092
+ "learning_rate": 2e-05,
1093
+ "loss": 3.0681,
1094
+ "step": 155
1095
+ },
1096
+ {
1097
+ "epoch": 0.1677870395267545,
1098
+ "grad_norm": 1.6193856000900269,
1099
+ "learning_rate": 2e-05,
1100
+ "loss": 2.8379,
1101
+ "step": 156
1102
+ },
1103
+ {
1104
+ "epoch": 0.16886259747243884,
1105
+ "grad_norm": 4.455961227416992,
1106
+ "learning_rate": 2e-05,
1107
+ "loss": 2.7098,
1108
+ "step": 157
1109
+ },
1110
+ {
1111
+ "epoch": 0.16993815541812315,
1112
+ "grad_norm": 1.417285680770874,
1113
+ "learning_rate": 2e-05,
1114
+ "loss": 1.7622,
1115
+ "step": 158
1116
+ },
1117
+ {
1118
+ "epoch": 0.17101371336380747,
1119
+ "grad_norm": 4.029820442199707,
1120
+ "learning_rate": 2e-05,
1121
+ "loss": 4.3837,
1122
+ "step": 159
1123
+ },
1124
+ {
1125
+ "epoch": 0.1720892713094918,
1126
+ "grad_norm": 1.492792010307312,
1127
+ "learning_rate": 2e-05,
1128
+ "loss": 2.9025,
1129
+ "step": 160
1130
+ },
1131
+ {
1132
+ "epoch": 0.17316482925517612,
1133
+ "grad_norm": 4.077206134796143,
1134
+ "learning_rate": 2e-05,
1135
+ "loss": 3.2485,
1136
+ "step": 161
1137
+ },
1138
+ {
1139
+ "epoch": 0.17424038720086044,
1140
+ "grad_norm": 3.3235838413238525,
1141
+ "learning_rate": 2e-05,
1142
+ "loss": 2.7973,
1143
+ "step": 162
1144
+ },
1145
+ {
1146
+ "epoch": 0.17531594514654478,
1147
+ "grad_norm": 3.8488590717315674,
1148
+ "learning_rate": 2e-05,
1149
+ "loss": 3.8877,
1150
+ "step": 163
1151
+ },
1152
+ {
1153
+ "epoch": 0.1763915030922291,
1154
+ "grad_norm": 2.2555062770843506,
1155
+ "learning_rate": 2e-05,
1156
+ "loss": 3.0205,
1157
+ "step": 164
1158
+ },
1159
+ {
1160
+ "epoch": 0.1774670610379134,
1161
+ "grad_norm": 1.7995167970657349,
1162
+ "learning_rate": 2e-05,
1163
+ "loss": 2.2315,
1164
+ "step": 165
1165
+ },
1166
+ {
1167
+ "epoch": 0.17854261898359775,
1168
+ "grad_norm": 1.2209360599517822,
1169
+ "learning_rate": 2e-05,
1170
+ "loss": 2.4366,
1171
+ "step": 166
1172
+ },
1173
+ {
1174
+ "epoch": 0.17961817692928206,
1175
+ "grad_norm": 1.4641145467758179,
1176
+ "learning_rate": 2e-05,
1177
+ "loss": 1.9656,
1178
+ "step": 167
1179
+ },
1180
+ {
1181
+ "epoch": 0.1806937348749664,
1182
+ "grad_norm": 2.6921463012695312,
1183
+ "learning_rate": 2e-05,
1184
+ "loss": 4.2631,
1185
+ "step": 168
1186
+ },
1187
+ {
1188
+ "epoch": 0.1817692928206507,
1189
+ "grad_norm": 4.518846035003662,
1190
+ "learning_rate": 2e-05,
1191
+ "loss": 3.7601,
1192
+ "step": 169
1193
+ },
1194
+ {
1195
+ "epoch": 0.18284485076633503,
1196
+ "grad_norm": 1.3827426433563232,
1197
+ "learning_rate": 2e-05,
1198
+ "loss": 2.3307,
1199
+ "step": 170
1200
+ },
1201
+ {
1202
+ "epoch": 0.18392040871201937,
1203
+ "grad_norm": 2.110506057739258,
1204
+ "learning_rate": 2e-05,
1205
+ "loss": 2.8058,
1206
+ "step": 171
1207
+ },
1208
+ {
1209
+ "epoch": 0.18499596665770368,
1210
+ "grad_norm": 1.1899346113204956,
1211
+ "learning_rate": 2e-05,
1212
+ "loss": 1.9843,
1213
+ "step": 172
1214
+ },
1215
+ {
1216
+ "epoch": 0.186071524603388,
1217
+ "grad_norm": 3.0731053352355957,
1218
+ "learning_rate": 2e-05,
1219
+ "loss": 2.3099,
1220
+ "step": 173
1221
+ },
1222
+ {
1223
+ "epoch": 0.18714708254907234,
1224
+ "grad_norm": 3.0657575130462646,
1225
+ "learning_rate": 2e-05,
1226
+ "loss": 2.8818,
1227
+ "step": 174
1228
+ },
1229
+ {
1230
+ "epoch": 0.18822264049475665,
1231
+ "grad_norm": 2.161705493927002,
1232
+ "learning_rate": 2e-05,
1233
+ "loss": 2.4349,
1234
+ "step": 175
1235
+ },
1236
+ {
1237
+ "epoch": 0.189298198440441,
1238
+ "grad_norm": 3.0202341079711914,
1239
+ "learning_rate": 2e-05,
1240
+ "loss": 2.1977,
1241
+ "step": 176
1242
+ },
1243
+ {
1244
+ "epoch": 0.1903737563861253,
1245
+ "grad_norm": 2.346747398376465,
1246
+ "learning_rate": 2e-05,
1247
+ "loss": 2.399,
1248
+ "step": 177
1249
+ },
1250
+ {
1251
+ "epoch": 0.19144931433180962,
1252
+ "grad_norm": 1.5756202936172485,
1253
+ "learning_rate": 2e-05,
1254
+ "loss": 2.2689,
1255
+ "step": 178
1256
+ },
1257
+ {
1258
+ "epoch": 0.19252487227749396,
1259
+ "grad_norm": 2.398148775100708,
1260
+ "learning_rate": 2e-05,
1261
+ "loss": 3.8229,
1262
+ "step": 179
1263
+ },
1264
+ {
1265
+ "epoch": 0.19360043022317827,
1266
+ "grad_norm": 3.2819738388061523,
1267
+ "learning_rate": 2e-05,
1268
+ "loss": 2.6033,
1269
+ "step": 180
1270
+ },
1271
+ {
1272
+ "epoch": 0.19467598816886258,
1273
+ "grad_norm": 5.887080192565918,
1274
+ "learning_rate": 2e-05,
1275
+ "loss": 2.4956,
1276
+ "step": 181
1277
+ },
1278
+ {
1279
+ "epoch": 0.19575154611454693,
1280
+ "grad_norm": 2.208775520324707,
1281
+ "learning_rate": 2e-05,
1282
+ "loss": 2.1286,
1283
+ "step": 182
1284
+ },
1285
+ {
1286
+ "epoch": 0.19682710406023124,
1287
+ "grad_norm": 1.817888617515564,
1288
+ "learning_rate": 2e-05,
1289
+ "loss": 2.4608,
1290
+ "step": 183
1291
+ },
1292
+ {
1293
+ "epoch": 0.19790266200591558,
1294
+ "grad_norm": 2.150890588760376,
1295
+ "learning_rate": 2e-05,
1296
+ "loss": 3.2177,
1297
+ "step": 184
1298
+ },
1299
+ {
1300
+ "epoch": 0.1989782199515999,
1301
+ "grad_norm": 6.783865928649902,
1302
+ "learning_rate": 2e-05,
1303
+ "loss": 3.6292,
1304
+ "step": 185
1305
+ },
1306
+ {
1307
+ "epoch": 0.2000537778972842,
1308
+ "grad_norm": 3.8008947372436523,
1309
+ "learning_rate": 2e-05,
1310
+ "loss": 2.9182,
1311
+ "step": 186
1312
+ },
1313
+ {
1314
+ "epoch": 0.20112933584296855,
1315
+ "grad_norm": 4.612886905670166,
1316
+ "learning_rate": 2e-05,
1317
+ "loss": 3.1909,
1318
+ "step": 187
1319
+ },
1320
+ {
1321
+ "epoch": 0.20220489378865286,
1322
+ "grad_norm": 5.594083786010742,
1323
+ "learning_rate": 2e-05,
1324
+ "loss": 3.0986,
1325
+ "step": 188
1326
+ },
1327
+ {
1328
+ "epoch": 0.20328045173433718,
1329
+ "grad_norm": 3.358548402786255,
1330
+ "learning_rate": 2e-05,
1331
+ "loss": 2.2792,
1332
+ "step": 189
1333
+ },
1334
+ {
1335
+ "epoch": 0.20435600968002152,
1336
+ "grad_norm": 2.6911849975585938,
1337
+ "learning_rate": 2e-05,
1338
+ "loss": 3.4548,
1339
+ "step": 190
1340
+ },
1341
+ {
1342
+ "epoch": 0.20543156762570583,
1343
+ "grad_norm": 2.4795117378234863,
1344
+ "learning_rate": 2e-05,
1345
+ "loss": 2.3414,
1346
+ "step": 191
1347
+ },
1348
+ {
1349
+ "epoch": 0.20650712557139017,
1350
+ "grad_norm": 1.7064529657363892,
1351
+ "learning_rate": 2e-05,
1352
+ "loss": 1.8158,
1353
+ "step": 192
1354
+ },
1355
+ {
1356
+ "epoch": 0.20758268351707448,
1357
+ "grad_norm": 7.239320755004883,
1358
+ "learning_rate": 2e-05,
1359
+ "loss": 2.6676,
1360
+ "step": 193
1361
+ },
1362
+ {
1363
+ "epoch": 0.2086582414627588,
1364
+ "grad_norm": 4.747775077819824,
1365
+ "learning_rate": 2e-05,
1366
+ "loss": 3.2929,
1367
+ "step": 194
1368
+ },
1369
+ {
1370
+ "epoch": 0.20973379940844314,
1371
+ "grad_norm": 1.850651502609253,
1372
+ "learning_rate": 2e-05,
1373
+ "loss": 2.3005,
1374
+ "step": 195
1375
+ },
1376
+ {
1377
+ "epoch": 0.21080935735412745,
1378
+ "grad_norm": 2.4029853343963623,
1379
+ "learning_rate": 2e-05,
1380
+ "loss": 3.7213,
1381
+ "step": 196
1382
+ },
1383
+ {
1384
+ "epoch": 0.21188491529981177,
1385
+ "grad_norm": 2.5401084423065186,
1386
+ "learning_rate": 2e-05,
1387
+ "loss": 2.7665,
1388
+ "step": 197
1389
+ },
1390
+ {
1391
+ "epoch": 0.2129604732454961,
1392
+ "grad_norm": 4.080130577087402,
1393
+ "learning_rate": 2e-05,
1394
+ "loss": 1.8886,
1395
+ "step": 198
1396
+ },
1397
+ {
1398
+ "epoch": 0.21403603119118042,
1399
+ "grad_norm": 1.7884845733642578,
1400
+ "learning_rate": 2e-05,
1401
+ "loss": 2.4168,
1402
+ "step": 199
1403
+ },
1404
+ {
1405
+ "epoch": 0.21511158913686476,
1406
+ "grad_norm": 3.1292009353637695,
1407
+ "learning_rate": 2e-05,
1408
+ "loss": 2.4442,
1409
+ "step": 200
1410
+ },
1411
+ {
1412
+ "epoch": 0.21618714708254907,
1413
+ "grad_norm": 6.860455513000488,
1414
+ "learning_rate": 2e-05,
1415
+ "loss": 3.103,
1416
+ "step": 201
1417
+ },
1418
+ {
1419
+ "epoch": 0.2172627050282334,
1420
+ "grad_norm": 4.3930511474609375,
1421
+ "learning_rate": 2e-05,
1422
+ "loss": 2.7579,
1423
+ "step": 202
1424
+ },
1425
+ {
1426
+ "epoch": 0.21833826297391773,
1427
+ "grad_norm": 5.641942024230957,
1428
+ "learning_rate": 2e-05,
1429
+ "loss": 2.657,
1430
+ "step": 203
1431
+ },
1432
+ {
1433
+ "epoch": 0.21941382091960204,
1434
+ "grad_norm": 2.2956748008728027,
1435
+ "learning_rate": 2e-05,
1436
+ "loss": 2.2479,
1437
+ "step": 204
1438
+ },
1439
+ {
1440
+ "epoch": 0.22048937886528636,
1441
+ "grad_norm": 2.176769971847534,
1442
+ "learning_rate": 2e-05,
1443
+ "loss": 1.2909,
1444
+ "step": 205
1445
+ },
1446
+ {
1447
+ "epoch": 0.2215649368109707,
1448
+ "grad_norm": 4.874899864196777,
1449
+ "learning_rate": 2e-05,
1450
+ "loss": 1.9261,
1451
+ "step": 206
1452
+ },
1453
+ {
1454
+ "epoch": 0.222640494756655,
1455
+ "grad_norm": 2.68208909034729,
1456
+ "learning_rate": 2e-05,
1457
+ "loss": 2.1664,
1458
+ "step": 207
1459
+ },
1460
+ {
1461
+ "epoch": 0.22371605270233935,
1462
+ "grad_norm": 3.6076104640960693,
1463
+ "learning_rate": 2e-05,
1464
+ "loss": 2.8906,
1465
+ "step": 208
1466
+ },
1467
+ {
1468
+ "epoch": 0.22479161064802367,
1469
+ "grad_norm": 2.173996925354004,
1470
+ "learning_rate": 2e-05,
1471
+ "loss": 2.3881,
1472
+ "step": 209
1473
+ },
1474
+ {
1475
+ "epoch": 0.22586716859370798,
1476
+ "grad_norm": 2.750606060028076,
1477
+ "learning_rate": 2e-05,
1478
+ "loss": 2.1685,
1479
+ "step": 210
1480
+ },
1481
+ {
1482
+ "epoch": 0.22694272653939232,
1483
+ "grad_norm": 3.133878469467163,
1484
+ "learning_rate": 2e-05,
1485
+ "loss": 2.5526,
1486
+ "step": 211
1487
+ },
1488
+ {
1489
+ "epoch": 0.22801828448507663,
1490
+ "grad_norm": 3.0615320205688477,
1491
+ "learning_rate": 2e-05,
1492
+ "loss": 2.7754,
1493
+ "step": 212
1494
+ },
1495
+ {
1496
+ "epoch": 0.22909384243076095,
1497
+ "grad_norm": 2.5888783931732178,
1498
+ "learning_rate": 2e-05,
1499
+ "loss": 2.0269,
1500
+ "step": 213
1501
+ },
1502
+ {
1503
+ "epoch": 0.2301694003764453,
1504
+ "grad_norm": 3.7221732139587402,
1505
+ "learning_rate": 2e-05,
1506
+ "loss": 1.7132,
1507
+ "step": 214
1508
+ },
1509
+ {
1510
+ "epoch": 0.2312449583221296,
1511
+ "grad_norm": 2.1723954677581787,
1512
+ "learning_rate": 2e-05,
1513
+ "loss": 1.3366,
1514
+ "step": 215
1515
+ },
1516
+ {
1517
+ "epoch": 0.23232051626781391,
1518
+ "grad_norm": 2.335772752761841,
1519
+ "learning_rate": 2e-05,
1520
+ "loss": 2.6138,
1521
+ "step": 216
1522
+ },
1523
+ {
1524
+ "epoch": 0.23339607421349826,
1525
+ "grad_norm": 4.438052654266357,
1526
+ "learning_rate": 2e-05,
1527
+ "loss": 3.134,
1528
+ "step": 217
1529
+ },
1530
+ {
1531
+ "epoch": 0.23447163215918257,
1532
+ "grad_norm": 1.6511766910552979,
1533
+ "learning_rate": 2e-05,
1534
+ "loss": 2.2954,
1535
+ "step": 218
1536
+ },
1537
+ {
1538
+ "epoch": 0.2355471901048669,
1539
+ "grad_norm": 1.8855592012405396,
1540
+ "learning_rate": 2e-05,
1541
+ "loss": 1.7853,
1542
+ "step": 219
1543
+ },
1544
+ {
1545
+ "epoch": 0.23662274805055122,
1546
+ "grad_norm": 3.8127474784851074,
1547
+ "learning_rate": 2e-05,
1548
+ "loss": 1.8774,
1549
+ "step": 220
1550
+ },
1551
+ {
1552
+ "epoch": 0.23769830599623554,
1553
+ "grad_norm": 5.802454471588135,
1554
+ "learning_rate": 2e-05,
1555
+ "loss": 3.2977,
1556
+ "step": 221
1557
+ },
1558
+ {
1559
+ "epoch": 0.23877386394191988,
1560
+ "grad_norm": 3.8824198246002197,
1561
+ "learning_rate": 2e-05,
1562
+ "loss": 2.715,
1563
+ "step": 222
1564
+ },
1565
+ {
1566
+ "epoch": 0.2398494218876042,
1567
+ "grad_norm": 3.3041539192199707,
1568
+ "learning_rate": 2e-05,
1569
+ "loss": 2.4187,
1570
+ "step": 223
1571
+ },
1572
+ {
1573
+ "epoch": 0.2409249798332885,
1574
+ "grad_norm": 2.7080237865448,
1575
+ "learning_rate": 2e-05,
1576
+ "loss": 2.0946,
1577
+ "step": 224
1578
+ },
1579
+ {
1580
+ "epoch": 0.24200053777897285,
1581
+ "grad_norm": 2.8399364948272705,
1582
+ "learning_rate": 2e-05,
1583
+ "loss": 1.994,
1584
+ "step": 225
1585
+ },
1586
+ {
1587
+ "epoch": 0.24307609572465716,
1588
+ "grad_norm": 2.896878480911255,
1589
+ "learning_rate": 2e-05,
1590
+ "loss": 2.2411,
1591
+ "step": 226
1592
+ },
1593
+ {
1594
+ "epoch": 0.2441516536703415,
1595
+ "grad_norm": 2.4240972995758057,
1596
+ "learning_rate": 2e-05,
1597
+ "loss": 2.5445,
1598
+ "step": 227
1599
+ },
1600
+ {
1601
+ "epoch": 0.24522721161602581,
1602
+ "grad_norm": 2.8553824424743652,
1603
+ "learning_rate": 2e-05,
1604
+ "loss": 2.0291,
1605
+ "step": 228
1606
+ },
1607
+ {
1608
+ "epoch": 0.24630276956171013,
1609
+ "grad_norm": 3.546189069747925,
1610
+ "learning_rate": 2e-05,
1611
+ "loss": 2.9868,
1612
+ "step": 229
1613
+ },
1614
+ {
1615
+ "epoch": 0.24737832750739447,
1616
+ "grad_norm": 3.4419355392456055,
1617
+ "learning_rate": 2e-05,
1618
+ "loss": 1.7061,
1619
+ "step": 230
1620
+ },
1621
+ {
1622
+ "epoch": 0.24845388545307878,
1623
+ "grad_norm": 2.98702073097229,
1624
+ "learning_rate": 2e-05,
1625
+ "loss": 2.3165,
1626
+ "step": 231
1627
+ },
1628
+ {
1629
+ "epoch": 0.2495294433987631,
1630
+ "grad_norm": 4.188878536224365,
1631
+ "learning_rate": 2e-05,
1632
+ "loss": 1.9139,
1633
+ "step": 232
1634
+ },
1635
+ {
1636
+ "epoch": 0.2506050013444474,
1637
+ "grad_norm": 3.832674026489258,
1638
+ "learning_rate": 2e-05,
1639
+ "loss": 2.1851,
1640
+ "step": 233
1641
+ },
1642
+ {
1643
+ "epoch": 0.2516805592901318,
1644
+ "grad_norm": 2.5292224884033203,
1645
+ "learning_rate": 2e-05,
1646
+ "loss": 1.395,
1647
+ "step": 234
1648
+ },
1649
+ {
1650
+ "epoch": 0.2527561172358161,
1651
+ "grad_norm": 5.447204113006592,
1652
+ "learning_rate": 2e-05,
1653
+ "loss": 1.6947,
1654
+ "step": 235
1655
+ },
1656
+ {
1657
+ "epoch": 0.2538316751815004,
1658
+ "grad_norm": 5.222197532653809,
1659
+ "learning_rate": 2e-05,
1660
+ "loss": 2.7788,
1661
+ "step": 236
1662
+ },
1663
+ {
1664
+ "epoch": 0.2549072331271847,
1665
+ "grad_norm": 6.383084297180176,
1666
+ "learning_rate": 2e-05,
1667
+ "loss": 2.7386,
1668
+ "step": 237
1669
+ },
1670
+ {
1671
+ "epoch": 0.25598279107286903,
1672
+ "grad_norm": 2.2934720516204834,
1673
+ "learning_rate": 2e-05,
1674
+ "loss": 2.5268,
1675
+ "step": 238
1676
+ },
1677
+ {
1678
+ "epoch": 0.2570583490185534,
1679
+ "grad_norm": 2.0854413509368896,
1680
+ "learning_rate": 2e-05,
1681
+ "loss": 1.5054,
1682
+ "step": 239
1683
+ },
1684
+ {
1685
+ "epoch": 0.2581339069642377,
1686
+ "grad_norm": 3.8236308097839355,
1687
+ "learning_rate": 2e-05,
1688
+ "loss": 2.2114,
1689
+ "step": 240
1690
+ },
1691
+ {
1692
+ "epoch": 0.259209464909922,
1693
+ "grad_norm": 3.419747829437256,
1694
+ "learning_rate": 2e-05,
1695
+ "loss": 1.7876,
1696
+ "step": 241
1697
+ },
1698
+ {
1699
+ "epoch": 0.26028502285560634,
1700
+ "grad_norm": 3.672929048538208,
1701
+ "learning_rate": 2e-05,
1702
+ "loss": 1.9878,
1703
+ "step": 242
1704
+ },
1705
+ {
1706
+ "epoch": 0.26136058080129065,
1707
+ "grad_norm": 1.8193111419677734,
1708
+ "learning_rate": 2e-05,
1709
+ "loss": 1.7618,
1710
+ "step": 243
1711
+ },
1712
+ {
1713
+ "epoch": 0.26243613874697497,
1714
+ "grad_norm": 7.668798446655273,
1715
+ "learning_rate": 2e-05,
1716
+ "loss": 1.8264,
1717
+ "step": 244
1718
+ },
1719
+ {
1720
+ "epoch": 0.26351169669265934,
1721
+ "grad_norm": 4.639315128326416,
1722
+ "learning_rate": 2e-05,
1723
+ "loss": 2.9063,
1724
+ "step": 245
1725
+ },
1726
+ {
1727
+ "epoch": 0.26458725463834365,
1728
+ "grad_norm": 2.3293251991271973,
1729
+ "learning_rate": 2e-05,
1730
+ "loss": 1.2018,
1731
+ "step": 246
1732
+ },
1733
+ {
1734
+ "epoch": 0.26566281258402796,
1735
+ "grad_norm": 4.38407564163208,
1736
+ "learning_rate": 2e-05,
1737
+ "loss": 1.6128,
1738
+ "step": 247
1739
+ },
1740
+ {
1741
+ "epoch": 0.2667383705297123,
1742
+ "grad_norm": 1.7634291648864746,
1743
+ "learning_rate": 2e-05,
1744
+ "loss": 1.4277,
1745
+ "step": 248
1746
+ },
1747
+ {
1748
+ "epoch": 0.2678139284753966,
1749
+ "grad_norm": 3.81402325630188,
1750
+ "learning_rate": 2e-05,
1751
+ "loss": 2.0313,
1752
+ "step": 249
1753
+ },
1754
+ {
1755
+ "epoch": 0.26888948642108096,
1756
+ "grad_norm": 2.318251848220825,
1757
+ "learning_rate": 2e-05,
1758
+ "loss": 1.7833,
1759
+ "step": 250
1760
+ },
1761
+ {
1762
+ "epoch": 0.26996504436676527,
1763
+ "grad_norm": 1.8340922594070435,
1764
+ "learning_rate": 2e-05,
1765
+ "loss": 1.6557,
1766
+ "step": 251
1767
+ },
1768
+ {
1769
+ "epoch": 0.2710406023124496,
1770
+ "grad_norm": 4.6209845542907715,
1771
+ "learning_rate": 2e-05,
1772
+ "loss": 2.166,
1773
+ "step": 252
1774
+ },
1775
+ {
1776
+ "epoch": 0.2721161602581339,
1777
+ "grad_norm": 3.370737075805664,
1778
+ "learning_rate": 2e-05,
1779
+ "loss": 1.9072,
1780
+ "step": 253
1781
+ },
1782
+ {
1783
+ "epoch": 0.2731917182038182,
1784
+ "grad_norm": 4.559715747833252,
1785
+ "learning_rate": 2e-05,
1786
+ "loss": 1.7178,
1787
+ "step": 254
1788
+ },
1789
+ {
1790
+ "epoch": 0.2742672761495026,
1791
+ "grad_norm": 4.001542568206787,
