arunasank commited on
Commit
3d9f822
·
verified ·
1 Parent(s): a75864e

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. .gitattributes +1 -0
  2. llm_judge_csvs/baseline_pre_recover.csv +0 -0
  3. llm_judge_csvs/baseline_pre_recover.scores.json +18 -0
  4. llm_judge_csvs/llm_judge_step_1000_avg.csv +0 -0
  5. llm_judge_csvs/llm_judge_step_1000_run1.csv +0 -0
  6. llm_judge_csvs/llm_judge_step_1000_run2.csv +0 -0
  7. llm_judge_csvs/llm_judge_step_1500_avg.csv +0 -0
  8. llm_judge_csvs/llm_judge_step_1500_run1.csv +0 -0
  9. llm_judge_csvs/llm_judge_step_1500_run2.csv +0 -0
  10. llm_judge_csvs/llm_judge_step_2000_run1.csv +0 -0
  11. llm_judge_csvs/llm_judge_step_2000_run2.csv +0 -0
  12. llm_judge_csvs/llm_judge_step_2500_avg.csv +0 -0
  13. llm_judge_csvs/llm_judge_step_2500_run1.csv +0 -0
  14. llm_judge_csvs/llm_judge_step_2500_run2.csv +0 -0
  15. llm_judge_csvs/llm_judge_step_3000_avg.csv +0 -0
  16. llm_judge_csvs/llm_judge_step_3000_run1.csv +0 -0
  17. llm_judge_csvs/llm_judge_step_500_avg.csv +0 -0
  18. llm_judge_csvs/llm_judge_step_500_run1.csv +0 -0
  19. llm_judge_csvs/llm_judge_step_500_run2.csv +0 -0
  20. recover-backup/checkpoint-1000/added_tokens.json +3 -0
  21. recover-backup/checkpoint-1000/chat_template.jinja +47 -0
  22. recover-backup/checkpoint-1000/config.json +110 -0
  23. recover-backup/checkpoint-1000/generation_config.json +13 -0
  24. recover-backup/checkpoint-1000/model.safetensors.index.json +0 -0
  25. recover-backup/checkpoint-1000/special_tokens_map.json +33 -0
  26. recover-backup/checkpoint-1000/tokenizer_config.json +0 -0
  27. recover-backup/checkpoint-1000/trainer_state.json +1474 -0
  28. recover-backup/checkpoint-2000/added_tokens.json +3 -0
  29. recover-backup/checkpoint-2000/chat_template.jinja +47 -0
  30. recover-backup/checkpoint-2000/config.json +110 -0
  31. recover-backup/checkpoint-2000/generation_config.json +13 -0
  32. recover-backup/checkpoint-2000/model.safetensors.index.json +0 -0
  33. recover-backup/checkpoint-2000/scheduler.pt +3 -0
  34. recover-backup/checkpoint-2000/special_tokens_map.json +33 -0
  35. recover-backup/checkpoint-2000/tokenizer.json +3 -0
  36. recover-backup/checkpoint-2000/tokenizer_config.json +0 -0
  37. recover-backup/checkpoint-2000/trainer_state.json +2914 -0
  38. recover/README.md +58 -0
  39. recover/checkpoint-3000/added_tokens.json +3 -0
  40. recover/checkpoint-3000/chat_template.jinja +47 -0
  41. recover/checkpoint-3000/generation_config.json +13 -0
  42. recover/checkpoint-3000/model.safetensors.index.json +0 -0
  43. recover/checkpoint-3000/special_tokens_map.json +33 -0
  44. recover/checkpoint-3000/tokenizer_config.json +0 -0
  45. recover/checkpoint-3000/trainer_state.json +0 -0
  46. recover/final/added_tokens.json +3 -0
  47. recover/final/chat_template.jinja +47 -0
  48. recover/final/config.json +110 -0
  49. recover/final/generation_config.json +13 -0
  50. recover/final/model.safetensors.index.json +0 -0
.gitattributes CHANGED
@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
+ recover-backup/checkpoint-2000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
llm_judge_csvs/baseline_pre_recover.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/baseline_pre_recover.scores.json ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "llm_judge/biology/in_scope/utility": 0.5333333333333333,
3
+ "llm_judge/biology/in_scope/answering": 1.0,
4
+ "llm_judge/biology/in_scope/factual_helpful": 0.6,
5
+ "llm_judge/biology/in_scope/precise": 0.0,
6
+ "llm_judge/chemistry/out_of_scope/utility": 0.26666666666666666,
7
+ "llm_judge/chemistry/out_of_scope/answering": 0.4,
8
+ "llm_judge/chemistry/out_of_scope/factual_helpful": 0.3,
9
+ "llm_judge/chemistry/out_of_scope/precise": 0.1,
10
+ "llm_judge/math/out_of_scope/utility": 0.03333333333333333,
11
+ "llm_judge/math/out_of_scope/answering": 0.1,
12
+ "llm_judge/math/out_of_scope/factual_helpful": 0.0,
13
+ "llm_judge/math/out_of_scope/precise": 0.0,
14
+ "llm_judge/cyber/out_of_scope/utility": 0.16666666666666666,
15
+ "llm_judge/cyber/out_of_scope/answering": 0.3,
16
+ "llm_judge/cyber/out_of_scope/factual_helpful": 0.2,
17
+ "llm_judge/cyber/out_of_scope/precise": 0.0
18
+ }
llm_judge_csvs/llm_judge_step_1000_avg.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_1000_run1.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_1000_run2.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_1500_avg.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_1500_run1.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_1500_run2.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_2000_run1.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_2000_run2.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_2500_avg.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_2500_run1.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_2500_run2.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_3000_avg.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_3000_run1.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_500_avg.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_500_run1.csv ADDED
The diff for this file is too large to render. See raw diff
 
llm_judge_csvs/llm_judge_step_500_run2.csv ADDED
The diff for this file is too large to render. See raw diff
 
recover-backup/checkpoint-1000/added_tokens.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "<image_soft_token>": 262144
3
+ }
recover-backup/checkpoint-1000/chat_template.jinja ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'image' -%}
33
+ {{ '<start_of_image>' }}
34
+ {%- elif item['type'] == 'text' -%}
35
+ {{ item['text'] | trim }}
36
+ {%- endif -%}
37
+ {%- endfor -%}
38
+ {%- else -%}
39
+ {{ raise_exception("Invalid content type") }}
40
+ {%- endif -%}
41
+ {{ '<end_of_turn>
42
+ ' }}
43
+ {%- endfor -%}
44
+ {%- if add_generation_prompt -%}
45
+ {{'<start_of_turn>model
46
+ '}}
47
+ {%- endif -%}
recover-backup/checkpoint-1000/config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "dtype": "bfloat16",
21
+ "final_logit_softcapping": null,
22
+ "head_dim": 256,
23
+ "hidden_activation": "gelu_pytorch_tanh",
24
+ "hidden_size": 3840,
25
+ "initializer_range": 0.02,
26
+ "intermediate_size": 15360,
27
+ "layer_types": [
28
+ "sliding_attention",
29
+ "sliding_attention",
30
+ "sliding_attention",
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "full_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "sliding_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "full_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "sliding_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "full_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "sliding_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "full_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "sliding_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "full_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "sliding_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "full_attention",
64
+ "sliding_attention",
65
+ "sliding_attention",
66
+ "sliding_attention",
67
+ "sliding_attention",
68
+ "sliding_attention",
69
+ "full_attention",
70
+ "sliding_attention",
71
+ "sliding_attention",
72
+ "sliding_attention",
73
+ "sliding_attention",
74
+ "sliding_attention",
75
+ "full_attention"
76
+ ],
77
+ "max_position_embeddings": 131072,
78
+ "model_type": "gemma3_text",
79
+ "num_attention_heads": 16,
80
+ "num_hidden_layers": 48,
81
+ "num_key_value_heads": 8,
82
+ "query_pre_attn_scalar": 256,
83
+ "rms_norm_eps": 1e-06,
84
+ "rope_local_base_freq": 10000.0,
85
+ "rope_scaling": {
86
+ "factor": 8.0,
87
+ "rope_type": "linear"
88
+ },
89
+ "rope_theta": 1000000.0,
90
+ "sliding_window": 1024,
91
+ "use_cache": true,
92
+ "vocab_size": 262208
93
+ },
94
+ "transformers_version": "4.56.1",
95
+ "vision_config": {
96
+ "attention_dropout": 0.0,
97
+ "dtype": "bfloat16",
98
+ "hidden_act": "gelu_pytorch_tanh",
99
+ "hidden_size": 1152,
100
+ "image_size": 896,
101
+ "intermediate_size": 4304,
102
+ "layer_norm_eps": 1e-06,
103
+ "model_type": "siglip_vision_model",
104
+ "num_attention_heads": 16,
105
+ "num_channels": 3,
106
+ "num_hidden_layers": 27,
107
+ "patch_size": 14,
108
+ "vision_use_head": false
109
+ }
110
+ }
recover-backup/checkpoint-1000/generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 1,
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "top_k": 64,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.56.1"
13
+ }
recover-backup/checkpoint-1000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
recover-backup/checkpoint-1000/special_tokens_map.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "boi_token": "<start_of_image>",
3
+ "bos_token": {
4
+ "content": "<bos>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "eoi_token": "<end_of_image>",
11
+ "eos_token": {
12
+ "content": "<eos>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false
17
+ },
18
+ "image_token": "<image_soft_token>",
19
+ "pad_token": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "unk_token": {
27
+ "content": "<unk>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ }
33
+ }
recover-backup/checkpoint-1000/tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff
 
recover-backup/checkpoint-1000/trainer_state.json ADDED
@@ -0,0 +1,1474 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.6,
6
+ "eval_steps": 100,
7
+ "global_step": 1000,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 0.6808024171739817,
14
+ "epoch": 0.016,
15
+ "grad_norm": 2096.0,
16
+ "learning_rate": 6.000000000000001e-07,
17
+ "loss": 81.3949,
18
+ "mean_token_accuracy": 0.469288082793355,
19
+ "num_tokens": 195524.0,
20
+ "step": 10
21
+ },
22
+ {
23
+ "entropy": 0.6932845771312713,
24
+ "epoch": 0.032,
25
+ "grad_norm": 1840.0,
26
+ "learning_rate": 1.2666666666666669e-06,
27
+ "loss": 79.1276,
28
+ "mean_token_accuracy": 0.4702936248853803,
29
+ "num_tokens": 390903.0,
30
+ "step": 20
31
+ },
32
+ {
33
+ "entropy": 0.7348822019994259,
34
+ "epoch": 0.048,
35
+ "grad_norm": 1896.0,
36
+ "learning_rate": 1.9333333333333336e-06,
37
+ "loss": 75.5796,
38
+ "mean_token_accuracy": 0.47059151269495486,
39
+ "num_tokens": 589868.0,
40
+ "step": 30
41
+ },
42
+ {
43
+ "entropy": 0.8692382726818323,
44
+ "epoch": 0.064,
45
+ "grad_norm": 868.0,
46
+ "learning_rate": 2.6e-06,
47
+ "loss": 66.3894,
48
+ "mean_token_accuracy": 0.4747353194281459,
49
+ "num_tokens": 791190.0,
50
+ "step": 40
51
+ },
52
+ {
53
+ "entropy": 1.1099284883588552,
54
+ "epoch": 0.08,
55
+ "grad_norm": 520.0,
56
+ "learning_rate": 3.266666666666667e-06,
57
+ "loss": 57.5001,
58
+ "mean_token_accuracy": 0.47747670076787474,
59
+ "num_tokens": 989860.0,
60
+ "step": 50
61
+ },
62
+ {
63
+ "entropy": 1.4874154418706893,
64
+ "epoch": 0.096,
65
+ "grad_norm": 144.0,
66
+ "learning_rate": 3.9333333333333335e-06,
67
+ "loss": 48.6006,
68
+ "mean_token_accuracy": 0.4924122652038932,
69
+ "num_tokens": 1181777.0,
70
+ "step": 60
71
+ },
72
+ {
73
+ "entropy": 2.022240047156811,
74
+ "epoch": 0.112,
75
+ "grad_norm": 52.0,
76
+ "learning_rate": 4.600000000000001e-06,
77
+ "loss": 41.7489,
78
+ "mean_token_accuracy": 0.5201752169057727,
79
+ "num_tokens": 1385513.0,
80
+ "step": 70
81
+ },
82
+ {
83
+ "entropy": 2.395130204409361,
84
+ "epoch": 0.128,
85
+ "grad_norm": 44.0,
86
+ "learning_rate": 5.2666666666666665e-06,
87
+ "loss": 37.2886,
88
+ "mean_token_accuracy": 0.5428294729441404,
89
+ "num_tokens": 1582368.0,
90
+ "step": 80
91
+ },
92
+ {
93
+ "entropy": 2.1368538811802864,
94
+ "epoch": 0.144,
95
+ "grad_norm": 29.5,
96
+ "learning_rate": 5.933333333333335e-06,
97
+ "loss": 33.907,
98
+ "mean_token_accuracy": 0.5704466678202152,
99
+ "num_tokens": 1773764.0,
100
+ "step": 90
101
+ },
102
+ {
103
+ "entropy": 1.9877845980226994,
104
+ "epoch": 0.16,
105
+ "grad_norm": 25.375,
106
+ "learning_rate": 6.600000000000001e-06,
107
+ "loss": 31.7141,
108
+ "mean_token_accuracy": 0.58747753277421,
109
+ "num_tokens": 1970077.0,
110
+ "step": 100
111
+ },
112
+ {
113
+ "epoch": 0.16,
114
+ "eval_biology_entropy": 1.911924614906311,
115
+ "eval_biology_loss": 1.831228256225586,
116
+ "eval_biology_mean_token_accuracy": 0.6091388630867004,
117
+ "eval_biology_num_tokens": 1970077.0,
118
+ "eval_biology_runtime": 48.4202,
119
+ "eval_biology_samples_per_second": 10.326,
120
+ "eval_biology_steps_per_second": 2.582,
121
+ "step": 100
122
+ },
123
+ {
124
+ "epoch": 0.16,
125
+ "eval_chemistry_entropy": 1.5279665517807006,
126
+ "eval_chemistry_loss": 1.507634162902832,
127
+ "eval_chemistry_mean_token_accuracy": 0.6661558594703675,
128
+ "eval_chemistry_num_tokens": 1970077.0,
129
+ "eval_chemistry_runtime": 60.0771,
130
+ "eval_chemistry_samples_per_second": 8.323,
131
+ "eval_chemistry_steps_per_second": 2.081,
132
+ "step": 100
133
+ },
134
+ {
135
+ "epoch": 0.16,
136
+ "eval_math_entropy": 1.3111717319488525,
137
+ "eval_math_loss": 1.5638481378555298,
138
+ "eval_math_mean_token_accuracy": 0.6615601582527161,
139
+ "eval_math_num_tokens": 1970077.0,
140
+ "eval_math_runtime": 61.5942,
141
+ "eval_math_samples_per_second": 8.118,
142
+ "eval_math_steps_per_second": 2.029,
143
+ "step": 100
144
+ },
145
+ {
146
+ "epoch": 0.16,
147
+ "eval_cyber_entropy": 2.3538206100463865,
148
+ "eval_cyber_loss": 3.4596033096313477,
149
+ "eval_cyber_mean_token_accuracy": 0.4573894315958023,
150
+ "eval_cyber_num_tokens": 1970077.0,
151
+ "eval_cyber_runtime": 58.7811,
152
+ "eval_cyber_samples_per_second": 6.754,
153
+ "eval_cyber_steps_per_second": 1.701,
154
+ "step": 100
155
+ },
156
+ {
157
+ "entropy": 1.8946144469082355,
158
+ "epoch": 0.176,
159
+ "grad_norm": 22.0,
160
+ "learning_rate": 7.266666666666668e-06,
161
+ "loss": 29.5004,
162
+ "mean_token_accuracy": 0.6055971266701817,
163
+ "num_tokens": 2168354.0,
164
+ "step": 110
165
+ },
166
+ {
167
+ "entropy": 1.7916729532182216,
168
+ "epoch": 0.192,
169
+ "grad_norm": 22.375,
170
+ "learning_rate": 7.933333333333334e-06,
171
+ "loss": 28.5473,
172
+ "mean_token_accuracy": 0.6124886509031058,
173
+ "num_tokens": 2365822.0,
174
+ "step": 120
175
+ },
176
+ {
177
+ "entropy": 1.7780573107302189,
178
+ "epoch": 0.208,
179
+ "grad_norm": 20.625,
180
+ "learning_rate": 8.6e-06,
181
+ "loss": 28.0106,
182
+ "mean_token_accuracy": 0.617958590388298,
183
+ "num_tokens": 2558762.0,
184
+ "step": 130
185
+ },
186
+ {
187
+ "entropy": 1.6913408383727073,
188
+ "epoch": 0.224,
189
+ "grad_norm": 19.875,
190
+ "learning_rate": 9.266666666666667e-06,
191
+ "loss": 27.182,
192
+ "mean_token_accuracy": 0.6272589169442654,
193
+ "num_tokens": 2755347.0,
194
+ "step": 140
195
+ },
196
+ {
197
+ "entropy": 1.6812096990644931,
198
+ "epoch": 0.24,
199
+ "grad_norm": 19.5,
200
+ "learning_rate": 9.933333333333334e-06,
201
+ "loss": 26.6205,
202
+ "mean_token_accuracy": 0.6303446084260941,
203
+ "num_tokens": 2947346.0,
204
+ "step": 150
205
+ },
206
+ {
207
+ "entropy": 1.6677771665155887,
208
+ "epoch": 0.256,
209
+ "grad_norm": 20.25,
210
+ "learning_rate": 1.0600000000000002e-05,
211
+ "loss": 26.6086,
212
+ "mean_token_accuracy": 0.6319835562258959,
213
+ "num_tokens": 3139957.0,
214
+ "step": 160
215
+ },
216
+ {
217
+ "entropy": 1.6161759495735168,
218
+ "epoch": 0.272,
219
+ "grad_norm": 20.375,
220
+ "learning_rate": 1.1266666666666668e-05,
221
+ "loss": 26.0026,
222
+ "mean_token_accuracy": 0.6365392006933689,
223
+ "num_tokens": 3335951.0,
224
+ "step": 170
225
+ },
226
+ {
227
+ "entropy": 1.6003736153244972,
228
+ "epoch": 0.288,
229
+ "grad_norm": 21.0,
230
+ "learning_rate": 1.1933333333333335e-05,
231
+ "loss": 25.4302,
232
+ "mean_token_accuracy": 0.6393310058861971,
233
+ "num_tokens": 3539731.0,
234
+ "step": 180
235
+ },
236
+ {
237
+ "entropy": 1.5776585705578328,
238
+ "epoch": 0.304,
239
+ "grad_norm": 18.875,
240
+ "learning_rate": 1.2600000000000001e-05,
241
+ "loss": 25.2177,
242
+ "mean_token_accuracy": 0.646138820424676,
243
+ "num_tokens": 3733488.0,
244
+ "step": 190
245
+ },
246
+ {
247
+ "entropy": 1.5823919720947743,
248
+ "epoch": 0.32,
249
+ "grad_norm": 21.75,
250
+ "learning_rate": 1.3266666666666668e-05,
251
+ "loss": 25.46,
252
+ "mean_token_accuracy": 0.6409369662404061,
253
+ "num_tokens": 3920545.0,
254
+ "step": 200
255
+ },
256
+ {
257
+ "epoch": 0.32,
258
+ "eval_biology_entropy": 1.5241963243484498,
259
+ "eval_biology_loss": 1.512497067451477,
260
+ "eval_biology_mean_token_accuracy": 0.656851170539856,
261
+ "eval_biology_num_tokens": 3920545.0,
262
+ "eval_biology_runtime": 48.4372,
263
+ "eval_biology_samples_per_second": 10.323,
264
+ "eval_biology_steps_per_second": 2.581,
265
+ "step": 200
266
+ },
267
+ {
268
+ "epoch": 0.32,
269
+ "eval_chemistry_entropy": 1.1822538471221924,
270