1792
+ "learning_rate": 2e-05,
1793
+ "loss": 1.4689,
1794
+ "step": 255
1795
+ },
1796
+ {
1797
+ "epoch": 0.2753428340951869,
1798
+ "grad_norm": 2.454691171646118,
1799
+ "learning_rate": 2e-05,
1800
+ "loss": 1.4426,
1801
+ "step": 256
1802
+ },
1803
+ {
1804
+ "epoch": 0.2764183920408712,
1805
+ "grad_norm": 2.3479011058807373,
1806
+ "learning_rate": 2e-05,
1807
+ "loss": 2.0009,
1808
+ "step": 257
1809
+ },
1810
+ {
1811
+ "epoch": 0.2774939499865555,
1812
+ "grad_norm": 5.103316307067871,
1813
+ "learning_rate": 2e-05,
1814
+ "loss": 1.9592,
1815
+ "step": 258
1816
+ },
1817
+ {
1818
+ "epoch": 0.27856950793223983,
1819
+ "grad_norm": 2.4349985122680664,
1820
+ "learning_rate": 2e-05,
1821
+ "loss": 2.0947,
1822
+ "step": 259
1823
+ },
1824
+ {
1825
+ "epoch": 0.27964506587792415,
1826
+ "grad_norm": 2.940519332885742,
1827
+ "learning_rate": 2e-05,
1828
+ "loss": 1.9484,
1829
+ "step": 260
1830
+ },
1831
+ {
1832
+ "epoch": 0.2807206238236085,
1833
+ "grad_norm": 4.070859432220459,
1834
+ "learning_rate": 2e-05,
1835
+ "loss": 2.1733,
1836
+ "step": 261
1837
+ },
1838
+ {
1839
+ "epoch": 0.28179618176929283,
1840
+ "grad_norm": 5.868851184844971,
1841
+ "learning_rate": 2e-05,
1842
+ "loss": 1.6656,
1843
+ "step": 262
1844
+ },
1845
+ {
1846
+ "epoch": 0.28287173971497714,
1847
+ "grad_norm": 4.625424861907959,
1848
+ "learning_rate": 2e-05,
1849
+ "loss": 2.3667,
1850
+ "step": 263
1851
+ },
1852
+ {
1853
+ "epoch": 0.28394729766066146,
1854
+ "grad_norm": 2.5272748470306396,
1855
+ "learning_rate": 2e-05,
1856
+ "loss": 2.1122,
1857
+ "step": 264
1858
+ },
1859
+ {
1860
+ "epoch": 0.28502285560634577,
1861
+ "grad_norm": 5.193785190582275,
1862
+ "learning_rate": 2e-05,
1863
+ "loss": 1.5063,
1864
+ "step": 265
1865
+ },
1866
+ {
1867
+ "epoch": 0.28609841355203014,
1868
+ "grad_norm": 1.6589148044586182,
1869
+ "learning_rate": 2e-05,
1870
+ "loss": 1.9044,
1871
+ "step": 266
1872
+ },
1873
+ {
1874
+ "epoch": 0.28717397149771445,
1875
+ "grad_norm": 3.423194408416748,
1876
+ "learning_rate": 2e-05,
1877
+ "loss": 2.2866,
1878
+ "step": 267
1879
+ },
1880
+ {
1881
+ "epoch": 0.28824952944339877,
1882
+ "grad_norm": 2.188795328140259,
1883
+ "learning_rate": 2e-05,
1884
+ "loss": 1.6161,
1885
+ "step": 268
1886
+ },
1887
+ {
1888
+ "epoch": 0.2893250873890831,
1889
+ "grad_norm": 3.441420078277588,
1890
+ "learning_rate": 2e-05,
1891
+ "loss": 1.42,
1892
+ "step": 269
1893
+ },
1894
+ {
1895
+ "epoch": 0.2904006453347674,
1896
+ "grad_norm": 4.066496849060059,
1897
+ "learning_rate": 2e-05,
1898
+ "loss": 1.9209,
1899
+ "step": 270
1900
+ },
1901
+ {
1902
+ "epoch": 0.2914762032804517,
1903
+ "grad_norm": 4.315361022949219,
1904
+ "learning_rate": 2e-05,
1905
+ "loss": 1.433,
1906
+ "step": 271
1907
+ },
1908
+ {
1909
+ "epoch": 0.2925517612261361,
1910
+ "grad_norm": 2.7533764839172363,
1911
+ "learning_rate": 2e-05,
1912
+ "loss": 1.3967,
1913
+ "step": 272
1914
+ },
1915
+ {
1916
+ "epoch": 0.2936273191718204,
1917
+ "grad_norm": 2.4276041984558105,
1918
+ "learning_rate": 2e-05,
1919
+ "loss": 2.5529,
1920
+ "step": 273
1921
+ },
1922
+ {
1923
+ "epoch": 0.2947028771175047,
1924
+ "grad_norm": 2.0695526599884033,
1925
+ "learning_rate": 2e-05,
1926
+ "loss": 1.7445,
1927
+ "step": 274
1928
+ },
1929
+ {
1930
+ "epoch": 0.295778435063189,
1931
+ "grad_norm": 3.037795066833496,
1932
+ "learning_rate": 2e-05,
1933
+ "loss": 2.4438,
1934
+ "step": 275
1935
+ },
1936
+ {
1937
+ "epoch": 0.29685399300887333,
1938
+ "grad_norm": 3.6190900802612305,
1939
+ "learning_rate": 2e-05,
1940
+ "loss": 1.6739,
1941
+ "step": 276
1942
+ },
1943
+ {
1944
+ "epoch": 0.2979295509545577,
1945
+ "grad_norm": 2.2492079734802246,
1946
+ "learning_rate": 2e-05,
1947
+ "loss": 1.4227,
1948
+ "step": 277
1949
+ },
1950
+ {
1951
+ "epoch": 0.299005108900242,
1952
+ "grad_norm": 2.544053554534912,
1953
+ "learning_rate": 2e-05,
1954
+ "loss": 2.1751,
1955
+ "step": 278
1956
+ },
1957
+ {
1958
+ "epoch": 0.3000806668459263,
1959
+ "grad_norm": 2.0408968925476074,
1960
+ "learning_rate": 2e-05,
1961
+ "loss": 1.6613,
1962
+ "step": 279
1963
+ },
1964
+ {
1965
+ "epoch": 0.30115622479161064,
1966
+ "grad_norm": 4.337264537811279,
1967
+ "learning_rate": 2e-05,
1968
+ "loss": 1.6372,
1969
+ "step": 280
1970
+ },
1971
+ {
1972
+ "epoch": 0.30223178273729495,
1973
+ "grad_norm": 4.142371654510498,
1974
+ "learning_rate": 2e-05,
1975
+ "loss": 1.4561,
1976
+ "step": 281
1977
+ },
1978
+ {
1979
+ "epoch": 0.3033073406829793,
1980
+ "grad_norm": 3.091535806655884,
1981
+ "learning_rate": 2e-05,
1982
+ "loss": 2.024,
1983
+ "step": 282
1984
+ },
1985
+ {
1986
+ "epoch": 0.30438289862866363,
1987
+ "grad_norm": 5.492683410644531,
1988
+ "learning_rate": 2e-05,
1989
+ "loss": 1.6435,
1990
+ "step": 283
1991
+ },
1992
+ {
1993
+ "epoch": 0.30545845657434795,
1994
+ "grad_norm": 6.648707389831543,
1995
+ "learning_rate": 2e-05,
1996
+ "loss": 2.6662,
1997
+ "step": 284
1998
+ },
1999
+ {
2000
+ "epoch": 0.30653401452003226,
2001
+ "grad_norm": 2.870140790939331,
2002
+ "learning_rate": 2e-05,
2003
+ "loss": 1.2809,
2004
+ "step": 285
2005
+ },
2006
+ {
2007
+ "epoch": 0.3076095724657166,
2008
+ "grad_norm": 1.6203242540359497,
2009
+ "learning_rate": 2e-05,
2010
+ "loss": 1.6137,
2011
+ "step": 286
2012
+ },
2013
+ {
2014
+ "epoch": 0.3086851304114009,
2015
+ "grad_norm": 3.1347341537475586,
2016
+ "learning_rate": 2e-05,
2017
+ "loss": 1.0086,
2018
+ "step": 287
2019
+ },
2020
+ {
2021
+ "epoch": 0.30976068835708526,
2022
+ "grad_norm": 1.8718631267547607,
2023
+ "learning_rate": 2e-05,
2024
+ "loss": 1.6824,
2025
+ "step": 288
2026
+ },
2027
+ {
2028
+ "epoch": 0.31083624630276957,
2029
+ "grad_norm": 3.668214797973633,
2030
+ "learning_rate": 2e-05,
2031
+ "loss": 2.3531,
2032
+ "step": 289
2033
+ },
2034
+ {
2035
+ "epoch": 0.3119118042484539,
2036
+ "grad_norm": 1.985969066619873,
2037
+ "learning_rate": 2e-05,
2038
+ "loss": 1.526,
2039
+ "step": 290
2040
+ },
2041
+ {
2042
+ "epoch": 0.3129873621941382,
2043
+ "grad_norm": 6.0746846199035645,
2044
+ "learning_rate": 2e-05,
2045
+ "loss": 2.3411,
2046
+ "step": 291
2047
+ },
2048
+ {
2049
+ "epoch": 0.3140629201398225,
2050
+ "grad_norm": 2.04298734664917,
2051
+ "learning_rate": 2e-05,
2052
+ "loss": 1.6556,
2053
+ "step": 292
2054
+ },
2055
+ {
2056
+ "epoch": 0.3151384780855069,
2057
+ "grad_norm": 4.120944499969482,
2058
+ "learning_rate": 2e-05,
2059
+ "loss": 1.788,
2060
+ "step": 293
2061
+ },
2062
+ {
2063
+ "epoch": 0.3162140360311912,
2064
+ "grad_norm": 5.191648006439209,
2065
+ "learning_rate": 2e-05,
2066
+ "loss": 1.6943,
2067
+ "step": 294
2068
+ },
2069
+ {
2070
+ "epoch": 0.3172895939768755,
2071
+ "grad_norm": 1.5654542446136475,
2072
+ "learning_rate": 2e-05,
2073
+ "loss": 1.7853,
2074
+ "step": 295
2075
+ },
2076
+ {
2077
+ "epoch": 0.3183651519225598,
2078
+ "grad_norm": 1.9325037002563477,
2079
+ "learning_rate": 2e-05,
2080
+ "loss": 1.4699,
2081
+ "step": 296
2082
+ },
2083
+ {
2084
+ "epoch": 0.31944070986824413,
2085
+ "grad_norm": 5.055520534515381,
2086
+ "learning_rate": 2e-05,
2087
+ "loss": 1.9544,
2088
+ "step": 297
2089
+ },
2090
+ {
2091
+ "epoch": 0.3205162678139285,
2092
+ "grad_norm": 3.137733221054077,
2093
+ "learning_rate": 2e-05,
2094
+ "loss": 2.2007,
2095
+ "step": 298
2096
+ },
2097
+ {
2098
+ "epoch": 0.3215918257596128,
2099
+ "grad_norm": 2.723611354827881,
2100
+ "learning_rate": 2e-05,
2101
+ "loss": 1.7762,
2102
+ "step": 299
2103
+ },
2104
+ {
2105
+ "epoch": 0.32266738370529713,
2106
+ "grad_norm": 2.2930235862731934,
2107
+ "learning_rate": 2e-05,
2108
+ "loss": 1.1551,
2109
+ "step": 300
2110
+ }
2111
+ ],
2112
+ "logging_steps": 1.0,
2113
+ "max_steps": 465,
2114
+ "num_input_tokens_seen": 0,
2115
+ "num_train_epochs": 1,
2116
+ "save_steps": 100,
2117
+ "stateful_callbacks": {
2118
+ "TrainerControl": {
2119
+ "args": {
2120
+ "should_epoch_stop": false,
2121
+ "should_evaluate": false,
2122
+ "should_log": false,
2123
+ "should_save": true,
2124
+ "should_training_stop": false
2125
+ },
2126
+ "attributes": {}
2127
+ }
2128
+ },
2129
+ "total_flos": 2769964824576.0,
2130
+ "train_batch_size": 4,
2131
+ "trial_name": null,
2132
+ "trial_params": null
2133
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-300/zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/adapter_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 16,
14
+ "lora_dropout": 0.05,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": null,
18
+ "peft_type": "LORA",
19
+ "r": 8,
20
+ "rank_pattern": {},
21
+ "revision": null,
22
+ "target_modules": [
23
+ "mlp.2",
24
+ "mlp.0"
25
+ ],
26
+ "task_type": "CAUSAL_LM",
27
+ "use_dora": false,
28
+ "use_rslora": false
29
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/added_tokens.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</tool_call>": 151658,
3
+ "<tool_call>": 151657,
4
+ "<|box_end|>": 151649,
5
+ "<|box_start|>": 151648,
6
+ "<|endoftext|>": 151643,
7
+ "<|file_sep|>": 151664,
8
+ "<|fim_middle|>": 151660,
9
+ "<|fim_pad|>": 151662,
10
+ "<|fim_prefix|>": 151659,
11
+ "<|fim_suffix|>": 151661,
12
+ "<|im_end|>": 151645,
13
+ "<|im_start|>": 151644,
14
+ "<|image_pad|>": 151655,
15
+ "<|object_ref_end|>": 151647,
16
+ "<|object_ref_start|>": 151646,
17
+ "<|quad_end|>": 151651,
18
+ "<|quad_start|>": 151650,
19
+ "<|repo_name|>": 151663,
20
+ "<|video_pad|>": 151656,
21
+ "<|vision_end|>": 151653,
22
+ "<|vision_pad|>": 151654,
23
+ "<|vision_start|>": 151652
24
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/config.json ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen2_5_VLForConditionalGeneration"
4
+ ],
5
+ "attention_dropout": 0.0,
6
+ "bos_token_id": 151643,
7
+ "eos_token_id": 151645,
8
+ "vision_start_token_id": 151652,
9
+ "vision_end_token_id": 151653,
10
+ "vision_token_id": 151654,
11
+ "image_token_id": 151655,
12
+ "video_token_id": 151656,
13
+ "hidden_act": "silu",
14
+ "hidden_size": 3584,
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 18944,
17
+ "max_position_embeddings": 128000,
18
+ "max_window_layers": 28,
19
+ "model_type": "qwen2_5_vl",
20
+ "num_attention_heads": 28,
21
+ "num_hidden_layers": 28,
22
+ "num_key_value_heads": 4,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_theta": 1000000.0,
25
+ "sliding_window": 32768,
26
+ "tie_word_embeddings": false,
27
+ "torch_dtype": "bfloat16",
28
+ "transformers_version": "4.41.2",
29
+ "use_cache": true,
30
+ "use_sliding_window": false,
31
+ "vision_config": {
32
+ "depth": 32,
33
+ "hidden_act": "silu",
34
+ "hidden_size": 1280,
35
+ "intermediate_size": 3420,
36
+ "num_heads": 16,
37
+ "in_chans": 3,
38
+ "out_hidden_size": 3584,
39
+ "patch_size": 14,
40
+ "spatial_merge_size": 2,
41
+ "spatial_patch_size": 14,
42
+ "window_size": 112,
43
+ "fullatt_block_indexes": [
44
+ 7,
45
+ 15,
46
+ 23,
47
+ 31
48
+ ],
49
+ "tokens_per_second": 2,
50
+ "temporal_patch_size": 2
51
+ },
52
+ "rope_scaling": {
53
+ "type": "mrope",
54
+ "mrope_section": [
55
+ 16,
56
+ 24,
57
+ 24
58
+ ]
59
+ },
60
+ "vocab_size": 152064
61
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step400
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/preprocessor_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "min_pixels": 3136,
3
+ "max_pixels": 12845056,
4
+ "patch_size": 14,
5
+ "temporal_patch_size": 2,
6
+ "merge_size": 2,
7
+ "image_mean": [
8
+ 0.48145466,
9
+ 0.4578275,
10
+ 0.40821073
11
+ ],
12
+ "image_std": [
13
+ 0.26862954,
14
+ 0.26130258,
15
+ 0.27577711
16
+ ],
17
+ "image_processor_type": "Qwen2VLImageProcessor",
18
+ "processor_class": "Qwen2_5_VLProcessor"
19
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/tokenizer_config.json ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "151643": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "151644": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "151645": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "151646": {
29
+ "content": "<|object_ref_start|>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "151647": {
37
+ "content": "<|object_ref_end|>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "151648": {
45
+ "content": "<|box_start|>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "151649": {
53
+ "content": "<|box_end|>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "151650": {
61
+ "content": "<|quad_start|>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "151651": {
69
+ "content": "<|quad_end|>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "151652": {
77
+ "content": "<|vision_start|>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "151653": {
85
+ "content": "<|vision_end|>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "151654": {
93
+ "content": "<|vision_pad|>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "151655": {
101
+ "content": "<|image_pad|>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "151656": {
109
+ "content": "<|video_pad|>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "151657": {
117
+ "content": "<tool_call>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": false
123
+ },
124
+ "151658": {
125
+ "content": "</tool_call>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": false
131
+ },
132
+ "151659": {
133
+ "content": "<|fim_prefix|>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": false
139
+ },
140
+ "151660": {
141
+ "content": "<|fim_middle|>",
142
+ "lstrip": false,
143
+ "normalized": false,
144
+ "rstrip": false,
145
+ "single_word": false,
146
+ "special": false
147
+ },
148
+ "151661": {
149
+ "content": "<|fim_suffix|>",
150
+ "lstrip": false,
151
+ "normalized": false,
152
+ "rstrip": false,
153
+ "single_word": false,
154
+ "special": false
155
+ },
156
+ "151662": {
157
+ "content": "<|fim_pad|>",
158
+ "lstrip": false,
159
+ "normalized": false,
160
+ "rstrip": false,
161
+ "single_word": false,
162
+ "special": false
163
+ },
164
+ "151663": {
165
+ "content": "<|repo_name|>",
166
+ "lstrip": false,
167
+ "normalized": false,
168
+ "rstrip": false,
169
+ "single_word": false,
170
+ "special": false
171
+ },
172
+ "151664": {
173
+ "content": "<|file_sep|>",
174
+ "lstrip": false,
175
+ "normalized": false,
176
+ "rstrip": false,
177
+ "single_word": false,
178
+ "special": false
179
+ }
180
+ },
181
+ "additional_special_tokens": [
182
+ "<|im_start|>",
183
+ "<|im_end|>",
184
+ "<|object_ref_start|>",
185
+ "<|object_ref_end|>",
186
+ "<|box_start|>",
187
+ "<|box_end|>",
188
+ "<|quad_start|>",
189
+ "<|quad_end|>",
190
+ "<|vision_start|>",
191
+ "<|vision_end|>",
192
+ "<|vision_pad|>",
193
+ "<|image_pad|>",
194
+ "<|video_pad|>"
195
+ ],
196
+ "bos_token": null,
197
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
198
+ "clean_up_tokenization_spaces": false,
199
+ "eos_token": "<|im_end|>",
200
+ "errors": "replace",
201
+ "model_max_length": 131072,
202
+ "pad_token": "<|endoftext|>",
203
+ "split_special_tokens": false,
204
+ "tokenizer_class": "Qwen2Tokenizer",
205
+ "unk_token": null,
206
+ "add_bos_token": false
207
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/trainer_state.json ADDED
@@ -0,0 +1,2833 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.4302231782737295,
5
+ "eval_steps": 500,
6
+ "global_step": 400,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0010755579456843238,
13
+ "grad_norm": 0.08796706795692444,
14
+ "learning_rate": 0.0,
15
+ "loss": 3.92,
16
+ "step": 1
17
+ },
18
+ {
19
+ "epoch": 0.0021511158913686476,
20
+ "grad_norm": 0.03759637102484703,
21
+ "learning_rate": 5.252990700743872e-06,
22
+ "loss": 2.2035,
23
+ "step": 2
24
+ },
25
+ {
26
+ "epoch": 0.0032266738370529714,
27
+ "grad_norm": 0.08365879952907562,
28
+ "learning_rate": 8.325793277315987e-06,
29
+ "loss": 4.949,
30
+ "step": 3
31
+ },
32
+ {
33
+ "epoch": 0.004302231782737295,
34
+ "grad_norm": 0.08474036306142807,
35
+ "learning_rate": 1.0505981401487744e-05,
36
+ "loss": 4.1587,
37
+ "step": 4
38
+ },
39
+ {
40
+ "epoch": 0.005377789728421619,
41
+ "grad_norm": 0.04265577346086502,
42
+ "learning_rate": 1.2197066690239248e-05,
43
+ "loss": 2.5989,
44
+ "step": 5
45
+ },
46
+ {
47
+ "epoch": 0.006453347674105943,
48
+ "grad_norm": 0.12465030699968338,
49
+ "learning_rate": 1.3578783978059857e-05,
50
+ "loss": 3.7898,
51
+ "step": 6
52
+ },
53
+ {
54
+ "epoch": 0.007528905619790266,
55
+ "grad_norm": 0.07569626718759537,
56
+ "learning_rate": 1.4747009299256131e-05,
57
+ "loss": 4.4848,
58
+ "step": 7
59
+ },
60
+ {
61
+ "epoch": 0.00860446356547459,
62
+ "grad_norm": 0.047666411846876144,
63
+ "learning_rate": 1.5758972102231614e-05,
64
+ "loss": 4.0286,
65
+ "step": 8
66
+ },
67
+ {
68
+ "epoch": 0.009680021511158914,
69
+ "grad_norm": 0.06569754332304001,
70
+ "learning_rate": 1.6651586554631973e-05,
71
+ "loss": 4.8475,
72
+ "step": 9
73
+ },
74
+ {
75
+ "epoch": 0.010755579456843238,
76
+ "grad_norm": 0.08362340182065964,
77
+ "learning_rate": 1.745005739098312e-05,
78
+ "loss": 4.1507,
79
+ "step": 10
80
+ },
81
+ {
82
+ "epoch": 0.011831137402527562,
83
+ "grad_norm": 0.0652509331703186,
84
+ "learning_rate": 1.8172362122561044e-05,
85
+ "loss": 3.1503,
86
+ "step": 11
87
+ },
88
+ {
89
+ "epoch": 0.012906695348211886,
90
+ "grad_norm": 0.07024404406547546,
91
+ "learning_rate": 1.883177467880373e-05,
92
+ "loss": 4.5793,
93
+ "step": 12
94
+ },
95
+ {
96
+ "epoch": 0.01398225329389621,
97
+ "grad_norm": 0.08304284512996674,
98
+ "learning_rate": 1.9438375428058432e-05,
99
+ "loss": 2.2916,
100
+ "step": 13
101
+ },
102
+ {
103
+ "epoch": 0.015057811239580532,
104
+ "grad_norm": 0.09187567979097366,
105
+ "learning_rate": 2e-05,
106
+ "loss": 4.3406,
107
+ "step": 14
108
+ },
109
+ {
110
+ "epoch": 0.016133369185264857,
111
+ "grad_norm": 0.09281238913536072,
112
+ "learning_rate": 2e-05,
113
+ "loss": 3.9966,
114
+ "step": 15
115
+ },
116
+ {
117
+ "epoch": 0.01720892713094918,
118
+ "grad_norm": 0.06343083083629608,
119
+ "learning_rate": 2e-05,
120
+ "loss": 3.1267,
121
+ "step": 16
122
+ },
123
+ {
124
+ "epoch": 0.018284485076633505,
125
+ "grad_norm": 0.05553126335144043,
126
+ "learning_rate": 2e-05,
127
+ "loss": 2.6709,
128
+ "step": 17
129
+ },
130
+ {
131
+ "epoch": 0.01936004302231783,
132
+ "grad_norm": 0.07925406098365784,
133
+ "learning_rate": 2e-05,
134
+ "loss": 3.0051,
135
+ "step": 18
136
+ },
137
+ {
138
+ "epoch": 0.020435600968002152,
139
+ "grad_norm": 0.07598422467708588,
140
+ "learning_rate": 2e-05,
141
+ "loss": 2.5423,
142
+ "step": 19
143
+ },
144
+ {
145
+ "epoch": 0.021511158913686476,
146
+ "grad_norm": 0.11351823806762695,
147
+ "learning_rate": 2e-05,
148
+ "loss": 4.2694,
149
+ "step": 20
150
+ },
151
+ {
152
+ "epoch": 0.0225867168593708,
153
+ "grad_norm": 0.07363948971033096,
154
+ "learning_rate": 2e-05,
155
+ "loss": 3.2626,
156
+ "step": 21
157
+ },
158
+ {
159
+ "epoch": 0.023662274805055124,
160
+ "grad_norm": 0.06679286062717438,
161
+ "learning_rate": 2e-05,
162
+ "loss": 3.476,
163
+ "step": 22
164
+ },
165
+ {
166
+ "epoch": 0.024737832750739448,
167
+ "grad_norm": 0.0646878108382225,
168
+ "learning_rate": 2e-05,
169
+ "loss": 2.6229,
170
+ "step": 23
171
+ },
172
+ {
173
+ "epoch": 0.02581339069642377,
174
+ "grad_norm": 0.14322379231452942,
175
+ "learning_rate": 2e-05,
176
+ "loss": 4.1941,
177
+ "step": 24
178
+ },
179
+ {
180
+ "epoch": 0.026888948642108095,
181
+ "grad_norm": 0.11759044229984283,
182
+ "learning_rate": 2e-05,
183
+ "loss": 4.4217,
184
+ "step": 25
185
+ },
186
+ {
187
+ "epoch": 0.02796450658779242,
188
+ "grad_norm": 0.11527741700410843,
189
+ "learning_rate": 2e-05,
190
+ "loss": 4.2243,
191
+ "step": 26
192
+ },
193
+ {
194
+ "epoch": 0.02904006453347674,
195
+ "grad_norm": 0.06782741844654083,
196
+ "learning_rate": 2e-05,
197
+ "loss": 2.8814,
198
+ "step": 27
199
+ },
200
+ {
201
+ "epoch": 0.030115622479161063,
202
+ "grad_norm": 0.1451217532157898,
203
+ "learning_rate": 2e-05,
204
+ "loss": 3.0934,
205
+ "step": 28
206
+ },
207
+ {
208
+ "epoch": 0.031191180424845387,
209
+ "grad_norm": 0.11510222405195236,
210
+ "learning_rate": 2e-05,
211
+ "loss": 2.9115,
212
+ "step": 29
213
+ },
214
+ {
215
+ "epoch": 0.032266738370529714,
216
+ "grad_norm": 0.14234976470470428,
217
+ "learning_rate": 2e-05,
218
+ "loss": 4.2921,
219
+ "step": 30
220
+ },
221
+ {
222
+ "epoch": 0.033342296316214035,
223
+ "grad_norm": 0.12387730926275253,
224
+ "learning_rate": 2e-05,
225
+ "loss": 3.8533,
226
+ "step": 31
227
+ },
228
+ {
229
+ "epoch": 0.03441785426189836,
230
+ "grad_norm": 0.08505982905626297,
231
+ "learning_rate": 2e-05,
232
+ "loss": 2.293,
233
+ "step": 32
234
+ },
235
+ {
236
+ "epoch": 0.03549341220758268,
237
+ "grad_norm": 0.1067776158452034,
238
+ "learning_rate": 2e-05,
239
+ "loss": 3.5162,
240
+ "step": 33
241
+ },
242
+ {
243
+ "epoch": 0.03656897015326701,
244
+ "grad_norm": 0.1792977899312973,
245
+ "learning_rate": 2e-05,
246
+ "loss": 5.5227,
247
+ "step": 34
248
+ },
249
+ {
250
+ "epoch": 0.03764452809895133,
251
+ "grad_norm": 0.1272369921207428,
252
+ "learning_rate": 2e-05,
253
+ "loss": 3.1357,
254
+ "step": 35
255
+ },
256
+ {
257
+ "epoch": 0.03872008604463566,
258
+ "grad_norm": 0.19525307416915894,
259
+ "learning_rate": 2e-05,
260
+ "loss": 3.4933,
261
+ "step": 36
262
+ },
263
+ {
264
+ "epoch": 0.03979564399031998,
265
+ "grad_norm": 0.15390488505363464,
266
+ "learning_rate": 2e-05,
267
+ "loss": 3.9449,
268
+ "step": 37
269
+ },
270
+ {
271
+ "epoch": 0.040871201936004305,
272
+ "grad_norm": 0.1725727617740631,
273
+ "learning_rate": 2e-05,
274
+ "loss": 4.6611,
275
+ "step": 38
276
+ },
277
+ {
278
+ "epoch": 0.041946759881688625,
279
+ "grad_norm": 0.14692844450473785,
280
+ "learning_rate": 2e-05,
281
+ "loss": 3.9717,
282
+ "step": 39
283
+ },
284
+ {
285
+ "epoch": 0.04302231782737295,
286
+ "grad_norm": 0.21558256447315216,
287
+ "learning_rate": 2e-05,
288
+ "loss": 4.4963,
289
+ "step": 40
290
+ },
291
+ {
292
+ "epoch": 0.04409787577305727,
293
+ "grad_norm": 0.16119123995304108,
294
+ "learning_rate": 2e-05,
295
+ "loss": 3.7143,
296
+ "step": 41
297
+ },
298
+ {
299
+ "epoch": 0.0451734337187416,
300
+ "grad_norm": 0.17820394039154053,
301
+ "learning_rate": 2e-05,
302
+ "loss": 3.6309,
303
+ "step": 42
304
+ },
305
+ {
306
+ "epoch": 0.04624899166442592,
307
+ "grad_norm": 0.2944539785385132,
308
+ "learning_rate": 2e-05,
309
+ "loss": 4.682,
310
+ "step": 43
311
+ },
312
+ {
313
+ "epoch": 0.04732454961011025,
314
+ "grad_norm": 0.24868269264698029,
315
+ "learning_rate": 2e-05,
316
+ "loss": 4.4821,
317
+ "step": 44
318
+ },
319
+ {
320
+ "epoch": 0.04840010755579457,
321
+ "grad_norm": 0.22005093097686768,
322
+ "learning_rate": 2e-05,
323
+ "loss": 3.8853,
324
+ "step": 45
325
+ },
326
+ {
327
+ "epoch": 0.049475665501478895,
328
+ "grad_norm": 0.219339519739151,
329
+ "learning_rate": 2e-05,
330
+ "loss": 3.1609,
331
+ "step": 46
332
+ },
333
+ {
334
+ "epoch": 0.050551223447163215,
335
+ "grad_norm": 0.18265879154205322,
336
+ "learning_rate": 2e-05,
337
+ "loss": 3.5927,
338
+ "step": 47
339
+ },
340
+ {
341
+ "epoch": 0.05162678139284754,
342
+ "grad_norm": 0.14955416321754456,
343
+ "learning_rate": 2e-05,
344
+ "loss": 3.2562,
345
+ "step": 48
346
+ },
347
+ {
348
+ "epoch": 0.05270233933853186,
349
+ "grad_norm": 0.15133006870746613,
350
+ "learning_rate": 2e-05,
351
+ "loss": 3.3774,
352
+ "step": 49
353
+ },
354
+ {
355
+ "epoch": 0.05377789728421619,
356
+ "grad_norm": 0.29029661417007446,
357
+ "learning_rate": 2e-05,
358
+ "loss": 4.0942,
359
+ "step": 50
360
+ },
361
+ {
362
+ "epoch": 0.05485345522990051,
363
+ "grad_norm": 0.22209279239177704,
364
+ "learning_rate": 2e-05,
365
+ "loss": 3.3751,
366
+ "step": 51
367
+ },
368
+ {
369
+ "epoch": 0.05592901317558484,
370
+ "grad_norm": 0.3011445701122284,
371
+ "learning_rate": 2e-05,
372
+ "loss": 3.2212,
373
+ "step": 52
374
+ },
375
+ {
376
+ "epoch": 0.05700457112126916,
377
+ "grad_norm": 0.16566476225852966,
378
+ "learning_rate": 2e-05,
379
+ "loss": 3.3951,
380
+ "step": 53
381
+ },
382
+ {
383
+ "epoch": 0.05808012906695348,
384
+ "grad_norm": 0.21326138079166412,
385
+ "learning_rate": 2e-05,
386
+ "loss": 2.4738,
387
+ "step": 54
388
+ },
389
+ {
390
+ "epoch": 0.059155687012637806,
391
+ "grad_norm": 0.37225082516670227,
392
+ "learning_rate": 2e-05,
393
+ "loss": 3.9215,
394
+ "step": 55
395
+ },
396
+ {
397
+ "epoch": 0.060231244958322126,
398
+ "grad_norm": 0.3678470849990845,
399
+ "learning_rate": 2e-05,
400
+ "loss": 5.1287,
401
+ "step": 56
402
+ },
403
+ {
404
+ "epoch": 0.061306802904006454,
405
+ "grad_norm": 0.3504056930541992,
406
+ "learning_rate": 2e-05,
407
+ "loss": 4.5698,
408
+ "step": 57
409
+ },
410
+ {
411
+ "epoch": 0.062382360849690774,
412
+ "grad_norm": 0.3031858503818512,
413
+ "learning_rate": 2e-05,
414
+ "loss": 3.0631,
415
+ "step": 58
416
+ },
417
+ {
418
+ "epoch": 0.0634579187953751,
419
+ "grad_norm": 0.19123384356498718,
420
+ "learning_rate": 2e-05,
421
+ "loss": 3.2144,
422
+ "step": 59
423
+ },
424
+ {
425
+ "epoch": 0.06453347674105943,
426
+ "grad_norm": 0.4370661973953247,
427
+ "learning_rate": 2e-05,
428
+ "loss": 4.4572,
429
+ "step": 60
430
+ },
431
+ {
432
+ "epoch": 0.06560903468674374,
433
+ "grad_norm": 0.22062987089157104,
434
+ "learning_rate": 2e-05,
435
+ "loss": 3.6537,
436
+ "step": 61
437
+ },
438
+ {
439
+ "epoch": 0.06668459263242807,
440
+ "grad_norm": 0.3883632719516754,
441
+ "learning_rate": 2e-05,
442
+ "loss": 4.5733,
443
+ "step": 62
444
+ },
445
+ {
446
+ "epoch": 0.0677601505781124,
447
+ "grad_norm": 0.5261430144309998,
448
+ "learning_rate": 2e-05,
449
+ "loss": 4.7935,
450
+ "step": 63
451
+ },
452
+ {
453
+ "epoch": 0.06883570852379672,
454
+ "grad_norm": 0.3224443197250366,
455
+ "learning_rate": 2e-05,
456
+ "loss": 4.2075,
457
+ "step": 64
458
+ },
459
+ {
460
+ "epoch": 0.06991126646948104,
461
+ "grad_norm": 0.31252115964889526,
462
+ "learning_rate": 2e-05,
463
+ "loss": 3.8278,
464
+ "step": 65
465
+ },
466
+ {
467
+ "epoch": 0.07098682441516536,
468
+ "grad_norm": 0.3743115961551666,
469
+ "learning_rate": 2e-05,
470
+ "loss": 4.0193,
471
+ "step": 66
472
+ },
473
+ {
474
+ "epoch": 0.07206238236084969,
475
+ "grad_norm": 0.08219049125909805,
476
+ "learning_rate": 2e-05,
477
+ "loss": 2.1053,
478
+ "step": 67
479
+ },
480
+ {
481
+ "epoch": 0.07313794030653402,
482
+ "grad_norm": 0.32555970549583435,
483
+ "learning_rate": 2e-05,
484
+ "loss": 3.6152,
485
+ "step": 68
486
+ },
487
+ {
488
+ "epoch": 0.07421349825221833,
489
+ "grad_norm": 0.5192983746528625,
490
+ "learning_rate": 2e-05,
491
+ "loss": 2.7418,
492
+ "step": 69
493
+ },
494
+ {
495
+ "epoch": 0.07528905619790266,
496
+ "grad_norm": 0.40597131848335266,
497
+ "learning_rate": 2e-05,
498
+ "loss": 3.8889,
499
+ "step": 70
500
+ },
501
+ {
502
+ "epoch": 0.07636461414358699,
503
+ "grad_norm": 0.2159370630979538,
504
+ "learning_rate": 2e-05,
505
+ "loss": 2.2449,
506
+ "step": 71
507
+ },
508
+ {
509
+ "epoch": 0.07744017208927131,
510
+ "grad_norm": 0.30155783891677856,
511
+ "learning_rate": 2e-05,
512
+ "loss": 3.261,
513
+ "step": 72
514
+ },
515
+ {
516
+ "epoch": 0.07851573003495563,
517
+ "grad_norm": 0.8526483774185181,
518
+ "learning_rate": 2e-05,
519
+ "loss": 4.2702,
520
+ "step": 73
521
+ },
522
+ {
523
+ "epoch": 0.07959128798063995,
524
+ "grad_norm": 0.20504102110862732,
525
+ "learning_rate": 2e-05,
526
+ "loss": 2.1707,
527
+ "step": 74
528
+ },
529
+ {
530
+ "epoch": 0.08066684592632428,
531
+ "grad_norm": 0.4856877028942108,
532
+ "learning_rate": 2e-05,
533
+ "loss": 2.8522,
534
+ "step": 75
535
+ },
536
+ {
537
+ "epoch": 0.08174240387200861,
538
+ "grad_norm": 0.5779041647911072,
539
+ "learning_rate": 2e-05,
540
+ "loss": 3.46,
541
+ "step": 76
542
+ },
543
+ {
544
+ "epoch": 0.08281796181769292,
545
+ "grad_norm": 0.440947949886322,
546
+ "learning_rate": 2e-05,
547
+ "loss": 3.5463,
548
+ "step": 77
549
+ },
550
+ {
551
+ "epoch": 0.08389351976337725,
552
+ "grad_norm": 0.20853711664676666,
553
+ "learning_rate": 2e-05,
554
+ "loss": 2.2373,
555
+ "step": 78
556
+ },
557
+ {
558
+ "epoch": 0.08496907770906158,
559
+ "grad_norm": 0.23959238827228546,
560
+ "learning_rate": 2e-05,
561
+ "loss": 2.2488,
562
+ "step": 79
563
+ },
564
+ {
565
+ "epoch": 0.0860446356547459,
566
+ "grad_norm": 0.5024415850639343,
567
+ "learning_rate": 2e-05,
568
+ "loss": 4.206,
569
+ "step": 80
570
+ },
571
+ {
572
+ "epoch": 0.08712019360043022,
573
+ "grad_norm": 0.559302806854248,
574
+ "learning_rate": 2e-05,
575
+ "loss": 3.4094,
576
+ "step": 81
577
+ },
578
+ {
579
+ "epoch": 0.08819575154611455,
580
+ "grad_norm": 0.6160392165184021,
581
+ "learning_rate": 2e-05,
582
+ "loss": 3.4671,
583
+ "step": 82
584
+ },
585
+ {
586
+ "epoch": 0.08927130949179887,
587
+ "grad_norm": 0.38725361227989197,
588
+ "learning_rate": 2e-05,
589
+ "loss": 2.444,
590
+ "step": 83
591
+ },
592
+ {
593
+ "epoch": 0.0903468674374832,
594
+ "grad_norm": 0.8837510943412781,
595
+ "learning_rate": 2e-05,
596
+ "loss": 2.8208,
597
+ "step": 84
598
+ },
599
+ {
600
+ "epoch": 0.09142242538316751,
601
+ "grad_norm": 0.7302711606025696,
602
+ "learning_rate": 2e-05,
603
+ "loss": 4.2269,
604
+ "step": 85
605
+ },
606
+ {
607
+ "epoch": 0.09249798332885184,
608
+ "grad_norm": 0.5721362233161926,
609
+ "learning_rate": 2e-05,
610
+ "loss": 3.7523,
611
+ "step": 86
612
+ },
613
+ {
614
+ "epoch": 0.09357354127453617,
615
+ "grad_norm": 0.5297729969024658,
616
+ "learning_rate": 2e-05,
617
+ "loss": 3.9402,
618
+ "step": 87
619
+ },
620
+ {
621
+ "epoch": 0.0946490992202205,
622
+ "grad_norm": 0.4848865270614624,
623
+ "learning_rate": 2e-05,
624
+ "loss": 4.0042,
625
+ "step": 88
626
+ },
627
+ {
628
+ "epoch": 0.09572465716590481,
629
+ "grad_norm": 0.7766829133033752,
630
+ "learning_rate": 2e-05,
631
+ "loss": 4.9007,
632
+ "step": 89
633
+ },
634
+ {
635
+ "epoch": 0.09680021511158914,
636
+ "grad_norm": 0.7517933249473572,
637
+ "learning_rate": 2e-05,
638
+ "loss": 3.4699,
639
+ "step": 90
640
+ },
641
+ {
642
+ "epoch": 0.09787577305727346,
643
+ "grad_norm": 0.8348935842514038,
644
+ "learning_rate": 2e-05,
645
+ "loss": 3.3264,
646
+ "step": 91
647
+ },
648
+ {
649
+ "epoch": 0.09895133100295779,
650
+ "grad_norm": 0.6233230233192444,
651
+ "learning_rate": 2e-05,
652
+ "loss": 3.0732,
653
+ "step": 92
654
+ },
655
+ {
656
+ "epoch": 0.1000268889486421,
657
+ "grad_norm": 0.7525995373725891,
658
+ "learning_rate": 2e-05,
659
+ "loss": 4.2947,
660
+ "step": 93
661
+ },
662
+ {
663
+ "epoch": 0.10110244689432643,
664
+ "grad_norm": 1.0709550380706787,
665
+ "learning_rate": 2e-05,
666
+ "loss": 3.9258,
667
+ "step": 94
668
+ },
669
+ {
670
+ "epoch": 0.10217800484001076,
671
+ "grad_norm": 0.5882276892662048,
672
+ "learning_rate": 2e-05,
673
+ "loss": 3.2486,
674
+ "step": 95
675
+ },
676
+ {
677
+ "epoch": 0.10325356278569509,
678
+ "grad_norm": 1.3848390579223633,
679
+ "learning_rate": 2e-05,
680
+ "loss": 3.1059,
681
+ "step": 96
682
+ },
683
+ {
684
+ "epoch": 0.1043291207313794,
685
+ "grad_norm": 1.0833474397659302,
686
+ "learning_rate": 2e-05,
687
+ "loss": 3.176,
688
+ "step": 97
689
+ },
690
+ {
691
+ "epoch": 0.10540467867706373,
692
+ "grad_norm": 0.6375811100006104,
693
+ "learning_rate": 2e-05,
694
+ "loss": 3.2412,
695
+ "step": 98
696
+ },
697
+ {
698
+ "epoch": 0.10648023662274805,
699
+ "grad_norm": 1.6618685722351074,
700
+ "learning_rate": 2e-05,
701
+ "loss": 3.2567,
702
+ "step": 99
703
+ },
704
+ {
705
+ "epoch": 0.10755579456843238,
706
+ "grad_norm": 0.978987991809845,
707
+ "learning_rate": 2e-05,
708
+ "loss": 4.2846,
709
+ "step": 100
710
+ },
711
+ {
712
+ "epoch": 0.1086313525141167,
713
+ "grad_norm": 0.857375979423523,
714
+ "learning_rate": 2e-05,
715
+ "loss": 3.5841,
716
+ "step": 101
717
+ },
718
+ {
719
+ "epoch": 0.10970691045980102,
720
+ "grad_norm": 1.7573585510253906,
721
+ "learning_rate": 2e-05,
722
+ "loss": 4.3449,
723
+ "step": 102
724
+ },
725
+ {
726
+ "epoch": 0.11078246840548535,
727
+ "grad_norm": 0.925139307975769,
728
+ "learning_rate": 2e-05,
729
+ "loss": 2.9236,
730
+ "step": 103
731
+ },
732
+ {
733
+ "epoch": 0.11185802635116968,
734
+ "grad_norm": 1.0588639974594116,
735
+ "learning_rate": 2e-05,
736
+ "loss": 3.5894,
737
+ "step": 104
738
+ },
739
+ {
740
+ "epoch": 0.11293358429685399,
741
+ "grad_norm": 2.0492420196533203,
742
+ "learning_rate": 2e-05,
743
+ "loss": 3.3241,
744
+ "step": 105
745
+ },
746
+ {
747
+ "epoch": 0.11400914224253832,
748
+ "grad_norm": 1.2681959867477417,
749
+ "learning_rate": 2e-05,
750
+ "loss": 2.9072,
751
+ "step": 106
752
+ },
753
+ {
754
+ "epoch": 0.11508470018822264,
755
+ "grad_norm": 0.8113515377044678,
756
+ "learning_rate": 2e-05,
757
+ "loss": 3.7567,
758
+ "step": 107
759
+ },
760
+ {
761
+ "epoch": 0.11616025813390696,
762
+ "grad_norm": 0.8880893588066101,
763
+ "learning_rate": 2e-05,
764
+ "loss": 2.9194,
765
+ "step": 108
766
+ },
767
+ {
768
+ "epoch": 0.11723581607959128,
769
+ "grad_norm": 0.3360120356082916,
770
+ "learning_rate": 2e-05,
771
+ "loss": 2.0488,
772
+ "step": 109
773
+ },
774
+ {
775
+ "epoch": 0.11831137402527561,
776
+ "grad_norm": 0.9331343173980713,
777
+ "learning_rate": 2e-05,
778
+ "loss": 3.0238,
779
+ "step": 110
780
+ },
781
+ {
782
+ "epoch": 0.11938693197095994,
783
+ "grad_norm": 1.8001103401184082,
784
+ "learning_rate": 2e-05,
785
+ "loss": 4.6799,
786
+ "step": 111
787
+ },
788
+ {
789
+ "epoch": 0.12046248991664425,
790
+ "grad_norm": 0.6512516140937805,
791
+ "learning_rate": 2e-05,
792
+ "loss": 2.1307,
793
+ "step": 112
794
+ },
795
+ {
796
+ "epoch": 0.12153804786232858,
797
+ "grad_norm": 1.2599865198135376,
798
+ "learning_rate": 2e-05,
799