+ "eval_chemistry_loss": 1.178497552871704,
271
+ "eval_chemistry_mean_token_accuracy": 0.7221880826950073,
272
+ "eval_chemistry_num_tokens": 3920545.0,
273
+ "eval_chemistry_runtime": 60.0879,
274
+ "eval_chemistry_samples_per_second": 8.321,
275
+ "eval_chemistry_steps_per_second": 2.08,
276
+ "step": 200
277
+ },
278
+ {
279
+ "epoch": 0.32,
280
+ "eval_math_entropy": 0.9796018509864807,
281
+ "eval_math_loss": 1.2802495956420898,
282
+ "eval_math_mean_token_accuracy": 0.713343710899353,
283
+ "eval_math_num_tokens": 3920545.0,
284
+ "eval_math_runtime": 61.6189,
285
+ "eval_math_samples_per_second": 8.114,
286
+ "eval_math_steps_per_second": 2.029,
287
+ "step": 200
288
+ },
289
+ {
290
+ "epoch": 0.32,
291
+ "eval_cyber_entropy": 2.516505275964737,
292
+ "eval_cyber_loss": 3.1510307788848877,
293
+ "eval_cyber_mean_token_accuracy": 0.47284564673900603,
294
+ "eval_cyber_num_tokens": 3920545.0,
295
+ "eval_cyber_runtime": 58.8071,
296
+ "eval_cyber_samples_per_second": 6.751,
297
+ "eval_cyber_steps_per_second": 1.7,
298
+ "step": 200
299
+ },
300
+ {
301
+ "entropy": 1.5317105814814567,
302
+ "epoch": 0.336,
303
+ "grad_norm": 18.375,
304
+ "learning_rate": 1.3933333333333334e-05,
305
+ "loss": 24.5282,
306
+ "mean_token_accuracy": 0.6502130385488272,
307
+ "num_tokens": 4114077.0,
308
+ "step": 210
309
+ },
310
+ {
311
+ "entropy": 1.5466628648340701,
312
+ "epoch": 0.352,
313
+ "grad_norm": 18.875,
314
+ "learning_rate": 1.46e-05,
315
+ "loss": 24.8233,
316
+ "mean_token_accuracy": 0.6490028414875268,
317
+ "num_tokens": 4306949.0,
318
+ "step": 220
319
+ },
320
+ {
321
+ "entropy": 1.5188672259449958,
322
+ "epoch": 0.368,
323
+ "grad_norm": 19.25,
324
+ "learning_rate": 1.5266666666666667e-05,
325
+ "loss": 24.4549,
326
+ "mean_token_accuracy": 0.6544424753636122,
327
+ "num_tokens": 4504001.0,
328
+ "step": 230
329
+ },
330
+ {
331
+ "entropy": 1.5070753492414952,
332
+ "epoch": 0.384,
333
+ "grad_norm": 21.75,
334
+ "learning_rate": 1.5933333333333336e-05,
335
+ "loss": 24.3379,
336
+ "mean_token_accuracy": 0.6532186731696129,
337
+ "num_tokens": 4693812.0,
338
+ "step": 240
339
+ },
340
+ {
341
+ "entropy": 1.5358249843120575,
342
+ "epoch": 0.4,
343
+ "grad_norm": 17.25,
344
+ "learning_rate": 1.66e-05,
345
+ "loss": 24.4225,
346
+ "mean_token_accuracy": 0.6525934401899576,
347
+ "num_tokens": 4887094.0,
348
+ "step": 250
349
+ },
350
+ {
351
+ "entropy": 1.5161145091056825,
352
+ "epoch": 0.416,
353
+ "grad_norm": 19.75,
354
+ "learning_rate": 1.726666666666667e-05,
355
+ "loss": 24.4554,
356
+ "mean_token_accuracy": 0.6528204523026944,
357
+ "num_tokens": 5085369.0,
358
+ "step": 260
359
+ },
360
+ {
361
+ "entropy": 1.5000567641109228,
362
+ "epoch": 0.432,
363
+ "grad_norm": 20.625,
364
+ "learning_rate": 1.7933333333333333e-05,
365
+ "loss": 24.2948,
366
+ "mean_token_accuracy": 0.6535403318703175,
367
+ "num_tokens": 5271275.0,
368
+ "step": 270
369
+ },
370
+ {
371
+ "entropy": 1.5077810816466808,
372
+ "epoch": 0.448,
373
+ "grad_norm": 22.375,
374
+ "learning_rate": 1.86e-05,
375
+ "loss": 24.2712,
376
+ "mean_token_accuracy": 0.6533999439328909,
377
+ "num_tokens": 5460559.0,
378
+ "step": 280
379
+ },
380
+ {
381
+ "entropy": 1.4563340097665787,
382
+ "epoch": 0.464,
383
+ "grad_norm": 22.0,
384
+ "learning_rate": 1.926666666666667e-05,
385
+ "loss": 23.2755,
386
+ "mean_token_accuracy": 0.6638390984386205,
387
+ "num_tokens": 5653809.0,
388
+ "step": 290
389
+ },
390
+ {
391
+ "entropy": 1.468742286041379,
392
+ "epoch": 0.48,
393
+ "grad_norm": 18.0,
394
+ "learning_rate": 1.9933333333333334e-05,
395
+ "loss": 23.7149,
396
+ "mean_token_accuracy": 0.6593498229980469,
397
+ "num_tokens": 5850176.0,
398
+ "step": 300
399
+ },
400
+ {
401
+ "epoch": 0.48,
402
+ "eval_biology_entropy": 1.4316115617752074,
403
+ "eval_biology_loss": 1.4177387952804565,
404
+ "eval_biology_mean_token_accuracy": 0.6717347936630249,
405
+ "eval_biology_num_tokens": 5850176.0,
406
+ "eval_biology_runtime": 48.4847,
407
+ "eval_biology_samples_per_second": 10.313,
408
+ "eval_biology_steps_per_second": 2.578,
409
+ "step": 300
410
+ },
411
+ {
412
+ "epoch": 0.48,
413
+ "eval_chemistry_entropy": 1.1160053358078004,
414
+ "eval_chemistry_loss": 1.0916916131973267,
415
+ "eval_chemistry_mean_token_accuracy": 0.7381541066169739,
416
+ "eval_chemistry_num_tokens": 5850176.0,
417
+ "eval_chemistry_runtime": 60.1537,
418
+ "eval_chemistry_samples_per_second": 8.312,
419
+ "eval_chemistry_steps_per_second": 2.078,
420
+ "step": 300
421
+ },
422
+ {
423
+ "epoch": 0.48,
424
+ "eval_math_entropy": 0.9393695564270019,
425
+ "eval_math_loss": 1.2016712427139282,
426
+ "eval_math_mean_token_accuracy": 0.7277262687683106,
427
+ "eval_math_num_tokens": 5850176.0,
428
+ "eval_math_runtime": 61.6699,
429
+ "eval_math_samples_per_second": 8.108,
430
+ "eval_math_steps_per_second": 2.027,
431
+ "step": 300
432
+ },
433
+ {
434
+ "epoch": 0.48,
435
+ "eval_cyber_entropy": 2.514738072156906,
436
+ "eval_cyber_loss": 2.9828875064849854,
437
+ "eval_cyber_mean_token_accuracy": 0.47654601037502287,
438
+ "eval_cyber_num_tokens": 5850176.0,
439
+ "eval_cyber_runtime": 58.8231,
440
+ "eval_cyber_samples_per_second": 6.749,
441
+ "eval_cyber_steps_per_second": 1.7,
442
+ "step": 300
443
+ },
444
+ {
445
+ "entropy": 1.3836464550346137,
446
+ "epoch": 0.496,
447
+ "grad_norm": 17.875,
448
+ "learning_rate": 1.9933333333333334e-05,
449
+ "loss": 22.405,
450
+ "mean_token_accuracy": 0.6713070213794708,
451
+ "num_tokens": 6046503.0,
452
+ "step": 310
453
+ },
454
+ {
455
+ "entropy": 1.4630893386900425,
456
+ "epoch": 0.512,
457
+ "grad_norm": 19.25,
458
+ "learning_rate": 1.985925925925926e-05,
459
+ "loss": 23.4697,
460
+ "mean_token_accuracy": 0.6617988537997007,
461
+ "num_tokens": 6240456.0,
462
+ "step": 320
463
+ },
464
+ {
465
+ "entropy": 1.448515933379531,
466
+ "epoch": 0.528,
467
+ "grad_norm": 18.75,
468
+ "learning_rate": 1.9785185185185187e-05,
469
+ "loss": 23.3355,
470
+ "mean_token_accuracy": 0.6626260321587324,
471
+ "num_tokens": 6430555.0,
472
+ "step": 330
473
+ },
474
+ {
475
+ "entropy": 1.44985730946064,
476
+ "epoch": 0.544,
477
+ "grad_norm": 18.625,
478
+ "learning_rate": 1.971111111111111e-05,
479
+ "loss": 23.3133,
480
+ "mean_token_accuracy": 0.665154778957367,
481
+ "num_tokens": 6626006.0,
482
+ "step": 340
483
+ },
484
+ {
485
+ "entropy": 1.4254157304763795,
486
+ "epoch": 0.56,
487
+ "grad_norm": 19.125,
488
+ "learning_rate": 1.963703703703704e-05,
489
+ "loss": 23.1032,
490
+ "mean_token_accuracy": 0.6651736661791802,
491
+ "num_tokens": 6820754.0,
492
+ "step": 350
493
+ },
494
+ {
495
+ "entropy": 1.4192099779844285,
496
+ "epoch": 0.576,
497
+ "grad_norm": 18.875,
498
+ "learning_rate": 1.9562962962962964e-05,
499
+ "loss": 22.7221,
500
+ "mean_token_accuracy": 0.6693361282348633,
501
+ "num_tokens": 7021844.0,
502
+ "step": 360
503
+ },
504
+ {
505
+ "entropy": 1.4211084496229887,
506
+ "epoch": 0.592,
507
+ "grad_norm": 19.0,
508
+ "learning_rate": 1.948888888888889e-05,
509
+ "loss": 23.0159,
510
+ "mean_token_accuracy": 0.6660285223275423,
511
+ "num_tokens": 7213951.0,
512
+ "step": 370
513
+ },
514
+ {
515
+ "entropy": 1.4022404573857785,
516
+ "epoch": 0.608,
517
+ "grad_norm": 18.625,
518
+ "learning_rate": 1.9414814814814817e-05,
519
+ "loss": 22.5685,
520
+ "mean_token_accuracy": 0.6711618881672621,
521
+ "num_tokens": 7416773.0,
522
+ "step": 380
523
+ },
524
+ {
525
+ "entropy": 1.4124153569340705,
526
+ "epoch": 0.624,
527
+ "grad_norm": 18.625,
528
+ "learning_rate": 1.9340740740740743e-05,
529
+ "loss": 22.8021,
530
+ "mean_token_accuracy": 0.6678782407194376,
531
+ "num_tokens": 7612843.0,
532
+ "step": 390
533
+ },
534
+ {
535
+ "entropy": 1.4238922595977783,
536
+ "epoch": 0.64,
537
+ "grad_norm": 20.375,
538
+ "learning_rate": 1.926666666666667e-05,
539
+ "loss": 22.9558,
540
+ "mean_token_accuracy": 0.6680531721562147,
541
+ "num_tokens": 7801633.0,
542
+ "step": 400
543
+ },
544
+ {
545
+ "epoch": 0.64,
546
+ "eval_biology_entropy": 1.3578908562660217,
547
+ "eval_biology_loss": 1.3675978183746338,
548
+ "eval_biology_mean_token_accuracy": 0.6797989273071289,
549
+ "eval_biology_num_tokens": 7801633.0,
550
+ "eval_biology_runtime": 48.4218,
551
+ "eval_biology_samples_per_second": 10.326,
552
+ "eval_biology_steps_per_second": 2.581,
553
+ "step": 400
554
+ },
555
+ {
556
+ "epoch": 0.64,
557
+ "eval_chemistry_entropy": 1.0446269502639771,
558
+ "eval_chemistry_loss": 1.0433638095855713,
559
+ "eval_chemistry_mean_token_accuracy": 0.7473979935646057,
560
+ "eval_chemistry_num_tokens": 7801633.0,
561
+ "eval_chemistry_runtime": 60.1024,
562
+ "eval_chemistry_samples_per_second": 8.319,
563
+ "eval_chemistry_steps_per_second": 2.08,
564
+ "step": 400
565
+ },
566
+ {
567
+ "epoch": 0.64,
568
+ "eval_math_entropy": 0.8829994263648987,
569
+ "eval_math_loss": 1.168383002281189,
570
+ "eval_math_mean_token_accuracy": 0.7337917404174805,
571
+ "eval_math_num_tokens": 7801633.0,
572
+ "eval_math_runtime": 61.66,
573
+ "eval_math_samples_per_second": 8.109,
574
+ "eval_math_steps_per_second": 2.027,
575
+ "step": 400
576
+ },
577
+ {
578
+ "epoch": 0.64,
579
+ "eval_cyber_entropy": 2.3425097274780273,
580
+ "eval_cyber_loss": 2.8961009979248047,
581
+ "eval_cyber_mean_token_accuracy": 0.48323330402374265,
582
+ "eval_cyber_num_tokens": 7801633.0,
583
+ "eval_cyber_runtime": 58.7869,
584
+ "eval_cyber_samples_per_second": 6.753,
585
+ "eval_cyber_steps_per_second": 1.701,
586
+ "step": 400
587
+ },
588
+ {
589
+ "entropy": 1.4169765036553144,
590
+ "epoch": 0.656,
591
+ "grad_norm": 21.125,
592
+ "learning_rate": 1.9192592592592593e-05,
593
+ "loss": 22.9523,
594
+ "mean_token_accuracy": 0.6662464819848537,
595
+ "num_tokens": 7995843.0,
596
+ "step": 410
597
+ },
598
+ {
599
+ "entropy": 1.43333672657609,
600
+ "epoch": 0.672,
601
+ "grad_norm": 18.625,
602
+ "learning_rate": 1.911851851851852e-05,
603
+ "loss": 22.9483,
604
+ "mean_token_accuracy": 0.667000836879015,
605
+ "num_tokens": 8183103.0,
606
+ "step": 420
607
+ },
608
+ {
609
+ "entropy": 1.4049694981426,
610
+ "epoch": 0.688,
611
+ "grad_norm": 17.75,
612
+ "learning_rate": 1.9044444444444446e-05,
613
+ "loss": 22.7023,
614
+ "mean_token_accuracy": 0.670426407828927,
615
+ "num_tokens": 8385976.0,
616
+ "step": 430
617
+ },
618
+ {
619
+ "entropy": 1.4090089831501245,
620
+ "epoch": 0.704,
621
+ "grad_norm": 17.875,
622
+ "learning_rate": 1.8970370370370372e-05,
623
+ "loss": 22.6381,
624
+ "mean_token_accuracy": 0.6711054440587759,
625
+ "num_tokens": 8578431.0,
626
+ "step": 440
627
+ },
628
+ {
629
+ "entropy": 1.3353864025324582,
630
+ "epoch": 0.72,
631
+ "grad_norm": 17.75,
632
+ "learning_rate": 1.8896296296296295e-05,
633
+ "loss": 21.6293,
634
+ "mean_token_accuracy": 0.681050893291831,
635
+ "num_tokens": 8781342.0,
636
+ "step": 450
637
+ },
638
+ {
639
+ "entropy": 1.3734276317059995,
640
+ "epoch": 0.736,
641
+ "grad_norm": 17.25,
642
+ "learning_rate": 1.8822222222222225e-05,
643
+ "loss": 22.0542,
644
+ "mean_token_accuracy": 0.6750697229057551,
645
+ "num_tokens": 8977918.0,
646
+ "step": 460
647
+ },
648
+ {
649
+ "entropy": 1.3878948390483856,
650
+ "epoch": 0.752,
651
+ "grad_norm": 17.875,
652
+ "learning_rate": 1.874814814814815e-05,
653
+ "loss": 22.4846,
654
+ "mean_token_accuracy": 0.6717681255191564,
655
+ "num_tokens": 9169322.0,
656
+ "step": 470
657
+ },
658
+ {
659
+ "entropy": 1.3424169197678566,
660
+ "epoch": 0.768,
661
+ "grad_norm": 20.125,
662
+ "learning_rate": 1.8674074074074075e-05,
663
+ "loss": 21.6359,
664
+ "mean_token_accuracy": 0.6805981863290071,
665
+ "num_tokens": 9368141.0,
666
+ "step": 480
667
+ },
668
+ {
669
+ "entropy": 1.3768138941377401,
670
+ "epoch": 0.784,
671
+ "grad_norm": 19.75,
672
+ "learning_rate": 1.86e-05,
673
+ "loss": 22.1917,
674
+ "mean_token_accuracy": 0.6750959828495979,
675
+ "num_tokens": 9565236.0,
676
+ "step": 490
677
+ },
678
+ {
679
+ "entropy": 1.3996582824736834,
680
+ "epoch": 0.8,
681
+ "grad_norm": 19.25,
682
+ "learning_rate": 1.8525925925925928e-05,
683
+ "loss": 22.6339,
684
+ "mean_token_accuracy": 0.6715325634926558,
685
+ "num_tokens": 9761227.0,
686
+ "step": 500
687
+ },
688
+ {
689
+ "epoch": 0.8,
690
+ "eval_biology_entropy": 1.3487377295494078,
691
+ "eval_biology_loss": 1.3396238088607788,
692
+ "eval_biology_mean_token_accuracy": 0.6843163523674011,
693
+ "eval_biology_num_tokens": 9761227.0,
694
+ "eval_biology_runtime": 48.4345,
695
+ "eval_biology_samples_per_second": 10.323,
696
+ "eval_biology_steps_per_second": 2.581,
697
+ "step": 500
698
+ },
699
+ {
700
+ "epoch": 0.8,
701
+ "eval_chemistry_entropy": 1.0342083148956298,
702
+ "eval_chemistry_loss": 1.015425682067871,
703
+ "eval_chemistry_mean_token_accuracy": 0.7529125623703002,
704
+ "eval_chemistry_num_tokens": 9761227.0,
705
+ "eval_chemistry_runtime": 59.7072,
706
+ "eval_chemistry_samples_per_second": 8.374,
707
+ "eval_chemistry_steps_per_second": 2.094,
708
+ "step": 500
709
+ },
710
+ {
711
+ "epoch": 0.8,
712
+ "eval_math_entropy": 0.9004156064987182,
713
+ "eval_math_loss": 1.1309455633163452,
714
+ "eval_math_mean_token_accuracy": 0.7394839000701904,
715
+ "eval_math_num_tokens": 9761227.0,
716
+ "eval_math_runtime": 61.1899,
717
+ "eval_math_samples_per_second": 8.171,
718
+ "eval_math_steps_per_second": 2.043,
719
+ "step": 500
720
+ },
721
+ {
722
+ "epoch": 0.8,
723
+ "eval_cyber_entropy": 2.320133538246155,
724
+ "eval_cyber_loss": 2.8678040504455566,
725
+ "eval_cyber_mean_token_accuracy": 0.4835522577166557,
726
+ "eval_cyber_num_tokens": 9761227.0,
727
+ "eval_cyber_runtime": 58.9104,
728
+ "eval_cyber_samples_per_second": 6.739,
729
+ "eval_cyber_steps_per_second": 1.697,
730
+ "step": 500
731
+ },
732
+ {
733
+ "entropy": 1.35246634632349,
734
+ "epoch": 0.816,
735
+ "grad_norm": 20.75,
736
+ "learning_rate": 1.8451851851851855e-05,
737
+ "loss": 21.7978,
738
+ "mean_token_accuracy": 0.6768883503973484,
739
+ "num_tokens": 9958727.0,
740
+ "step": 510
741
+ },
742
+ {
743
+ "entropy": 1.3554580289870501,
744
+ "epoch": 0.832,
745
+ "grad_norm": 17.5,
746
+ "learning_rate": 1.8377777777777778e-05,
747
+ "loss": 21.7881,
748
+ "mean_token_accuracy": 0.6783689457923174,
749
+ "num_tokens": 10155771.0,
750
+ "step": 520
751
+ },
752
+ {
753
+ "entropy": 1.342598542943597,
754
+ "epoch": 0.848,
755
+ "grad_norm": 19.5,
756
+ "learning_rate": 1.8303703703703704e-05,
757
+ "loss": 21.6264,
758
+ "mean_token_accuracy": 0.679912555962801,
759
+ "num_tokens": 10357721.0,
760
+ "step": 530
761
+ },
762
+ {
763
+ "entropy": 1.3717901095747949,
764
+ "epoch": 0.864,
765
+ "grad_norm": 18.375,
766
+ "learning_rate": 1.822962962962963e-05,
767
+ "loss": 22.2423,
768
+ "mean_token_accuracy": 0.6749696064740419,
769
+ "num_tokens": 10552495.0,
770
+ "step": 540
771
+ },
772
+ {
773
+ "entropy": 1.3606136079877615,
774
+ "epoch": 0.88,
775
+ "grad_norm": 21.0,
776
+ "learning_rate": 1.8155555555555557e-05,
777
+ "loss": 21.921,
778
+ "mean_token_accuracy": 0.6776851717382669,
779
+ "num_tokens": 10748749.0,
780
+ "step": 550
781
+ },
782
+ {
783
+ "entropy": 1.3807077493518591,
784
+ "epoch": 0.896,
785
+ "grad_norm": 19.0,
786
+ "learning_rate": 1.8081481481481484e-05,
787
+ "loss": 22.306,
788
+ "mean_token_accuracy": 0.6740046124905348,
789
+ "num_tokens": 10943319.0,
790
+ "step": 560
791
+ },
792
+ {
793
+ "entropy": 1.3541915334761143,
794
+ "epoch": 0.912,
795
+ "grad_norm": 20.5,
796
+ "learning_rate": 1.800740740740741e-05,
797
+ "loss": 21.989,
798
+ "mean_token_accuracy": 0.6777493793517351,
799
+ "num_tokens": 11136935.0,
800
+ "step": 570
801
+ },
802
+ {
803
+ "entropy": 1.3871366024017333,
804
+ "epoch": 0.928,
805
+ "grad_norm": 19.75,
806
+ "learning_rate": 1.7933333333333333e-05,
807
+ "loss": 22.1135,
808
+ "mean_token_accuracy": 0.6744728099554778,
809
+ "num_tokens": 11331098.0,
810
+ "step": 580
811
+ },
812
+ {
813
+ "entropy": 1.3096800699830056,
814
+ "epoch": 0.944,
815
+ "grad_norm": 17.5,
816
+ "learning_rate": 1.785925925925926e-05,
817
+ "loss": 21.2939,
818
+ "mean_token_accuracy": 0.6831198997795582,
819
+ "num_tokens": 11530550.0,
820
+ "step": 590
821
+ },
822
+ {
823
+ "entropy": 1.3506321482360364,
824
+ "epoch": 0.96,
825
+ "grad_norm": 20.875,
826
+ "learning_rate": 1.7785185185185186e-05,
827
+ "loss": 21.8226,
828
+ "mean_token_accuracy": 0.6785171415656805,
829
+ "num_tokens": 11729645.0,
830
+ "step": 600
831
+ },
832
+ {
833
+ "epoch": 0.96,
834
+ "eval_biology_entropy": 1.3259480624198914,
835
+ "eval_biology_loss": 1.3215161561965942,
836
+ "eval_biology_mean_token_accuracy": 0.6879564213752747,
837
+ "eval_biology_num_tokens": 11729645.0,
838
+ "eval_biology_runtime": 48.4324,
839
+ "eval_biology_samples_per_second": 10.324,
840
+ "eval_biology_steps_per_second": 2.581,
841
+ "step": 600
842
+ },
843
+ {
844
+ "epoch": 0.96,
845
+ "eval_chemistry_entropy": 1.02067849111557,
846
+ "eval_chemistry_loss": 0.9999639987945557,
847
+ "eval_chemistry_mean_token_accuracy": 0.7563390741348267,
848
+ "eval_chemistry_num_tokens": 11729645.0,
849
+ "eval_chemistry_runtime": 60.1019,
850
+ "eval_chemistry_samples_per_second": 8.319,
851
+ "eval_chemistry_steps_per_second": 2.08,
852
+ "step": 600
853
+ },
854
+ {
855
+ "epoch": 0.96,
856
+ "eval_math_entropy": 0.895684769153595,
857
+ "eval_math_loss": 1.1203279495239258,
858
+ "eval_math_mean_token_accuracy": 0.7420237326622009,
859
+ "eval_math_num_tokens": 11729645.0,
860
+ "eval_math_runtime": 61.6071,
861
+ "eval_math_samples_per_second": 8.116,
862
+ "eval_math_steps_per_second": 2.029,
863
+ "step": 600
864
+ },
865
+ {
866
+ "epoch": 0.96,
867
+ "eval_cyber_entropy": 2.2757809734344483,
868
+ "eval_cyber_loss": 2.8872666358947754,
869
+ "eval_cyber_mean_token_accuracy": 0.4848860424757004,
870
+ "eval_cyber_num_tokens": 11729645.0,
871
+ "eval_cyber_runtime": 58.7914,
872
+ "eval_cyber_samples_per_second": 6.753,
873
+ "eval_cyber_steps_per_second": 1.701,
874
+ "step": 600
875
+ },
876
+ {
877
+ "entropy": 1.3319925643503665,
878
+ "epoch": 0.976,
879
+ "grad_norm": 18.875,
880
+ "learning_rate": 1.7711111111111113e-05,
881
+ "loss": 21.4623,
882
+ "mean_token_accuracy": 0.6810844399034977,
883
+ "num_tokens": 11924644.0,
884
+ "step": 610
885
+ },
886
+ {
887
+ "entropy": 1.354709730297327,
888
+ "epoch": 0.992,
889
+ "grad_norm": 17.875,
890
+ "learning_rate": 1.763703703703704e-05,
891
+ "loss": 22.0021,
892
+ "mean_token_accuracy": 0.6767893016338349,
893
+ "num_tokens": 12123059.0,
894
+ "step": 620
895
+ },
896
+ {
897
+ "entropy": 1.358278013765812,
898
+ "epoch": 1.008,
899
+ "grad_norm": 17.625,
900
+ "learning_rate": 1.7562962962962962e-05,
901
+ "loss": 21.6771,
902
+ "mean_token_accuracy": 0.6793475031852723,
903
+ "num_tokens": 12319366.0,
904
+ "step": 630
905
+ },
906
+ {
907
+ "entropy": 1.2945445731282235,
908
+ "epoch": 1.024,
909
+ "grad_norm": 19.875,
910
+ "learning_rate": 1.7488888888888892e-05,
911
+ "loss": 20.9621,
912
+ "mean_token_accuracy": 0.6880514930933714,
913
+ "num_tokens": 12524183.0,
914
+ "step": 640
915
+ },
916
+ {
917
+ "entropy": 1.3366071730852127,
918
+ "epoch": 1.04,
919
+ "grad_norm": 19.375,
920
+ "learning_rate": 1.7414814814814815e-05,
921
+ "loss": 21.6169,
922
+ "mean_token_accuracy": 0.6800738513469696,
923
+ "num_tokens": 12718593.0,
924
+ "step": 650
925
+ },
926
+ {
927
+ "entropy": 1.3035322044044733,
928
+ "epoch": 1.056,
929
+ "grad_norm": 18.875,
930
+ "learning_rate": 1.7340740740740742e-05,
931
+ "loss": 21.0086,
932
+ "mean_token_accuracy": 0.6872079599648714,
933
+ "num_tokens": 12917803.0,
934
+ "step": 660
935
+ },
936
+ {
937
+ "entropy": 1.3336386241018772,
938
+ "epoch": 1.072,
939
+ "grad_norm": 20.125,
940
+ "learning_rate": 1.726666666666667e-05,
941
+ "loss": 21.3783,
942
+ "mean_token_accuracy": 0.6830162953585386,
943
+ "num_tokens": 13105826.0,
944
+ "step": 670
945
+ },
946
+ {
947
+ "entropy": 1.3145085237920284,
948
+ "epoch": 1.088,
949
+ "grad_norm": 19.875,
950
+ "learning_rate": 1.7192592592592595e-05,
951
+ "loss": 21.2325,
952
+ "mean_token_accuracy": 0.6849405582994222,
953
+ "num_tokens": 13298619.0,
954
+ "step": 680
955
+ },
956
+ {
957
+ "entropy": 1.3143333733081817,
958
+ "epoch": 1.104,
959
+ "grad_norm": 19.625,
960
+ "learning_rate": 1.711851851851852e-05,
961
+ "loss": 21.2224,
962
+ "mean_token_accuracy": 0.6856034275144338,
963
+ "num_tokens": 13491486.0,
964
+ "step": 690
965
+ },
966
+ {
967
+ "entropy": 1.3378163430839778,
968
+ "epoch": 1.12,
969
+ "grad_norm": 21.125,
970
+ "learning_rate": 1.7044444444444445e-05,
971
+ "loss": 21.5282,
972
+ "mean_token_accuracy": 0.6820732779800892,
973
+ "num_tokens": 13674663.0,
974
+ "step": 700
975
+ },
976
+ {
977
+ "epoch": 1.12,
978
+ "eval_biology_entropy": 1.2879770975112914,
979
+ "eval_biology_loss": 1.3086122274398804,
980
+ "eval_biology_mean_token_accuracy": 0.6899549188613892,
981
+ "eval_biology_num_tokens": 13674663.0,
982
+ "eval_biology_runtime": 48.4315,
983
+ "eval_biology_samples_per_second": 10.324,
984
+ "eval_biology_steps_per_second": 2.581,
985
+ "step": 700
986
+ },
987
+ {
988
+ "epoch": 1.12,
989
+ "eval_chemistry_entropy": 0.9838369536399841,
990
+ "eval_chemistry_loss": 0.9902806878089905,
991
+ "eval_chemistry_mean_token_accuracy": 0.757919919013977,
992
+ "eval_chemistry_num_tokens": 13674663.0,
993
+ "eval_chemistry_runtime": 60.0982,
994
+ "eval_chemistry_samples_per_second": 8.32,
995
+ "eval_chemistry_steps_per_second": 2.08,
996
+ "step": 700
997
+ },
998
+ {
999
+ "epoch": 1.12,
1000
+ "eval_math_entropy": 0.8821793942451477,
1001
+ "eval_math_loss": 1.108689308166504,
1002
+ "eval_math_mean_token_accuracy": 0.7436572351455688,
1003
+ "eval_math_num_tokens": 13674663.0,
1004
+ "eval_math_runtime": 61.6134,
1005
+ "eval_math_samples_per_second": 8.115,
1006
+ "eval_math_steps_per_second": 2.029,
1007
+ "step": 700
1008
+ },
1009
+ {
1010
+ "epoch": 1.12,
1011
+ "eval_cyber_entropy": 2.3264354813098906,
1012
+ "eval_cyber_loss": 2.8054211139678955,
1013
+ "eval_cyber_mean_token_accuracy": 0.49246201038360593,
1014
+ "eval_cyber_num_tokens": 13674663.0,
1015
+ "eval_cyber_runtime": 58.7919,
1016
+ "eval_cyber_samples_per_second": 6.753,
1017
+ "eval_cyber_steps_per_second": 1.701,
1018
+ "step": 700
1019
+ },
1020
+ {
1021
+ "entropy": 1.3018307507038116,
1022
+ "epoch": 1.1360000000000001,
1023
+ "grad_norm": 20.625,
1024
+ "learning_rate": 1.697037037037037e-05,
1025
+ "loss": 21.2057,
1026
+ "mean_token_accuracy": 0.685313081741333,
1027
+ "num_tokens": 13869134.0,
1028
+ "step": 710
1029
+ },
1030
+ {
1031
+ "entropy": 1.2973876625299454,
1032
+ "epoch": 1.152,
1033
+ "grad_norm": 18.0,
1034
+ "learning_rate": 1.6896296296296298e-05,
1035
+ "loss": 20.8346,
1036
+ "mean_token_accuracy": 0.6878448143601418,
1037
+ "num_tokens": 14078365.0,
1038
+ "step": 720
1039
+ },
1040
+ {
1041
+ "entropy": 1.3114512529224158,
1042
+ "epoch": 1.168,
1043
+ "grad_norm": 17.75,
1044
+ "learning_rate": 1.6822222222222224e-05,
1045
+ "loss": 21.086,
1046
+ "mean_token_accuracy": 0.684901175647974,
1047
+ "num_tokens": 14266831.0,
1048
+ "step": 730
1049
+ },
1050
+ {
1051
+ "entropy": 1.2921015951782464,
1052
+ "epoch": 1.184,
1053
+ "grad_norm": 19.875,
1054
+ "learning_rate": 1.6748148148148147e-05,
1055
+ "loss": 21.1084,
1056
+ "mean_token_accuracy": 0.6854452688246966,
1057
+ "num_tokens": 14465660.0,
1058
+ "step": 740
1059
+ },
1060
+ {
1061
+ "entropy": 1.3100581634789705,
1062
+ "epoch": 1.2,
1063
+ "grad_norm": 20.625,
1064
+ "learning_rate": 1.6674074074074077e-05,
1065
+ "loss": 20.9256,
1066
+ "mean_token_accuracy": 0.6867778114974499,
1067
+ "num_tokens": 14653228.0,
1068
+ "step": 750
1069
+ },
1070
+ {
1071
+ "entropy": 1.2675646111369132,
1072
+ "epoch": 1.216,
1073
+ "grad_norm": 19.5,
1074
+ "learning_rate": 1.66e-05,
1075
+ "loss": 20.4105,
1076
+ "mean_token_accuracy": 0.6919024847447872,
1077
+ "num_tokens": 14857782.0,
1078
+ "step": 760
1079
+ },
1080
+ {
1081
+ "entropy": 1.291072677075863,
1082
+ "epoch": 1.232,
1083
+ "grad_norm": 17.875,
1084
+ "learning_rate": 1.6525925925925927e-05,
1085
+ "loss": 20.8356,
1086
+ "mean_token_accuracy": 0.689462835341692,
1087
+ "num_tokens": 15047356.0,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "entropy": 1.296948005259037,
1092
+ "epoch": 1.248,
1093
+ "grad_norm": 19.0,
1094
+ "learning_rate": 1.6451851851851853e-05,
1095
+ "loss": 21.0571,
1096
+ "mean_token_accuracy": 0.6855845946818591,
1097
+ "num_tokens": 15241098.0,
1098
+ "step": 780
1099
+ },
1100
+ {
1101
+ "entropy": 1.3225075900554657,
1102
+ "epoch": 1.264,
1103
+ "grad_norm": 18.25,
1104
+ "learning_rate": 1.637777777777778e-05,
1105
+ "loss": 21.1821,
1106
+ "mean_token_accuracy": 0.6856160722672939,
1107
+ "num_tokens": 15437657.0,
1108
+ "step": 790
1109
+ },
1110
+ {
1111
+ "entropy": 1.3229579757899046,
1112
+ "epoch": 1.28,
1113
+ "grad_norm": 19.75,
1114
+ "learning_rate": 1.6303703703703706e-05,
1115
+ "loss": 21.3985,
1116
+ "mean_token_accuracy": 0.681986927613616,
1117
+ "num_tokens": 15630795.0,
1118
+ "step": 800
1119
+ },
1120
+ {
1121
+ "epoch": 1.28,
1122
+ "eval_biology_entropy": 1.2697611575126648,
1123
+ "eval_biology_loss": 1.299078106880188,
1124
+ "eval_biology_mean_token_accuracy": 0.6915725932121277,
1125
+ "eval_biology_num_tokens": 15630795.0,
1126
+ "eval_biology_runtime": 48.5582,
1127
+ "eval_biology_samples_per_second": 10.297,
1128
+ "eval_biology_steps_per_second": 2.574,
1129
+ "step": 800
1130
+ },
1131
+ {
1132
+ "epoch": 1.28,
1133
+ "eval_chemistry_entropy": 0.9691643033027649,
1134
+ "eval_chemistry_loss": 0.9838607311248779,
1135
+ "eval_chemistry_mean_token_accuracy": 0.7591533994674683,
1136
+ "eval_chemistry_num_tokens": 15630795.0,
1137
+ "eval_chemistry_runtime": 60.2581,
1138
+ "eval_chemistry_samples_per_second": 8.298,
1139
+ "eval_chemistry_steps_per_second": 2.074,
1140
+ "step": 800
1141
+ },
1142
+ {
1143
+ "epoch": 1.28,
1144
+ "eval_math_entropy": 0.863752941608429,
1145
+ "eval_math_loss": 1.1061811447143555,
1146
+ "eval_math_mean_token_accuracy": 0.7452329030036926,
1147
+ "eval_math_num_tokens": 15630795.0,
1148
+ "eval_math_runtime": 61.8563,
1149
+ "eval_math_samples_per_second": 8.083,
1150
+ "eval_math_steps_per_second": 2.021,
1151
+ "step": 800
1152
+ },
1153
+ {
1154
+ "epoch": 1.28,
1155
+ "eval_cyber_entropy": 2.3142928409576418,
1156
+ "eval_cyber_loss": 2.8583574295043945,
1157
+ "eval_cyber_mean_token_accuracy": 0.4875562259554863,
1158
+ "eval_cyber_num_tokens": 15630795.0,
1159
+ "eval_cyber_runtime": 58.9374,
1160
+ "eval_cyber_samples_per_second": 6.736,
1161
+ "eval_cyber_steps_per_second": 1.697,
1162
+ "step": 800
1163
+ },
1164
+ {
1165
+ "entropy": 1.322736281901598,
1166
+ "epoch": 1.296,
1167
+ "grad_norm": 18.375,
1168
+ "learning_rate": 1.622962962962963e-05,
1169
+ "loss": 21.2994,
1170
+ "mean_token_accuracy": 0.6834880676120519,
1171
+ "num_tokens": 15827105.0,
1172
+ "step": 810
1173
+ },
1174
+ {
1175
+ "entropy": 1.3064148597419263,
1176
+ "epoch": 1.312,
1177
+ "grad_norm": 20.375,
1178
+ "learning_rate": 1.6155555555555556e-05,
1179
+ "loss": 21.2908,
1180
+ "mean_token_accuracy": 0.6831366084516048,
1181
+ "num_tokens": 16019645.0,
1182
+ "step": 820
1183
+ },
1184
+ {
1185
+ "entropy": 1.3229932710528374,
1186
+ "epoch": 1.328,
1187
+ "grad_norm": 19.0,
1188
+ "learning_rate": 1.6081481481481482e-05,
1189
+ "loss": 21.2517,
1190
+ "mean_token_accuracy": 0.6835683796554803,
1191
+ "num_tokens": 16221726.0,
1192
+ "step": 830
1193
+ },
1194
+ {
1195
+ "entropy": 1.2925970274955034,
1196
+ "epoch": 1.3439999999999999,
1197
+ "grad_norm": 17.5,
1198
+ "learning_rate": 1.600740740740741e-05,
1199
+ "loss": 20.7376,
1200
+ "mean_token_accuracy": 0.6901313602924347,
1201
+ "num_tokens": 16427594.0,
1202
+ "step": 840
1203
+ },
1204
+ {
1205
+ "entropy": 1.2990364480763674,
1206
+ "epoch": 1.3599999999999999,
1207
+ "grad_norm": 19.875,
1208
+ "learning_rate": 1.5933333333333336e-05,
1209
+ "loss": 21.0638,
1210
+ "mean_token_accuracy": 0.6867326147854328,
1211
+ "num_tokens": 16621605.0,
1212
+ "step": 850
1213
+ },
1214
+ {
1215
+ "entropy": 1.3395817268639802,
1216
+ "epoch": 1.376,
1217
+ "grad_norm": 21.5,
1218
+ "learning_rate": 1.5859259259259262e-05,
1219
+ "loss": 21.6332,
1220
+ "mean_token_accuracy": 0.6804220989346504,
1221
+ "num_tokens": 16813444.0,
1222
+ "step": 860
1223
+ },
1224
+ {
1225
+ "entropy": 1.3104124892503024,
1226
+ "epoch": 1.392,
1227
+ "grad_norm": 18.625,
1228
+ "learning_rate": 1.5785185185185185e-05,
1229
+ "loss": 21.1663,
1230
+ "mean_token_accuracy": 0.6856168929487467,
1231
+ "num_tokens": 17006509.0,
1232
+ "step": 870
1233
+ },
1234
+ {
1235
+ "entropy": 1.3050751704722643,
1236
+ "epoch": 1.408,
1237
+ "grad_norm": 16.625,
1238
+ "learning_rate": 1.571111111111111e-05,
1239
+ "loss": 20.8645,
1240
+ "mean_token_accuracy": 0.6886320147663355,
1241
+ "num_tokens": 17197870.0,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "entropy": 1.2993885681033135,
1246
+ "epoch": 1.424,
1247
+ "grad_norm": 19.625,
1248
+ "learning_rate": 1.5637037037037038e-05,
1249
+ "loss": 21.0292,
1250
+ "mean_token_accuracy": 0.6871409460902214,
1251
+ "num_tokens": 17394390.0,
1252
+ "step": 890
1253
+ },
1254
+ {
1255
+ "entropy": 1.2986160211265088,
1256
+ "epoch": 1.44,
1257
+ "grad_norm": 22.75,
1258
+ "learning_rate": 1.5562962962962965e-05,
1259
+ "loss": 21.064,
1260
+ "mean_token_accuracy": 0.6858013048768044,
1261
+ "num_tokens": 17587777.0,
1262
+ "step": 900
1263
+ },
1264
+ {
1265
+ "epoch": 1.44,
1266
+ "eval_biology_entropy": 1.2633282842636109,
1267
+ "eval_biology_loss": 1.2911479473114014,
1268
+ "eval_biology_mean_token_accuracy": 0.6930723128318786,
1269
+ "eval_biology_num_tokens": 17587777.0,
1270
+ "eval_biology_runtime": 48.5747,
1271
+ "eval_biology_samples_per_second": 10.293,
1272
+ "eval_biology_steps_per_second": 2.573,
1273
+ "step": 900
1274
+ },
1275
+ {
1276
+ "epoch": 1.44,
1277
+ "eval_chemistry_entropy": 0.9724984226226807,
1278
+ "eval_chemistry_loss": 0.9782924056053162,
1279
+ "eval_chemistry_mean_token_accuracy": 0.7604495301246643,
1280
+ "eval_chemistry_num_tokens": 17587777.0,
1281
+ "eval_chemistry_runtime": 60.2279,
1282
+ "eval_chemistry_samples_per_second": 8.302,
1283
+ "eval_chemistry_steps_per_second": 2.075,
1284
+ "step": 900
1285
+ },
1286
+ {
1287
+ "epoch": 1.44,
1288
+ "eval_math_entropy": 0.878196352481842,
1289
+ "eval_math_loss": 1.0984141826629639,
1290
+ "eval_math_mean_token_accuracy": 0.7456621165275574,
1291
+ "eval_math_num_tokens": 17587777.0,
1292
+ "eval_math_runtime": 61.7802,
1293
+ "eval_math_samples_per_second": 8.093,
1294
+ "eval_math_steps_per_second": 2.023,
1295
+ "step": 900
1296
+ },
1297
+ {
1298
+ "epoch": 1.44,
1299
+ "eval_cyber_entropy": 2.317098902463913,
1300
+ "eval_cyber_loss": 2.8237366676330566,
1301
+ "eval_cyber_mean_token_accuracy": 0.49122542947530745,
1302
+ "eval_cyber_num_tokens": 17587777.0,
1303
+ "eval_cyber_runtime": 58.9196,
1304
+ "eval_cyber_samples_per_second": 6.738,
1305
+ "eval_cyber_steps_per_second": 1.697,
1306
+ "step": 900
1307
+ },
1308
+ {
1309
+ "entropy": 1.2987288374453783,
1310
+ "epoch": 1.456,
1311
+ "grad_norm": 18.375,
1312
+ "learning_rate": 1.548888888888889e-05,
1313
+ "loss": 20.8783,
1314
+ "mean_token_accuracy": 0.6883518882095814,
1315
+ "num_tokens": 17788456.0,
1316
+ "step": 910
1317
+ },
1318
+ {
1319
+ "entropy": 1.292324574291706,
1320
+ "epoch": 1.472,
1321
+ "grad_norm": 17.375,
1322
+ "learning_rate": 1.5414814814814814e-05,
1323
+ "loss": 20.8447,
1324
+ "mean_token_accuracy": 0.6893134098500013,
1325
+ "num_tokens": 17984063.0,
1326
+ "step": 920
1327
+ },
1328
+ {
1329
+ "entropy": 1.2839893605560064,
1330
+ "epoch": 1.488,
1331
+ "grad_norm": 19.125,
1332
+ "learning_rate": 1.5340740740740744e-05,
1333
+ "loss": 20.8773,
1334
+ "mean_token_accuracy": 0.6899567134678364,
1335
+ "num_tokens": 18175640.0,
1336
+ "step": 930
1337
+ },
1338
+ {
1339
+ "entropy": 1.3172472290694714,
1340
+ "epoch": 1.504,
1341
+ "grad_norm": 19.25,
1342
+ "learning_rate": 1.5266666666666667e-05,
1343
+ "loss": 21.141,
1344
+ "mean_token_accuracy": 0.687034023180604,
1345
+ "num_tokens": 18367857.0,
1346
+ "step": 940
1347
+ },
1348
+ {
1349
+ "entropy": 1.288551079109311,
1350
+ "epoch": 1.52,
1351
+ "grad_norm": 18.375,
1352
+ "learning_rate": 1.5192592592592594e-05,
1353
+ "loss": 21.025,
1354
+ "mean_token_accuracy": 0.6869576655328273,
1355
+ "num_tokens": 18569146.0,
1356
+ "step": 950
1357
+ },
1358
+ {
1359
+ "entropy": 1.3221289869397879,
1360
+ "epoch": 1.536,
1361
+ "grad_norm": 19.75,
1362
+ "learning_rate": 1.5118518518518519e-05,
1363
+ "loss": 21.2,
1364
+ "mean_token_accuracy": 0.6844527788460255,
1365
+ "num_tokens": 18755079.0,
1366
+ "step": 960
1367
+ },
1368
+ {
1369
+ "entropy": 1.2898046210408212,
1370
+ "epoch": 1.552,
1371
+ "grad_norm": 20.375,
1372
+ "learning_rate": 1.5044444444444445e-05,
1373
+ "loss": 20.8166,
1374
+ "mean_token_accuracy": 0.690209424123168,
1375
+ "num_tokens": 18956248.0,
1376
+ "step": 970
1377
+ },
1378
+ {
1379
+ "entropy": 1.280243619531393,
1380
+ "epoch": 1.568,
1381
+ "grad_norm": 19.875,
1382
+ "learning_rate": 1.497037037037037e-05,
1383
+ "loss": 20.6218,
1384
+ "mean_token_accuracy": 0.6917802631855011,
1385
+ "num_tokens": 19150315.0,
1386
+ "step": 980
1387
+ },
1388
+ {
1389
+ "entropy": 1.2912988126277924,
1390
+ "epoch": 1.584,
1391
+ "grad_norm": 18.875,
1392
+ "learning_rate": 1.4896296296296298e-05,
1393
+ "loss": 20.7886,
1394