+ "loss": 2.6622,
800
+ "step": 113
801
+ },
802
+ {
803
+ "epoch": 0.12261360580801291,
804
+ "grad_norm": 0.7627917528152466,
805
+ "learning_rate": 2e-05,
806
+ "loss": 3.5136,
807
+ "step": 114
808
+ },
809
+ {
810
+ "epoch": 0.12368916375369723,
811
+ "grad_norm": 0.6828168630599976,
812
+ "learning_rate": 2e-05,
813
+ "loss": 2.471,
814
+ "step": 115
815
+ },
816
+ {
817
+ "epoch": 0.12476472169938155,
818
+ "grad_norm": 2.2847108840942383,
819
+ "learning_rate": 2e-05,
820
+ "loss": 5.1293,
821
+ "step": 116
822
+ },
823
+ {
824
+ "epoch": 0.1258402796450659,
825
+ "grad_norm": 1.5811688899993896,
826
+ "learning_rate": 2e-05,
827
+ "loss": 3.4057,
828
+ "step": 117
829
+ },
830
+ {
831
+ "epoch": 0.1269158375907502,
832
+ "grad_norm": 0.6133345365524292,
833
+ "learning_rate": 2e-05,
834
+ "loss": 2.1214,
835
+ "step": 118
836
+ },
837
+ {
838
+ "epoch": 0.12799139553643452,
839
+ "grad_norm": 1.0904330015182495,
840
+ "learning_rate": 2e-05,
841
+ "loss": 3.5764,
842
+ "step": 119
843
+ },
844
+ {
845
+ "epoch": 0.12906695348211886,
846
+ "grad_norm": 3.196547269821167,
847
+ "learning_rate": 2e-05,
848
+ "loss": 4.751,
849
+ "step": 120
850
+ },
851
+ {
852
+ "epoch": 0.13014251142780317,
853
+ "grad_norm": 2.6566030979156494,
854
+ "learning_rate": 2e-05,
855
+ "loss": 4.1988,
856
+ "step": 121
857
+ },
858
+ {
859
+ "epoch": 0.13121806937348748,
860
+ "grad_norm": 2.3683555126190186,
861
+ "learning_rate": 2e-05,
862
+ "loss": 4.2292,
863
+ "step": 122
864
+ },
865
+ {
866
+ "epoch": 0.13229362731917182,
867
+ "grad_norm": 2.0530343055725098,
868
+ "learning_rate": 2e-05,
869
+ "loss": 3.3207,
870
+ "step": 123
871
+ },
872
+ {
873
+ "epoch": 0.13336918526485614,
874
+ "grad_norm": 1.4857006072998047,
875
+ "learning_rate": 2e-05,
876
+ "loss": 3.1832,
877
+ "step": 124
878
+ },
879
+ {
880
+ "epoch": 0.13444474321054048,
881
+ "grad_norm": 1.1598551273345947,
882
+ "learning_rate": 2e-05,
883
+ "loss": 3.7383,
884
+ "step": 125
885
+ },
886
+ {
887
+ "epoch": 0.1355203011562248,
888
+ "grad_norm": 1.7023690938949585,
889
+ "learning_rate": 2e-05,
890
+ "loss": 4.4133,
891
+ "step": 126
892
+ },
893
+ {
894
+ "epoch": 0.1365958591019091,
895
+ "grad_norm": 2.516563653945923,
896
+ "learning_rate": 2e-05,
897
+ "loss": 4.4345,
898
+ "step": 127
899
+ },
900
+ {
901
+ "epoch": 0.13767141704759345,
902
+ "grad_norm": 1.2092735767364502,
903
+ "learning_rate": 2e-05,
904
+ "loss": 2.0301,
905
+ "step": 128
906
+ },
907
+ {
908
+ "epoch": 0.13874697499327776,
909
+ "grad_norm": 3.6382813453674316,
910
+ "learning_rate": 2e-05,
911
+ "loss": 3.1083,
912
+ "step": 129
913
+ },
914
+ {
915
+ "epoch": 0.13982253293896207,
916
+ "grad_norm": 1.815604567527771,
917
+ "learning_rate": 2e-05,
918
+ "loss": 3.6486,
919
+ "step": 130
920
+ },
921
+ {
922
+ "epoch": 0.14089809088464642,
923
+ "grad_norm": 2.0142934322357178,
924
+ "learning_rate": 2e-05,
925
+ "loss": 3.137,
926
+ "step": 131
927
+ },
928
+ {
929
+ "epoch": 0.14197364883033073,
930
+ "grad_norm": 1.4500945806503296,
931
+ "learning_rate": 2e-05,
932
+ "loss": 2.9764,
933
+ "step": 132
934
+ },
935
+ {
936
+ "epoch": 0.14304920677601507,
937
+ "grad_norm": 3.060540199279785,
938
+ "learning_rate": 2e-05,
939
+ "loss": 3.8098,
940
+ "step": 133
941
+ },
942
+ {
943
+ "epoch": 0.14412476472169938,
944
+ "grad_norm": 1.1867421865463257,
945
+ "learning_rate": 2e-05,
946
+ "loss": 2.3144,
947
+ "step": 134
948
+ },
949
+ {
950
+ "epoch": 0.1452003226673837,
951
+ "grad_norm": 2.2631044387817383,
952
+ "learning_rate": 2e-05,
953
+ "loss": 3.6758,
954
+ "step": 135
955
+ },
956
+ {
957
+ "epoch": 0.14627588061306804,
958
+ "grad_norm": 1.878165364265442,
959
+ "learning_rate": 2e-05,
960
+ "loss": 2.954,
961
+ "step": 136
962
+ },
963
+ {
964
+ "epoch": 0.14735143855875235,
965
+ "grad_norm": 2.5874857902526855,
966
+ "learning_rate": 2e-05,
967
+ "loss": 2.9404,
968
+ "step": 137
969
+ },
970
+ {
971
+ "epoch": 0.14842699650443666,
972
+ "grad_norm": 1.9505641460418701,
973
+ "learning_rate": 2e-05,
974
+ "loss": 3.8377,
975
+ "step": 138
976
+ },
977
+ {
978
+ "epoch": 0.149502554450121,
979
+ "grad_norm": 3.316041946411133,
980
+ "learning_rate": 2e-05,
981
+ "loss": 3.3591,
982
+ "step": 139
983
+ },
984
+ {
985
+ "epoch": 0.15057811239580532,
986
+ "grad_norm": 1.034047245979309,
987
+ "learning_rate": 2e-05,
988
+ "loss": 2.7709,
989
+ "step": 140
990
+ },
991
+ {
992
+ "epoch": 0.15165367034148966,
993
+ "grad_norm": 3.9413716793060303,
994
+ "learning_rate": 2e-05,
995
+ "loss": 4.6814,
996
+ "step": 141
997
+ },
998
+ {
999
+ "epoch": 0.15272922828717397,
1000
+ "grad_norm": 1.6777610778808594,
1001
+ "learning_rate": 2e-05,
1002
+ "loss": 3.0112,
1003
+ "step": 142
1004
+ },
1005
+ {
1006
+ "epoch": 0.1538047862328583,
1007
+ "grad_norm": 2.4318885803222656,
1008
+ "learning_rate": 2e-05,
1009
+ "loss": 3.0604,
1010
+ "step": 143
1011
+ },
1012
+ {
1013
+ "epoch": 0.15488034417854263,
1014
+ "grad_norm": 1.0361096858978271,
1015
+ "learning_rate": 2e-05,
1016
+ "loss": 2.4662,
1017
+ "step": 144
1018
+ },
1019
+ {
1020
+ "epoch": 0.15595590212422694,
1021
+ "grad_norm": 1.5604453086853027,
1022
+ "learning_rate": 2e-05,
1023
+ "loss": 2.6922,
1024
+ "step": 145
1025
+ },
1026
+ {
1027
+ "epoch": 0.15703146006991126,
1028
+ "grad_norm": 2.758985996246338,
1029
+ "learning_rate": 2e-05,
1030
+ "loss": 4.1986,
1031
+ "step": 146
1032
+ },
1033
+ {
1034
+ "epoch": 0.1581070180155956,
1035
+ "grad_norm": 3.1454544067382812,
1036
+ "learning_rate": 2e-05,
1037
+ "loss": 3.3037,
1038
+ "step": 147
1039
+ },
1040
+ {
1041
+ "epoch": 0.1591825759612799,
1042
+ "grad_norm": 5.285416126251221,
1043
+ "learning_rate": 2e-05,
1044
+ "loss": 4.41,
1045
+ "step": 148
1046
+ },
1047
+ {
1048
+ "epoch": 0.16025813390696425,
1049
+ "grad_norm": 4.258360385894775,
1050
+ "learning_rate": 2e-05,
1051
+ "loss": 3.552,
1052
+ "step": 149
1053
+ },
1054
+ {
1055
+ "epoch": 0.16133369185264856,
1056
+ "grad_norm": 3.8160877227783203,
1057
+ "learning_rate": 2e-05,
1058
+ "loss": 2.3886,
1059
+ "step": 150
1060
+ },
1061
+ {
1062
+ "epoch": 0.16240924979833288,
1063
+ "grad_norm": 2.338573694229126,
1064
+ "learning_rate": 2e-05,
1065
+ "loss": 2.7356,
1066
+ "step": 151
1067
+ },
1068
+ {
1069
+ "epoch": 0.16348480774401722,
1070
+ "grad_norm": 4.280124664306641,
1071
+ "learning_rate": 2e-05,
1072
+ "loss": 3.6049,
1073
+ "step": 152
1074
+ },
1075
+ {
1076
+ "epoch": 0.16456036568970153,
1077
+ "grad_norm": 2.4771525859832764,
1078
+ "learning_rate": 2e-05,
1079
+ "loss": 4.0302,
1080
+ "step": 153
1081
+ },
1082
+ {
1083
+ "epoch": 0.16563592363538585,
1084
+ "grad_norm": 1.8418432474136353,
1085
+ "learning_rate": 2e-05,
1086
+ "loss": 3.6569,
1087
+ "step": 154
1088
+ },
1089
+ {
1090
+ "epoch": 0.1667114815810702,
1091
+ "grad_norm": 2.0972683429718018,
1092
+ "learning_rate": 2e-05,
1093
+ "loss": 3.0681,
1094
+ "step": 155
1095
+ },
1096
+ {
1097
+ "epoch": 0.1677870395267545,
1098
+ "grad_norm": 1.6193856000900269,
1099
+ "learning_rate": 2e-05,
1100
+ "loss": 2.8379,
1101
+ "step": 156
1102
+ },
1103
+ {
1104
+ "epoch": 0.16886259747243884,
1105
+ "grad_norm": 4.455961227416992,
1106
+ "learning_rate": 2e-05,
1107
+ "loss": 2.7098,
1108
+ "step": 157
1109
+ },
1110
+ {
1111
+ "epoch": 0.16993815541812315,
1112
+ "grad_norm": 1.417285680770874,
1113
+ "learning_rate": 2e-05,
1114
+ "loss": 1.7622,
1115
+ "step": 158
1116
+ },
1117
+ {
1118
+ "epoch": 0.17101371336380747,
1119
+ "grad_norm": 4.029820442199707,
1120
+ "learning_rate": 2e-05,
1121
+ "loss": 4.3837,
1122
+ "step": 159
1123
+ },
1124
+ {
1125
+ "epoch": 0.1720892713094918,
1126
+ "grad_norm": 1.492792010307312,
1127
+ "learning_rate": 2e-05,
1128
+ "loss": 2.9025,
1129
+ "step": 160
1130
+ },
1131
+ {
1132
+ "epoch": 0.17316482925517612,
1133
+ "grad_norm": 4.077206134796143,
1134
+ "learning_rate": 2e-05,
1135
+ "loss": 3.2485,
1136
+ "step": 161
1137
+ },
1138
+ {
1139
+ "epoch": 0.17424038720086044,
1140
+ "grad_norm": 3.3235838413238525,
1141
+ "learning_rate": 2e-05,
1142
+ "loss": 2.7973,
1143
+ "step": 162
1144
+ },
1145
+ {
1146
+ "epoch": 0.17531594514654478,
1147
+ "grad_norm": 3.8488590717315674,
1148
+ "learning_rate": 2e-05,
1149
+ "loss": 3.8877,
1150
+ "step": 163
1151
+ },
1152
+ {
1153
+ "epoch": 0.1763915030922291,
1154
+ "grad_norm": 2.2555062770843506,
1155
+ "learning_rate": 2e-05,
1156
+ "loss": 3.0205,
1157
+ "step": 164
1158
+ },
1159
+ {
1160
+ "epoch": 0.1774670610379134,
1161
+ "grad_norm": 1.7995167970657349,
1162
+ "learning_rate": 2e-05,
1163
+ "loss": 2.2315,
1164
+ "step": 165
1165
+ },
1166
+ {
1167
+ "epoch": 0.17854261898359775,
1168
+ "grad_norm": 1.2209360599517822,
1169
+ "learning_rate": 2e-05,
1170
+ "loss": 2.4366,
1171
+ "step": 166
1172
+ },
1173
+ {
1174
+ "epoch": 0.17961817692928206,
1175
+ "grad_norm": 1.4641145467758179,
1176
+ "learning_rate": 2e-05,
1177
+ "loss": 1.9656,
1178
+ "step": 167
1179
+ },
1180
+ {
1181
+ "epoch": 0.1806937348749664,
1182
+ "grad_norm": 2.6921463012695312,
1183
+ "learning_rate": 2e-05,
1184
+ "loss": 4.2631,
1185
+ "step": 168
1186
+ },
1187
+ {
1188
+ "epoch": 0.1817692928206507,
1189
+ "grad_norm": 4.518846035003662,
1190
+ "learning_rate": 2e-05,
1191
+ "loss": 3.7601,
1192
+ "step": 169
1193
+ },
1194
+ {
1195
+ "epoch": 0.18284485076633503,
1196
+ "grad_norm": 1.3827426433563232,
1197
+ "learning_rate": 2e-05,
1198
+ "loss": 2.3307,
1199
+ "step": 170
1200
+ },
1201
+ {
1202
+ "epoch": 0.18392040871201937,
1203
+ "grad_norm": 2.110506057739258,
1204
+ "learning_rate": 2e-05,
1205
+ "loss": 2.8058,
1206
+ "step": 171
1207
+ },
1208
+ {
1209
+ "epoch": 0.18499596665770368,
1210
+ "grad_norm": 1.1899346113204956,
1211
+ "learning_rate": 2e-05,
1212
+ "loss": 1.9843,
1213
+ "step": 172
1214
+ },
1215
+ {
1216
+ "epoch": 0.186071524603388,
1217
+ "grad_norm": 3.0731053352355957,
1218
+ "learning_rate": 2e-05,
1219
+ "loss": 2.3099,
1220
+ "step": 173
1221
+ },
1222
+ {
1223
+ "epoch": 0.18714708254907234,
1224
+ "grad_norm": 3.0657575130462646,
1225
+ "learning_rate": 2e-05,
1226
+ "loss": 2.8818,
1227
+ "step": 174
1228
+ },
1229
+ {
1230
+ "epoch": 0.18822264049475665,
1231
+ "grad_norm": 2.161705493927002,
1232
+ "learning_rate": 2e-05,
1233
+ "loss": 2.4349,
1234
+ "step": 175
1235
+ },
1236
+ {
1237
+ "epoch": 0.189298198440441,
1238
+ "grad_norm": 3.0202341079711914,
1239
+ "learning_rate": 2e-05,
1240
+ "loss": 2.1977,
1241
+ "step": 176
1242
+ },
1243
+ {
1244
+ "epoch": 0.1903737563861253,
1245
+ "grad_norm": 2.346747398376465,
1246
+ "learning_rate": 2e-05,
1247
+ "loss": 2.399,
1248
+ "step": 177
1249
+ },
1250
+ {
1251
+ "epoch": 0.19144931433180962,
1252
+ "grad_norm": 1.5756202936172485,
1253
+ "learning_rate": 2e-05,
1254
+ "loss": 2.2689,
1255
+ "step": 178
1256
+ },
1257
+ {
1258
+ "epoch": 0.19252487227749396,
1259
+ "grad_norm": 2.398148775100708,
1260
+ "learning_rate": 2e-05,
1261
+ "loss": 3.8229,
1262
+ "step": 179
1263
+ },
1264
+ {
1265
+ "epoch": 0.19360043022317827,
1266
+ "grad_norm": 3.2819738388061523,
1267
+ "learning_rate": 2e-05,
1268
+ "loss": 2.6033,
1269
+ "step": 180
1270
+ },
1271
+ {
1272
+ "epoch": 0.19467598816886258,
1273
+ "grad_norm": 5.887080192565918,
1274
+ "learning_rate": 2e-05,
1275
+ "loss": 2.4956,
1276
+ "step": 181
1277
+ },
1278
+ {
1279
+ "epoch": 0.19575154611454693,
1280
+ "grad_norm": 2.208775520324707,
1281
+ "learning_rate": 2e-05,
1282
+ "loss": 2.1286,
1283
+ "step": 182
1284
+ },
1285
+ {
1286
+ "epoch": 0.19682710406023124,
1287
+ "grad_norm": 1.817888617515564,
1288
+ "learning_rate": 2e-05,
1289
+ "loss": 2.4608,
1290
+ "step": 183
1291
+ },
1292
+ {
1293
+ "epoch": 0.19790266200591558,
1294
+ "grad_norm": 2.150890588760376,
1295
+ "learning_rate": 2e-05,
1296
+ "loss": 3.2177,
1297
+ "step": 184
1298
+ },
1299
+ {
1300
+ "epoch": 0.1989782199515999,
1301
+ "grad_norm": 6.783865928649902,
1302
+ "learning_rate": 2e-05,
1303
+ "loss": 3.6292,
1304
+ "step": 185
1305
+ },
1306
+ {
1307
+ "epoch": 0.2000537778972842,
1308
+ "grad_norm": 3.8008947372436523,
1309
+ "learning_rate": 2e-05,
1310
+ "loss": 2.9182,
1311
+ "step": 186
1312
+ },
1313
+ {
1314
+ "epoch": 0.20112933584296855,
1315
+ "grad_norm": 4.612886905670166,
1316
+ "learning_rate": 2e-05,
1317
+ "loss": 3.1909,
1318
+ "step": 187
1319
+ },
1320
+ {
1321
+ "epoch": 0.20220489378865286,
1322
+ "grad_norm": 5.594083786010742,
1323
+ "learning_rate": 2e-05,
1324
+ "loss": 3.0986,
1325
+ "step": 188
1326
+ },
1327
+ {
1328
+ "epoch": 0.20328045173433718,
1329
+ "grad_norm": 3.358548402786255,
1330
+ "learning_rate": 2e-05,
1331
+ "loss": 2.2792,
1332
+ "step": 189
1333
+ },
1334
+ {
1335
+ "epoch": 0.20435600968002152,
1336
+ "grad_norm": 2.6911849975585938,
1337
+ "learning_rate": 2e-05,
1338
+ "loss": 3.4548,
1339
+ "step": 190
1340
+ },
1341
+ {
1342
+ "epoch": 0.20543156762570583,
1343
+ "grad_norm": 2.4795117378234863,
1344
+ "learning_rate": 2e-05,
1345
+ "loss": 2.3414,
1346
+ "step": 191
1347
+ },
1348
+ {
1349
+ "epoch": 0.20650712557139017,
1350
+ "grad_norm": 1.7064529657363892,
1351
+ "learning_rate": 2e-05,
1352
+ "loss": 1.8158,
1353
+ "step": 192
1354
+ },
1355
+ {
1356
+ "epoch": 0.20758268351707448,
1357
+ "grad_norm": 7.239320755004883,
1358
+ "learning_rate": 2e-05,
1359
+ "loss": 2.6676,
1360
+ "step": 193
1361
+ },
1362
+ {
1363
+ "epoch": 0.2086582414627588,
1364
+ "grad_norm": 4.747775077819824,
1365
+ "learning_rate": 2e-05,
1366
+ "loss": 3.2929,
1367
+ "step": 194
1368
+ },
1369
+ {
1370
+ "epoch": 0.20973379940844314,
1371
+ "grad_norm": 1.850651502609253,
1372
+ "learning_rate": 2e-05,
1373
+ "loss": 2.3005,
1374
+ "step": 195
1375
+ },
1376
+ {
1377
+ "epoch": 0.21080935735412745,
1378
+ "grad_norm": 2.4029853343963623,
1379
+ "learning_rate": 2e-05,
1380
+ "loss": 3.7213,
1381
+ "step": 196
1382
+ },
1383
+ {
1384
+ "epoch": 0.21188491529981177,
1385
+ "grad_norm": 2.5401084423065186,
1386
+ "learning_rate": 2e-05,
1387
+ "loss": 2.7665,
1388
+ "step": 197
1389
+ },
1390
+ {
1391
+ "epoch": 0.2129604732454961,
1392
+ "grad_norm": 4.080130577087402,
1393
+ "learning_rate": 2e-05,
1394
+ "loss": 1.8886,
1395
+ "step": 198
1396
+ },
1397
+ {
1398
+ "epoch": 0.21403603119118042,
1399
+ "grad_norm": 1.7884845733642578,
1400
+ "learning_rate": 2e-05,
1401
+ "loss": 2.4168,
1402
+ "step": 199
1403
+ },
1404
+ {
1405
+ "epoch": 0.21511158913686476,
1406
+ "grad_norm": 3.1292009353637695,
1407
+ "learning_rate": 2e-05,
1408
+ "loss": 2.4442,
1409
+ "step": 200
1410
+ },
1411
+ {
1412
+ "epoch": 0.21618714708254907,
1413
+ "grad_norm": 6.860455513000488,
1414
+ "learning_rate": 2e-05,
1415
+ "loss": 3.103,
1416
+ "step": 201
1417
+ },
1418
+ {
1419
+ "epoch": 0.2172627050282334,
1420
+ "grad_norm": 4.3930511474609375,
1421
+ "learning_rate": 2e-05,
1422
+ "loss": 2.7579,
1423
+ "step": 202
1424
+ },
1425
+ {
1426
+ "epoch": 0.21833826297391773,
1427
+ "grad_norm": 5.641942024230957,
1428
+ "learning_rate": 2e-05,
1429
+ "loss": 2.657,
1430
+ "step": 203
1431
+ },
1432
+ {
1433
+ "epoch": 0.21941382091960204,
1434
+ "grad_norm": 2.2956748008728027,
1435
+ "learning_rate": 2e-05,
1436
+ "loss": 2.2479,
1437
+ "step": 204
1438
+ },
1439
+ {
1440
+ "epoch": 0.22048937886528636,
1441
+ "grad_norm": 2.176769971847534,
1442
+ "learning_rate": 2e-05,
1443
+ "loss": 1.2909,
1444
+ "step": 205
1445
+ },
1446
+ {
1447
+ "epoch": 0.2215649368109707,
1448
+ "grad_norm": 4.874899864196777,
1449
+ "learning_rate": 2e-05,
1450
+ "loss": 1.9261,
1451
+ "step": 206
1452
+ },
1453
+ {
1454
+ "epoch": 0.222640494756655,
1455
+ "grad_norm": 2.68208909034729,
1456
+ "learning_rate": 2e-05,
1457
+ "loss": 2.1664,
1458
+ "step": 207
1459
+ },
1460
+ {
1461
+ "epoch": 0.22371605270233935,
1462
+ "grad_norm": 3.6076104640960693,
1463
+ "learning_rate": 2e-05,
1464
+ "loss": 2.8906,
1465
+ "step": 208
1466
+ },
1467
+ {
1468
+ "epoch": 0.22479161064802367,
1469
+ "grad_norm": 2.173996925354004,
1470
+ "learning_rate": 2e-05,
1471
+ "loss": 2.3881,
1472
+ "step": 209
1473
+ },
1474
+ {
1475
+ "epoch": 0.22586716859370798,
1476
+ "grad_norm": 2.750606060028076,
1477
+ "learning_rate": 2e-05,
1478
+ "loss": 2.1685,
1479
+ "step": 210
1480
+ },
1481
+ {
1482
+ "epoch": 0.22694272653939232,
1483
+ "grad_norm": 3.133878469467163,
1484
+ "learning_rate": 2e-05,
1485
+ "loss": 2.5526,
1486
+ "step": 211
1487
+ },
1488
+ {
1489
+ "epoch": 0.22801828448507663,
1490
+ "grad_norm": 3.0615320205688477,
1491
+ "learning_rate": 2e-05,
1492
+ "loss": 2.7754,
1493
+ "step": 212
1494
+ },
1495
+ {
1496
+ "epoch": 0.22909384243076095,
1497
+ "grad_norm": 2.5888783931732178,
1498
+ "learning_rate": 2e-05,
1499
+ "loss": 2.0269,
1500
+ "step": 213
1501
+ },
1502
+ {
1503
+ "epoch": 0.2301694003764453,
1504
+ "grad_norm": 3.7221732139587402,
1505
+ "learning_rate": 2e-05,
1506
+ "loss": 1.7132,