+ "mean_token_accuracy": 0.6892261151224375,
1395
+ "num_tokens": 19344260.0,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "entropy": 1.3189360287040472,
1400
+ "epoch": 1.6,
1401
+ "grad_norm": 19.0,
1402
+ "learning_rate": 1.4822222222222225e-05,
1403
+ "loss": 21.2912,
1404
+ "mean_token_accuracy": 0.6828928265720606,
1405
+ "num_tokens": 19532552.0,
1406
+ "step": 1000
1407
+ },
1408
+ {
1409
+ "epoch": 1.6,
1410
+ "eval_biology_entropy": 1.263430886745453,
1411
+ "eval_biology_loss": 1.2847552299499512,
1412
+ "eval_biology_mean_token_accuracy": 0.6937462825775147,
1413
+ "eval_biology_num_tokens": 19532552.0,
1414
+ "eval_biology_runtime": 48.616,
1415
+ "eval_biology_samples_per_second": 10.285,
1416
+ "eval_biology_steps_per_second": 2.571,
1417
+ "step": 1000
1418
+ },
1419
+ {
1420
+ "epoch": 1.6,
1421
+ "eval_chemistry_entropy": 0.9613517146110535,
1422
+ "eval_chemistry_loss": 0.9752106070518494,
1423
+ "eval_chemistry_mean_token_accuracy": 0.7608753662109375,
1424
+ "eval_chemistry_num_tokens": 19532552.0,
1425
+ "eval_chemistry_runtime": 60.0131,
1426
+ "eval_chemistry_samples_per_second": 8.332,
1427
+ "eval_chemistry_steps_per_second": 2.083,
1428
+ "step": 1000
1429
+ },
1430
+ {
1431
+ "epoch": 1.6,
1432
+ "eval_math_entropy": 0.8507145104408265,
1433
+ "eval_math_loss": 1.0975427627563477,
1434
+ "eval_math_mean_token_accuracy": 0.7471181530952453,
1435
+ "eval_math_num_tokens": 19532552.0,
1436
+ "eval_math_runtime": 61.5437,
1437
+ "eval_math_samples_per_second": 8.124,
1438
+ "eval_math_steps_per_second": 2.031,
1439
+ "step": 1000
1440
+ },
1441
+ {
1442
+ "epoch": 1.6,
1443
+ "eval_cyber_entropy": 2.262136605978012,
1444
+ "eval_cyber_loss": 2.8179190158843994,
1445
+ "eval_cyber_mean_token_accuracy": 0.4946762239933014,
1446
+ "eval_cyber_num_tokens": 19532552.0,
1447
+ "eval_cyber_runtime": 58.9033,
1448
+ "eval_cyber_samples_per_second": 6.74,
1449
+ "eval_cyber_steps_per_second": 1.698,
1450
+ "step": 1000
1451
+ }
1452
+ ],
1453
+ "logging_steps": 10,
1454
+ "max_steps": 3000,
1455
+ "num_input_tokens_seen": 0,
1456
+ "num_train_epochs": 5,
1457
+ "save_steps": 1000,
1458
+ "stateful_callbacks": {
1459
+ "TrainerControl": {
1460
+ "args": {
1461
+ "should_epoch_stop": false,
1462
+ "should_evaluate": false,
1463
+ "should_log": false,
1464
+ "should_save": true,
1465
+ "should_training_stop": false
1466
+ },
1467
+ "attributes": {}
1468
+ }
1469
+ },
1470
+ "total_flos": 2.2020838985302374e+18,
1471
+ "train_batch_size": 4,
1472
+ "trial_name": null,
1473
+ "trial_params": null
1474
+ }
recover-backup/checkpoint-2000/added_tokens.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "<image_soft_token>": 262144
3
+ }
recover-backup/checkpoint-2000/chat_template.jinja ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'image' -%}
33
+ {{ '<start_of_image>' }}
34
+ {%- elif item['type'] == 'text' -%}
35
+ {{ item['text'] | trim }}
36
+ {%- endif -%}
37
+ {%- endfor -%}
38
+ {%- else -%}
39
+ {{ raise_exception("Invalid content type") }}
40
+ {%- endif -%}
41
+ {{ '<end_of_turn>
42
+ ' }}
43
+ {%- endfor -%}
44
+ {%- if add_generation_prompt -%}
45
+ {{'<start_of_turn>model
46
+ '}}
47
+ {%- endif -%}
recover-backup/checkpoint-2000/config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "dtype": "bfloat16",
21
+ "final_logit_softcapping": null,
22
+ "head_dim": 256,
23
+ "hidden_activation": "gelu_pytorch_tanh",
24
+ "hidden_size": 3840,
25
+ "initializer_range": 0.02,
26
+ "intermediate_size": 15360,
27
+ "layer_types": [
28
+ "sliding_attention",
29
+ "sliding_attention",
30
+ "sliding_attention",
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "full_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "sliding_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "full_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "sliding_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "full_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "sliding_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "full_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "sliding_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "full_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "sliding_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "full_attention",
64
+ "sliding_attention",
65
+ "sliding_attention",
66
+ "sliding_attention",
67
+ "sliding_attention",
68
+ "sliding_attention",
69
+ "full_attention",
70
+ "sliding_attention",
71
+ "sliding_attention",
72
+ "sliding_attention",
73
+ "sliding_attention",
74
+ "sliding_attention",
75
+ "full_attention"
76
+ ],
77
+ "max_position_embeddings": 131072,
78
+ "model_type": "gemma3_text",
79
+ "num_attention_heads": 16,
80
+ "num_hidden_layers": 48,
81
+ "num_key_value_heads": 8,
82
+ "query_pre_attn_scalar": 256,
83
+ "rms_norm_eps": 1e-06,
84
+ "rope_local_base_freq": 10000.0,
85
+ "rope_scaling": {
86
+ "factor": 8.0,
87
+ "rope_type": "linear"
88
+ },
89
+ "rope_theta": 1000000.0,
90
+ "sliding_window": 1024,
91
+ "use_cache": true,
92
+ "vocab_size": 262208
93
+ },
94
+ "transformers_version": "4.56.1",
95
+ "vision_config": {
96
+ "attention_dropout": 0.0,
97
+ "dtype": "bfloat16",
98
+ "hidden_act": "gelu_pytorch_tanh",
99
+ "hidden_size": 1152,
100
+ "image_size": 896,
101
+ "intermediate_size": 4304,
102
+ "layer_norm_eps": 1e-06,
103
+ "model_type": "siglip_vision_model",
104
+ "num_attention_heads": 16,
105
+ "num_channels": 3,
106
+ "num_hidden_layers": 27,
107
+ "patch_size": 14,
108
+ "vision_use_head": false
109
+ }
110
+ }
recover-backup/checkpoint-2000/generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 1,
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "top_k": 64,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.56.1"
13
+ }
recover-backup/checkpoint-2000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
recover-backup/checkpoint-2000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c56cc7d2ab3dfe9edf7a28cb038a85004572c59f841d9bdfd8a137d5d874b19
3
+ size 1465
recover-backup/checkpoint-2000/special_tokens_map.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "boi_token": "<start_of_image>",
3
+ "bos_token": {
4
+ "content": "<bos>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "eoi_token": "<end_of_image>",
11
+ "eos_token": {
12
+ "content": "<eos>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false
17
+ },
18
+ "image_token": "<image_soft_token>",
19
+ "pad_token": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "unk_token": {
27
+ "content": "<unk>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ }
33
+ }
recover-backup/checkpoint-2000/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4667f2089529e8e7657cfb6d1c19910ae71ff5f28aa7ab2ff2763330affad795
3
+ size 33384568
recover-backup/checkpoint-2000/tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff
 
recover-backup/checkpoint-2000/trainer_state.json ADDED
@@ -0,0 +1,2914 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 3.2,
6
+ "eval_steps": 100,
7
+ "global_step": 2000,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "entropy": 0.6808024171739817,
14
+ "epoch": 0.016,
15
+ "grad_norm": 2096.0,
16
+ "learning_rate": 6.000000000000001e-07,
17
+ "loss": 81.3949,
18
+ "mean_token_accuracy": 0.469288082793355,
19
+ "num_tokens": 195524.0,
20
+ "step": 10
21
+ },
22
+ {
23
+ "entropy": 0.6932845771312713,
24
+ "epoch": 0.032,
25
+ "grad_norm": 1840.0,
26
+ "learning_rate": 1.2666666666666669e-06,
27
+ "loss": 79.1276,
28
+ "mean_token_accuracy": 0.4702936248853803,
29
+ "num_tokens": 390903.0,
30
+ "step": 20
31
+ },
32
+ {
33
+ "entropy": 0.7348822019994259,
34
+ "epoch": 0.048,
35
+ "grad_norm": 1896.0,
36
+ "learning_rate": 1.9333333333333336e-06,
37
+ "loss": 75.5796,
38
+ "mean_token_accuracy": 0.47059151269495486,
39
+ "num_tokens": 589868.0,
40
+ "step": 30
41
+ },
42
+ {
43
+ "entropy": 0.8692382726818323,
44
+ "epoch": 0.064,
45
+ "grad_norm": 868.0,
46
+ "learning_rate": 2.6e-06,
47
+ "loss": 66.3894,
48
+ "mean_token_accuracy": 0.4747353194281459,
49
+ "num_tokens": 791190.0,
50
+ "step": 40
51
+ },
52
+ {
53
+ "entropy": 1.1099284883588552,
54
+ "epoch": 0.08,
55
+ "grad_norm": 520.0,
56
+ "learning_rate": 3.266666666666667e-06,
57
+ "loss": 57.5001,
58
+ "mean_token_accuracy": 0.47747670076787474,
59
+ "num_tokens": 989860.0,
60
+ "step": 50
61
+ },
62
+ {
63
+ "entropy": 1.4874154418706893,
64
+ "epoch": 0.096,
65
+ "grad_norm": 144.0,
66
+ "learning_rate": 3.9333333333333335e-06,
67
+ "loss": 48.6006,
68
+ "mean_token_accuracy": 0.4924122652038932,
69
+ "num_tokens": 1181777.0,
70
+ "step": 60
71
+ },
72
+ {
73
+ "entropy": 2.022240047156811,
74
+ "epoch": 0.112,
75
+ "grad_norm": 52.0,
76
+ "learning_rate": 4.600000000000001e-06,
77
+ "loss": 41.7489,
78
+ "mean_token_accuracy": 0.5201752169057727,
79
+ "num_tokens": 1385513.0,
80
+ "step": 70
81
+ },
82
+ {
83
+ "entropy": 2.395130204409361,
84
+ "epoch": 0.128,
85
+ "grad_norm": 44.0,
86
+ "learning_rate": 5.2666666666666665e-06,
87
+ "loss": 37.2886,
88
+ "mean_token_accuracy": 0.5428294729441404,
89
+ "num_tokens": 1582368.0,
90
+ "step": 80
91
+ },
92
+ {
93
+ "entropy": 2.1368538811802864,
94
+ "epoch": 0.144,
95
+ "grad_norm": 29.5,
96
+ "learning_rate": 5.933333333333335e-06,
97
+ "loss": 33.907,
98
+ "mean_token_accuracy": 0.5704466678202152,
99
+ "num_tokens": 1773764.0,
100
+ "step": 90
101
+ },
102
+ {
103
+ "entropy": 1.9877845980226994,
104
+ "epoch": 0.16,
105
+ "grad_norm": 25.375,
106
+ "learning_rate": 6.600000000000001e-06,
107
+ "loss": 31.7141,
108
+ "mean_token_accuracy": 0.58747753277421,
109
+ "num_tokens": 1970077.0,
110
+ "step": 100
111
+ },
112
+ {
113
+ "epoch": 0.16,
114
+ "eval_biology_entropy": 1.911924614906311,
115
+ "eval_biology_loss": 1.831228256225586,
116
+ "eval_biology_mean_token_accuracy": 0.6091388630867004,
117
+ "eval_biology_num_tokens": 1970077.0,
118
+ "eval_biology_runtime": 48.4202,
119
+ "eval_biology_samples_per_second": 10.326,
120
+ "eval_biology_steps_per_second": 2.582,
121
+ "step": 100
122
+ },
123
+ {
124
+ "epoch": 0.16,
125
+ "eval_chemistry_entropy": 1.5279665517807006,
126
+ "eval_chemistry_loss": 1.507634162902832,
127
+ "eval_chemistry_mean_token_accuracy": 0.6661558594703675,
128
+ "eval_chemistry_num_tokens": 1970077.0,
129
+ "eval_chemistry_runtime": 60.0771,
130
+ "eval_chemistry_samples_per_second": 8.323,
131
+ "eval_chemistry_steps_per_second": 2.081,
132
+ "step": 100
133
+ },
134
+ {
135
+ "epoch": 0.16,
136
+ "eval_math_entropy": 1.3111717319488525,
137
+ "eval_math_loss": 1.5638481378555298,
138
+ "eval_math_mean_token_accuracy": 0.6615601582527161,
139
+ "eval_math_num_tokens": 1970077.0,
140
+ "eval_math_runtime": 61.5942,
141
+ "eval_math_samples_per_second": 8.118,
142
+ "eval_math_steps_per_second": 2.029,
143
+ "step": 100
144
+ },
145
+ {
146
+ "epoch": 0.16,
147
+ "eval_cyber_entropy": 2.3538206100463865,
148
+ "eval_cyber_loss": 3.4596033096313477,
149
+ "eval_cyber_mean_token_accuracy": 0.4573894315958023,
150
+ "eval_cyber_num_tokens": 1970077.0,
151
+ "eval_cyber_runtime": 58.7811,
152
+ "eval_cyber_samples_per_second": 6.754,
153
+ "eval_cyber_steps_per_second": 1.701,
154
+ "step": 100
155
+ },
156
+ {
157
+ "entropy": 1.8946144469082355,
158
+ "epoch": 0.176,
159
+ "grad_norm": 22.0,
160
+ "learning_rate": 7.266666666666668e-06,
161
+ "loss": 29.5004,
162
+ "mean_token_accuracy": 0.6055971266701817,
163
+ "num_tokens": 2168354.0,
164
+ "step": 110
165
+ },
166
+ {
167
+ "entropy": 1.7916729532182216,
168
+ "epoch": 0.192,
169
+ "grad_norm": 22.375,
170
+ "learning_rate": 7.933333333333334e-06,
171
+ "loss": 28.5473,
172
+ "mean_token_accuracy": 0.6124886509031058,
173
+ "num_tokens": 2365822.0,
174
+ "step": 120
175
+ },
176
+ {
177
+ "entropy": 1.7780573107302189,
178
+ "epoch": 0.208,
179
+ "grad_norm": 20.625,
180
+ "learning_rate": 8.6e-06,
181
+ "loss": 28.0106,
182
+ "mean_token_accuracy": 0.617958590388298,
183
+ "num_tokens": 2558762.0,
184
+ "step": 130
185
+ },
186
+ {
187
+ "entropy": 1.6913408383727073,
188
+ "epoch": 0.224,
189
+ "grad_norm": 19.875,
190
+ "learning_rate": 9.266666666666667e-06,
191
+ "loss": 27.182,
192
+ "mean_token_accuracy": 0.6272589169442654,
193
+ "num_tokens": 2755347.0,
194
+ "step": 140
195
+ },
196
+ {
197
+ "entropy": 1.6812096990644931,
198
+ "epoch": 0.24,
199
+ "grad_norm": 19.5,
200
+ "learning_rate": 9.933333333333334e-06,
201
+ "loss": 26.6205,
202
+ "mean_token_accuracy": 0.6303446084260941,
203
+ "num_tokens": 2947346.0,
204
+ "step": 150
205
+ },
206
+ {
207
+ "entropy": 1.6677771665155887,
208
+ "epoch": 0.256,
209
+ "grad_norm": 20.25,
210
+ "learning_rate": 1.0600000000000002e-05,
211
+ "loss": 26.6086,
212
+ "mean_token_accuracy": 0.6319835562258959,
213
+ "num_tokens": 3139957.0,
214
+ "step": 160
215
+ },
216
+ {
217
+ "entropy": 1.6161759495735168,
218
+ "epoch": 0.272,
219
+ "grad_norm": 20.375,
220
+ "learning_rate": 1.1266666666666668e-05,
221
+ "loss": 26.0026,
222
+ "mean_token_accuracy": 0.6365392006933689,
223
+ "num_tokens": 3335951.0,
224
+ "step": 170
225
+ },
226
+ {
227
+ "entropy": 1.6003736153244972,
228
+ "epoch": 0.288,
229
+ "grad_norm": 21.0,
230
+ "learning_rate": 1.1933333333333335e-05,
231
+ "loss": 25.4302,
232
+ "mean_token_accuracy": 0.6393310058861971,
233
+ "num_tokens": 3539731.0,
234
+ "step": 180
235
+ },
236
+ {
237
+ "entropy": 1.5776585705578328,
238
+ "epoch": 0.304,
239
+ "grad_norm": 18.875,
240
+ "learning_rate": 1.2600000000000001e-05,
241
+ "loss": 25.2177,
242
+ "mean_token_accuracy": 0.646138820424676,
243
+ "num_tokens": 3733488.0,
244
+ "step": 190
245
+ },
246
+ {
247
+ "entropy": 1.5823919720947743,
248
+ "epoch": 0.32,
249
+ "grad_norm": 21.75,
250
+ "learning_rate": 1.3266666666666668e-05,
251
+ "loss": 25.46,
252
+ "mean_token_accuracy": 0.6409369662404061,
253
+ "num_tokens": 3920545.0,
254
+ "step": 200
255
+ },
256
+ {
257
+ "epoch": 0.32,
258
+ "eval_biology_entropy": 1.5241963243484498,
259
+ "eval_biology_loss": 1.512497067451477,
260
+ "eval_biology_mean_token_accuracy": 0.656851170539856,
261
+ "eval_biology_num_tokens": 3920545.0,
262
+ "eval_biology_runtime": 48.4372,
263
+ "eval_biology_samples_per_second": 10.323,
264
+ "eval_biology_steps_per_second": 2.581,
265
+ "step": 200
266
+ },
267
+ {
268
+ "epoch": 0.32,
269
+ "eval_chemistry_entropy": 1.1822538471221924,
270
+ "eval_chemistry_loss": 1.178497552871704,
271
+ "eval_chemistry_mean_token_accuracy": 0.7221880826950073,
272
+ "eval_chemistry_num_tokens": 3920545.0,
273
+ "eval_chemistry_runtime": 60.0879,
274
+ "eval_chemistry_samples_per_second": 8.321,
275
+ "eval_chemistry_steps_per_second": 2.08,
276
+ "step": 200
277
+ },
278
+ {
279
+ "epoch": 0.32,
280
+ "eval_math_entropy": 0.9796018509864807,
281
+ "eval_math_loss": 1.2802495956420898,
282
+ "eval_math_mean_token_accuracy": 0.713343710899353,
283
+ "eval_math_num_tokens": 3920545.0,
284
+ "eval_math_runtime": 61.6189,
285
+ "eval_math_samples_per_second": 8.114,
286
+ "eval_math_steps_per_second": 2.029,
287
+ "step": 200
288
+ },
289
+ {
290
+ "epoch": 0.32,
291
+ "eval_cyber_entropy": 2.516505275964737,
292
+ "eval_cyber_loss": 3.1510307788848877,
293
+ "eval_cyber_mean_token_accuracy": 0.47284564673900603,
294
+ "eval_cyber_num_tokens": 3920545.0,
295
+ "eval_cyber_runtime": 58.8071,
296
+ "eval_cyber_samples_per_second": 6.751,
297
+ "eval_cyber_steps_per_second": 1.7,
298
+ "step": 200
299
+ },
300
+ {
301
+ "entropy": 1.5317105814814567,
302
+ "epoch": 0.336,
303
+ "grad_norm": 18.375,
304
+ "learning_rate": 1.3933333333333334e-05,
305
+ "loss": 24.5282,
306
+ "mean_token_accuracy": 0.6502130385488272,
307
+ "num_tokens": 4114077.0,
308
+ "step": 210
309
+ },
310
+ {
311
+ "entropy": 1.5466628648340701,
312
+ "epoch": 0.352,
313
+ "grad_norm": 18.875,
314
+ "learning_rate": 1.46e-05,
315
+ "loss": 24.8233,
316
+ "mean_token_accuracy": 0.6490028414875268,
317
+ "num_tokens": 4306949.0,
318
+ "step": 220
319
+ },
320
+ {
321
+ "entropy": 1.5188672259449958,
322
+ "epoch": 0.368,
323
+ "grad_norm": 19.25,
324
+ "learning_rate": 1.5266666666666667e-05,
325
+ "loss": 24.4549,
326
+ "mean_token_accuracy": 0.6544424753636122,
327
+ "num_tokens": 4504001.0,
328
+ "step": 230
329
+ },
330
+ {
331
+ "entropy": 1.5070753492414952,
332
+ "epoch": 0.384,
333
+ "grad_norm": 21.75,
334
+ "learning_rate": 1.5933333333333336e-05,
335
+ "loss": 24.3379,
336
+ "mean_token_accuracy": 0.6532186731696129,
337
+ "num_tokens": 4693812.0,
338
+ "step": 240
339
+ },
340
+ {
341
+ "entropy": 1.5358249843120575,
342
+ "epoch": 0.4,
343
+ "grad_norm": 17.25,
344
+ "learning_rate": 1.66e-05,
345
+ "loss": 24.4225,
346
+ "mean_token_accuracy": 0.6525934401899576,
347
+ "num_tokens": 4887094.0,
348
+ "step": 250
349
+ },
350
+ {
351
+ "entropy": 1.5161145091056825,
352
+ "epoch": 0.416,
353
+ "grad_norm": 19.75,
354
+ "learning_rate": 1.726666666666667e-05,
355
+ "loss": 24.4554,
356
+ "mean_token_accuracy": 0.6528204523026944,
357
+ "num_tokens": 5085369.0,
358
+ "step": 260
359
+ },
360
+ {
361
+ "entropy": 1.5000567641109228,
362
+ "epoch": 0.432,
363
+ "grad_norm": 20.625,
364
+ "learning_rate": 1.7933333333333333e-05,
365
+ "loss": 24.2948,
366
+ "mean_token_accuracy": 0.6535403318703175,
367
+ "num_tokens": 5271275.0,
368
+ "step": 270
369
+ },
370
+ {
371
+ "entropy": 1.5077810816466808,
372
+ "epoch": 0.448,
373
+ "grad_norm": 22.375,
374
+ "learning_rate": 1.86e-05,
375
+ "loss": 24.2712,
376
+ "mean_token_accuracy": 0.6533999439328909,
377
+ "num_tokens": 5460559.0,
378
+ "step": 280
379
+ },
380
+ {
381
+ "entropy": 1.4563340097665787,
382
+ "epoch": 0.464,
383
+ "grad_norm": 22.0,
384
+ "learning_rate": 1.926666666666667e-05,
385
+ "loss": 23.2755,
386
+ "mean_token_accuracy": 0.6638390984386205,
387
+ "num_tokens": 5653809.0,
388
+ "step": 290
389
+ },
390
+ {
391
+ "entropy": 1.468742286041379,
392
+ "epoch": 0.48,
393
+ "grad_norm": 18.0,
394
+ "learning_rate": 1.9933333333333334e-05,
395
+ "loss": 23.7149,
396
+ "mean_token_accuracy": 0.6593498229980469,
397
+ "num_tokens": 5850176.0,
398
+ "step": 300
399
+ },
400
+ {
401
+ "epoch": 0.48,
402
+ "eval_biology_entropy": 1.4316115617752074,
403
+ "eval_biology_loss": 1.4177387952804565,
404
+ "eval_biology_mean_token_accuracy": 0.6717347936630249,
405
+ "eval_biology_num_tokens": 5850176.0,
406
+ "eval_biology_runtime": 48.4847,
407
+ "eval_biology_samples_per_second": 10.313,
408
+ "eval_biology_steps_per_second": 2.578,
409
+ "step": 300
410
+ },
411
+ {
412
+ "epoch": 0.48,
413
+ "eval_chemistry_entropy": 1.1160053358078004,
414
+ "eval_chemistry_loss": 1.0916916131973267,
415
+ "eval_chemistry_mean_token_accuracy": 0.7381541066169739,
416
+ "eval_chemistry_num_tokens": 5850176.0,
417
+ "eval_chemistry_runtime": 60.1537,
418
+ "eval_chemistry_samples_per_second": 8.312,
419
+ "eval_chemistry_steps_per_second": 2.078,
420
+ "step": 300
421
+ },
422
+ {
423
+ "epoch": 0.48,
424
+ "eval_math_entropy": 0.9393695564270019,
425
+ "eval_math_loss": 1.2016712427139282,
426
+ "eval_math_mean_token_accuracy": 0.7277262687683106,
427
+ "eval_math_num_tokens": 5850176.0,
428
+ "eval_math_runtime": 61.6699,
429
+ "eval_math_samples_per_second": 8.108,
430
+ "eval_math_steps_per_second": 2.027,
431
+ "step": 300
432
+ },
433
+ {
434
+ "epoch": 0.48,
435
+ "eval_cyber_entropy": 2.514738072156906,
436
+ "eval_cyber_loss": 2.9828875064849854,
437
+ "eval_cyber_mean_token_accuracy": 0.47654601037502287,
438
+ "eval_cyber_num_tokens": 5850176.0,
439
+ "eval_cyber_runtime": 58.8231,
440
+ "eval_cyber_samples_per_second": 6.749,
441
+ "eval_cyber_steps_per_second": 1.7,
442
+ "step": 300
443
+ },
444
+ {
445
+ "entropy": 1.3836464550346137,
446
+ "epoch": 0.496,
447
+ "grad_norm": 17.875,
448
+ "learning_rate": 1.9933333333333334e-05,
449
+ "loss": 22.405,
450
+ "mean_token_accuracy": 0.6713070213794708,
451
+ "num_tokens": 6046503.0,
452
+ "step": 310
453
+ },
454
+ {
455
+ "entropy": 1.4630893386900425,
456
+ "epoch": 0.512,
457
+ "grad_norm": 19.25,
458
+ "learning_rate": 1.985925925925926e-05,
459
+ "loss": 23.4697,
460
+ "mean_token_accuracy": 0.6617988537997007,
461
+ "num_tokens": 6240456.0,
462
+ "step": 320
463
+ },
464
+ {
465
+ "entropy": 1.448515933379531,
466
+ "epoch": 0.528,
467
+ "grad_norm": 18.75,
468
+ "learning_rate": 1.9785185185185187e-05,
469
+ "loss": 23.3355,
470
+ "mean_token_accuracy": 0.6626260321587324,
471
+ "num_tokens": 6430555.0,
472
+ "step": 330
473
+ },
474
+ {
475
+ "entropy": 1.44985730946064,
476
+ "epoch": 0.544,
477
+ "grad_norm": 18.625,
478
+ "learning_rate": 1.971111111111111e-05,
479
+ "loss": 23.3133,
480
+ "mean_token_accuracy": 0.665154778957367,
481
+ "num_tokens": 6626006.0,
482
+ "step": 340
483
+ },
484
+ {
485
+ "entropy": 1.4254157304763795,
486
+ "epoch": 0.56,
487
+ "grad_norm": 19.125,
488
+ "learning_rate": 1.963703703703704e-05,
489
+ "loss": 23.1032,
490
+ "mean_token_accuracy": 0.6651736661791802,
491
+ "num_tokens": 6820754.0,
492
+ "step": 350
493
+ },
494
+ {
495
+ "entropy": 1.4192099779844285,
496
+ "epoch": 0.576,
497
+ "grad_norm": 18.875,
498
+ "learning_rate": 1.9562962962962964e-05,
499
+ "loss": 22.7221,
500
+ "mean_token_accuracy": 0.6693361282348633,
501
+ "num_tokens": 7021844.0,
502
+ "step": 360
503
+ },
504
+ {
505
+ "entropy": 1.4211084496229887,
506
+ "epoch": 0.592,
507
+ "grad_norm": 19.0,
508
+ "learning_rate": 1.948888888888889e-05,
509
+ "loss": 23.0159,
510
+ "mean_token_accuracy": 0.6660285223275423,
511
+ "num_tokens": 7213951.0,
512
+ "step": 370
513
+ },
514
+ {
515
+ "entropy": 1.4022404573857785,
516
+ "epoch": 0.608,
517
+ "grad_norm": 18.625,
518
+ "learning_rate": 1.9414814814814817e-05,
519
+ "loss": 22.5685,
520
+ "mean_token_accuracy": 0.6711618881672621,
521
+ "num_tokens": 7416773.0,
522
+ "step": 380
523
+ },
524
+ {
525
+ "entropy": 1.4124153569340705,
526
+ "epoch": 0.624,
527
+ "grad_norm": 18.625,
528
+ "learning_rate": 1.9340740740740743e-05,
529
+ "loss": 22.8021,
530
+ "mean_token_accuracy": 0.6678782407194376,
531
+ "num_tokens": 7612843.0,
532
+ "step": 390
533
+ },
534
+ {
535
+ "entropy": 1.4238922595977783,
536
+ "epoch": 0.64,
537
+ "grad_norm": 20.375,
538
+ "learning_rate": 1.926666666666667e-05,
539
+ "loss": 22.9558,
540
+ "mean_token_accuracy": 0.6680531721562147,
541
+ "num_tokens": 7801633.0,
542
+ "step": 400
543
+ },
544
+ {
545
+ "epoch": 0.64,
546
+ "eval_biology_entropy": 1.3578908562660217,
547
+ "eval_biology_loss": 1.3675978183746338,
548
+ "eval_biology_mean_token_accuracy": 0.6797989273071289,
549
+ "eval_biology_num_tokens": 7801633.0,
550
+ "eval_biology_runtime": 48.4218,
551
+ "eval_biology_samples_per_second": 10.326,
552
+ "eval_biology_steps_per_second": 2.581,
553
+ "step": 400
554
+ },
555
+ {
556
+ "epoch": 0.64,
557
+ "eval_chemistry_entropy": 1.0446269502639771,
558
+ "eval_chemistry_loss": 1.0433638095855713,
559
+ "eval_chemistry_mean_token_accuracy": 0.7473979935646057,
560
+ "eval_chemistry_num_tokens": 7801633.0,
561
+ "eval_chemistry_runtime": 60.1024,
562
+ "eval_chemistry_samples_per_second": 8.319,
563
+ "eval_chemistry_steps_per_second": 2.08,
564
+ "step": 400
565
+ },
566
+ {
567
+ "epoch": 0.64,
568
+ "eval_math_entropy": 0.8829994263648987,
569
+ "eval_math_loss": 1.168383002281189,
570
+ "eval_math_mean_token_accuracy": 0.7337917404174805,
571
+ "eval_math_num_tokens": 7801633.0,
572
+ "eval_math_runtime": 61.66,
573
+ "eval_math_samples_per_second": 8.109,
574
+ "eval_math_steps_per_second": 2.027,
575
+ "step": 400
576
+ },
577
+ {
578
+ "epoch": 0.64,
579
+ "eval_cyber_entropy": 2.3425097274780273,
580
+ "eval_cyber_loss": 2.8961009979248047,
581
+ "eval_cyber_mean_token_accuracy": 0.48323330402374265,
582
+ "eval_cyber_num_tokens": 7801633.0,
583
+ "eval_cyber_runtime": 58.7869,
584
+ "eval_cyber_samples_per_second": 6.753,
585
+ "eval_cyber_steps_per_second": 1.701,
586
+ "step": 400
587
+ },
588
+ {
589
+ "entropy": 1.4169765036553144,
590
+ "epoch": 0.656,
591
+ "grad_norm": 21.125,
592
+ "learning_rate": 1.9192592592592593e-05,
593
+ "loss": 22.9523,
594
+ "mean_token_accuracy": 0.6662464819848537,
595
+ "num_tokens": 7995843.0,
596
+ "step": 410
597
+ },
598
+ {
599
+ "entropy": 1.43333672657609,
600
+ "epoch": 0.672,
601
+ "grad_norm": 18.625,
602
+ "learning_rate": 1.911851851851852e-05,
603
+ "loss": 22.9483,
604
+ "mean_token_accuracy": 0.667000836879015,
605
+ "num_tokens": 8183103.0,
606
+ "step": 420
607
+ },
608
+ {
609
+ "entropy": 1.4049694981426,
610
+ "epoch": 0.688,
611
+ "grad_norm": 17.75,
612
+ "learning_rate": 1.9044444444444446e-05,
613
+ "loss": 22.7023,
614
+ "mean_token_accuracy": 0.670426407828927,
615
+ "num_tokens": 8385976.0,
616
+ "step": 430
617
+ },
618
+ {
619
+ "entropy": 1.4090089831501245,
620
+ "epoch": 0.704,
621
+ "grad_norm": 17.875,
622
+ "learning_rate": 1.8970370370370372e-05,
623
+ "loss": 22.6381,
624
+ "mean_token_accuracy": 0.6711054440587759,
625
+ "num_tokens": 8578431.0,
626
+ "step": 440
627
+ },
628
+ {
629
+ "entropy": 1.3353864025324582,
630
+ "epoch": 0.72,
631
+ "grad_norm": 17.75,
632
+ "learning_rate": 1.8896296296296295e-05,
633
+ "loss": 21.6293,
634
+ "mean_token_accuracy": 0.681050893291831,
635
+ "num_tokens": 8781342.0,
636
+ "step": 450
637
+ },
638
+ {
639
+ "entropy": 1.3734276317059995,
640
+ "epoch": 0.736,
641
+ "grad_norm": 17.25,
642
+ "learning_rate": 1.8822222222222225e-05,
643
+ "loss": 22.0542,
644
+ "mean_token_accuracy": 0.6750697229057551,
645
+ "num_tokens": 8977918.0,
646
+ "step": 460
647
+ },
648
+ {
649
+ "entropy": 1.3878948390483856,
650
+ "epoch": 0.752,
651
+ "grad_norm": 17.875,
652
+ "learning_rate": 1.874814814814815e-05,
653
+ "loss": 22.4846,
654
+ "mean_token_accuracy": 0.6717681255191564,
655
+ "num_tokens": 9169322.0,
656
+ "step": 470
657
+ },
658
+ {
659
+ "entropy": 1.3424169197678566,
660
+ "epoch": 0.768,
661
+ "grad_norm": 20.125,
662
+ "learning_rate": 1.8674074074074075e-05,
663
+ "loss": 21.6359,
664
+ "mean_token_accuracy": 0.6805981863290071,
665
+ "num_tokens": 9368141.0,
666
+ "step": 480
667
+ },
668
+ {
669
+ "entropy": 1.3768138941377401,
670
+ "epoch": 0.784,
671
+ "grad_norm": 19.75,
672
+ "learning_rate": 1.86e-05,
673
+ "loss": 22.1917,
674
+ "mean_token_accuracy": 0.6750959828495979,
675
+ "num_tokens": 9565236.0,
676
+ "step": 490
677
+ },
678
+ {
679
+ "entropy": 1.3996582824736834,
680
+ "epoch": 0.8,
681
+ "grad_norm": 19.25,
682
+ "learning_rate": 1.8525925925925928e-05,
683
+ "loss": 22.6339,
684
+ "mean_token_accuracy": 0.6715325634926558,
685
+ "num_tokens": 9761227.0,
686
+ "step": 500
687
+ },
688
+ {
689
+ "epoch": 0.8,
690
+ "eval_biology_entropy": 1.3487377295494078,
691
+ "eval_biology_loss": 1.3396238088607788,
692
+ "eval_biology_mean_token_accuracy": 0.6843163523674011,
693
+ "eval_biology_num_tokens": 9761227.0,
694
+ "eval_biology_runtime": 48.4345,
695
+ "eval_biology_samples_per_second": 10.323,
696
+ "eval_biology_steps_per_second": 2.581,
697
+ "step": 500
698
+ },
699
+ {
700
+ "epoch": 0.8,
701
+ "eval_chemistry_entropy": 1.0342083148956298,
702
+ "eval_chemistry_loss": 1.015425682067871,
703
+ "eval_chemistry_mean_token_accuracy": 0.7529125623703002,
704
+ "eval_chemistry_num_tokens": 9761227.0,
705
+ "eval_chemistry_runtime": 59.7072,
706
+ "eval_chemistry_samples_per_second": 8.374,
707
+ "eval_chemistry_steps_per_second": 2.094,
708
+ "step": 500
709
+ },
710
+ {
711
+ "epoch": 0.8,
712
+ "eval_math_entropy": 0.9004156064987182,
713
+ "eval_math_loss": 1.1309455633163452,
714
+ "eval_math_mean_token_accuracy": 0.7394839000701904,
715
+ "eval_math_num_tokens": 9761227.0,
716
+ "eval_math_runtime": 61.1899,
717
+ "eval_math_samples_per_second": 8.171,
718
+ "eval_math_steps_per_second": 2.043,
719
+ "step": 500
720
+ },
721
+ {
722
+ "epoch": 0.8,
723
+ "eval_cyber_entropy": 2.320133538246155,
724
+ "eval_cyber_loss": 2.8678040504455566,
725
+ "eval_cyber_mean_token_accuracy": 0.4835522577166557,
726
+ "eval_cyber_num_tokens": 9761227.0,
727
+ "eval_cyber_runtime": 58.9104,
728
+ "eval_cyber_samples_per_second": 6.739,
729
+ "eval_cyber_steps_per_second": 1.697,
730
+ "step": 500
731
+ },
732
+ {
733
+ "entropy": 1.35246634632349,
734
+ "epoch": 0.816,
735
+ "grad_norm": 20.75,
736
+ "learning_rate": 1.8451851851851855e-05,
737
+ "loss": 21.7978,
738
+ "mean_token_accuracy": 0.6768883503973484,
739
+ "num_tokens": 9958727.0,
740
+ "step": 510
741
+ },
742
+ {
743
+ "entropy": 1.3554580289870501,
744
+ "epoch": 0.832,
745
+ "grad_norm": 17.5,
746
+ "learning_rate": 1.8377777777777778e-05,
747
+ "loss": 21.7881,
748
+ "mean_token_accuracy": 0.6783689457923174,
749
+ "num_tokens": 10155771.0,
750
+ "step": 520
751
+ },
752
+ {
753
+ "entropy": 1.342598542943597,
754
+ "epoch": 0.848,
755
+ "grad_norm": 19.5,
756
+ "learning_rate": 1.8303703703703704e-05,
757
+ "loss": 21.6264,
758
+ "mean_token_accuracy": 0.679912555962801,
759
+ "num_tokens": 10357721.0,
760
+ "step": 530
761
+ },
762
+ {
763
+ "entropy": 1.3717901095747949,
764
+ "epoch": 0.864,
765
+ "grad_norm": 18.375,
766
+ "learning_rate": 1.822962962962963e-05,
767
+ "loss": 22.2423,
768
+ "mean_token_accuracy": 0.6749696064740419,
769
+ "num_tokens": 10552495.0,
770
+ "step": 540
771
+ },
772
+ {
773
+ "entropy": 1.3606136079877615,
774
+ "epoch": 0.88,
775
+ "grad_norm": 21.0,
776
+ "learning_rate": 1.8155555555555557e-05,
777
+ "loss": 21.921,
778
+ "mean_token_accuracy": 0.6776851717382669,
779
+ "num_tokens": 10748749.0,
780
+ "step": 550
781
+ },
782
+ {
783
+ "entropy": 1.3807077493518591,
784
+ "epoch": 0.896,
785
+ "grad_norm": 19.0,
786
+ "learning_rate": 1.8081481481481484e-05,
787
+ "loss": 22.306,
788
+ "mean_token_accuracy": 0.6740046124905348,
789
+ "num_tokens": 10943319.0,
790
+ "step": 560
791
+ },
792
+ {
793
+ "entropy": 1.3541915334761143,
794
+ "epoch": 0.912,
795
+ "grad_norm": 20.5,
796
+ "learning_rate": 1.800740740740741e-05,
797
+ "loss": 21.989,
798
+ "mean_token_accuracy": 0.6777493793517351,
799
+ "num_tokens": 11136935.0,
800
+ "step": 570
801
+ },
802
+ {
803
+ "entropy": 1.3871366024017333,
804
+ "epoch": 0.928,
805
+ "grad_norm": 19.75,
806
+ "learning_rate": 1.7933333333333333e-05,
807
+ "loss": 22.1135,
808
+ "mean_token_accuracy": 0.6744728099554778,
809
+ "num_tokens": 11331098.0,
810
+ "step": 580
811
+ },
812
+ {
813
+ "entropy": 1.3096800699830056,
814
+ "epoch": 0.944,
815
+ "grad_norm": 17.5,
816
+ "learning_rate": 1.785925925925926e-05,
817
+ "loss": 21.2939,
818
+ "mean_token_accuracy": 0.6831198997795582,
819
+ "num_tokens": 11530550.0,
820
+ "step": 590
821
+ },
822
+ {
823
+ "entropy": 1.3506321482360364,
824
+ "epoch": 0.96,
825
+ "grad_norm": 20.875,
826
+ "learning_rate": 1.7785185185185186e-05,
827
+ "loss": 21.8226,
828
+ "mean_token_accuracy": 0.6785171415656805,
829
+ "num_tokens": 11729645.0,
830
+ "step": 600
831
+ },
832
+ {
833
+ "epoch": 0.96,
834
+ "eval_biology_entropy": 1.3259480624198914,
835
+ "eval_biology_loss": 1.3215161561965942,
836
+ "eval_biology_mean_token_accuracy": 0.6879564213752747,
837
+ "eval_biology_num_tokens": 11729645.0,
838
+ "eval_biology_runtime": 48.4324,
839
+ "eval_biology_samples_per_second": 10.324,
840
+ "eval_biology_steps_per_second": 2.581,
841
+ "step": 600
842
+ },
843
+ {
844
+ "epoch": 0.96,
845
+ "eval_chemistry_entropy": 1.02067849111557,
846
+ "eval_chemistry_loss": 0.9999639987945557,
847
+ "eval_chemistry_mean_token_accuracy": 0.7563390741348267,
848
+ "eval_chemistry_num_tokens": 11729645.0,
849
+ "eval_chemistry_runtime": 60.1019,
850
+ "eval_chemistry_samples_per_second": 8.319,
851
+ "eval_chemistry_steps_per_second": 2.08,
852
+ "step": 600
853
+ },
854
+ {
855
+ "epoch": 0.96,
856
+ "eval_math_entropy": 0.895684769153595,
857
+ "eval_math_loss": 1.1203279495239258,
858
+ "eval_math_mean_token_accuracy": 0.7420237326622009,
859
+ "eval_math_num_tokens": 11729645.0,
860
+ "eval_math_runtime": 61.6071,
861
+ "eval_math_samples_per_second": 8.116,
862
+ "eval_math_steps_per_second": 2.029,
863
+ "step": 600
864
+ },
865
+ {
866
+ "epoch": 0.96,
867
+ "eval_cyber_entropy": 2.2757809734344483,
868
+ "eval_cyber_loss": 2.8872666358947754,
869
+ "eval_cyber_mean_token_accuracy": 0.4848860424757004,
870
+ "eval_cyber_num_tokens": 11729645.0,
871
+ "eval_cyber_runtime": 58.7914,
872
+ "eval_cyber_samples_per_second": 6.753,
873
+ "eval_cyber_steps_per_second": 1.701,
874
+ "step": 600
875
+ },
876
+ {
877
+ "entropy": 1.3319925643503665,
878
+ "epoch": 0.976,
879
+ "grad_norm": 18.875,
880
+ "learning_rate": 1.7711111111111113e-05,
881
+ "loss": 21.4623,
882
+ "mean_token_accuracy": 0.6810844399034977,
883
+ "num_tokens": 11924644.0,
884
+ "step": 610
885
+ },
886
+ {
887
+ "entropy": 1.354709730297327,
888
+ "epoch": 0.992,
889
+ "grad_norm": 17.875,
890
+ "learning_rate": 1.763703703703704e-05,
891
+ "loss": 22.0021,
892
+ "mean_token_accuracy": 0.6767893016338349,
893
+ "num_tokens": 12123059.0,
894
+ "step": 620
895
+ },
896
+ {
897
+ "entropy": 1.358278013765812,
898
+ "epoch": 1.008,
899
+ "grad_norm": 17.625,
900
+ "learning_rate": 1.7562962962962962e-05,
901
+ "loss": 21.6771,
902
+ "mean_token_accuracy": 0.6793475031852723,
903
+ "num_tokens": 12319366.0,
904
+ "step": 630
905
+ },
906
+ {
907
+ "entropy": 1.2945445731282235,
908
+ "epoch": 1.024,
909
+ "grad_norm": 19.875,
910
+ "learning_rate": 1.7488888888888892e-05,
911
+ "loss": 20.9621,
912
+ "mean_token_accuracy": 0.6880514930933714,
913
+ "num_tokens": 12524183.0,
914
+ "step": 640
915
+ },
916
+ {
917
+ "entropy": 1.3366071730852127,
918
+ "epoch": 1.04,
919
+ "grad_norm": 19.375,
920
+ "learning_rate": 1.7414814814814815e-05,