1507
+ "step": 214
1508
+ },
1509
+ {
1510
+ "epoch": 0.2312449583221296,
1511
+ "grad_norm": 2.1723954677581787,
1512
+ "learning_rate": 2e-05,
1513
+ "loss": 1.3366,
1514
+ "step": 215
1515
+ },
1516
+ {
1517
+ "epoch": 0.23232051626781391,
1518
+ "grad_norm": 2.335772752761841,
1519
+ "learning_rate": 2e-05,
1520
+ "loss": 2.6138,
1521
+ "step": 216
1522
+ },
1523
+ {
1524
+ "epoch": 0.23339607421349826,
1525
+ "grad_norm": 4.438052654266357,
1526
+ "learning_rate": 2e-05,
1527
+ "loss": 3.134,
1528
+ "step": 217
1529
+ },
1530
+ {
1531
+ "epoch": 0.23447163215918257,
1532
+ "grad_norm": 1.6511766910552979,
1533
+ "learning_rate": 2e-05,
1534
+ "loss": 2.2954,
1535
+ "step": 218
1536
+ },
1537
+ {
1538
+ "epoch": 0.2355471901048669,
1539
+ "grad_norm": 1.8855592012405396,
1540
+ "learning_rate": 2e-05,
1541
+ "loss": 1.7853,
1542
+ "step": 219
1543
+ },
1544
+ {
1545
+ "epoch": 0.23662274805055122,
1546
+ "grad_norm": 3.8127474784851074,
1547
+ "learning_rate": 2e-05,
1548
+ "loss": 1.8774,
1549
+ "step": 220
1550
+ },
1551
+ {
1552
+ "epoch": 0.23769830599623554,
1553
+ "grad_norm": 5.802454471588135,
1554
+ "learning_rate": 2e-05,
1555
+ "loss": 3.2977,
1556
+ "step": 221
1557
+ },
1558
+ {
1559
+ "epoch": 0.23877386394191988,
1560
+ "grad_norm": 3.8824198246002197,
1561
+ "learning_rate": 2e-05,
1562
+ "loss": 2.715,
1563
+ "step": 222
1564
+ },
1565
+ {
1566
+ "epoch": 0.2398494218876042,
1567
+ "grad_norm": 3.3041539192199707,
1568
+ "learning_rate": 2e-05,
1569
+ "loss": 2.4187,
1570
+ "step": 223
1571
+ },
1572
+ {
1573
+ "epoch": 0.2409249798332885,
1574
+ "grad_norm": 2.7080237865448,
1575
+ "learning_rate": 2e-05,
1576
+ "loss": 2.0946,
1577
+ "step": 224
1578
+ },
1579
+ {
1580
+ "epoch": 0.24200053777897285,
1581
+ "grad_norm": 2.8399364948272705,
1582
+ "learning_rate": 2e-05,
1583
+ "loss": 1.994,
1584
+ "step": 225
1585
+ },
1586
+ {
1587
+ "epoch": 0.24307609572465716,
1588
+ "grad_norm": 2.896878480911255,
1589
+ "learning_rate": 2e-05,
1590
+ "loss": 2.2411,
1591
+ "step": 226
1592
+ },
1593
+ {
1594
+ "epoch": 0.2441516536703415,
1595
+ "grad_norm": 2.4240972995758057,
1596
+ "learning_rate": 2e-05,
1597
+ "loss": 2.5445,
1598
+ "step": 227
1599
+ },
1600
+ {
1601
+ "epoch": 0.24522721161602581,
1602
+ "grad_norm": 2.8553824424743652,
1603
+ "learning_rate": 2e-05,
1604
+ "loss": 2.0291,
1605
+ "step": 228
1606
+ },
1607
+ {
1608
+ "epoch": 0.24630276956171013,
1609
+ "grad_norm": 3.546189069747925,
1610
+ "learning_rate": 2e-05,
1611
+ "loss": 2.9868,
1612
+ "step": 229
1613
+ },
1614
+ {
1615
+ "epoch": 0.24737832750739447,
1616
+ "grad_norm": 3.4419355392456055,
1617
+ "learning_rate": 2e-05,
1618
+ "loss": 1.7061,
1619
+ "step": 230
1620
+ },
1621
+ {
1622
+ "epoch": 0.24845388545307878,
1623
+ "grad_norm": 2.98702073097229,
1624
+ "learning_rate": 2e-05,
1625
+ "loss": 2.3165,
1626
+ "step": 231
1627
+ },
1628
+ {
1629
+ "epoch": 0.2495294433987631,
1630
+ "grad_norm": 4.188878536224365,
1631
+ "learning_rate": 2e-05,
1632
+ "loss": 1.9139,
1633
+ "step": 232
1634
+ },
1635
+ {
1636
+ "epoch": 0.2506050013444474,
1637
+ "grad_norm": 3.832674026489258,
1638
+ "learning_rate": 2e-05,
1639
+ "loss": 2.1851,
1640
+ "step": 233
1641
+ },
1642
+ {
1643
+ "epoch": 0.2516805592901318,
1644
+ "grad_norm": 2.5292224884033203,
1645
+ "learning_rate": 2e-05,
1646
+ "loss": 1.395,
1647
+ "step": 234
1648
+ },
1649
+ {
1650
+ "epoch": 0.2527561172358161,
1651
+ "grad_norm": 5.447204113006592,
1652
+ "learning_rate": 2e-05,
1653
+ "loss": 1.6947,
1654
+ "step": 235
1655
+ },
1656
+ {
1657
+ "epoch": 0.2538316751815004,
1658
+ "grad_norm": 5.222197532653809,
1659
+ "learning_rate": 2e-05,
1660
+ "loss": 2.7788,
1661
+ "step": 236
1662
+ },
1663
+ {
1664
+ "epoch": 0.2549072331271847,
1665
+ "grad_norm": 6.383084297180176,
1666
+ "learning_rate": 2e-05,
1667
+ "loss": 2.7386,
1668
+ "step": 237
1669
+ },
1670
+ {
1671
+ "epoch": 0.25598279107286903,
1672
+ "grad_norm": 2.2934720516204834,
1673
+ "learning_rate": 2e-05,
1674
+ "loss": 2.5268,
1675
+ "step": 238
1676
+ },
1677
+ {
1678
+ "epoch": 0.2570583490185534,
1679
+ "grad_norm": 2.0854413509368896,
1680
+ "learning_rate": 2e-05,
1681
+ "loss": 1.5054,
1682
+ "step": 239
1683
+ },
1684
+ {
1685
+ "epoch": 0.2581339069642377,
1686
+ "grad_norm": 3.8236308097839355,
1687
+ "learning_rate": 2e-05,
1688
+ "loss": 2.2114,
1689
+ "step": 240
1690
+ },
1691
+ {
1692
+ "epoch": 0.259209464909922,
1693
+ "grad_norm": 3.419747829437256,
1694
+ "learning_rate": 2e-05,
1695
+ "loss": 1.7876,
1696
+ "step": 241
1697
+ },
1698
+ {
1699
+ "epoch": 0.26028502285560634,
1700
+ "grad_norm": 3.672929048538208,
1701
+ "learning_rate": 2e-05,
1702
+ "loss": 1.9878,
1703
+ "step": 242
1704
+ },
1705
+ {
1706
+ "epoch": 0.26136058080129065,
1707
+ "grad_norm": 1.8193111419677734,
1708
+ "learning_rate": 2e-05,
1709
+ "loss": 1.7618,
1710
+ "step": 243
1711
+ },
1712
+ {
1713
+ "epoch": 0.26243613874697497,
1714
+ "grad_norm": 7.668798446655273,
1715
+ "learning_rate": 2e-05,
1716
+ "loss": 1.8264,
1717
+ "step": 244
1718
+ },
1719
+ {
1720
+ "epoch": 0.26351169669265934,
1721
+ "grad_norm": 4.639315128326416,
1722
+ "learning_rate": 2e-05,
1723
+ "loss": 2.9063,
1724
+ "step": 245
1725
+ },
1726
+ {
1727
+ "epoch": 0.26458725463834365,
1728
+ "grad_norm": 2.3293251991271973,
1729
+ "learning_rate": 2e-05,
1730
+ "loss": 1.2018,
1731
+ "step": 246
1732
+ },
1733
+ {
1734
+ "epoch": 0.26566281258402796,
1735
+ "grad_norm": 4.38407564163208,
1736
+ "learning_rate": 2e-05,
1737
+ "loss": 1.6128,
1738
+ "step": 247
1739
+ },
1740
+ {
1741
+ "epoch": 0.2667383705297123,
1742
+ "grad_norm": 1.7634291648864746,
1743
+ "learning_rate": 2e-05,
1744
+ "loss": 1.4277,
1745
+ "step": 248
1746
+ },
1747
+ {
1748
+ "epoch": 0.2678139284753966,
1749
+ "grad_norm": 3.81402325630188,
1750
+ "learning_rate": 2e-05,
1751
+ "loss": 2.0313,
1752
+ "step": 249
1753
+ },
1754
+ {
1755
+ "epoch": 0.26888948642108096,
1756
+ "grad_norm": 2.318251848220825,
1757
+ "learning_rate": 2e-05,
1758
+ "loss": 1.7833,
1759
+ "step": 250
1760
+ },
1761
+ {
1762
+ "epoch": 0.26996504436676527,
1763
+ "grad_norm": 1.8340922594070435,
1764
+ "learning_rate": 2e-05,
1765
+ "loss": 1.6557,
1766
+ "step": 251
1767
+ },
1768
+ {
1769
+ "epoch": 0.2710406023124496,
1770
+ "grad_norm": 4.6209845542907715,
1771
+ "learning_rate": 2e-05,
1772
+ "loss": 2.166,
1773
+ "step": 252
1774
+ },
1775
+ {
1776
+ "epoch": 0.2721161602581339,
1777
+ "grad_norm": 3.370737075805664,
1778
+ "learning_rate": 2e-05,
1779
+ "loss": 1.9072,
1780
+ "step": 253
1781
+ },
1782
+ {
1783
+ "epoch": 0.2731917182038182,
1784
+ "grad_norm": 4.559715747833252,
1785
+ "learning_rate": 2e-05,
1786
+ "loss": 1.7178,
1787
+ "step": 254
1788
+ },
1789
+ {
1790
+ "epoch": 0.2742672761495026,
1791
+ "grad_norm": 4.001542568206787,
1792
+ "learning_rate": 2e-05,
1793
+ "loss": 1.4689,
1794
+ "step": 255
1795
+ },
1796
+ {
1797
+ "epoch": 0.2753428340951869,
1798
+ "grad_norm": 2.454691171646118,
1799
+ "learning_rate": 2e-05,
1800
+ "loss": 1.4426,
1801
+ "step": 256
1802
+ },
1803
+ {
1804
+ "epoch": 0.2764183920408712,
1805
+ "grad_norm": 2.3479011058807373,
1806
+ "learning_rate": 2e-05,
1807
+ "loss": 2.0009,
1808
+ "step": 257
1809
+ },
1810
+ {
1811
+ "epoch": 0.2774939499865555,
1812
+ "grad_norm": 5.103316307067871,
1813
+ "learning_rate": 2e-05,
1814
+ "loss": 1.9592,
1815
+ "step": 258
1816
+ },
1817
+ {
1818
+ "epoch": 0.27856950793223983,
1819
+ "grad_norm": 2.4349985122680664,
1820
+ "learning_rate": 2e-05,
1821
+ "loss": 2.0947,
1822
+ "step": 259
1823
+ },
1824
+ {
1825
+ "epoch": 0.27964506587792415,
1826
+ "grad_norm": 2.940519332885742,
1827
+ "learning_rate": 2e-05,
1828
+ "loss": 1.9484,
1829
+ "step": 260
1830
+ },
1831
+ {
1832
+ "epoch": 0.2807206238236085,
1833
+ "grad_norm": 4.070859432220459,
1834
+ "learning_rate": 2e-05,
1835
+ "loss": 2.1733,
1836
+ "step": 261
1837
+ },
1838
+ {
1839
+ "epoch": 0.28179618176929283,
1840
+ "grad_norm": 5.868851184844971,
1841
+ "learning_rate": 2e-05,
1842
+ "loss": 1.6656,
1843
+ "step": 262
1844
+ },
1845
+ {
1846
+ "epoch": 0.28287173971497714,
1847
+ "grad_norm": 4.625424861907959,
1848
+ "learning_rate": 2e-05,
1849
+ "loss": 2.3667,
1850
+ "step": 263
1851
+ },
1852
+ {
1853
+ "epoch": 0.28394729766066146,
1854
+ "grad_norm": 2.5272748470306396,
1855
+ "learning_rate": 2e-05,
1856
+ "loss": 2.1122,
1857
+ "step": 264
1858
+ },
1859
+ {
1860
+ "epoch": 0.28502285560634577,
1861
+ "grad_norm": 5.193785190582275,
1862
+ "learning_rate": 2e-05,
1863
+ "loss": 1.5063,
1864
+ "step": 265
1865
+ },
1866
+ {
1867
+ "epoch": 0.28609841355203014,
1868
+ "grad_norm": 1.6589148044586182,
1869
+ "learning_rate": 2e-05,
1870
+ "loss": 1.9044,
1871
+ "step": 266
1872
+ },
1873
+ {
1874
+ "epoch": 0.28717397149771445,
1875
+ "grad_norm": 3.423194408416748,
1876
+ "learning_rate": 2e-05,
1877
+ "loss": 2.2866,
1878
+ "step": 267
1879
+ },
1880
+ {
1881
+ "epoch": 0.28824952944339877,
1882
+ "grad_norm": 2.188795328140259,
1883
+ "learning_rate": 2e-05,
1884
+ "loss": 1.6161,
1885
+ "step": 268
1886
+ },
1887
+ {
1888
+ "epoch": 0.2893250873890831,
1889
+ "grad_norm": 3.441420078277588,
1890
+ "learning_rate": 2e-05,
1891
+ "loss": 1.42,
1892
+ "step": 269
1893
+ },
1894
+ {
1895
+ "epoch": 0.2904006453347674,
1896
+ "grad_norm": 4.066496849060059,
1897
+ "learning_rate": 2e-05,
1898
+ "loss": 1.9209,
1899
+ "step": 270
1900
+ },
1901
+ {
1902
+ "epoch": 0.2914762032804517,
1903
+ "grad_norm": 4.315361022949219,
1904
+ "learning_rate": 2e-05,
1905
+ "loss": 1.433,
1906
+ "step": 271
1907
+ },
1908
+ {
1909
+ "epoch": 0.2925517612261361,
1910
+ "grad_norm": 2.7533764839172363,
1911
+ "learning_rate": 2e-05,
1912
+ "loss": 1.3967,
1913
+ "step": 272
1914
+ },
1915
+ {
1916
+ "epoch": 0.2936273191718204,
1917
+ "grad_norm": 2.4276041984558105,
1918
+ "learning_rate": 2e-05,
1919
+ "loss": 2.5529,
1920
+ "step": 273
1921
+ },
1922
+ {
1923
+ "epoch": 0.2947028771175047,
1924
+ "grad_norm": 2.0695526599884033,
1925
+ "learning_rate": 2e-05,
1926
+ "loss": 1.7445,
1927
+ "step": 274
1928
+ },
1929
+ {
1930
+ "epoch": 0.295778435063189,
1931
+ "grad_norm": 3.037795066833496,
1932
+ "learning_rate": 2e-05,
1933
+ "loss": 2.4438,
1934
+ "step": 275
1935
+ },
1936
+ {
1937
+ "epoch": 0.29685399300887333,
1938
+ "grad_norm": 3.6190900802612305,
1939
+ "learning_rate": 2e-05,
1940
+ "loss": 1.6739,
1941
+ "step": 276
1942
+ },
1943
+ {
1944
+ "epoch": 0.2979295509545577,
1945
+ "grad_norm": 2.2492079734802246,
1946
+ "learning_rate": 2e-05,
1947
+ "loss": 1.4227,
1948
+ "step": 277
1949
+ },
1950
+ {
1951
+ "epoch": 0.299005108900242,
1952
+ "grad_norm": 2.544053554534912,
1953
+ "learning_rate": 2e-05,
1954
+ "loss": 2.1751,
1955
+ "step": 278
1956
+ },
1957
+ {
1958
+ "epoch": 0.3000806668459263,
1959
+ "grad_norm": 2.0408968925476074,
1960
+ "learning_rate": 2e-05,
1961
+ "loss": 1.6613,
1962
+ "step": 279
1963
+ },
1964
+ {
1965
+ "epoch": 0.30115622479161064,
1966
+ "grad_norm": 4.337264537811279,
1967
+ "learning_rate": 2e-05,
1968
+ "loss": 1.6372,
1969
+ "step": 280
1970
+ },
1971
+ {
1972
+ "epoch": 0.30223178273729495,
1973
+ "grad_norm": 4.142371654510498,
1974
+ "learning_rate": 2e-05,
1975
+ "loss": 1.4561,
1976
+ "step": 281
1977
+ },
1978
+ {
1979
+ "epoch": 0.3033073406829793,
1980
+ "grad_norm": 3.091535806655884,
1981
+ "learning_rate": 2e-05,
1982
+ "loss": 2.024,
1983
+ "step": 282
1984
+ },
1985
+ {
1986
+ "epoch": 0.30438289862866363,
1987
+ "grad_norm": 5.492683410644531,
1988
+ "learning_rate": 2e-05,
1989
+ "loss": 1.6435,
1990
+ "step": 283
1991
+ },
1992
+ {
1993
+ "epoch": 0.30545845657434795,
1994
+ "grad_norm": 6.648707389831543,
1995
+ "learning_rate": 2e-05,
1996
+ "loss": 2.6662,
1997
+ "step": 284
1998
+ },
1999
+ {
2000
+ "epoch": 0.30653401452003226,
2001
+ "grad_norm": 2.870140790939331,
2002
+ "learning_rate": 2e-05,
2003
+ "loss": 1.2809,
2004
+ "step": 285
2005
+ },
2006
+ {
2007
+ "epoch": 0.3076095724657166,
2008
+ "grad_norm": 1.6203242540359497,
2009
+ "learning_rate": 2e-05,
2010
+ "loss": 1.6137,
2011
+ "step": 286
2012
+ },
2013
+ {
2014
+ "epoch": 0.3086851304114009,
2015
+ "grad_norm": 3.1347341537475586,
2016
+ "learning_rate": 2e-05,
2017
+ "loss": 1.0086,
2018
+ "step": 287
2019
+ },
2020
+ {
2021
+ "epoch": 0.30976068835708526,
2022
+ "grad_norm": 1.8718631267547607,
2023
+ "learning_rate": 2e-05,
2024
+ "loss": 1.6824,
2025
+ "step": 288
2026
+ },
2027
+ {
2028
+ "epoch": 0.31083624630276957,
2029
+ "grad_norm": 3.668214797973633,
2030
+ "learning_rate": 2e-05,
2031
+ "loss": 2.3531,
2032
+ "step": 289
2033
+ },
2034
+ {
2035
+ "epoch": 0.3119118042484539,
2036
+ "grad_norm": 1.985969066619873,
2037
+ "learning_rate": 2e-05,
2038
+ "loss": 1.526,
2039
+ "step": 290
2040
+ },
2041
+ {
2042
+ "epoch": 0.3129873621941382,
2043
+ "grad_norm": 6.0746846199035645,
2044
+ "learning_rate": 2e-05,
2045
+ "loss": 2.3411,
2046
+ "step": 291
2047
+ },
2048
+ {
2049
+ "epoch": 0.3140629201398225,
2050
+ "grad_norm": 2.04298734664917,
2051
+ "learning_rate": 2e-05,
2052
+ "loss": 1.6556,
2053
+ "step": 292
2054
+ },
2055
+ {
2056
+ "epoch": 0.3151384780855069,
2057
+ "grad_norm": 4.120944499969482,
2058
+ "learning_rate": 2e-05,
2059
+ "loss": 1.788,
2060
+ "step": 293
2061
+ },
2062
+ {
2063
+ "epoch": 0.3162140360311912,
2064
+ "grad_norm": 5.191648006439209,
2065
+ "learning_rate": 2e-05,
2066
+ "loss": 1.6943,
2067
+ "step": 294
2068
+ },
2069
+ {
2070
+ "epoch": 0.3172895939768755,
2071
+ "grad_norm": 1.5654542446136475,
2072
+ "learning_rate": 2e-05,
2073
+ "loss": 1.7853,
2074
+ "step": 295
2075
+ },
2076
+ {
2077
+ "epoch": 0.3183651519225598,
2078
+ "grad_norm": 1.9325037002563477,
2079
+ "learning_rate": 2e-05,
2080
+ "loss": 1.4699,
2081
+ "step": 296
2082
+ },
2083
+ {
2084
+ "epoch": 0.31944070986824413,
2085
+ "grad_norm": 5.055520534515381,
2086
+ "learning_rate": 2e-05,
2087
+ "loss": 1.9544,
2088
+ "step": 297
2089
+ },
2090
+ {
2091
+ "epoch": 0.3205162678139285,
2092
+ "grad_norm": 3.137733221054077,
2093
+ "learning_rate": 2e-05,
2094
+ "loss": 2.2007,
2095
+ "step": 298
2096
+ },
2097
+ {
2098
+ "epoch": 0.3215918257596128,
2099
+ "grad_norm": 2.723611354827881,
2100
+ "learning_rate": 2e-05,
2101
+ "loss": 1.7762,
2102
+ "step": 299
2103
+ },
2104
+ {
2105
+ "epoch": 0.32266738370529713,
2106
+ "grad_norm": 2.2930235862731934,
2107
+ "learning_rate": 2e-05,
2108
+ "loss": 1.1551,
2109
+ "step": 300
2110
+ },
2111
+ {
2112
+ "epoch": 0.32374294165098144,
2113
+ "grad_norm": 2.1257314682006836,
2114
+ "learning_rate": 2e-05,
2115
+ "loss": 1.9487,
2116
+ "step": 301
2117
+ },
2118
+ {
2119
+ "epoch": 0.32481849959666576,
2120
+ "grad_norm": 1.8796786069869995,
2121
+ "learning_rate": 2e-05,
2122
+ "loss": 1.4238,
2123
+ "step": 302
2124
+ },
2125
+ {
2126
+ "epoch": 0.32589405754235007,
2127
+ "grad_norm": 2.0493640899658203,
2128
+ "learning_rate": 2e-05,
2129
+ "loss": 1.6256,
2130
+ "step": 303
2131
+ },
2132
+ {
2133
+ "epoch": 0.32696961548803444,
2134
+ "grad_norm": 2.2541568279266357,
2135
+ "learning_rate": 2e-05,
2136
+ "loss": 1.2289,
2137
+ "step": 304
2138
+ },
2139
+ {
2140
+ "epoch": 0.32804517343371875,
2141
+ "grad_norm": 2.2614448070526123,
2142
+ "learning_rate": 2e-05,
2143
+ "loss": 1.1208,
2144
+ "step": 305
2145
+ },
2146
+ {
2147
+ "epoch": 0.32912073137940306,
2148
+ "grad_norm": 2.197634220123291,
2149
+ "learning_rate": 2e-05,
2150
+ "loss": 0.9634,
2151
+ "step": 306
2152
+ },
2153
+ {
2154
+ "epoch": 0.3301962893250874,
2155
+ "grad_norm": 1.8836265802383423,
2156
+ "learning_rate": 2e-05,
2157
+ "loss": 1.2566,
2158
+ "step": 307
2159
+ },
2160
+ {
2161
+ "epoch": 0.3312718472707717,
2162
+ "grad_norm": 1.3131864070892334,
2163
+ "learning_rate": 2e-05,
2164
+ "loss": 1.6519,
2165
+ "step": 308
2166
+ },
2167
+ {
2168
+ "epoch": 0.33234740521645606,
2169
+ "grad_norm": 2.0432112216949463,
2170
+ "learning_rate": 2e-05,
2171
+ "loss": 1.5373,
2172
+ "step": 309
2173
+ },
2174
+ {
2175
+ "epoch": 0.3334229631621404,
2176
+ "grad_norm": 1.9313534498214722,
2177
+ "learning_rate": 2e-05,
2178
+ "loss": 1.0634,
2179
+ "step": 310
2180
+ },
2181
+ {
2182
+ "epoch": 0.3344985211078247,
2183
+ "grad_norm": 1.6917750835418701,
2184
+ "learning_rate": 2e-05,
2185
+ "loss": 1.2723,
2186
+ "step": 311
2187
+ },
2188
+ {
2189
+ "epoch": 0.335574079053509,
2190
+ "grad_norm": 3.3362669944763184,
2191
+ "learning_rate": 2e-05,
2192
+ "loss": 2.1054,
2193
+ "step": 312
2194
+ },
2195
+ {
2196
+ "epoch": 0.3366496369991933,
2197
+ "grad_norm": 3.4973859786987305,
2198
+ "learning_rate": 2e-05,
2199
+ "loss": 1.9061,
2200
+ "step": 313
2201
+ },
2202
+ {
2203
+ "epoch": 0.3377251949448777,
2204
+ "grad_norm": 2.9965860843658447,
2205
+ "learning_rate": 2e-05,
2206
+ "loss": 1.7292,
2207