921
+ "loss": 21.6169,
922
+ "mean_token_accuracy": 0.6800738513469696,
923
+ "num_tokens": 12718593.0,
924
+ "step": 650
925
+ },
926
+ {
927
+ "entropy": 1.3035322044044733,
928
+ "epoch": 1.056,
929
+ "grad_norm": 18.875,
930
+ "learning_rate": 1.7340740740740742e-05,
931
+ "loss": 21.0086,
932
+ "mean_token_accuracy": 0.6872079599648714,
933
+ "num_tokens": 12917803.0,
934
+ "step": 660
935
+ },
936
+ {
937
+ "entropy": 1.3336386241018772,
938
+ "epoch": 1.072,
939
+ "grad_norm": 20.125,
940
+ "learning_rate": 1.726666666666667e-05,
941
+ "loss": 21.3783,
942
+ "mean_token_accuracy": 0.6830162953585386,
943
+ "num_tokens": 13105826.0,
944
+ "step": 670
945
+ },
946
+ {
947
+ "entropy": 1.3145085237920284,
948
+ "epoch": 1.088,
949
+ "grad_norm": 19.875,
950
+ "learning_rate": 1.7192592592592595e-05,
951
+ "loss": 21.2325,
952
+ "mean_token_accuracy": 0.6849405582994222,
953
+ "num_tokens": 13298619.0,
954
+ "step": 680
955
+ },
956
+ {
957
+ "entropy": 1.3143333733081817,
958
+ "epoch": 1.104,
959
+ "grad_norm": 19.625,
960
+ "learning_rate": 1.711851851851852e-05,
961
+ "loss": 21.2224,
962
+ "mean_token_accuracy": 0.6856034275144338,
963
+ "num_tokens": 13491486.0,
964
+ "step": 690
965
+ },
966
+ {
967
+ "entropy": 1.3378163430839778,
968
+ "epoch": 1.12,
969
+ "grad_norm": 21.125,
970
+ "learning_rate": 1.7044444444444445e-05,
971
+ "loss": 21.5282,
972
+ "mean_token_accuracy": 0.6820732779800892,
973
+ "num_tokens": 13674663.0,
974
+ "step": 700
975
+ },
976
+ {
977
+ "epoch": 1.12,
978
+ "eval_biology_entropy": 1.2879770975112914,
979
+ "eval_biology_loss": 1.3086122274398804,
980
+ "eval_biology_mean_token_accuracy": 0.6899549188613892,
981
+ "eval_biology_num_tokens": 13674663.0,
982
+ "eval_biology_runtime": 48.4315,
983
+ "eval_biology_samples_per_second": 10.324,
984
+ "eval_biology_steps_per_second": 2.581,
985
+ "step": 700
986
+ },
987
+ {
988
+ "epoch": 1.12,
989
+ "eval_chemistry_entropy": 0.9838369536399841,
990
+ "eval_chemistry_loss": 0.9902806878089905,
991
+ "eval_chemistry_mean_token_accuracy": 0.757919919013977,
992
+ "eval_chemistry_num_tokens": 13674663.0,
993
+ "eval_chemistry_runtime": 60.0982,
994
+ "eval_chemistry_samples_per_second": 8.32,
995
+ "eval_chemistry_steps_per_second": 2.08,
996
+ "step": 700
997
+ },
998
+ {
999
+ "epoch": 1.12,
1000
+ "eval_math_entropy": 0.8821793942451477,
1001
+ "eval_math_loss": 1.108689308166504,
1002
+ "eval_math_mean_token_accuracy": 0.7436572351455688,
1003
+ "eval_math_num_tokens": 13674663.0,
1004
+ "eval_math_runtime": 61.6134,
1005
+ "eval_math_samples_per_second": 8.115,
1006
+ "eval_math_steps_per_second": 2.029,
1007
+ "step": 700
1008
+ },
1009
+ {
1010
+ "epoch": 1.12,
1011
+ "eval_cyber_entropy": 2.3264354813098906,
1012
+ "eval_cyber_loss": 2.8054211139678955,
1013
+ "eval_cyber_mean_token_accuracy": 0.49246201038360593,
1014
+ "eval_cyber_num_tokens": 13674663.0,
1015
+ "eval_cyber_runtime": 58.7919,
1016
+ "eval_cyber_samples_per_second": 6.753,
1017
+ "eval_cyber_steps_per_second": 1.701,
1018
+ "step": 700
1019
+ },
1020
+ {
1021
+ "entropy": 1.3018307507038116,
1022
+ "epoch": 1.1360000000000001,
1023
+ "grad_norm": 20.625,
1024
+ "learning_rate": 1.697037037037037e-05,
1025
+ "loss": 21.2057,
1026
+ "mean_token_accuracy": 0.685313081741333,
1027
+ "num_tokens": 13869134.0,
1028
+ "step": 710
1029
+ },
1030
+ {
1031
+ "entropy": 1.2973876625299454,
1032
+ "epoch": 1.152,
1033
+ "grad_norm": 18.0,
1034
+ "learning_rate": 1.6896296296296298e-05,
1035
+ "loss": 20.8346,
1036
+ "mean_token_accuracy": 0.6878448143601418,
1037
+ "num_tokens": 14078365.0,
1038
+ "step": 720
1039
+ },
1040
+ {
1041
+ "entropy": 1.3114512529224158,
1042
+ "epoch": 1.168,
1043
+ "grad_norm": 17.75,
1044
+ "learning_rate": 1.6822222222222224e-05,
1045
+ "loss": 21.086,
1046
+ "mean_token_accuracy": 0.684901175647974,
1047
+ "num_tokens": 14266831.0,
1048
+ "step": 730
1049
+ },
1050
+ {
1051
+ "entropy": 1.2921015951782464,
1052
+ "epoch": 1.184,
1053
+ "grad_norm": 19.875,
1054
+ "learning_rate": 1.6748148148148147e-05,
1055
+ "loss": 21.1084,
1056
+ "mean_token_accuracy": 0.6854452688246966,
1057
+ "num_tokens": 14465660.0,
1058
+ "step": 740
1059
+ },
1060
+ {
1061
+ "entropy": 1.3100581634789705,
1062
+ "epoch": 1.2,
1063
+ "grad_norm": 20.625,
1064
+ "learning_rate": 1.6674074074074077e-05,
1065
+ "loss": 20.9256,
1066
+ "mean_token_accuracy": 0.6867778114974499,
1067
+ "num_tokens": 14653228.0,
1068
+ "step": 750
1069
+ },
1070
+ {
1071
+ "entropy": 1.2675646111369132,
1072
+ "epoch": 1.216,
1073
+ "grad_norm": 19.5,
1074
+ "learning_rate": 1.66e-05,
1075
+ "loss": 20.4105,
1076
+ "mean_token_accuracy": 0.6919024847447872,
1077
+ "num_tokens": 14857782.0,
1078
+ "step": 760
1079
+ },
1080
+ {
1081
+ "entropy": 1.291072677075863,
1082
+ "epoch": 1.232,
1083
+ "grad_norm": 17.875,
1084
+ "learning_rate": 1.6525925925925927e-05,
1085
+ "loss": 20.8356,
1086
+ "mean_token_accuracy": 0.689462835341692,
1087
+ "num_tokens": 15047356.0,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "entropy": 1.296948005259037,
1092
+ "epoch": 1.248,
1093
+ "grad_norm": 19.0,
1094
+ "learning_rate": 1.6451851851851853e-05,
1095
+ "loss": 21.0571,
1096
+ "mean_token_accuracy": 0.6855845946818591,
1097
+ "num_tokens": 15241098.0,
1098
+ "step": 780
1099
+ },
1100
+ {
1101
+ "entropy": 1.3225075900554657,
1102
+ "epoch": 1.264,
1103
+ "grad_norm": 18.25,
1104
+ "learning_rate": 1.637777777777778e-05,
1105
+ "loss": 21.1821,
1106
+ "mean_token_accuracy": 0.6856160722672939,
1107
+ "num_tokens": 15437657.0,
1108
+ "step": 790
1109
+ },
1110
+ {
1111
+ "entropy": 1.3229579757899046,
1112
+ "epoch": 1.28,
1113
+ "grad_norm": 19.75,
1114
+ "learning_rate": 1.6303703703703706e-05,
1115
+ "loss": 21.3985,
1116
+ "mean_token_accuracy": 0.681986927613616,
1117
+ "num_tokens": 15630795.0,
1118
+ "step": 800
1119
+ },
1120
+ {
1121
+ "epoch": 1.28,
1122
+ "eval_biology_entropy": 1.2697611575126648,
1123
+ "eval_biology_loss": 1.299078106880188,
1124
+ "eval_biology_mean_token_accuracy": 0.6915725932121277,
1125
+ "eval_biology_num_tokens": 15630795.0,
1126
+ "eval_biology_runtime": 48.5582,
1127
+ "eval_biology_samples_per_second": 10.297,
1128
+ "eval_biology_steps_per_second": 2.574,
1129
+ "step": 800
1130
+ },
1131
+ {
1132
+ "epoch": 1.28,
1133
+ "eval_chemistry_entropy": 0.9691643033027649,
1134
+ "eval_chemistry_loss": 0.9838607311248779,
1135
+ "eval_chemistry_mean_token_accuracy": 0.7591533994674683,
1136
+ "eval_chemistry_num_tokens": 15630795.0,
1137
+ "eval_chemistry_runtime": 60.2581,
1138
+ "eval_chemistry_samples_per_second": 8.298,
1139
+ "eval_chemistry_steps_per_second": 2.074,
1140
+ "step": 800
1141
+ },
1142
+ {
1143
+ "epoch": 1.28,
1144
+ "eval_math_entropy": 0.863752941608429,
1145
+ "eval_math_loss": 1.1061811447143555,
1146
+ "eval_math_mean_token_accuracy": 0.7452329030036926,
1147
+ "eval_math_num_tokens": 15630795.0,
1148
+ "eval_math_runtime": 61.8563,
1149
+ "eval_math_samples_per_second": 8.083,
1150
+ "eval_math_steps_per_second": 2.021,
1151
+ "step": 800
1152
+ },
1153
+ {
1154
+ "epoch": 1.28,
1155
+ "eval_cyber_entropy": 2.3142928409576418,
1156
+ "eval_cyber_loss": 2.8583574295043945,
1157
+ "eval_cyber_mean_token_accuracy": 0.4875562259554863,
1158
+ "eval_cyber_num_tokens": 15630795.0,
1159
+ "eval_cyber_runtime": 58.9374,
1160
+ "eval_cyber_samples_per_second": 6.736,
1161
+ "eval_cyber_steps_per_second": 1.697,
1162
+ "step": 800
1163
+ },
1164
+ {
1165
+ "entropy": 1.322736281901598,
1166
+ "epoch": 1.296,
1167
+ "grad_norm": 18.375,
1168
+ "learning_rate": 1.622962962962963e-05,
1169
+ "loss": 21.2994,
1170
+ "mean_token_accuracy": 0.6834880676120519,
1171
+ "num_tokens": 15827105.0,
1172
+ "step": 810
1173
+ },
1174
+ {
1175
+ "entropy": 1.3064148597419263,
1176
+ "epoch": 1.312,
1177
+ "grad_norm": 20.375,
1178
+ "learning_rate": 1.6155555555555556e-05,
1179
+ "loss": 21.2908,
1180
+ "mean_token_accuracy": 0.6831366084516048,
1181
+ "num_tokens": 16019645.0,
1182
+ "step": 820
1183
+ },
1184
+ {
1185
+ "entropy": 1.3229932710528374,
1186
+ "epoch": 1.328,
1187
+ "grad_norm": 19.0,
1188
+ "learning_rate": 1.6081481481481482e-05,
1189
+ "loss": 21.2517,
1190
+ "mean_token_accuracy": 0.6835683796554803,
1191
+ "num_tokens": 16221726.0,
1192
+ "step": 830
1193
+ },
1194
+ {
1195
+ "entropy": 1.2925970274955034,
1196
+ "epoch": 1.3439999999999999,
1197
+ "grad_norm": 17.5,
1198
+ "learning_rate": 1.600740740740741e-05,
1199
+ "loss": 20.7376,
1200
+ "mean_token_accuracy": 0.6901313602924347,
1201
+ "num_tokens": 16427594.0,
1202
+ "step": 840
1203
+ },
1204
+ {
1205
+ "entropy": 1.2990364480763674,
1206
+ "epoch": 1.3599999999999999,
1207
+ "grad_norm": 19.875,
1208
+ "learning_rate": 1.5933333333333336e-05,
1209
+ "loss": 21.0638,
1210
+ "mean_token_accuracy": 0.6867326147854328,
1211
+ "num_tokens": 16621605.0,
1212
+ "step": 850
1213
+ },
1214
+ {
1215
+ "entropy": 1.3395817268639802,
1216
+ "epoch": 1.376,
1217
+ "grad_norm": 21.5,
1218
+ "learning_rate": 1.5859259259259262e-05,
1219
+ "loss": 21.6332,
1220
+ "mean_token_accuracy": 0.6804220989346504,
1221
+ "num_tokens": 16813444.0,
1222
+ "step": 860
1223
+ },
1224
+ {
1225
+ "entropy": 1.3104124892503024,
1226
+ "epoch": 1.392,
1227
+ "grad_norm": 18.625,
1228
+ "learning_rate": 1.5785185185185185e-05,
1229
+ "loss": 21.1663,
1230
+ "mean_token_accuracy": 0.6856168929487467,
1231
+ "num_tokens": 17006509.0,
1232
+ "step": 870
1233
+ },
1234
+ {
1235
+ "entropy": 1.3050751704722643,
1236
+ "epoch": 1.408,
1237
+ "grad_norm": 16.625,
1238
+ "learning_rate": 1.571111111111111e-05,
1239
+ "loss": 20.8645,
1240
+ "mean_token_accuracy": 0.6886320147663355,
1241
+ "num_tokens": 17197870.0,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "entropy": 1.2993885681033135,
1246
+ "epoch": 1.424,
1247
+ "grad_norm": 19.625,
1248
+ "learning_rate": 1.5637037037037038e-05,
1249
+ "loss": 21.0292,
1250
+ "mean_token_accuracy": 0.6871409460902214,
1251
+ "num_tokens": 17394390.0,
1252
+ "step": 890
1253
+ },
1254
+ {
1255
+ "entropy": 1.2986160211265088,
1256
+ "epoch": 1.44,
1257
+ "grad_norm": 22.75,
1258
+ "learning_rate": 1.5562962962962965e-05,
1259
+ "loss": 21.064,
1260
+ "mean_token_accuracy": 0.6858013048768044,
1261
+ "num_tokens": 17587777.0,
1262
+ "step": 900
1263
+ },
1264
+ {
1265
+ "epoch": 1.44,
1266
+ "eval_biology_entropy": 1.2633282842636109,
1267
+ "eval_biology_loss": 1.2911479473114014,
1268
+ "eval_biology_mean_token_accuracy": 0.6930723128318786,
1269
+ "eval_biology_num_tokens": 17587777.0,
1270
+ "eval_biology_runtime": 48.5747,
1271
+ "eval_biology_samples_per_second": 10.293,
1272
+ "eval_biology_steps_per_second": 2.573,
1273
+ "step": 900
1274
+ },
1275
+ {
1276
+ "epoch": 1.44,
1277
+ "eval_chemistry_entropy": 0.9724984226226807,
1278
+ "eval_chemistry_loss": 0.9782924056053162,
1279
+ "eval_chemistry_mean_token_accuracy": 0.7604495301246643,
1280
+ "eval_chemistry_num_tokens": 17587777.0,
1281
+ "eval_chemistry_runtime": 60.2279,
1282
+ "eval_chemistry_samples_per_second": 8.302,
1283
+ "eval_chemistry_steps_per_second": 2.075,
1284
+ "step": 900
1285
+ },
1286
+ {
1287
+ "epoch": 1.44,
1288
+ "eval_math_entropy": 0.878196352481842,
1289
+ "eval_math_loss": 1.0984141826629639,
1290
+ "eval_math_mean_token_accuracy": 0.7456621165275574,
1291
+ "eval_math_num_tokens": 17587777.0,
1292
+ "eval_math_runtime": 61.7802,
1293
+ "eval_math_samples_per_second": 8.093,
1294
+ "eval_math_steps_per_second": 2.023,
1295
+ "step": 900
1296
+ },
1297
+ {
1298
+ "epoch": 1.44,
1299
+ "eval_cyber_entropy": 2.317098902463913,
1300
+ "eval_cyber_loss": 2.8237366676330566,
1301
+ "eval_cyber_mean_token_accuracy": 0.49122542947530745,
1302
+ "eval_cyber_num_tokens": 17587777.0,
1303
+ "eval_cyber_runtime": 58.9196,
1304
+ "eval_cyber_samples_per_second": 6.738,
1305
+ "eval_cyber_steps_per_second": 1.697,
1306
+ "step": 900
1307
+ },
1308
+ {
1309
+ "entropy": 1.2987288374453783,
1310
+ "epoch": 1.456,
1311
+ "grad_norm": 18.375,
1312
+ "learning_rate": 1.548888888888889e-05,
1313
+ "loss": 20.8783,
1314
+ "mean_token_accuracy": 0.6883518882095814,
1315
+ "num_tokens": 17788456.0,
1316
+ "step": 910
1317
+ },
1318
+ {
1319
+ "entropy": 1.292324574291706,
1320
+ "epoch": 1.472,
1321
+ "grad_norm": 17.375,
1322
+ "learning_rate": 1.5414814814814814e-05,
1323
+ "loss": 20.8447,
1324
+ "mean_token_accuracy": 0.6893134098500013,
1325
+ "num_tokens": 17984063.0,
1326
+ "step": 920
1327
+ },
1328
+ {
1329
+ "entropy": 1.2839893605560064,
1330
+ "epoch": 1.488,
1331
+ "grad_norm": 19.125,
1332
+ "learning_rate": 1.5340740740740744e-05,
1333
+ "loss": 20.8773,
1334
+ "mean_token_accuracy": 0.6899567134678364,
1335
+ "num_tokens": 18175640.0,
1336
+ "step": 930
1337
+ },
1338
+ {
1339
+ "entropy": 1.3172472290694714,
1340
+ "epoch": 1.504,
1341
+ "grad_norm": 19.25,
1342
+ "learning_rate": 1.5266666666666667e-05,
1343
+ "loss": 21.141,
1344
+ "mean_token_accuracy": 0.687034023180604,
1345
+ "num_tokens": 18367857.0,
1346
+ "step": 940
1347
+ },
1348
+ {
1349
+ "entropy": 1.288551079109311,
1350
+ "epoch": 1.52,
1351
+ "grad_norm": 18.375,
1352
+ "learning_rate": 1.5192592592592594e-05,
1353
+ "loss": 21.025,
1354
+ "mean_token_accuracy": 0.6869576655328273,
1355
+ "num_tokens": 18569146.0,
1356
+ "step": 950
1357
+ },
1358
+ {
1359
+ "entropy": 1.3221289869397879,
1360
+ "epoch": 1.536,
1361
+ "grad_norm": 19.75,
1362
+ "learning_rate": 1.5118518518518519e-05,
1363
+ "loss": 21.2,
1364
+ "mean_token_accuracy": 0.6844527788460255,
1365
+ "num_tokens": 18755079.0,
1366
+ "step": 960
1367
+ },
1368
+ {
1369
+ "entropy": 1.2898046210408212,
1370
+ "epoch": 1.552,
1371
+ "grad_norm": 20.375,
1372
+ "learning_rate": 1.5044444444444445e-05,
1373
+ "loss": 20.8166,
1374
+ "mean_token_accuracy": 0.690209424123168,
1375
+ "num_tokens": 18956248.0,
1376
+ "step": 970
1377
+ },
1378
+ {
1379
+ "entropy": 1.280243619531393,
1380
+ "epoch": 1.568,
1381
+ "grad_norm": 19.875,
1382
+ "learning_rate": 1.497037037037037e-05,
1383
+ "loss": 20.6218,
1384
+ "mean_token_accuracy": 0.6917802631855011,
1385
+ "num_tokens": 19150315.0,
1386
+ "step": 980
1387
+ },
1388
+ {
1389
+ "entropy": 1.2912988126277924,
1390
+ "epoch": 1.584,
1391
+ "grad_norm": 18.875,
1392
+ "learning_rate": 1.4896296296296298e-05,
1393
+ "loss": 20.7886,
1394
+ "mean_token_accuracy": 0.6892261151224375,
1395
+ "num_tokens": 19344260.0,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "entropy": 1.3189360287040472,
1400
+ "epoch": 1.6,
1401
+ "grad_norm": 19.0,
1402
+ "learning_rate": 1.4822222222222225e-05,
1403
+ "loss": 21.2912,
1404
+ "mean_token_accuracy": 0.6828928265720606,
1405
+ "num_tokens": 19532552.0,
1406
+ "step": 1000
1407
+ },
1408
+ {
1409
+ "epoch": 1.6,
1410
+ "eval_biology_entropy": 1.263430886745453,
1411
+ "eval_biology_loss": 1.2847552299499512,
1412
+ "eval_biology_mean_token_accuracy": 0.6937462825775147,
1413
+ "eval_biology_num_tokens": 19532552.0,
1414
+ "eval_biology_runtime": 48.616,
1415
+ "eval_biology_samples_per_second": 10.285,
1416
+ "eval_biology_steps_per_second": 2.571,
1417
+ "step": 1000
1418
+ },
1419
+ {
1420
+ "epoch": 1.6,
1421
+ "eval_chemistry_entropy": 0.9613517146110535,
1422
+ "eval_chemistry_loss": 0.9752106070518494,
1423
+ "eval_chemistry_mean_token_accuracy": 0.7608753662109375,
1424
+ "eval_chemistry_num_tokens": 19532552.0,
1425
+ "eval_chemistry_runtime": 60.0131,
1426
+ "eval_chemistry_samples_per_second": 8.332,
1427
+ "eval_chemistry_steps_per_second": 2.083,
1428
+ "step": 1000
1429
+ },
1430
+ {
1431
+ "epoch": 1.6,
1432
+ "eval_math_entropy": 0.8507145104408265,
1433
+ "eval_math_loss": 1.0975427627563477,
1434
+ "eval_math_mean_token_accuracy": 0.7471181530952453,
1435
+ "eval_math_num_tokens": 19532552.0,
1436
+ "eval_math_runtime": 61.5437,
1437
+ "eval_math_samples_per_second": 8.124,
1438
+ "eval_math_steps_per_second": 2.031,
1439
+ "step": 1000
1440
+ },
1441
+ {
1442
+ "epoch": 1.6,
1443
+ "eval_cyber_entropy": 2.262136605978012,
1444
+ "eval_cyber_loss": 2.8179190158843994,
1445
+ "eval_cyber_mean_token_accuracy": 0.4946762239933014,
1446
+ "eval_cyber_num_tokens": 19532552.0,
1447
+ "eval_cyber_runtime": 58.9033,
1448
+ "eval_cyber_samples_per_second": 6.74,
1449
+ "eval_cyber_steps_per_second": 1.698,
1450
+ "step": 1000
1451
+ },
1452
+ {
1453
+ "entropy": 1.2780998092144729,
1454
+ "epoch": 1.616,
1455
+ "grad_norm": 18.25,
1456
+ "learning_rate": 1.474814814814815e-05,
1457
+ "loss": 20.7217,
1458
+ "mean_token_accuracy": 0.690490011498332,
1459
+ "num_tokens": 19732719.0,
1460
+ "step": 1010
1461
+ },
1462
+ {
1463
+ "entropy": 1.322769870609045,
1464
+ "epoch": 1.6320000000000001,
1465
+ "grad_norm": 21.375,
1466
+ "learning_rate": 1.4674074074074076e-05,
1467
+ "loss": 21.3196,