+ "step": 314
2208
+ },
2209
+ {
2210
+ "epoch": 0.338800752890562,
2211
+ "grad_norm": 1.1123921871185303,
2212
+ "learning_rate": 2e-05,
2213
+ "loss": 1.024,
2214
+ "step": 315
2215
+ },
2216
+ {
2217
+ "epoch": 0.3398763108362463,
2218
+ "grad_norm": 1.8463563919067383,
2219
+ "learning_rate": 2e-05,
2220
+ "loss": 1.4357,
2221
+ "step": 316
2222
+ },
2223
+ {
2224
+ "epoch": 0.3409518687819306,
2225
+ "grad_norm": 1.1670702695846558,
2226
+ "learning_rate": 2e-05,
2227
+ "loss": 0.9946,
2228
+ "step": 317
2229
+ },
2230
+ {
2231
+ "epoch": 0.34202742672761494,
2232
+ "grad_norm": 1.933604121208191,
2233
+ "learning_rate": 2e-05,
2234
+ "loss": 1.919,
2235
+ "step": 318
2236
+ },
2237
+ {
2238
+ "epoch": 0.34310298467329925,
2239
+ "grad_norm": 2.2463345527648926,
2240
+ "learning_rate": 2e-05,
2241
+ "loss": 1.5173,
2242
+ "step": 319
2243
+ },
2244
+ {
2245
+ "epoch": 0.3441785426189836,
2246
+ "grad_norm": 2.4717116355895996,
2247
+ "learning_rate": 2e-05,
2248
+ "loss": 1.4675,
2249
+ "step": 320
2250
+ },
2251
+ {
2252
+ "epoch": 0.34525410056466793,
2253
+ "grad_norm": 2.748335599899292,
2254
+ "learning_rate": 2e-05,
2255
+ "loss": 1.2585,
2256
+ "step": 321
2257
+ },
2258
+ {
2259
+ "epoch": 0.34632965851035225,
2260
+ "grad_norm": 2.388786554336548,
2261
+ "learning_rate": 2e-05,
2262
+ "loss": 1.7388,
2263
+ "step": 322
2264
+ },
2265
+ {
2266
+ "epoch": 0.34740521645603656,
2267
+ "grad_norm": 2.1550583839416504,
2268
+ "learning_rate": 2e-05,
2269
+ "loss": 2.2277,
2270
+ "step": 323
2271
+ },
2272
+ {
2273
+ "epoch": 0.34848077440172087,
2274
+ "grad_norm": 1.5130534172058105,
2275
+ "learning_rate": 2e-05,
2276
+ "loss": 1.1991,
2277
+ "step": 324
2278
+ },
2279
+ {
2280
+ "epoch": 0.34955633234740524,
2281
+ "grad_norm": 1.768734335899353,
2282
+ "learning_rate": 2e-05,
2283
+ "loss": 1.2314,
2284
+ "step": 325
2285
+ },
2286
+ {
2287
+ "epoch": 0.35063189029308955,
2288
+ "grad_norm": 2.1945996284484863,
2289
+ "learning_rate": 2e-05,
2290
+ "loss": 1.6976,
2291
+ "step": 326
2292
+ },
2293
+ {
2294
+ "epoch": 0.35170744823877387,
2295
+ "grad_norm": 3.313124179840088,
2296
+ "learning_rate": 2e-05,
2297
+ "loss": 2.1373,
2298
+ "step": 327
2299
+ },
2300
+ {
2301
+ "epoch": 0.3527830061844582,
2302
+ "grad_norm": 1.6641830205917358,
2303
+ "learning_rate": 2e-05,
2304
+ "loss": 1.0339,
2305
+ "step": 328
2306
+ },
2307
+ {
2308
+ "epoch": 0.3538585641301425,
2309
+ "grad_norm": 2.551252841949463,
2310
+ "learning_rate": 2e-05,
2311
+ "loss": 1.0713,
2312
+ "step": 329
2313
+ },
2314
+ {
2315
+ "epoch": 0.3549341220758268,
2316
+ "grad_norm": 3.3957738876342773,
2317
+ "learning_rate": 2e-05,
2318
+ "loss": 2.2178,
2319
+ "step": 330
2320
+ },
2321
+ {
2322
+ "epoch": 0.3560096800215112,
2323
+ "grad_norm": 1.5952703952789307,
2324
+ "learning_rate": 2e-05,
2325
+ "loss": 1.6338,
2326
+ "step": 331
2327
+ },
2328
+ {
2329
+ "epoch": 0.3570852379671955,
2330
+ "grad_norm": 2.4693539142608643,
2331
+ "learning_rate": 2e-05,
2332
+ "loss": 2.2655,
2333
+ "step": 332
2334
+ },
2335
+ {
2336
+ "epoch": 0.3581607959128798,
2337
+ "grad_norm": 3.20023250579834,
2338
+ "learning_rate": 2e-05,
2339
+ "loss": 1.4743,
2340
+ "step": 333
2341
+ },
2342
+ {
2343
+ "epoch": 0.3592363538585641,
2344
+ "grad_norm": 3.136413335800171,
2345
+ "learning_rate": 2e-05,
2346
+ "loss": 2.3212,
2347
+ "step": 334
2348
+ },
2349
+ {
2350
+ "epoch": 0.36031191180424843,
2351
+ "grad_norm": 2.068697452545166,
2352
+ "learning_rate": 2e-05,
2353
+ "loss": 2.2536,
2354
+ "step": 335
2355
+ },
2356
+ {
2357
+ "epoch": 0.3613874697499328,
2358
+ "grad_norm": 2.8290975093841553,
2359
+ "learning_rate": 2e-05,
2360
+ "loss": 1.6358,
2361
+ "step": 336
2362
+ },
2363
+ {
2364
+ "epoch": 0.3624630276956171,
2365
+ "grad_norm": 5.349557399749756,
2366
+ "learning_rate": 2e-05,
2367
+ "loss": 2.2619,
2368
+ "step": 337
2369
+ },
2370
+ {
2371
+ "epoch": 0.3635385856413014,
2372
+ "grad_norm": 5.981229782104492,
2373
+ "learning_rate": 2e-05,
2374
+ "loss": 1.0677,
2375
+ "step": 338
2376
+ },
2377
+ {
2378
+ "epoch": 0.36461414358698574,
2379
+ "grad_norm": 1.3054362535476685,
2380
+ "learning_rate": 2e-05,
2381
+ "loss": 0.9468,
2382
+ "step": 339
2383
+ },
2384
+ {
2385
+ "epoch": 0.36568970153267005,
2386
+ "grad_norm": 3.109434127807617,
2387
+ "learning_rate": 2e-05,
2388
+ "loss": 1.5973,
2389
+ "step": 340
2390
+ },
2391
+ {
2392
+ "epoch": 0.3667652594783544,
2393
+ "grad_norm": 3.0365335941314697,
2394
+ "learning_rate": 2e-05,
2395
+ "loss": 1.7926,
2396
+ "step": 341
2397
+ },
2398
+ {
2399
+ "epoch": 0.36784081742403874,
2400
+ "grad_norm": 2.4590301513671875,
2401
+ "learning_rate": 2e-05,
2402
+ "loss": 1.2485,
2403
+ "step": 342
2404
+ },
2405
+ {
2406
+ "epoch": 0.36891637536972305,
2407
+ "grad_norm": 1.305474042892456,
2408
+ "learning_rate": 2e-05,
2409
+ "loss": 1.3799,
2410
+ "step": 343
2411
+ },
2412
+ {
2413
+ "epoch": 0.36999193331540736,
2414
+ "grad_norm": 1.8500665426254272,
2415
+ "learning_rate": 2e-05,
2416
+ "loss": 1.0873,
2417
+ "step": 344
2418
+ },
2419
+ {
2420
+ "epoch": 0.3710674912610917,
2421
+ "grad_norm": 2.907775402069092,
2422
+ "learning_rate": 2e-05,
2423
+ "loss": 1.694,
2424
+ "step": 345
2425
+ },
2426
+ {
2427
+ "epoch": 0.372143049206776,
2428
+ "grad_norm": 1.907683253288269,
2429
+ "learning_rate": 2e-05,
2430
+ "loss": 1.5214,
2431
+ "step": 346
2432
+ },
2433
+ {
2434
+ "epoch": 0.37321860715246036,
2435
+ "grad_norm": 2.173938512802124,
2436
+ "learning_rate": 2e-05,
2437
+ "loss": 1.6287,
2438
+ "step": 347
2439
+ },
2440
+ {
2441
+ "epoch": 0.37429416509814467,
2442
+ "grad_norm": 1.8932238817214966,
2443
+ "learning_rate": 2e-05,
2444
+ "loss": 1.9984,
2445
+ "step": 348
2446
+ },
2447
+ {
2448
+ "epoch": 0.375369723043829,
2449
+ "grad_norm": 2.37819766998291,
2450
+ "learning_rate": 2e-05,
2451
+ "loss": 2.3142,
2452
+ "step": 349
2453
+ },
2454
+ {
2455
+ "epoch": 0.3764452809895133,
2456
+ "grad_norm": 2.919982671737671,
2457
+ "learning_rate": 2e-05,
2458
+ "loss": 1.2937,
2459
+ "step": 350
2460
+ },
2461
+ {
2462
+ "epoch": 0.3775208389351976,
2463
+ "grad_norm": 1.182470679283142,
2464
+ "learning_rate": 2e-05,
2465
+ "loss": 1.17,
2466
+ "step": 351
2467
+ },
2468
+ {
2469
+ "epoch": 0.378596396880882,
2470
+ "grad_norm": 4.175192356109619,
2471
+ "learning_rate": 2e-05,
2472
+ "loss": 2.5641,
2473
+ "step": 352
2474
+ },
2475
+ {
2476
+ "epoch": 0.3796719548265663,
2477
+ "grad_norm": 2.881838321685791,
2478
+ "learning_rate": 2e-05,
2479
+ "loss": 1.2848,
2480
+ "step": 353
2481
+ },
2482
+ {
2483
+ "epoch": 0.3807475127722506,
2484
+ "grad_norm": 1.533673882484436,
2485
+ "learning_rate": 2e-05,
2486
+ "loss": 1.0104,
2487
+ "step": 354
2488
+ },
2489
+ {
2490
+ "epoch": 0.3818230707179349,
2491
+ "grad_norm": 2.300402879714966,
2492
+ "learning_rate": 2e-05,
2493
+ "loss": 1.665,
2494
+ "step": 355
2495
+ },
2496
+ {
2497
+ "epoch": 0.38289862866361923,
2498
+ "grad_norm": 2.2758076190948486,
2499
+ "learning_rate": 2e-05,
2500
+ "loss": 1.2572,
2501
+ "step": 356
2502
+ },
2503
+ {
2504
+ "epoch": 0.3839741866093036,
2505
+ "grad_norm": 1.2889113426208496,
2506
+ "learning_rate": 2e-05,
2507
+ "loss": 1.382,
2508
+ "step": 357
2509
+ },
2510
+ {
2511
+ "epoch": 0.3850497445549879,
2512
+ "grad_norm": 1.9991205930709839,
2513
+ "learning_rate": 2e-05,
2514
+ "loss": 1.2349,
2515
+ "step": 358
2516
+ },
2517
+ {
2518
+ "epoch": 0.38612530250067223,
2519
+ "grad_norm": 1.4156253337860107,
2520
+ "learning_rate": 2e-05,
2521
+ "loss": 1.3388,
2522
+ "step": 359
2523
+ },
2524
+ {
2525
+ "epoch": 0.38720086044635654,
2526
+ "grad_norm": 1.4127274751663208,
2527
+ "learning_rate": 2e-05,
2528
+ "loss": 1.1782,
2529
+ "step": 360
2530
+ },
2531
+ {
2532
+ "epoch": 0.38827641839204086,
2533
+ "grad_norm": 2.555490493774414,
2534
+ "learning_rate": 2e-05,
2535
+ "loss": 1.3054,
2536
+ "step": 361
2537
+ },
2538
+ {
2539
+ "epoch": 0.38935197633772517,
2540
+ "grad_norm": 1.8390058279037476,
2541
+ "learning_rate": 2e-05,
2542
+ "loss": 1.6844,
2543
+ "step": 362
2544
+ },
2545
+ {
2546
+ "epoch": 0.39042753428340954,
2547
+ "grad_norm": 1.543318510055542,
2548
+ "learning_rate": 2e-05,
2549
+ "loss": 1.2689,
2550
+ "step": 363
2551
+ },
2552
+ {
2553
+ "epoch": 0.39150309222909385,
2554
+ "grad_norm": 1.3956406116485596,
2555
+ "learning_rate": 2e-05,
2556
+ "loss": 1.3261,
2557
+ "step": 364
2558
+ },
2559
+ {
2560
+ "epoch": 0.39257865017477817,
2561
+ "grad_norm": 1.5241447687149048,
2562
+ "learning_rate": 2e-05,
2563
+ "loss": 1.7081,
2564
+ "step": 365
2565
+ },
2566
+ {
2567
+ "epoch": 0.3936542081204625,
2568
+ "grad_norm": 3.3806824684143066,
2569
+ "learning_rate": 2e-05,
2570
+ "loss": 1.0843,
2571
+ "step": 366
2572
+ },
2573
+ {
2574
+ "epoch": 0.3947297660661468,
2575
+ "grad_norm": 3.4040732383728027,
2576
+ "learning_rate": 2e-05,
2577
+ "loss": 1.9527,
2578
+ "step": 367
2579
+ },
2580
+ {
2581
+ "epoch": 0.39580532401183116,
2582
+ "grad_norm": 2.287287712097168,
2583
+ "learning_rate": 2e-05,
2584
+ "loss": 1.523,
2585
+ "step": 368
2586
+ },
2587
+ {
2588
+ "epoch": 0.3968808819575155,
2589
+ "grad_norm": 2.720411539077759,
2590
+ "learning_rate": 2e-05,
2591
+ "loss": 1.5933,
2592
+ "step": 369
2593
+ },
2594
+ {
2595
+ "epoch": 0.3979564399031998,
2596
+ "grad_norm": 2.0351316928863525,
2597
+ "learning_rate": 2e-05,
2598
+ "loss": 1.2118,
2599
+ "step": 370
2600
+ },
2601
+ {
2602
+ "epoch": 0.3990319978488841,
2603
+ "grad_norm": 2.3920750617980957,
2604
+ "learning_rate": 2e-05,
2605
+ "loss": 1.8173,
2606
+ "step": 371
2607
+ },
2608
+ {
2609
+ "epoch": 0.4001075557945684,
2610
+ "grad_norm": 1.955111026763916,
2611
+ "learning_rate": 2e-05,
2612
+ "loss": 1.2801,
2613
+ "step": 372
2614
+ },
2615
+ {
2616
+ "epoch": 0.4011831137402528,
2617
+ "grad_norm": 2.1078410148620605,
2618
+ "learning_rate": 2e-05,
2619
+ "loss": 2.3026,
2620
+ "step": 373
2621
+ },
2622
+ {
2623
+ "epoch": 0.4022586716859371,
2624
+ "grad_norm": 3.355013847351074,
2625
+ "learning_rate": 2e-05,
2626
+ "loss": 1.8343,
2627
+ "step": 374
2628
+ },
2629
+ {
2630
+ "epoch": 0.4033342296316214,
2631
+ "grad_norm": 1.319663643836975,
2632
+ "learning_rate": 2e-05,
2633
+ "loss": 1.3011,
2634
+ "step": 375
2635
+ },
2636
+ {
2637
+ "epoch": 0.4044097875773057,
2638
+ "grad_norm": 3.6383869647979736,
2639
+ "learning_rate": 2e-05,
2640
+ "loss": 1.813,
2641
+ "step": 376
2642
+ },
2643
+ {
2644
+ "epoch": 0.40548534552299004,
2645
+ "grad_norm": 2.878934621810913,
2646
+ "learning_rate": 2e-05,
2647
+ "loss": 1.3719,
2648
+ "step": 377
2649
+ },
2650
+ {
2651
+ "epoch": 0.40656090346867435,
2652
+ "grad_norm": 4.608090400695801,
2653
+ "learning_rate": 2e-05,
2654
+ "loss": 1.004,
2655
+ "step": 378
2656
+ },
2657
+ {
2658
+ "epoch": 0.4076364614143587,
2659
+ "grad_norm": 3.73118257522583,
2660
+ "learning_rate": 2e-05,
2661
+ "loss": 0.8164,
2662
+ "step": 379
2663
+ },
2664
+ {
2665
+ "epoch": 0.40871201936004303,
2666
+ "grad_norm": 4.62899112701416,
2667
+ "learning_rate": 2e-05,
2668
+ "loss": 1.6774,
2669
+ "step": 380
2670
+ },
2671
+ {
2672
+ "epoch": 0.40978757730572735,
2673
+ "grad_norm": 1.8718538284301758,
2674
+ "learning_rate": 2e-05,
2675
+ "loss": 1.8125,
2676
+ "step": 381
2677
+ },
2678
+ {
2679
+ "epoch": 0.41086313525141166,
2680
+ "grad_norm": 1.3654688596725464,
2681
+ "learning_rate": 2e-05,
2682
+ "loss": 1.0739,
2683
+ "step": 382
2684
+ },
2685
+ {
2686
+ "epoch": 0.411938693197096,
2687
+ "grad_norm": 3.5833754539489746,
2688
+ "learning_rate": 2e-05,
2689
+ "loss": 1.3718,
2690
+ "step": 383
2691
+ },
2692
+ {
2693
+ "epoch": 0.41301425114278034,
2694
+ "grad_norm": 1.8415347337722778,
2695
+ "learning_rate": 2e-05,
2696
+ "loss": 1.7924,
2697
+ "step": 384
2698
+ },
2699
+ {
2700
+ "epoch": 0.41408980908846466,
2701
+ "grad_norm": 2.1915903091430664,
2702
+ "learning_rate": 2e-05,
2703
+ "loss": 1.9869,
2704
+ "step": 385
2705
+ },
2706
+ {
2707
+ "epoch": 0.41516536703414897,
2708
+ "grad_norm": 2.770104169845581,
2709
+ "learning_rate": 2e-05,
2710
+ "loss": 1.0986,
2711
+ "step": 386
2712
+ },
2713
+ {
2714
+ "epoch": 0.4162409249798333,
2715
+ "grad_norm": 1.8012187480926514,
2716
+ "learning_rate": 2e-05,
2717
+ "loss": 1.1557,
2718
+ "step": 387
2719
+ },
2720
+ {
2721
+ "epoch": 0.4173164829255176,
2722
+ "grad_norm": 2.36631178855896,
2723
+ "learning_rate": 2e-05,
2724
+ "loss": 1.6846,
2725
+ "step": 388
2726
+ },
2727
+ {
2728
+ "epoch": 0.4183920408712019,
2729
+ "grad_norm": 4.233100414276123,
2730
+ "learning_rate": 2e-05,
2731
+ "loss": 1.5626,
2732
+ "step": 389
2733
+ },
2734
+ {
2735
+ "epoch": 0.4194675988168863,
2736
+ "grad_norm": 2.602360248565674,
2737
+ "learning_rate": 2e-05,
2738
+ "loss": 1.5002,
2739
+ "step": 390
2740
+ },
2741
+ {
2742
+ "epoch": 0.4205431567625706,
2743
+ "grad_norm": 2.060886859893799,
2744
+ "learning_rate": 2e-05,
2745
+ "loss": 1.4001,
2746
+ "step": 391
2747
+ },
2748
+ {
2749
+ "epoch": 0.4216187147082549,
2750
+ "grad_norm": 3.412872791290283,
2751
+ "learning_rate": 2e-05,
2752
+ "loss": 1.7084,
2753
+ "step": 392
2754
+ },
2755
+ {
2756
+ "epoch": 0.4226942726539392,
2757
+ "grad_norm": 2.384026050567627,
2758
+ "learning_rate": 2e-05,
2759
+ "loss": 1.3846,
2760
+ "step": 393
2761
+ },
2762
+ {
2763
+ "epoch": 0.42376983059962353,
2764
+ "grad_norm": 1.424104928970337,
2765
+ "learning_rate": 2e-05,
2766
+ "loss": 1.3321,
2767
+ "step": 394
2768
+ },
2769
+ {
2770
+ "epoch": 0.4248453885453079,
2771
+ "grad_norm": 2.0102243423461914,
2772
+ "learning_rate": 2e-05,
2773
+ "loss": 1.605,
2774
+ "step": 395
2775
+ },
2776
+ {
2777
+ "epoch": 0.4259209464909922,
2778
+ "grad_norm": 1.9879999160766602,
2779
+ "learning_rate": 2e-05,
2780
+ "loss": 1.6769,
2781
+ "step": 396
2782
+ },
2783
+ {
2784
+ "epoch": 0.4269965044366765,
2785
+ "grad_norm": 1.3937056064605713,
2786
+ "learning_rate": 2e-05,
2787
+ "loss": 1.153,
2788
+ "step": 397
2789
+ },
2790
+ {
2791
+ "epoch": 0.42807206238236084,
2792
+ "grad_norm": 3.6724510192871094,
2793
+ "learning_rate": 2e-05,
2794
+ "loss": 1.7874,
2795
+ "step": 398
2796
+ },
2797
+ {
2798
+ "epoch": 0.42914762032804515,
2799
+ "grad_norm": 1.6098462343215942,
2800
+ "learning_rate": 2e-05,
2801
+ "loss": 1.021,
2802
+ "step": 399
2803
+ },
2804
+ {
2805
+ "epoch": 0.4302231782737295,
2806
+ "grad_norm": 2.969383955001831,
2807
+ "learning_rate": 2e-05,
2808
+ "loss": 1.5014,
2809
+ "step": 400
2810
+ }
2811
+ ],
2812
+ "logging_steps": 1.0,
2813
+ "max_steps": 465,
2814
+ "num_input_tokens_seen": 0,
2815
+ "num_train_epochs": 1,
2816
+ "save_steps": 100,
2817
+ "stateful_callbacks": {
2818
+ "TrainerControl": {
2819
+ "args": {
2820
+ "should_epoch_stop": false,
2821
+ "should_evaluate": false,
2822
+ "should_log": false,
2823
+ "should_save": true,
2824
+ "should_training_stop": false
2825
+ },
2826
+ "attributes": {}
2827
+ }
2828
+ },
2829
+ "total_flos": 3670319790080.0,
2830
+ "train_batch_size": 4,
2831
+ "trial_name": null,
2832
+ "trial_params": null
2833
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch0.5-mlp/checkpoint-400/zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch1-mlp/checkpoint-100/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch1-mlp/checkpoint-100/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-epoch1-mlp/checkpoint-100/zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 16,
14
+ "lora_dropout": 0.05,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": null,
18
+ "peft_type": "LORA",
19
+ "r": 8,
20
+ "rank_pattern": {},
21
+ "revision": null,
22
+ "target_modules": [
23
+ "model.layers.26.mlp.up_proj",
24
+ "model.layers.20.mlp.down_proj",
25
+ "model.layers.10.mlp.gate_proj",
26
+ "model.layers.21.mlp.up_proj",
27
+ "model.layers.1.mlp.gate_proj",
28
+ "model.layers.8.mlp.down_proj",
29
+ "model.layers.15.mlp.down_proj",
30
+ "model.layers.20.mlp.gate_proj",
31
+ "model.layers.19.mlp.up_proj",
32
+ "model.layers.12.mlp.gate_proj",
33
+ "model.layers.5.mlp.gate_proj",
34
+ "model.layers.23.mlp.up_proj",
35
+ "model.layers.27.mlp.down_proj",
36
+ "model.layers.21.mlp.down_proj",
37
+ "model.layers.22.mlp.down_proj",
38
+ "model.layers.11.mlp.down_proj",
39
+ "model.layers.6.mlp.gate_proj",
40
+ "model.layers.17.mlp.gate_proj",
41
+ "model.layers.6.mlp.up_proj",
42
+ "model.layers.25.mlp.down_proj",
43
+ "model.layers.5.mlp.down_proj",
44
+ "model.layers.17.mlp.down_proj",
45
+ "model.layers.10.mlp.up_proj",
46
+ "model.layers.14.mlp.gate_proj",
47
+ "model.layers.6.mlp.down_proj",
48
+ "model.layers.3.mlp.up_proj",
49
+ "model.layers.27.mlp.up_proj",
50
+ "model.layers.8.mlp.up_proj",
51
+ "model.layers.16.mlp.up_proj",
52
+ "model.layers.27.mlp.gate_proj",
53
+ "model.layers.26.mlp.gate_proj",
54
+ "model.layers.23.mlp.gate_proj",
55
+ "model.layers.19.mlp.down_proj",
56
+ "model.layers.13.mlp.down_proj",
57
+ "model.layers.17.mlp.up_proj",
58
+ "model.layers.8.mlp.gate_proj",
59
+ "model.layers.2.mlp.gate_proj",
60
+ "model.layers.12.mlp.down_proj",
61
+ "model.layers.23.mlp.down_proj",
62
+ "model.layers.3.mlp.gate_proj",
63
+ "model.layers.22.mlp.up_proj",
64
+ "model.layers.12.mlp.up_proj",
65
+ "model.layers.2.mlp.down_proj",
66
+ "model.layers.7.mlp.up_proj",
67
+ "model.layers.25.mlp.up_proj",
68
+ "model.layers.11.mlp.up_proj",
69
+ "model.layers.4.mlp.gate_proj",
70
+ "model.layers.16.mlp.down_proj",
71
+ "model.layers.1.mlp.up_proj",
72
+ "model.layers.10.mlp.down_proj",
73
+ "model.layers.2.mlp.up_proj",
74
+ "model.layers.1.mlp.down_proj",
75
+ "model.layers.15.mlp.gate_proj",
76
+ "model.layers.24.mlp.gate_proj",
77
+ "model.layers.13.mlp.up_proj",
78
+ "model.layers.0.mlp.up_proj",
79
+ "model.layers.24.mlp.up_proj",
80
+ "model.layers.19.mlp.gate_proj",
81
+ "model.layers.9.mlp.down_proj",
82
+ "model.layers.18.mlp.down_proj",
83
+ "model.layers.7.mlp.gate_proj",
84
+ "model.layers.9.mlp.up_proj",
85
+ "model.layers.15.mlp.up_proj",
86
+ "model.layers.22.mlp.gate_proj",
87
+ "model.layers.11.mlp.gate_proj",
88
+ "model.layers.18.mlp.gate_proj",
89
+ "model.layers.9.mlp.gate_proj",
90
+ "model.layers.20.mlp.up_proj",
91
+ "model.layers.0.mlp.gate_proj",
92
+ "model.layers.13.mlp.gate_proj",
93
+ "model.layers.16.mlp.gate_proj",
94
+ "model.layers.7.mlp.down_proj",
95
+ "model.layers.3.mlp.down_proj",
96
+ "model.layers.18.mlp.up_proj",
97
+ "model.layers.26.mlp.down_proj",
98
+ "model.layers.25.mlp.gate_proj",
99
+ "model.layers.24.mlp.down_proj",
100
+ "model.layers.21.mlp.gate_proj",
101
+ "model.layers.4.mlp.down_proj",
102
+ "model.layers.4.mlp.up_proj",
103
+ "model.layers.14.mlp.up_proj",
104
+ "model.layers.14.mlp.down_proj",
105
+ "model.layers.0.mlp.down_proj",
106
+ "model.layers.5.mlp.up_proj"
107
+ ],
108
+ "task_type": "CAUSAL_LM",
109
+ "use_dora": false,
110
+ "use_rslora": false
111
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/added_tokens.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</tool_call>": 151658,
3
+ "<tool_call>": 151657,
4
+ "<|box_end|>": 151649,
5
+ "<|box_start|>": 151648,
6
+ "<|endoftext|>": 151643,
7
+ "<|file_sep|>": 151664,
8
+ "<|fim_middle|>": 151660,
9
+ "<|fim_pad|>": 151662,
10
+ "<|fim_prefix|>": 151659,
11
+ "<|fim_suffix|>": 151661,
12
+ "<|im_end|>": 151645,
13
+ "<|im_start|>": 151644,
14
+ "<|image_pad|>": 151655,
15
+ "<|object_ref_end|>": 151647,
16
+ "<|object_ref_start|>": 151646,
17
+ "<|quad_end|>": 151651,
18
+ "<|quad_start|>": 151650,
19
+ "<|repo_name|>": 151663,
20
+ "<|video_pad|>": 151656,
21
+ "<|vision_end|>": 151653,
22
+ "<|vision_pad|>": 151654,
23
+ "<|vision_start|>": 151652
24
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/chat_template.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}"
3
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/adapter_config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alpha_pattern": {},
3
+ "auto_mapping": null,
4
+ "base_model_name_or_path": "/home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct",
5
+ "bias": "none",
6
+ "fan_in_fan_out": false,
7
+ "inference_mode": true,
8
+ "init_lora_weights": true,
9
+ "layer_replication": null,
10
+ "layers_pattern": null,
11
+ "layers_to_transform": null,
12
+ "loftq_config": {},
13
+ "lora_alpha": 16,
14
+ "lora_dropout": 0.05,
15
+ "megatron_config": null,
16
+ "megatron_core": "megatron.core",
17
+ "modules_to_save": null,
18
+ "peft_type": "LORA",
19
+ "r": 8,
20
+ "rank_pattern": {},
21
+ "revision": null,
22
+ "target_modules": [
23
+ "model.layers.26.mlp.up_proj",
24
+ "model.layers.20.mlp.down_proj",
25
+ "model.layers.10.mlp.gate_proj",
26
+ "model.layers.21.mlp.up_proj",
27
+ "model.layers.1.mlp.gate_proj",
28
+ "model.layers.8.mlp.down_proj",
29
+ "model.layers.15.mlp.down_proj",
30
+ "model.layers.20.mlp.gate_proj",
31
+ "model.layers.19.mlp.up_proj",
32
+ "model.layers.12.mlp.gate_proj",
33
+ "model.layers.5.mlp.gate_proj",
34
+ "model.layers.23.mlp.up_proj",
35
+ "model.layers.27.mlp.down_proj",
36
+ "model.layers.21.mlp.down_proj",
37
+ "model.layers.22.mlp.down_proj",
38
+ "model.layers.11.mlp.down_proj",
39
+ "model.layers.6.mlp.gate_proj",
40
+ "model.layers.17.mlp.gate_proj",
41
+ "model.layers.6.mlp.up_proj",
42
+ "model.layers.25.mlp.down_proj",
43
+ "model.layers.5.mlp.down_proj",
44
+ "model.layers.17.mlp.down_proj",
45
+ "model.layers.10.mlp.up_proj",
46
+ "model.layers.14.mlp.gate_proj",
47
+ "model.layers.6.mlp.down_proj",
48
+ "model.layers.3.mlp.up_proj",
49
+ "model.layers.27.mlp.up_proj",
50
+ "model.layers.8.mlp.up_proj",
51
+ "model.layers.16.mlp.up_proj",
52
+ "model.layers.27.mlp.gate_proj",
53
+ "model.layers.26.mlp.gate_proj",
54
+ "model.layers.23.mlp.gate_proj",
55
+ "model.layers.19.mlp.down_proj",
56
+ "model.layers.13.mlp.down_proj",
57
+ "model.layers.17.mlp.up_proj",
58
+ "model.layers.8.mlp.gate_proj",
59
+ "model.layers.2.mlp.gate_proj",
60
+ "model.layers.12.mlp.down_proj",
61
+ "model.layers.23.mlp.down_proj",
62
+ "model.layers.3.mlp.gate_proj",
63
+ "model.layers.22.mlp.up_proj",
64
+ "model.layers.12.mlp.up_proj",
65
+ "model.layers.2.mlp.down_proj",
66
+ "model.layers.7.mlp.up_proj",
67
+ "model.layers.25.mlp.up_proj",
68
+ "model.layers.11.mlp.up_proj",
69
+ "model.layers.4.mlp.gate_proj",
70
+ "model.layers.16.mlp.down_proj",
71
+ "model.layers.1.mlp.up_proj",
72
+ "model.layers.10.mlp.down_proj",
73
+ "model.layers.2.mlp.up_proj",
74
+ "model.layers.1.mlp.down_proj",
75
+ "model.layers.15.mlp.gate_proj",
76
+ "model.layers.24.mlp.gate_proj",
77
+ "model.layers.13.mlp.up_proj",
78
+ "model.layers.0.mlp.up_proj",
79
+ "model.layers.24.mlp.up_proj",
80
+ "model.layers.19.mlp.gate_proj",
81
+ "model.layers.9.mlp.down_proj",
82
+ "model.layers.18.mlp.down_proj",
83
+ "model.layers.7.mlp.gate_proj",
84
+ "model.layers.9.mlp.up_proj",
85
+ "model.layers.15.mlp.up_proj",
86
+ "model.layers.22.mlp.gate_proj",
87
+ "model.layers.11.mlp.gate_proj",
88
+ "model.layers.18.mlp.gate_proj",
89
+ "model.layers.9.mlp.gate_proj",
90
+ "model.layers.20.mlp.up_proj",
91
+ "model.layers.0.mlp.gate_proj",
92
+ "model.layers.13.mlp.gate_proj",
93
+ "model.layers.16.mlp.gate_proj",
94
+ "model.layers.7.mlp.down_proj",
95
+ "model.layers.3.mlp.down_proj",
96
+ "model.layers.18.mlp.up_proj",
97
+ "model.layers.26.mlp.down_proj",
98
+ "model.layers.25.mlp.gate_proj",
99
+ "model.layers.24.mlp.down_proj",
100
+ "model.layers.21.mlp.gate_proj",
101
+ "model.layers.4.mlp.down_proj",
102
+ "model.layers.4.mlp.up_proj",
103
+ "model.layers.14.mlp.up_proj",
104
+ "model.layers.14.mlp.down_proj",
105
+ "model.layers.0.mlp.down_proj",
106
+ "model.layers.5.mlp.up_proj"
107
+ ],
108
+ "task_type": "CAUSAL_LM",
109
+ "use_dora": false,
110
+ "use_rslora": false
111
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/added_tokens.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "</tool_call>": 151658,
3
+ "<tool_call>": 151657,
4
+ "<|box_end|>": 151649,
5
+ "<|box_start|>": 151648,
6
+ "<|endoftext|>": 151643,
7
+ "<|file_sep|>": 151664,
8
+ "<|fim_middle|>": 151660,
9
+ "<|fim_pad|>": 151662,
10
+ "<|fim_prefix|>": 151659,
11
+ "<|fim_suffix|>": 151661,
12
+ "<|im_end|>": 151645,
13
+ "<|im_start|>": 151644,
14
+ "<|image_pad|>": 151655,
15
+ "<|object_ref_end|>": 151647,
16
+ "<|object_ref_start|>": 151646,
17
+ "<|quad_end|>": 151651,
18
+ "<|quad_start|>": 151650,
19
+ "<|repo_name|>": 151663,
20
+ "<|video_pad|>": 151656,
21
+ "<|vision_end|>": 151653,
22
+ "<|vision_pad|>": 151654,
23
+ "<|vision_start|>": 151652
24
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step100
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/special_tokens_map.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>",
5
+ "<|object_ref_start|>",
6
+ "<|object_ref_end|>",
7
+ "<|box_start|>",
8
+ "<|box_end|>",
9
+ "<|quad_start|>",
10
+ "<|quad_end|>",
11
+ "<|vision_start|>",
12
+ "<|vision_end|>",
13
+ "<|vision_pad|>",
14
+ "<|image_pad|>",
15
+ "<|video_pad|>"
16
+ ],
17
+ "eos_token": {
18
+ "content": "<|im_end|>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ },
24
+ "pad_token": {
25
+ "content": "<|endoftext|>",
26
+ "lstrip": false,
27
+ "normalized": false,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ }
31
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/tokenizer_config.json ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ }
181
+ },
182
+ "additional_special_tokens": [
183
+ "<|im_start|>",
184
+ "<|im_end|>",
185
+ "<|object_ref_start|>",
186
+ "<|object_ref_end|>",
187
+ "<|box_start|>",
188
+ "<|box_end|>",
189
+ "<|quad_start|>",
190
+ "<|quad_end|>",
191
+ "<|vision_start|>",
192
+ "<|vision_end|>",
193
+ "<|vision_pad|>",
194
+ "<|image_pad|>",
195
+ "<|video_pad|>"
196
+ ],
197
+ "bos_token": null,
198
+ "chat_template": "{% set image_count = namespace(value=0) %}{% set video_count = namespace(value=0) %}{% for message in messages %}{% if loop.first and message['role'] != 'system' %}<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n{% endif %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}<|im_end|>\n{% else %}{% for content in message['content'] %}{% if content['type'] == 'image' or 'image' in content or 'image_url' in content %}{% set image_count.value = image_count.value + 1 %}{% if add_vision_id %}Picture {{ image_count.value }}: {% endif %}<|vision_start|><|image_pad|><|vision_end|>{% elif content['type'] == 'video' or 'video' in content %}{% set video_count.value = video_count.value + 1 %}{% if add_vision_id %}Video {{ video_count.value }}: {% endif %}<|vision_start|><|video_pad|><|vision_end|>{% elif 'text' in content %}{{ content['text'] }}{% endif %}{% endfor %}<|im_end|>\n{% endif %}{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
199
+ "clean_up_tokenization_spaces": false,
200
+ "eos_token": "<|im_end|>",
201
+ "errors": "replace",
202
+ "extra_special_tokens": {},
203
+ "model_max_length": 8192,
204
+ "pad_token": "<|endoftext|>",
205
+ "padding_side": "right",
206
+ "split_special_tokens": false,
207
+ "tokenizer_class": "Qwen2Tokenizer",
208
+ "truncation": true,
209
+ "unk_token": null
210
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/trainer_state.json ADDED
@@ -0,0 +1,733 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.21505376344086022,
5
+ "eval_steps": 500,
6
+ "global_step": 100,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.002150537634408602,
13
+ "grad_norm": 1.028775291281574,
14
+ "learning_rate": 2.8571428571428573e-06,
15
+ "loss": 2.715,
16
+ "step": 1
17
+ },
18
+ {
19
+ "epoch": 0.004301075268817204,
20
+ "grad_norm": 1.266146082989029,
21
+ "learning_rate": 5.7142857142857145e-06,
22
+ "loss": 2.7625,
23
+ "step": 2
24
+ },
25
+ {
26
+ "epoch": 0.0064516129032258064,
27
+ "grad_norm": 1.3851458971137458,
28
+ "learning_rate": 8.571428571428571e-06,
29
+ "loss": 2.896,
30
+ "step": 3
31
+ },
32
+ {
33
+ "epoch": 0.008602150537634409,
34
+ "grad_norm": 0.5674885630538635,
35
+ "learning_rate": 1.1428571428571429e-05,
36
+ "loss": 1.873,
37
+ "step": 4
38
+ },
39
+ {
40
+ "epoch": 0.010752688172043012,
41
+ "grad_norm": 0.9123372587725512,
42
+ "learning_rate": 1.4285714285714287e-05,
43
+ "loss": 2.3067,
44
+ "step": 5
45
+ },
46
+ {
47
+ "epoch": 0.012903225806451613,
48
+ "grad_norm": 1.515297354758064,
49
+ "learning_rate": 1.7142857142857142e-05,
50
+ "loss": 3.0951,
51
+ "step": 6
52
+ },
53
+ {
54
+ "epoch": 0.015053763440860216,
55
+ "grad_norm": 1.2120983331352562,
56
+ "learning_rate": 2e-05,
57
+ "loss": 2.4116,
58
+ "step": 7
59
+ },
60
+ {
61
+ "epoch": 0.017204301075268817,
62
+ "grad_norm": 1.3093476695142727,
63
+ "learning_rate": 1.9999033847063813e-05,
64
+ "loss": 3.0779,
65
+ "step": 8
66
+ },
67
+ {
68
+ "epoch": 0.01935483870967742,
69
+ "grad_norm": 0.928908221297413,
70
+ "learning_rate": 1.9996135574945543e-05,
71
+ "loss": 2.2262,
72
+ "step": 9
73
+ },
74
+ {
75
+ "epoch": 0.021505376344086023,
76
+ "grad_norm": 2.052272976973174,
77
+ "learning_rate": 1.9991305743680013e-05,
78
+ "loss": 3.2259,
79
+ "step": 10
80
+ },
81
+ {
82
+ "epoch": 0.023655913978494623,
83
+ "grad_norm": 1.5575663384495326,
84
+ "learning_rate": 1.9984545286538362e-05,
85
+ "loss": 2.9672,
86
+ "step": 11
87
+ },
88
+ {
89
+ "epoch": 0.025806451612903226,
90
+ "grad_norm": 2.2344175402247153,
91
+ "learning_rate": 1.9975855509847688e-05,
92
+ "loss": 2.8637,
93
+ "step": 12
94
+ },
95
+ {
96
+ "epoch": 0.02795698924731183,
97
+ "grad_norm": 1.291084829324887,
98
+ "learning_rate": 1.9965238092738643e-05,
99
+ "loss": 2.5383,
100
+ "step": 13
101
+ },
102
+ {
103
+ "epoch": 0.030107526881720432,
104
+ "grad_norm": 1.1948462215586322,
105
+ "learning_rate": 1.9952695086820975e-05,
106
+ "loss": 1.8373,
107
+ "step": 14
108
+ },
109
+ {
110
+ "epoch": 0.03225806451612903,
111
+ "grad_norm": 2.875201355103465,
112
+ "learning_rate": 1.993822891578708e-05,
113
+ "loss": 2.6141,
114
+ "step": 15
115
+ },
116
+ {
117
+ "epoch": 0.034408602150537634,
118
+ "grad_norm": 3.138127792864874,
119
+ "learning_rate": 1.9921842374943682e-05,
120
+ "loss": 2.9965,
121
+ "step": 16
122
+ },
123
+ {
124
+ "epoch": 0.03655913978494624,
125
+ "grad_norm": 2.1535994216149765,
126
+ "learning_rate": 1.9903538630671687e-05,
127
+ "loss": 2.622,
128
+ "step": 17
129
+ },
130
+ {
131
+ "epoch": 0.03870967741935484,
132
+ "grad_norm": 1.95618716789349,
133
+ "learning_rate": 1.988332121981436e-05,
134
+ "loss": 1.9772,
135
+ "step": 18
136
+ },
137
+ {
138
+ "epoch": 0.04086021505376344,
139
+ "grad_norm": 2.420473463150693,
140
+ "learning_rate": 1.9861194048993865e-05,
141
+ "loss": 2.1071,
142
+ "step": 19
143
+ },
144
+ {
145
+ "epoch": 0.043010752688172046,
146
+ "grad_norm": 1.9745929108775373,
147
+ "learning_rate": 1.9837161393856413e-05,
148
+ "loss": 1.7868,
149
+ "step": 20
150
+ },
151
+ {
152
+ "epoch": 0.04516129032258064,
153
+ "grad_norm": 1.9816697704173123,
154
+ "learning_rate": 1.9811227898246072e-05,
155
+ "loss": 1.95,
156
+ "step": 21
157
+ },
158
+ {
159
+ "epoch": 0.047311827956989246,
160
+ "grad_norm": 2.3176350354606994,
161
+ "learning_rate": 1.978339857330743e-05,
162
+ "loss": 2.163,
163
+ "step": 22
164
+ },
165
+ {
166
+ "epoch": 0.04946236559139785,
167
+ "grad_norm": 1.7838340912469495,
168
+ "learning_rate": 1.975367879651728e-05,
169
+ "loss": 2.1613,
170
+ "step": 23
171
+ },
172
+ {
173
+ "epoch": 0.05161290322580645,
174
+ "grad_norm": 1.5618077475296102,
175
+ "learning_rate": 1.9722074310645553e-05,
176
+ "loss": 1.6462,
177
+ "step": 24
178
+ },
179
+ {
180
+ "epoch": 0.053763440860215055,
181
+ "grad_norm": 2.0482724129048915,
182
+ "learning_rate": 1.9688591222645607e-05,
183
+ "loss": 1.8061,
184
+ "step": 25
185
+ },
186
+ {
187
+ "epoch": 0.05591397849462366,
188
+ "grad_norm": 2.635346759837836,
189
+ "learning_rate": 1.9653236002474202e-05,
190
+ "loss": 2.2056,
191
+ "step": 26
192
+ },
193
+ {
194
+ "epoch": 0.05806451612903226,
195
+ "grad_norm": 0.9868363545154378,
196
+ "learning_rate": 1.9616015481841293e-05,
197
+ "loss": 1.7104,
198
+ "step": 27
199
+ },
200
+ {
201
+ "epoch": 0.060215053763440864,
202
+ "grad_norm": 2.153089415805509,
203
+ "learning_rate": 1.9576936852889937e-05,
204
+ "loss": 1.4439,
205
+ "step": 28
206
+ },
207
+ {
208
+ "epoch": 0.06236559139784946,
209
+ "grad_norm": 1.509335304272552,
210
+ "learning_rate": 1.9536007666806555e-05,
211
+ "loss": 1.764,
212
+ "step": 29
213
+ },
214
+ {
215
+ "epoch": 0.06451612903225806,
216
+ "grad_norm": 0.9232815858817741,
217
+ "learning_rate": 1.9493235832361812e-05,
218
+ "loss": 1.6303,
219
+ "step": 30
220
+ },
221
+ {
222
+ "epoch": 0.06666666666666667,
223
+ "grad_norm": 1.3776304755623878,
224
+ "learning_rate": 1.9448629614382394e-05,
225
+ "loss": 1.6826,
226
+ "step": 31
227
+ },
228
+ {
229
+ "epoch": 0.06881720430107527,
230
+ "grad_norm": 1.5485228920731133,
231
+ "learning_rate": 1.940219763215399e-05,
232
+ "loss": 1.6536,
233
+ "step": 32
234
+ },
235
+ {
236
+ "epoch": 0.07096774193548387,
237
+ "grad_norm": 1.065751280033005,
238
+ "learning_rate": 1.93539488577558e-05,
239
+ "loss": 1.385,
240
+ "step": 33
241
+ },
242
+ {
243
+ "epoch": 0.07311827956989247,
244
+ "grad_norm": 2.0448812482483465,
245
+ "learning_rate": 1.9303892614326835e-05,
246
+ "loss": 1.564,
247
+ "step": 34
248
+ },
249
+ {
250
+ "epoch": 0.07526881720430108,
251
+ "grad_norm": 0.6452066748687381,
252
+ "learning_rate": 1.9252038574264403e-05,
253
+ "loss": 1.564,
254
+ "step": 35
255
+ },
256
+ {
257
+ "epoch": 0.07741935483870968,
258
+ "grad_norm": 0.9508761639354778,
259
+ "learning_rate": 1.9198396757355118e-05,
260
+ "loss": 1.673,
261
+ "step": 36
262
+ },
263
+ {
264
+ "epoch": 0.07956989247311828,
265