1468
+ "mean_token_accuracy": 0.6848675295710563,
1469
+ "num_tokens": 19926830.0,
1470
+ "step": 1020
1471
+ },
1472
+ {
1473
+ "entropy": 1.299769001826644,
1474
+ "epoch": 1.6480000000000001,
1475
+ "grad_norm": 17.0,
1476
+ "learning_rate": 1.46e-05,
1477
+ "loss": 21.0578,
1478
+ "mean_token_accuracy": 0.685557559132576,
1479
+ "num_tokens": 20118800.0,
1480
+ "step": 1030
1481
+ },
1482
+ {
1483
+ "entropy": 1.2741328407078982,
1484
+ "epoch": 1.6640000000000001,
1485
+ "grad_norm": 18.0,
1486
+ "learning_rate": 1.4525925925925927e-05,
1487
+ "loss": 20.5087,
1488
+ "mean_token_accuracy": 0.6927142184227705,
1489
+ "num_tokens": 20320511.0,
1490
+ "step": 1040
1491
+ },
1492
+ {
1493
+ "entropy": 1.2958957750350237,
1494
+ "epoch": 1.6800000000000002,
1495
+ "grad_norm": 20.875,
1496
+ "learning_rate": 1.4451851851851852e-05,
1497
+ "loss": 20.8825,
1498
+ "mean_token_accuracy": 0.688383724913001,
1499
+ "num_tokens": 20513393.0,
1500
+ "step": 1050
1501
+ },
1502
+ {
1503
+ "entropy": 1.2880746740847826,
1504
+ "epoch": 1.696,
1505
+ "grad_norm": 19.875,
1506
+ "learning_rate": 1.4377777777777779e-05,
1507
+ "loss": 20.8343,
1508
+ "mean_token_accuracy": 0.6903862621635198,
1509
+ "num_tokens": 20707237.0,
1510
+ "step": 1060
1511
+ },
1512
+ {
1513
+ "entropy": 1.2862251173704862,
1514
+ "epoch": 1.712,
1515
+ "grad_norm": 20.0,
1516
+ "learning_rate": 1.4303703703703703e-05,
1517
+ "loss": 20.8599,
1518
+ "mean_token_accuracy": 0.6900554195046424,
1519
+ "num_tokens": 20910419.0,
1520
+ "step": 1070
1521
+ },
1522
+ {
1523
+ "entropy": 1.3037602450698613,
1524
+ "epoch": 1.728,
1525
+ "grad_norm": 18.75,
1526
+ "learning_rate": 1.4229629629629632e-05,
1527
+ "loss": 21.0116,
1528
+ "mean_token_accuracy": 0.6866883236914874,
1529
+ "num_tokens": 21107498.0,
1530
+ "step": 1080
1531
+ },
1532
+ {
1533
+ "entropy": 1.268125656619668,
1534
+ "epoch": 1.744,
1535
+ "grad_norm": 19.5,
1536
+ "learning_rate": 1.4155555555555556e-05,
1537
+ "loss": 20.3869,
1538
+ "mean_token_accuracy": 0.6919578000903129,
1539
+ "num_tokens": 21303955.0,
1540
+ "step": 1090
1541
+ },
1542
+ {
1543
+ "entropy": 1.277770683541894,
1544
+ "epoch": 1.76,
1545
+ "grad_norm": 20.5,
1546
+ "learning_rate": 1.4081481481481483e-05,
1547
+ "loss": 20.7842,
1548
+ "mean_token_accuracy": 0.6904734123498202,
1549
+ "num_tokens": 21499572.0,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 1.76,
1554
+ "eval_biology_entropy": 1.2706038541793823,
1555
+ "eval_biology_loss": 1.2793399095535278,
1556
+ "eval_biology_mean_token_accuracy": 0.6948989968299866,
1557
+ "eval_biology_num_tokens": 21499572.0,
1558
+ "eval_biology_runtime": 48.5261,
1559
+ "eval_biology_samples_per_second": 10.304,
1560
+ "eval_biology_steps_per_second": 2.576,
1561
+ "step": 1100
1562
+ },
1563
+ {
1564
+ "epoch": 1.76,
1565
+ "eval_chemistry_entropy": 0.9732482833862305,
1566
+ "eval_chemistry_loss": 0.9693844318389893,
1567
+ "eval_chemistry_mean_token_accuracy": 0.7617205648422242,
1568
+ "eval_chemistry_num_tokens": 21499572.0,
1569
+ "eval_chemistry_runtime": 60.2906,
1570
+ "eval_chemistry_samples_per_second": 8.293,
1571
+ "eval_chemistry_steps_per_second": 2.073,
1572
+ "step": 1100
1573
+ },
1574
+ {
1575
+ "epoch": 1.76,
1576
+ "eval_math_entropy": 0.8668387780189514,
1577
+ "eval_math_loss": 1.0918337106704712,
1578
+ "eval_math_mean_token_accuracy": 0.7475174107551574,
1579
+ "eval_math_num_tokens": 21499572.0,
1580
+ "eval_math_runtime": 61.8179,
1581
+ "eval_math_samples_per_second": 8.088,
1582
+ "eval_math_steps_per_second": 2.022,
1583
+ "step": 1100
1584
+ },
1585
+ {
1586
+ "epoch": 1.76,
1587
+ "eval_cyber_entropy": 2.2727494060993196,
1588
+ "eval_cyber_loss": 2.8152451515197754,
1589
+ "eval_cyber_mean_token_accuracy": 0.496353834271431,
1590
+ "eval_cyber_num_tokens": 21499572.0,
1591
+ "eval_cyber_runtime": 58.9408,
1592
+ "eval_cyber_samples_per_second": 6.736,
1593
+ "eval_cyber_steps_per_second": 1.697,
1594
+ "step": 1100
1595
+ },
1596
+ {
1597
+ "entropy": 1.2966676279902458,
1598
+ "epoch": 1.776,
1599
+ "grad_norm": 20.125,
1600
+ "learning_rate": 1.400740740740741e-05,
1601
+ "loss": 20.8811,
1602
+ "mean_token_accuracy": 0.6886047091335058,
1603
+ "num_tokens": 21692804.0,
1604
+ "step": 1110
1605
+ },
1606
+ {
1607
+ "entropy": 1.2577556148171425,
1608
+ "epoch": 1.792,
1609
+ "grad_norm": 20.375,
1610
+ "learning_rate": 1.3933333333333334e-05,
1611
+ "loss": 20.3406,
1612
+ "mean_token_accuracy": 0.6924801960587501,
1613
+ "num_tokens": 21894218.0,
1614
+ "step": 1120
1615
+ },
1616
+ {
1617
+ "entropy": 1.2880206000059844,
1618
+ "epoch": 1.808,
1619
+ "grad_norm": 21.5,
1620
+ "learning_rate": 1.385925925925926e-05,
1621
+ "loss": 20.7402,
1622
+ "mean_token_accuracy": 0.6916475351899862,
1623
+ "num_tokens": 22082522.0,
1624
+ "step": 1130
1625
+ },
1626
+ {
1627
+ "entropy": 1.31001619361341,
1628
+ "epoch": 1.8239999999999998,
1629
+ "grad_norm": 19.875,
1630
+ "learning_rate": 1.3785185185185186e-05,
1631
+ "loss": 21.1628,
1632
+ "mean_token_accuracy": 0.6851416889578104,
1633
+ "num_tokens": 22278933.0,
1634
+ "step": 1140
1635
+ },
1636
+ {
1637
+ "entropy": 1.3205205973237752,
1638
+ "epoch": 1.8399999999999999,
1639
+ "grad_norm": 21.0,
1640
+ "learning_rate": 1.3711111111111112e-05,
1641
+ "loss": 21.3264,
1642
+ "mean_token_accuracy": 0.6827714014798403,
1643
+ "num_tokens": 22473801.0,
1644
+ "step": 1150
1645
+ },
1646
+ {
1647
+ "entropy": 1.2739585246890783,
1648
+ "epoch": 1.8559999999999999,
1649
+ "grad_norm": 20.0,
1650
+ "learning_rate": 1.3637037037037037e-05,
1651
+ "loss": 20.5874,
1652
+ "mean_token_accuracy": 0.6894957963377237,
1653
+ "num_tokens": 22677853.0,
1654
+ "step": 1160
1655
+ },
1656
+ {
1657
+ "entropy": 1.2596167303621768,
1658
+ "epoch": 1.8719999999999999,
1659
+ "grad_norm": 18.5,
1660
+ "learning_rate": 1.3562962962962965e-05,
1661
+ "loss": 20.2631,
1662
+ "mean_token_accuracy": 0.69476717710495,
1663
+ "num_tokens": 22874965.0,
1664
+ "step": 1170
1665
+ },
1666
+ {
1667
+ "entropy": 1.3122890371829272,
1668
+ "epoch": 1.888,
1669
+ "grad_norm": 21.875,
1670
+ "learning_rate": 1.3488888888888888e-05,
1671
+ "loss": 21.2153,
1672
+ "mean_token_accuracy": 0.6857765819877386,
1673
+ "num_tokens": 23068892.0,
1674
+ "step": 1180
1675
+ },
1676
+ {
1677
+ "entropy": 1.3009861066937447,
1678
+ "epoch": 1.904,
1679
+ "grad_norm": 19.75,
1680
+ "learning_rate": 1.3414814814814817e-05,
1681
+ "loss": 21.0824,
1682
+ "mean_token_accuracy": 0.6862182442098856,
1683
+ "num_tokens": 23263827.0,
1684
+ "step": 1190
1685
+ },
1686
+ {
1687
+ "entropy": 1.3086694907397032,
1688
+ "epoch": 1.92,
1689
+ "grad_norm": 19.5,
1690
+ "learning_rate": 1.3340740740740741e-05,
1691
+ "loss": 21.0366,
1692
+ "mean_token_accuracy": 0.6864289414137602,
1693
+ "num_tokens": 23463627.0,
1694
+ "step": 1200
1695
+ },
1696
+ {
1697
+ "epoch": 1.92,
1698
+ "eval_biology_entropy": 1.2436800727844237,
1699
+ "eval_biology_loss": 1.2742465734481812,
1700
+ "eval_biology_mean_token_accuracy": 0.6959740309715271,
1701
+ "eval_biology_num_tokens": 23463627.0,
1702
+ "eval_biology_runtime": 48.7215,
1703
+ "eval_biology_samples_per_second": 10.262,
1704
+ "eval_biology_steps_per_second": 2.566,
1705
+ "step": 1200
1706
+ },
1707
+ {
1708
+ "epoch": 1.92,
1709
+ "eval_chemistry_entropy": 0.9520102243423462,
1710
+ "eval_chemistry_loss": 0.9654216170310974,
1711
+ "eval_chemistry_mean_token_accuracy": 0.7620544514656067,
1712
+ "eval_chemistry_num_tokens": 23463627.0,
1713
+ "eval_chemistry_runtime": 60.3322,
1714
+ "eval_chemistry_samples_per_second": 8.287,
1715
+ "eval_chemistry_steps_per_second": 2.072,
1716
+ "step": 1200
1717
+ },
1718
+ {
1719
+ "epoch": 1.92,
1720
+ "eval_math_entropy": 0.8506893858909607,
1721
+ "eval_math_loss": 1.089040756225586,
1722
+ "eval_math_mean_token_accuracy": 0.7481771049499512,
1723
+ "eval_math_num_tokens": 23463627.0,
1724
+ "eval_math_runtime": 62.2172,
1725
+ "eval_math_samples_per_second": 8.036,
1726
+ "eval_math_steps_per_second": 2.009,
1727
+ "step": 1200
1728
+ },
1729
+ {
1730
+ "epoch": 1.92,
1731
+ "eval_cyber_entropy": 2.270911729335785,
1732
+ "eval_cyber_loss": 2.815645933151245,
1733
+ "eval_cyber_mean_token_accuracy": 0.49721296519041064,
1734
+ "eval_cyber_num_tokens": 23463627.0,
1735
+ "eval_cyber_runtime": 59.0925,
1736
+ "eval_cyber_samples_per_second": 6.718,
1737
+ "eval_cyber_steps_per_second": 1.692,
1738
+ "step": 1200
1739
+ },
1740
+ {
1741
+ "entropy": 1.258543035387993,
1742
+ "epoch": 1.936,
1743
+ "grad_norm": 20.0,
1744
+ "learning_rate": 1.3266666666666668e-05,
1745
+ "loss": 20.3282,
1746
+ "mean_token_accuracy": 0.6942144498229027,
1747
+ "num_tokens": 23660418.0,
1748
+ "step": 1210
1749
+ },
1750
+ {
1751
+ "entropy": 1.2887091431766748,
1752
+ "epoch": 1.952,
1753
+ "grad_norm": 20.125,
1754
+ "learning_rate": 1.3192592592592594e-05,
1755
+ "loss": 20.7017,
1756
+ "mean_token_accuracy": 0.6914095990359783,
1757
+ "num_tokens": 23858145.0,
1758
+ "step": 1220
1759
+ },
1760
+ {
1761
+ "entropy": 1.2908520311117173,
1762
+ "epoch": 1.968,
1763
+ "grad_norm": 20.375,
1764
+ "learning_rate": 1.311851851851852e-05,
1765
+ "loss": 20.993,
1766
+ "mean_token_accuracy": 0.6882839858531952,
1767
+ "num_tokens": 24053364.0,
1768
+ "step": 1230
1769
+ },
1770
+ {
1771
+ "entropy": 1.2901352025568484,
1772
+ "epoch": 1.984,
1773
+ "grad_norm": 18.5,
1774
+ "learning_rate": 1.3044444444444446e-05,
1775
+ "loss": 20.7839,
1776
+ "mean_token_accuracy": 0.6889098271727562,
1777
+ "num_tokens": 24249465.0,
1778
+ "step": 1240
1779
+ },
1780
+ {
1781
+ "entropy": 1.2941514786332846,
1782
+ "epoch": 2.0,
1783
+ "grad_norm": 20.875,
1784
+ "learning_rate": 1.297037037037037e-05,
1785
+ "loss": 20.8924,
1786
+ "mean_token_accuracy": 0.690009992942214,
1787
+ "num_tokens": 24442582.0,
1788
+ "step": 1250
1789
+ },
1790
+ {
1791
+ "entropy": 1.2745149232447148,
1792
+ "epoch": 2.016,
1793
+ "grad_norm": 21.0,
1794
+ "learning_rate": 1.2896296296296299e-05,
1795
+ "loss": 20.4904,
1796
+ "mean_token_accuracy": 0.6926420453935862,
1797
+ "num_tokens": 24632353.0,
1798
+ "step": 1260
1799
+ },
1800
+ {
1801
+ "entropy": 1.2860474549233913,
1802
+ "epoch": 2.032,
1803
+ "grad_norm": 20.375,
1804
+ "learning_rate": 1.2822222222222222e-05,
1805
+ "loss": 20.7466,
1806
+ "mean_token_accuracy": 0.6908769007772207,
1807
+ "num_tokens": 24822715.0,
1808
+ "step": 1270
1809
+ },
1810
+ {
1811
+ "entropy": 1.260865818709135,
1812
+ "epoch": 2.048,
1813
+ "grad_norm": 22.5,
1814
+ "learning_rate": 1.274814814814815e-05,
1815
+ "loss": 20.2952,
1816
+ "mean_token_accuracy": 0.6937311202287674,
1817
+ "num_tokens": 25023118.0,
1818
+ "step": 1280
1819
+ },
1820
+ {
1821
+ "entropy": 1.2737747061997653,
1822
+ "epoch": 2.064,
1823
+ "grad_norm": 20.875,
1824
+ "learning_rate": 1.2674074074074075e-05,
1825
+ "loss": 20.6082,
1826
+ "mean_token_accuracy": 0.6917234756052494,
1827
+ "num_tokens": 25217414.0,
1828
+ "step": 1290
1829
+ },
1830
+ {
1831
+ "entropy": 1.2580339554697275,
1832
+ "epoch": 2.08,
1833
+ "grad_norm": 21.375,
1834
+ "learning_rate": 1.2600000000000001e-05,
1835
+ "loss": 20.3045,
1836
+ "mean_token_accuracy": 0.6938596669584512,
1837
+ "num_tokens": 25408961.0,
1838
+ "step": 1300
1839
+ },
1840
+ {
1841
+ "epoch": 2.08,
1842
+ "eval_biology_entropy": 1.2317639932632447,
1843
+ "eval_biology_loss": 1.272905707359314,
1844
+ "eval_biology_mean_token_accuracy": 0.6965393185615539,
1845
+ "eval_biology_num_tokens": 25408961.0,
1846
+ "eval_biology_runtime": 48.5992,
1847
+ "eval_biology_samples_per_second": 10.288,
1848
+ "eval_biology_steps_per_second": 2.572,
1849
+ "step": 1300
1850
+ },
1851
+ {
1852
+ "epoch": 2.08,
1853
+ "eval_chemistry_entropy": 0.9386696972846985,
1854
+ "eval_chemistry_loss": 0.9654728770256042,
1855
+ "eval_chemistry_mean_token_accuracy": 0.7624025421142578,
1856
+ "eval_chemistry_num_tokens": 25408961.0,
1857
+ "eval_chemistry_runtime": 60.2391,
1858
+ "eval_chemistry_samples_per_second": 8.3,
1859
+ "eval_chemistry_steps_per_second": 2.075,
1860
+ "step": 1300
1861
+ },
1862
+ {
1863
+ "epoch": 2.08,
1864
+ "eval_math_entropy": 0.8434405541419983,
1865
+ "eval_math_loss": 1.0893279314041138,
1866
+ "eval_math_mean_token_accuracy": 0.7487460594177247,
1867
+ "eval_math_num_tokens": 25408961.0,
1868
+ "eval_math_runtime": 61.7496,
1869
+ "eval_math_samples_per_second": 8.097,
1870
+ "eval_math_steps_per_second": 2.024,
1871
+ "step": 1300
1872
+ },
1873
+ {
1874
+ "epoch": 2.08,
1875
+ "eval_cyber_entropy": 2.222271492481232,
1876
+ "eval_cyber_loss": 2.839128017425537,
1877
+ "eval_cyber_mean_token_accuracy": 0.49531962275505065,
1878
+ "eval_cyber_num_tokens": 25408961.0,
1879
+ "eval_cyber_runtime": 58.8697,
1880
+ "eval_cyber_samples_per_second": 6.744,
1881
+ "eval_cyber_steps_per_second": 1.699,
1882
+ "step": 1300
1883
+ },
1884
+ {
1885
+ "entropy": 1.2651004053652286,
1886
+ "epoch": 2.096,
1887
+ "grad_norm": 21.75,
1888
+ "learning_rate": 1.2525925925925928e-05,
1889
+ "loss": 20.4274,
1890
+ "mean_token_accuracy": 0.6951500400900841,
1891
+ "num_tokens": 25600511.0,
1892
+ "step": 1310
1893
+ },
1894
+ {
1895
+ "entropy": 1.244707338884473,
1896
+ "epoch": 2.112,
1897
+ "grad_norm": 19.75,
1898
+ "learning_rate": 1.2451851851851853e-05,
1899
+ "loss": 19.9294,
1900
+ "mean_token_accuracy": 0.6981283750385046,
1901
+ "num_tokens": 25796565.0,
1902
+ "step": 1320
1903
+ },
1904
+ {
1905
+ "entropy": 1.2410834614187478,
1906
+ "epoch": 2.128,
1907
+ "grad_norm": 20.875,
1908
+ "learning_rate": 1.237777777777778e-05,
1909
+ "loss": 20.0032,
1910
+ "mean_token_accuracy": 0.6971660844981671,
1911
+ "num_tokens": 25991156.0,
1912
+ "step": 1330
1913
+ },
1914
+ {
1915
+ "entropy": 1.24519634693861,
1916
+ "epoch": 2.144,
1917
+ "grad_norm": 23.0,
1918
+ "learning_rate": 1.2303703703703704e-05,
1919
+ "loss": 20.1622,
1920
+ "mean_token_accuracy": 0.6968274012207984,
1921
+ "num_tokens": 26193237.0,
1922
+ "step": 1340
1923
+ },
1924
+ {
1925
+ "entropy": 1.2762737173587084,
1926
+ "epoch": 2.16,
1927
+ "grad_norm": 20.875,
1928
+ "learning_rate": 1.222962962962963e-05,
1929
+ "loss": 20.5746,
1930
+ "mean_token_accuracy": 0.6938918314874172,
1931
+ "num_tokens": 26387993.0,
1932
+ "step": 1350
1933
+ },
1934
+ {
1935
+ "entropy": 1.2816904935985804,
1936
+ "epoch": 2.176,
1937
+ "grad_norm": 20.625,
1938
+ "learning_rate": 1.2155555555555555e-05,
1939
+ "loss": 20.694,
1940
+ "mean_token_accuracy": 0.6903725199401378,
1941
+ "num_tokens": 26585269.0,
1942
+ "step": 1360
1943
+ },
1944
+ {
1945
+ "entropy": 1.23365030400455,
1946
+ "epoch": 2.192,
1947
+ "grad_norm": 18.25,
1948
+ "learning_rate": 1.2081481481481484e-05,
1949
+ "loss": 19.9231,
1950
+ "mean_token_accuracy": 0.7000154417008162,
1951
+ "num_tokens": 26776318.0,
1952
+ "step": 1370
1953
+ },
1954
+ {
1955
+ "entropy": 1.270122117176652,
1956
+ "epoch": 2.208,
1957
+ "grad_norm": 19.875,
1958
+ "learning_rate": 1.2007407407407408e-05,
1959
+ "loss": 20.347,
1960
+ "mean_token_accuracy": 0.6952163562178612,
1961
+ "num_tokens": 26974420.0,
1962
+ "step": 1380
1963
+ },
1964
+ {
1965
+ "entropy": 1.277249312400818,
1966
+ "epoch": 2.224,
1967
+ "grad_norm": 20.0,
1968
+ "learning_rate": 1.1933333333333335e-05,
1969
+ "loss": 20.8068,
1970
+ "mean_token_accuracy": 0.6900401685386897,
1971
+ "num_tokens": 27167577.0,
1972
+ "step": 1390
1973
+ },
1974
+ {
1975
+ "entropy": 1.2444671079516412,
1976
+ "epoch": 2.24,
1977
+ "grad_norm": 19.625,
1978
+ "learning_rate": 1.185925925925926e-05,
1979
+ "loss": 19.8147,
1980
+ "mean_token_accuracy": 0.699263571202755,
1981
+ "num_tokens": 27364189.0,
1982
+ "step": 1400
1983
+ },
1984
+ {
1985
+ "epoch": 2.24,
1986
+ "eval_biology_entropy": 1.2131686201095582,
1987
+ "eval_biology_loss": 1.2700260877609253,
1988
+ "eval_biology_mean_token_accuracy": 0.696834128856659,
1989
+ "eval_biology_num_tokens": 27364189.0,
1990
+ "eval_biology_runtime": 48.5544,
1991
+ "eval_biology_samples_per_second": 10.298,
1992
+ "eval_biology_steps_per_second": 2.574,
1993
+ "step": 1400
1994
+ },
1995
+ {
1996
+ "epoch": 2.24,
1997
+ "eval_chemistry_entropy": 0.9270022196769714,
1998
+ "eval_chemistry_loss": 0.9647746682167053,
1999
+ "eval_chemistry_mean_token_accuracy": 0.7625881099700927,
2000
+ "eval_chemistry_num_tokens": 27364189.0,
2001
+ "eval_chemistry_runtime": 60.2368,
2002
+ "eval_chemistry_samples_per_second": 8.301,
2003
+ "eval_chemistry_steps_per_second": 2.075,
2004
+ "step": 1400
2005
+ },
2006
+ {
2007
+ "epoch": 2.24,
2008
+ "eval_math_entropy": 0.8296263842582703,
2009
+ "eval_math_loss": 1.0910688638687134,
2010
+ "eval_math_mean_token_accuracy": 0.7489140472412109,
2011
+ "eval_math_num_tokens": 27364189.0,
2012
+ "eval_math_runtime": 61.7781,