+ "grad_norm": 0.8634819581273068,
266
+ "learning_rate": 1.9142977528838763e-05,
267
+ "loss": 1.5554,
268
+ "step": 37
269
+ },
270
+ {
271
+ "epoch": 0.08172043010752689,
272
+ "grad_norm": 0.8949043267226644,
273
+ "learning_rate": 1.9085791597405404e-05,
274
+ "loss": 1.7128,
275
+ "step": 38
276
+ },
277
+ {
278
+ "epoch": 0.08387096774193549,
279
+ "grad_norm": 1.10486255302291,
280
+ "learning_rate": 1.902685001312616e-05,
281
+ "loss": 1.6443,
282
+ "step": 39
283
+ },
284
+ {
285
+ "epoch": 0.08602150537634409,
286
+ "grad_norm": 0.9013353885296563,
287
+ "learning_rate": 1.8966164165317968e-05,
288
+ "loss": 1.4231,
289
+ "step": 40
290
+ },
291
+ {
292
+ "epoch": 0.08817204301075268,
293
+ "grad_norm": 0.9823275438703282,
294
+ "learning_rate": 1.8903745780342838e-05,
295
+ "loss": 1.443,
296
+ "step": 41
297
+ },
298
+ {
299
+ "epoch": 0.09032258064516129,
300
+ "grad_norm": 0.9186730751194769,
301
+ "learning_rate": 1.883960691934196e-05,
302
+ "loss": 1.5533,
303
+ "step": 42
304
+ },
305
+ {
306
+ "epoch": 0.09247311827956989,
307
+ "grad_norm": 0.9738337615893106,
308
+ "learning_rate": 1.8773759975905098e-05,
309
+ "loss": 1.7712,
310
+ "step": 43
311
+ },
312
+ {
313
+ "epoch": 0.09462365591397849,
314
+ "grad_norm": 0.8077228399501873,
315
+ "learning_rate": 1.8706217673675813e-05,
316
+ "loss": 1.5709,
317
+ "step": 44
318
+ },
319
+ {
320
+ "epoch": 0.0967741935483871,
321
+ "grad_norm": 0.7962799187209563,
322
+ "learning_rate": 1.8636993063892822e-05,
323
+ "loss": 1.4815,
324
+ "step": 45
325
+ },
326
+ {
327
+ "epoch": 0.0989247311827957,
328
+ "grad_norm": 0.7355643742784329,
329
+ "learning_rate": 1.8566099522868118e-05,
330
+ "loss": 1.3242,
331
+ "step": 46
332
+ },
333
+ {
334
+ "epoch": 0.1010752688172043,
335
+ "grad_norm": 0.8260087812542498,
336
+ "learning_rate": 1.8493550749402278e-05,
337
+ "loss": 1.3694,
338
+ "step": 47
339
+ },
340
+ {
341
+ "epoch": 0.1032258064516129,
342
+ "grad_norm": 0.8248563942827724,
343
+ "learning_rate": 1.8419360762137395e-05,
344
+ "loss": 1.5249,
345
+ "step": 48
346
+ },
347
+ {
348
+ "epoch": 0.1053763440860215,
349
+ "grad_norm": 0.9460042403500621,
350
+ "learning_rate": 1.8343543896848275e-05,
351
+ "loss": 1.5733,
352
+ "step": 49
353
+ },
354
+ {
355
+ "epoch": 0.10752688172043011,
356
+ "grad_norm": 1.0720194683533255,
357
+ "learning_rate": 1.826611480367232e-05,
358
+ "loss": 1.6769,
359
+ "step": 50
360
+ },
361
+ {
362
+ "epoch": 0.10967741935483871,
363
+ "grad_norm": 0.702058029783758,
364
+ "learning_rate": 1.8187088444278675e-05,
365
+ "loss": 1.3673,
366
+ "step": 51
367
+ },
368
+ {
369
+ "epoch": 0.11182795698924732,
370
+ "grad_norm": 1.0517510818973002,
371
+ "learning_rate": 1.8106480088977174e-05,
372
+ "loss": 1.1166,
373
+ "step": 52
374
+ },
375
+ {
376
+ "epoch": 0.11397849462365592,
377
+ "grad_norm": 1.5406409470877767,
378
+ "learning_rate": 1.8024305313767648e-05,
379
+ "loss": 1.5848,
380
+ "step": 53
381
+ },
382
+ {
383
+ "epoch": 0.11612903225806452,
384
+ "grad_norm": 0.8618867226265116,
385
+ "learning_rate": 1.7940579997330167e-05,
386
+ "loss": 1.4745,
387
+ "step": 54
388
+ },
389
+ {
390
+ "epoch": 0.11827956989247312,
391
+ "grad_norm": 0.6633326950074517,
392
+ "learning_rate": 1.7855320317956785e-05,
393
+ "loss": 1.5198,
394
+ "step": 55
395
+ },
396
+ {
397
+ "epoch": 0.12043010752688173,
398
+ "grad_norm": 0.7398449096547106,
399
+ "learning_rate": 1.7768542750425427e-05,
400
+ "loss": 1.3486,
401
+ "step": 56
402
+ },
403
+ {
404
+ "epoch": 0.12258064516129032,
405
+ "grad_norm": 0.8155703087336776,
406
+ "learning_rate": 1.768026406281642e-05,
407
+ "loss": 1.5291,
408
+ "step": 57
409
+ },
410
+ {
411
+ "epoch": 0.12473118279569892,
412
+ "grad_norm": 0.7800572343365931,
413
+ "learning_rate": 1.7590501313272415e-05,
414
+ "loss": 1.2441,
415
+ "step": 58
416
+ },
417
+ {
418
+ "epoch": 0.12688172043010754,
419
+ "grad_norm": 0.6936361669036757,
420
+ "learning_rate": 1.7499271846702216e-05,
421
+ "loss": 1.4205,
422
+ "step": 59
423
+ },
424
+ {
425
+ "epoch": 0.12903225806451613,
426
+ "grad_norm": 0.699424086290134,
427
+ "learning_rate": 1.740659329142922e-05,
428
+ "loss": 1.3268,
429
+ "step": 60
430
+ },
431
+ {
432
+ "epoch": 0.13118279569892474,
433
+ "grad_norm": 0.6130388326539034,
434
+ "learning_rate": 1.7312483555785087e-05,
435
+ "loss": 1.4501,
436
+ "step": 61
437
+ },
438
+ {
439
+ "epoch": 0.13333333333333333,
440
+ "grad_norm": 0.7485605942529221,
441
+ "learning_rate": 1.7216960824649304e-05,
442
+ "loss": 1.5062,
443
+ "step": 62
444
+ },
445
+ {
446
+ "epoch": 0.13548387096774195,
447
+ "grad_norm": 0.6671686223696721,
448
+ "learning_rate": 1.71200435559353e-05,
449
+ "loss": 1.2843,
450
+ "step": 63
451
+ },
452
+ {
453
+ "epoch": 0.13763440860215054,
454
+ "grad_norm": 0.9744442977718708,
455
+ "learning_rate": 1.7021750477023823e-05,
456
+ "loss": 1.5766,
457
+ "step": 64
458
+ },
459
+ {
460
+ "epoch": 0.13978494623655913,
461
+ "grad_norm": 0.6622529599922087,
462
+ "learning_rate": 1.6922100581144228e-05,
463
+ "loss": 1.3552,
464
+ "step": 65
465
+ },
466
+ {
467
+ "epoch": 0.14193548387096774,
468
+ "grad_norm": 0.7428301920959937,
469
+ "learning_rate": 1.6821113123704425e-05,
470
+ "loss": 1.423,
471
+ "step": 66
472
+ },
473
+ {
474
+ "epoch": 0.14408602150537633,
475
+ "grad_norm": 0.743916766078708,
476
+ "learning_rate": 1.671880761857011e-05,
477
+ "loss": 1.353,
478
+ "step": 67
479
+ },
480
+ {
481
+ "epoch": 0.14623655913978495,
482
+ "grad_norm": 0.6712833874716213,
483
+ "learning_rate": 1.661520383429412e-05,
484
+ "loss": 1.41,
485
+ "step": 68
486
+ },
487
+ {
488
+ "epoch": 0.14838709677419354,
489
+ "grad_norm": 1.0241065570826338,
490
+ "learning_rate": 1.6510321790296527e-05,
491
+ "loss": 1.701,
492
+ "step": 69
493
+ },
494
+ {
495
+ "epoch": 0.15053763440860216,
496
+ "grad_norm": 0.70329681139763,
497
+ "learning_rate": 1.6404181752996287e-05,
498
+ "loss": 1.3304,
499
+ "step": 70
500
+ },
501
+ {
502
+ "epoch": 0.15268817204301074,
503
+ "grad_norm": 0.7850717333768847,
504
+ "learning_rate": 1.629680423189514e-05,
505
+ "loss": 1.4618,
506
+ "step": 71
507
+ },
508
+ {
509
+ "epoch": 0.15483870967741936,
510
+ "grad_norm": 0.8582680765419307,
511
+ "learning_rate": 1.618820997561454e-05,
512
+ "loss": 1.6404,
513
+ "step": 72
514
+ },
515
+ {
516
+ "epoch": 0.15698924731182795,
517
+ "grad_norm": 0.9628131681536289,
518
+ "learning_rate": 1.6078419967886402e-05,
519
+ "loss": 1.5491,
520
+ "step": 73
521
+ },
522
+ {
523
+ "epoch": 0.15913978494623657,
524
+ "grad_norm": 1.078269260552029,
525
+ "learning_rate": 1.5967455423498387e-05,
526
+ "loss": 1.4008,
527
+ "step": 74
528
+ },
529
+ {
530
+ "epoch": 0.16129032258064516,
531
+ "grad_norm": 1.1623275690656032,
532
+ "learning_rate": 1.5855337784194576e-05,
533
+ "loss": 1.1836,
534
+ "step": 75
535
+ },
536
+ {
537
+ "epoch": 0.16344086021505377,
538
+ "grad_norm": 0.6228213495971294,
539
+ "learning_rate": 1.5742088714532247e-05,
540
+ "loss": 1.2516,
541
+ "step": 76
542
+ },
543
+ {
544
+ "epoch": 0.16559139784946236,
545
+ "grad_norm": 0.755068217142837,
546
+ "learning_rate": 1.562773009769564e-05,
547
+ "loss": 1.446,
548
+ "step": 77
549
+ },
550
+ {
551
+ "epoch": 0.16774193548387098,
552
+ "grad_norm": 0.9555557047128778,
553
+ "learning_rate": 1.551228403126744e-05,
554
+ "loss": 1.4804,
555
+ "step": 78
556
+ },
557
+ {
558
+ "epoch": 0.16989247311827957,
559
+ "grad_norm": 1.0071598497600383,
560
+ "learning_rate": 1.5395772822958844e-05,
561
+ "loss": 1.4,
562
+ "step": 79
563
+ },
564
+ {
565
+ "epoch": 0.17204301075268819,
566
+ "grad_norm": 0.9947465871321163,
567
+ "learning_rate": 1.5278218986299074e-05,
568
+ "loss": 1.4018,
569
+ "step": 80
570
+ },
571
+ {
572
+ "epoch": 0.17419354838709677,
573
+ "grad_norm": 0.7374223168455896,
574
+ "learning_rate": 1.515964523628501e-05,
575
+ "loss": 1.3364,
576
+ "step": 81
577
+ },
578
+ {
579
+ "epoch": 0.17634408602150536,
580
+ "grad_norm": 0.6054573297102332,
581
+ "learning_rate": 1.5040074484992e-05,
582
+ "loss": 1.3257,
583
+ "step": 82
584
+ },
585
+ {
586
+ "epoch": 0.17849462365591398,
587
+ "grad_norm": 0.6976335106890922,
588
+ "learning_rate": 1.4919529837146529e-05,
589
+ "loss": 1.4394,
590
+ "step": 83
591
+ },
592
+ {
593
+ "epoch": 0.18064516129032257,
594
+ "grad_norm": 0.7886961885602806,
595
+ "learning_rate": 1.4798034585661696e-05,
596
+ "loss": 1.4789,
597
+ "step": 84
598
+ },
599
+ {
600
+ "epoch": 0.1827956989247312,
601
+ "grad_norm": 0.751839328109352,
602
+ "learning_rate": 1.4675612207136283e-05,
603
+ "loss": 1.3444,
604
+ "step": 85
605
+ },
606
+ {
607
+ "epoch": 0.18494623655913978,
608
+ "grad_norm": 0.8929358539574412,
609
+ "learning_rate": 1.455228635731839e-05,
610
+ "loss": 1.3253,
611
+ "step": 86
612
+ },
613
+ {
614
+ "epoch": 0.1870967741935484,
615
+ "grad_norm": 0.7845523329556532,
616
+ "learning_rate": 1.4428080866534397e-05,
617
+ "loss": 1.4701,
618
+ "step": 87
619
+ },
620
+ {
621
+ "epoch": 0.18924731182795698,
622
+ "grad_norm": 0.8799266694008883,
623
+ "learning_rate": 1.4303019735084225e-05,
624
+ "loss": 1.3015,
625
+ "step": 88
626
+ },
627
+ {
628
+ "epoch": 0.1913978494623656,
629
+ "grad_norm": 1.0615758722787252,
630
+ "learning_rate": 1.4177127128603748e-05,
631
+ "loss": 1.6194,
632
+ "step": 89
633
+ },
634
+ {
635
+ "epoch": 0.1935483870967742,
636
+ "grad_norm": 0.9450936022711952,
637
+ "learning_rate": 1.4050427373395241e-05,
638
+ "loss": 1.3971,
639
+ "step": 90
640
+ },
641
+ {
642
+ "epoch": 0.1956989247311828,
643
+ "grad_norm": 0.7530041768052385,
644
+ "learning_rate": 1.3922944951726811e-05,
645
+ "loss": 1.3859,
646
+ "step": 91
647
+ },
648
+ {
649
+ "epoch": 0.1978494623655914,
650
+ "grad_norm": 0.8243741259823598,
651
+ "learning_rate": 1.3794704497101656e-05,
652
+ "loss": 1.2362,
653
+ "step": 92
654
+ },
655
+ {
656
+ "epoch": 0.2,
657
+ "grad_norm": 0.9011204125275231,
658
+ "learning_rate": 1.366573078949813e-05,
659
+ "loss": 1.5122,
660
+ "step": 93
661
+ },
662
+ {
663
+ "epoch": 0.2021505376344086,
664
+ "grad_norm": 0.7661850660795133,
665
+ "learning_rate": 1.3536048750581494e-05,
666
+ "loss": 1.3072,
667
+ "step": 94
668
+ },
669
+ {
670
+ "epoch": 0.20430107526881722,
671
+ "grad_norm": 0.8427897450324289,
672
+ "learning_rate": 1.3405683438888281e-05,
673
+ "loss": 1.4549,
674
+ "step": 95
675
+ },
676
+ {
677
+ "epoch": 0.2064516129032258,
678
+ "grad_norm": 1.007048569687895,
679
+ "learning_rate": 1.3274660044984225e-05,
680
+ "loss": 1.7148,
681
+ "step": 96
682
+ },
683
+ {
684
+ "epoch": 0.2086021505376344,
685
+ "grad_norm": 0.6474391620005616,
686
+ "learning_rate": 1.314300388659667e-05,
687
+ "loss": 1.4839,
688
+ "step": 97
689
+ },
690
+ {
691
+ "epoch": 0.210752688172043,
692
+ "grad_norm": 0.6908140970468737,
693
+ "learning_rate": 1.301074040372242e-05,
694
+ "loss": 1.4624,
695
+ "step": 98
696
+ },
697
+ {
698
+ "epoch": 0.2129032258064516,
699
+ "grad_norm": 0.8287286090888102,
700
+ "learning_rate": 1.2877895153711935e-05,
701
+ "loss": 1.2145,
702
+ "step": 99
703
+ },
704
+ {
705
+ "epoch": 0.21505376344086022,
706
+ "grad_norm": 0.9198148395415796,
707
+ "learning_rate": 1.274449380633089e-05,
708
+ "loss": 1.5292,
709
+ "step": 100
710
+ }
711
+ ],
712
+ "logging_steps": 1.0,
713
+ "max_steps": 233,
714
+ "num_input_tokens_seen": 0,
715
+ "num_train_epochs": 1,
716
+ "save_steps": 100,
717
+ "stateful_callbacks": {
718
+ "TrainerControl": {
719
+ "args": {
720
+ "should_epoch_stop": false,
721
+ "should_evaluate": false,
722
+ "should_log": false,
723
+ "should_save": true,
724
+ "should_training_stop": false
725
+ },
726
+ "attributes": {}
727
+ }
728
+ },
729
+ "total_flos": 7049250738176.0,
730
+ "train_batch_size": 8,
731
+ "trial_name": null,
732
+ "trial_params": null
733
+ }
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-100/zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)
InternVL/Qwen2.5-VL/finetune_checkpoints/Qwen2.5-vl-7b-lora-vision-origianl/checkpoint-200/README.md ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: /home1/gmh/InternVL/Qwen2.5-VL/models/Qwen2.5-VL-7B-Instruct
3
+ library_name: peft
4
+ ---
5
+
6
+ # Model Card for Model ID
7
+
8
+ <!-- Provide a quick summary of what the model is/does. -->
9
+
10
+
11
+
12
+ ## Model Details
13
+
14
+ ### Model Description
15
+
16
+ <!-- Provide a longer summary of what this model is. -->
17
+
18
+
19
+
20
+ - **Developed by:** [More Information Needed]
21
+ - **Funded by [optional]:** [More Information Needed]
22
+ - **Shared by [optional]:** [More Information Needed]
23
+ - **Model type:** [More Information Needed]
24
+ - **Language(s) (NLP):** [More Information Needed]
25
+ - **License:** [More Information Needed]
26
+ - **Finetuned from model [optional]:** [More Information Needed]
27
+
28
+ ### Model Sources [optional]
29
+
30
+ <!-- Provide the basic links for the model. -->
31
+
32
+ - **Repository:** [More Information Needed]
33
+ - **Paper [optional]:** [More Information Needed]
34
+ - **Demo [optional]:** [More Information Needed]
35
+
36
+ ## Uses
37
+
38
+ <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
+
40
+ ### Direct Use
41
+
42
+ <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
+
44
+ [More Information Needed]
45
+
46
+ ### Downstream Use [optional]
47
+
48
+ <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
+
50
+ [More Information Needed]
51
+
52
+ ### Out-of-Scope Use
53
+
54
+ <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
+
56
+ [More Information Needed]
57
+
58
+ ## Bias, Risks, and Limitations
59
+
60
+ <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
+
62
+ [More Information Needed]
63
+
64
+ ### Recommendations
65
+
66
+ <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
+
68
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
+
70
+ ## How to Get Started with the Model
71
+
72
+ Use the code below to get started with the model.
73
+
74
+ [More Information Needed]
75
+
76
+ ## Training Details
77
+
78
+ ### Training Data
79
+
80
+ <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
+
82
+ [More Information Needed]
83
+
84
+ ### Training Procedure
85
+
86
+ <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
+
88
+ #### Preprocessing [optional]
89
+
90
+ [More Information Needed]
91
+
92
+
93
+ #### Training Hyperparameters
94
+
95
+ - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
+
97
+ #### Speeds, Sizes, Times [optional]
98
+
99
+ <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
+
101
+ [More Information Needed]
102
+
103
+ ## Evaluation
104
+
105
+ <!-- This section describes the evaluation protocols and provides the results. -->
106
+
107
+ ### Testing Data, Factors & Metrics
108
+
109
+ #### Testing Data
110
+
111
+ <!-- This should link to a Dataset Card if possible. -->
112
+
113
+ [More Information Needed]
114
+
115
+ #### Factors
116
+
117
+ <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
+
119
+ [More Information Needed]
120
+
121
+ #### Metrics
122
+
123
+ <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
+
125
+ [More Information Needed]
126
+
127
+ ### Results
128
+
129
+ [More Information Needed]
130
+
131
+ #### Summary
132
+
133
+
134
+
135
+ ## Model Examination [optional]
136
+
137
+ <!-- Relevant interpretability work for the model goes here -->
138
+
139
+ [More Information Needed]
140
+
141
+ ## Environmental Impact
142
+
143
+ <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
+
145
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
+
147
+ - **Hardware Type:** [More Information Needed]
148
+ - **Hours used:** [More Information Needed]
149
+ - **Cloud Provider:** [More Information Needed]
150
+ - **Compute Region:** [More Information Needed]
151
+ - **Carbon Emitted:** [More Information Needed]
152
+
153
+ ## Technical Specifications [optional]
154
+
155
+ ### Model Architecture and Objective
156
+
157
+ [More Information Needed]
158
+
159
+ ### Compute Infrastructure
160
+
161
+ [More Information Needed]
162
+
163
+ #### Hardware
164
+
165
+ [More Information Needed]
166
+
167
+ #### Software
168
+
169
+ [More Information Needed]
170
+
171
+ ## Citation [optional]
172
+
173
+ <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
+
175
+ **BibTeX:**
176
+
177
+ [More Information Needed]
178
+
179
+ **APA:**
180
+
181
+ [More Information Needed]
182
+
183
+ ## Glossary [optional]
184
+
185
+ <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
+
187
+ [More Information Needed]
188
+
189
+ ## More Information [optional]
190
+
191
+ [More Information Needed]
192
+
193
+ ## Model Card Authors [optional]
194
+
195
+ [More Information Needed]
196
+
197
+ ## Model Card Contact
198
+
199
+ [More Information Needed]
200
+ ### Framework versions
201
+
202
+ - PEFT 0.12.0