2013
+ "eval_math_samples_per_second": 8.093,
2014
+ "eval_math_steps_per_second": 2.023,
2015
+ "step": 1400
2016
+ },
2017
+ {
2018
+ "epoch": 2.24,
2019
+ "eval_cyber_entropy": 2.2271901178359985,
2020
+ "eval_cyber_loss": 2.8551981449127197,
2021
+ "eval_cyber_mean_token_accuracy": 0.4925511845946312,
2022
+ "eval_cyber_num_tokens": 27364189.0,
2023
+ "eval_cyber_runtime": 58.9077,
2024
+ "eval_cyber_samples_per_second": 6.739,
2025
+ "eval_cyber_steps_per_second": 1.698,
2026
+ "step": 1400
2027
+ },
2028
+ {
2029
+ "entropy": 1.224042547494173,
2030
+ "epoch": 2.2560000000000002,
2031
+ "grad_norm": 20.375,
2032
+ "learning_rate": 1.1785185185185186e-05,
2033
+ "loss": 19.8235,
2034
+ "mean_token_accuracy": 0.7005104530602694,
2035
+ "num_tokens": 27557387.0,
2036
+ "step": 1410
2037
+ },
2038
+ {
2039
+ "entropy": 1.2278712920844554,
2040
+ "epoch": 2.2720000000000002,
2041
+ "grad_norm": 20.375,
2042
+ "learning_rate": 1.1711111111111113e-05,
2043
+ "loss": 19.8862,
2044
+ "mean_token_accuracy": 0.6978729262948036,
2045
+ "num_tokens": 27755725.0,
2046
+ "step": 1420
2047
+ },
2048
+ {
2049
+ "entropy": 1.2763480294495821,
2050
+ "epoch": 2.288,
2051
+ "grad_norm": 20.25,
2052
+ "learning_rate": 1.1637037037037037e-05,
2053
+ "loss": 20.5831,
2054
+ "mean_token_accuracy": 0.6889720026403665,
2055
+ "num_tokens": 27950694.0,
2056
+ "step": 1430
2057
+ },
2058
+ {
2059
+ "entropy": 1.2638509940356015,
2060
+ "epoch": 2.304,
2061
+ "grad_norm": 18.75,
2062
+ "learning_rate": 1.1562962962962964e-05,
2063
+ "loss": 20.5489,
2064
+ "mean_token_accuracy": 0.6938521608710289,
2065
+ "num_tokens": 28150719.0,
2066
+ "step": 1440
2067
+ },
2068
+ {
2069
+ "entropy": 1.2345790199935436,
2070
+ "epoch": 2.32,
2071
+ "grad_norm": 18.5,
2072
+ "learning_rate": 1.1488888888888889e-05,
2073
+ "loss": 19.71,
2074
+ "mean_token_accuracy": 0.7009611383080483,
2075
+ "num_tokens": 28348652.0,
2076
+ "step": 1450
2077
+ },
2078
+ {
2079
+ "entropy": 1.2490899875760078,
2080
+ "epoch": 2.336,
2081
+ "grad_norm": 20.25,
2082
+ "learning_rate": 1.1414814814814817e-05,
2083
+ "loss": 20.2585,
2084
+ "mean_token_accuracy": 0.6959190309047699,
2085
+ "num_tokens": 28542945.0,
2086
+ "step": 1460
2087
+ },
2088
+ {
2089
+ "entropy": 1.2702395316213368,
2090
+ "epoch": 2.352,
2091
+ "grad_norm": 20.375,
2092
+ "learning_rate": 1.1340740740740742e-05,
2093
+ "loss": 20.5325,
2094
+ "mean_token_accuracy": 0.6922014791518449,
2095
+ "num_tokens": 28731927.0,
2096
+ "step": 1470
2097
+ },
2098
+ {
2099
+ "entropy": 1.257371624931693,
2100
+ "epoch": 2.368,
2101
+ "grad_norm": 20.375,
2102
+ "learning_rate": 1.1266666666666668e-05,
2103
+ "loss": 20.2317,
2104
+ "mean_token_accuracy": 0.6950671311467886,
2105
+ "num_tokens": 28929298.0,
2106
+ "step": 1480
2107
+ },
2108
+ {
2109
+ "entropy": 1.2586353100836276,
2110
+ "epoch": 2.384,
2111
+ "grad_norm": 22.0,
2112
+ "learning_rate": 1.1192592592592593e-05,
2113
+ "loss": 20.4017,
2114
+ "mean_token_accuracy": 0.6933799501508474,
2115
+ "num_tokens": 29121142.0,
2116
+ "step": 1490
2117
+ },
2118
+ {
2119
+ "entropy": 1.2538471952080728,
2120
+ "epoch": 2.4,
2121
+ "grad_norm": 19.0,
2122
+ "learning_rate": 1.111851851851852e-05,
2123
+ "loss": 20.1414,
2124
+ "mean_token_accuracy": 0.6949946410953999,
2125
+ "num_tokens": 29317919.0,
2126
+ "step": 1500
2127
+ },
2128
+ {
2129
+ "epoch": 2.4,
2130
+ "eval_biology_entropy": 1.2187503614425659,
2131
+ "eval_biology_loss": 1.2677440643310547,
2132
+ "eval_biology_mean_token_accuracy": 0.6970518484115601,
2133
+ "eval_biology_num_tokens": 29317919.0,
2134
+ "eval_biology_runtime": 48.5785,
2135
+ "eval_biology_samples_per_second": 10.293,
2136
+ "eval_biology_steps_per_second": 2.573,
2137
+ "step": 1500
2138
+ },
2139
+ {
2140
+ "epoch": 2.4,
2141
+ "eval_chemistry_entropy": 0.9279513311386108,
2142
+ "eval_chemistry_loss": 0.9621784090995789,
2143
+ "eval_chemistry_mean_token_accuracy": 0.7627601904869079,
2144
+ "eval_chemistry_num_tokens": 29317919.0,
2145
+ "eval_chemistry_runtime": 59.9609,
2146
+ "eval_chemistry_samples_per_second": 8.339,
2147
+ "eval_chemistry_steps_per_second": 2.085,
2148
+ "step": 1500
2149
+ },
2150
+ {
2151
+ "epoch": 2.4,
2152
+ "eval_math_entropy": 0.829472065448761,
2153
+ "eval_math_loss": 1.0883209705352783,
2154
+ "eval_math_mean_token_accuracy": 0.7494171533584595,
2155
+ "eval_math_num_tokens": 29317919.0,
2156
+ "eval_math_runtime": 61.4914,
2157
+ "eval_math_samples_per_second": 8.131,
2158
+ "eval_math_steps_per_second": 2.033,
2159
+ "step": 1500
2160
+ },
2161
+ {
2162
+ "epoch": 2.4,
2163
+ "eval_cyber_entropy": 2.212998685836792,
2164
+ "eval_cyber_loss": 2.869270086288452,
2165
+ "eval_cyber_mean_token_accuracy": 0.49230658441782,
2166
+ "eval_cyber_num_tokens": 29317919.0,
2167
+ "eval_cyber_runtime": 58.8902,
2168
+ "eval_cyber_samples_per_second": 6.741,
2169
+ "eval_cyber_steps_per_second": 1.698,
2170
+ "step": 1500
2171
+ },
2172
+ {
2173
+ "entropy": 1.2400090254843235,
2174
+ "epoch": 2.416,
2175
+ "grad_norm": 21.25,
2176
+ "learning_rate": 1.1044444444444444e-05,
2177
+ "loss": 20.0928,
2178
+ "mean_token_accuracy": 0.6977616097778082,
2179
+ "num_tokens": 29518541.0,
2180
+ "step": 1510
2181
+ },
2182
+ {
2183
+ "entropy": 1.240101160109043,
2184
+ "epoch": 2.432,
2185
+ "grad_norm": 20.5,
2186
+ "learning_rate": 1.0970370370370371e-05,
2187
+ "loss": 20.0068,
2188
+ "mean_token_accuracy": 0.697515693679452,
2189
+ "num_tokens": 29720768.0,
2190
+ "step": 1520
2191
+ },
2192
+ {
2193
+ "entropy": 1.2756904646754266,
2194
+ "epoch": 2.448,
2195
+ "grad_norm": 21.5,
2196
+ "learning_rate": 1.0896296296296298e-05,
2197
+ "loss": 20.6782,
2198
+ "mean_token_accuracy": 0.6915484260767698,
2199
+ "num_tokens": 29916619.0,
2200
+ "step": 1530
2201
+ },
2202
+ {
2203
+ "entropy": 1.224426954984665,
2204
+ "epoch": 2.464,
2205
+ "grad_norm": 22.125,
2206
+ "learning_rate": 1.0822222222222222e-05,
2207
+ "loss": 19.6834,
2208
+ "mean_token_accuracy": 0.6991619531065225,
2209
+ "num_tokens": 30121198.0,
2210
+ "step": 1540
2211
+ },
2212
+ {
2213
+ "entropy": 1.2454459190368652,
2214
+ "epoch": 2.48,
2215
+ "grad_norm": 19.5,
2216
+ "learning_rate": 1.074814814814815e-05,
2217
+ "loss": 20.0401,
2218
+ "mean_token_accuracy": 0.6991795245558023,
2219
+ "num_tokens": 30325508.0,
2220
+ "step": 1550
2221
+ },
2222
+ {
2223
+ "entropy": 1.242332585528493,
2224
+ "epoch": 2.496,
2225
+ "grad_norm": 19.625,
2226
+ "learning_rate": 1.0674074074074074e-05,
2227
+ "loss": 20.0988,
2228
+ "mean_token_accuracy": 0.6955976873636246,
2229
+ "num_tokens": 30517445.0,
2230
+ "step": 1560
2231
+ },
2232
+ {
2233
+ "entropy": 1.2685849387198687,
2234
+ "epoch": 2.512,
2235
+ "grad_norm": 22.125,
2236
+ "learning_rate": 1.0600000000000002e-05,
2237
+ "loss": 20.372,
2238
+ "mean_token_accuracy": 0.6936233077198267,
2239
+ "num_tokens": 30714049.0,
2240
+ "step": 1570
2241
+ },
2242
+ {
2243
+ "entropy": 1.2411126971244812,
2244
+ "epoch": 2.528,
2245
+ "grad_norm": 21.25,
2246
+ "learning_rate": 1.0525925925925927e-05,
2247
+ "loss": 20.1713,
2248
+ "mean_token_accuracy": 0.6966881100088358,
2249
+ "num_tokens": 30911317.0,
2250
+ "step": 1580
2251
+ },
2252
+ {
2253
+ "entropy": 1.2535224214196206,
2254
+ "epoch": 2.544,
2255
+ "grad_norm": 18.625,
2256
+ "learning_rate": 1.0451851851851853e-05,
2257
+ "loss": 20.2186,
2258
+ "mean_token_accuracy": 0.6961116347461939,
2259
+ "num_tokens": 31108983.0,
2260
+ "step": 1590
2261
+ },
2262
+ {
2263
+ "entropy": 1.2615187034010886,
2264
+ "epoch": 2.56,
2265
+ "grad_norm": 21.125,
2266
+ "learning_rate": 1.0377777777777778e-05,
2267
+ "loss": 20.379,
2268
+ "mean_token_accuracy": 0.6933579590171576,
2269
+ "num_tokens": 31298994.0,
2270
+ "step": 1600
2271
+ },
2272
+ {
2273
+ "epoch": 2.56,
2274
+ "eval_biology_entropy": 1.2317106494903565,
2275
+ "eval_biology_loss": 1.2648837566375732,
2276
+ "eval_biology_mean_token_accuracy": 0.6972949070930481,
2277
+ "eval_biology_num_tokens": 31298994.0,
2278
+ "eval_biology_runtime": 48.5625,
2279
+ "eval_biology_samples_per_second": 10.296,
2280
+ "eval_biology_steps_per_second": 2.574,
2281
+ "step": 1600
2282
+ },
2283
+ {
2284
+ "epoch": 2.56,
2285
+ "eval_chemistry_entropy": 0.9419839086532593,
2286
+ "eval_chemistry_loss": 0.9602895379066467,
2287
+ "eval_chemistry_mean_token_accuracy": 0.7632233667373657,
2288
+ "eval_chemistry_num_tokens": 31298994.0,
2289
+ "eval_chemistry_runtime": 60.2567,
2290
+ "eval_chemistry_samples_per_second": 8.298,
2291
+ "eval_chemistry_steps_per_second": 2.074,
2292
+ "step": 1600
2293
+ },
2294
+ {
2295
+ "epoch": 2.56,
2296
+ "eval_math_entropy": 0.8460503368377685,
2297
+ "eval_math_loss": 1.0832732915878296,
2298
+ "eval_math_mean_token_accuracy": 0.749913236618042,
2299
+ "eval_math_num_tokens": 31298994.0,
2300
+ "eval_math_runtime": 61.768,
2301
+ "eval_math_samples_per_second": 8.095,
2302
+ "eval_math_steps_per_second": 2.024,
2303
+ "step": 1600
2304
+ },
2305
+ {
2306
+ "epoch": 2.56,
2307
+ "eval_cyber_entropy": 2.238676429986954,
2308
+ "eval_cyber_loss": 2.8418731689453125,
2309
+ "eval_cyber_mean_token_accuracy": 0.4932973644137382,
2310
+ "eval_cyber_num_tokens": 31298994.0,
2311
+ "eval_cyber_runtime": 58.9355,
2312
+ "eval_cyber_samples_per_second": 6.736,
2313
+ "eval_cyber_steps_per_second": 1.697,
2314
+ "step": 1600
2315
+ },
2316
+ {
2317
+ "entropy": 1.2670236613601447,
2318
+ "epoch": 2.576,
2319
+ "grad_norm": 19.125,
2320
+ "learning_rate": 1.0303703703703705e-05,
2321
+ "loss": 20.4162,
2322
+ "mean_token_accuracy": 0.6929084740579128,
2323
+ "num_tokens": 31488512.0,
2324
+ "step": 1610
2325
+ },
2326
+ {
2327
+ "entropy": 1.2848519578576088,
2328
+ "epoch": 2.592,
2329
+ "grad_norm": 20.125,
2330
+ "learning_rate": 1.0229629629629631e-05,
2331
+ "loss": 20.8122,
2332
+ "mean_token_accuracy": 0.6882869061082602,
2333
+ "num_tokens": 31684651.0,
2334
+ "step": 1620
2335
+ },
2336
+ {
2337
+ "entropy": 1.2756942443549633,
2338
+ "epoch": 2.608,
2339
+ "grad_norm": 21.625,
2340
+ "learning_rate": 1.0155555555555556e-05,
2341
+ "loss": 20.5742,
2342
+ "mean_token_accuracy": 0.6907225508242846,
2343
+ "num_tokens": 31876008.0,
2344
+ "step": 1630
2345
+ },
2346
+ {
2347
+ "entropy": 1.2478659953922033,
2348
+ "epoch": 2.624,
2349
+ "grad_norm": 21.375,
2350
+ "learning_rate": 1.0081481481481484e-05,
2351
+ "loss": 20.006,
2352
+ "mean_token_accuracy": 0.6965215235948563,
2353
+ "num_tokens": 32069617.0,
2354
+ "step": 1640
2355
+ },
2356
+ {
2357
+ "entropy": 1.2640259351581336,
2358
+ "epoch": 2.64,
2359
+ "grad_norm": 20.0,
2360
+ "learning_rate": 1.0007407407407407e-05,
2361
+ "loss": 20.6387,
2362
+ "mean_token_accuracy": 0.6902830567210912,
2363
+ "num_tokens": 32262932.0,
2364
+ "step": 1650
2365
+ },
2366
+ {
2367
+ "entropy": 1.2446098852902652,
2368
+ "epoch": 2.656,
2369
+ "grad_norm": 19.625,
2370
+ "learning_rate": 9.933333333333334e-06,
2371
+ "loss": 19.859,
2372
+ "mean_token_accuracy": 0.6993511699140071,
2373
+ "num_tokens": 32465495.0,
2374
+ "step": 1660
2375
+ },
2376
+ {
2377
+ "entropy": 1.2263234194368124,
2378
+ "epoch": 2.672,
2379
+ "grad_norm": 20.5,
2380
+ "learning_rate": 9.85925925925926e-06,
2381
+ "loss": 19.9066,
2382
+ "mean_token_accuracy": 0.6994441974908113,
2383
+ "num_tokens": 32665458.0,
2384
+ "step": 1670
2385
+ },
2386
+ {
2387
+ "entropy": 1.2391818948090076,
2388
+ "epoch": 2.6879999999999997,
2389
+ "grad_norm": 19.125,
2390
+ "learning_rate": 9.785185185185187e-06,
2391
+ "loss": 20.0315,
2392
+ "mean_token_accuracy": 0.6982862666249275,
2393
+ "num_tokens": 32867135.0,
2394
+ "step": 1680
2395
+ },
2396
+ {
2397
+ "entropy": 1.2510778598487378,
2398
+ "epoch": 2.7039999999999997,
2399
+ "grad_norm": 19.625,
2400
+ "learning_rate": 9.711111111111111e-06,
2401
+ "loss": 20.1242,
2402
+ "mean_token_accuracy": 0.6958760701119899,
2403
+ "num_tokens": 33070366.0,
2404
+ "step": 1690
2405
+ },
2406
+ {
2407
+ "entropy": 1.2359105788171292,
2408
+ "epoch": 2.7199999999999998,
2409
+ "grad_norm": 22.5,
2410
+ "learning_rate": 9.637037037037038e-06,
2411
+ "loss": 19.9727,
2412
+ "mean_token_accuracy": 0.6961525905877352,
2413
+ "num_tokens": 33269122.0,
2414
+ "step": 1700
2415
+ },
2416
+ {
2417
+ "epoch": 2.7199999999999998,
2418
+ "eval_biology_entropy": 1.220011395931244,
2419
+ "eval_biology_loss": 1.2638740539550781,
2420
+ "eval_biology_mean_token_accuracy": 0.6976135740280152,
2421
+ "eval_biology_num_tokens": 33269122.0,
2422
+ "eval_biology_runtime": 48.5623,
2423
+ "eval_biology_samples_per_second": 10.296,
2424
+ "eval_biology_steps_per_second": 2.574,
2425
+ "step": 1700
2426
+ },
2427
+ {
2428
+ "epoch": 2.7199999999999998,
2429
+ "eval_chemistry_entropy": 0.9310768895149231,
2430
+ "eval_chemistry_loss": 0.9585283398628235,
2431
+ "eval_chemistry_mean_token_accuracy": 0.7632611804008483,
2432
+ "eval_chemistry_num_tokens": 33269122.0,
2433
+ "eval_chemistry_runtime": 60.2482,
2434
+ "eval_chemistry_samples_per_second": 8.299,
2435
+ "eval_chemistry_steps_per_second": 2.075,
2436
+ "step": 1700
2437
+ },
2438
+ {
2439
+ "epoch": 2.7199999999999998,
2440
+ "eval_math_entropy": 0.8383772811889648,
2441
+ "eval_math_loss": 1.0849062204360962,
2442
+ "eval_math_mean_token_accuracy": 0.7499314618110656,
2443
+ "eval_math_num_tokens": 33269122.0,
2444
+ "eval_math_runtime": 61.8032,
2445
+ "eval_math_samples_per_second": 8.09,
2446
+ "eval_math_steps_per_second": 2.023,
2447
+ "step": 1700
2448
+ },
2449
+ {
2450
+ "epoch": 2.7199999999999998,
2451
+ "eval_cyber_entropy": 2.2346273112297057,
2452
+ "eval_cyber_loss": 2.8241565227508545,
2453
+ "eval_cyber_mean_token_accuracy": 0.4965496668219566,
2454
+ "eval_cyber_num_tokens": 33269122.0,
2455
+ "eval_cyber_runtime": 58.9306,
2456
+ "eval_cyber_samples_per_second": 6.737,
2457
+ "eval_cyber_steps_per_second": 1.697,
2458
+ "step": 1700
2459
+ },
2460
+ {
2461
+ "entropy": 1.275526064261794,
2462
+ "epoch": 2.7359999999999998,
2463
+ "grad_norm": 19.75,
2464
+ "learning_rate": 9.562962962962965e-06,
2465
+ "loss": 20.5536,
2466
+ "mean_token_accuracy": 0.6915018800646067,
2467
+ "num_tokens": 33456592.0,
2468
+ "step": 1710
2469
+ },
2470
+ {
2471
+ "entropy": 1.258423050120473,
2472
+ "epoch": 2.752,
2473
+ "grad_norm": 20.875,
2474
+ "learning_rate": 9.48888888888889e-06,
2475
+ "loss": 20.2345,
2476
+ "mean_token_accuracy": 0.6949015587568284,
2477
+ "num_tokens": 33645862.0,
2478
+ "step": 1720
2479
+ },
2480
+ {
2481
+ "entropy": 1.2810617201030254,
2482
+ "epoch": 2.768,
2483
+ "grad_norm": 20.625,
2484
+ "learning_rate": 9.414814814814816e-06,
2485
+ "loss": 20.7419,
2486
+ "mean_token_accuracy": 0.6883704375475646,
2487
+ "num_tokens": 33835740.0,
2488
+ "step": 1730
2489
+ },
2490
+ {
2491
+ "entropy": 1.235444600507617,
2492
+ "epoch": 2.784,
2493
+ "grad_norm": 18.25,
2494
+ "learning_rate": 9.34074074074074e-06,
2495
+ "loss": 19.9614,
2496
+ "mean_token_accuracy": 0.698374417796731,
2497
+ "num_tokens": 34034702.0,
2498
+ "step": 1740
2499
+ },
2500
+ {
2501
+ "entropy": 1.240835690125823,
2502
+ "epoch": 2.8,
2503
+ "grad_norm": 18.875,
2504
+ "learning_rate": 9.266666666666667e-06,
2505
+ "loss": 20.0363,
2506
+ "mean_token_accuracy": 0.6977193936705589,
2507
+ "num_tokens": 34235603.0,
2508
+ "step": 1750
2509
+ },
2510
+ {
2511
+ "entropy": 1.2478354915976524,
2512
+ "epoch": 2.816,
2513
+ "grad_norm": 20.125,
2514
+ "learning_rate": 9.192592592592594e-06,
2515
+ "loss": 20.1876,
2516
+ "mean_token_accuracy": 0.6962190333753824,
2517
+ "num_tokens": 34434833.0,
2518
+ "step": 1760
2519
+ },
2520
+ {
2521
+ "entropy": 1.2712427478283643,
2522
+ "epoch": 2.832,
2523
+ "grad_norm": 21.5,
2524
+ "learning_rate": 9.118518518518518e-06,
2525
+ "loss": 20.6114,
2526
+ "mean_token_accuracy": 0.6929421681910753,
2527
+ "num_tokens": 34625581.0,
2528
+ "step": 1770
2529
+ },
2530
+ {
2531
+ "entropy": 1.2750334855169059,
2532
+ "epoch": 2.848,
2533
+ "grad_norm": 21.75,
2534
+ "learning_rate": 9.044444444444445e-06,
2535
+ "loss": 20.4708,
2536
+ "mean_token_accuracy": 0.6927024837583303,
2537
+ "num_tokens": 34816521.0,
2538
+ "step": 1780
2539
+ },
2540
+ {
2541
+ "entropy": 1.240677573531866,
2542
+ "epoch": 2.864,
2543
+ "grad_norm": 20.375,
2544
+ "learning_rate": 8.970370370370372e-06,
2545
+ "loss": 20.0913,
2546
+ "mean_token_accuracy": 0.6966684833168983,
2547
+ "num_tokens": 35013631.0,
2548
+ "step": 1790
2549
+ },
2550
+ {
2551
+ "entropy": 1.2785068407654763,
2552
+ "epoch": 2.88,
2553
+ "grad_norm": 21.625,
2554
+ "learning_rate": 8.896296296296298e-06,
2555
+ "loss": 20.5855,
2556
+ "mean_token_accuracy": 0.6912990726530552,
2557
+ "num_tokens": 35207800.0,
2558
+ "step": 1800
2559
+ },
2560
+ {
2561
+ "epoch": 2.88,
2562
+ "eval_biology_entropy": 1.2326566462516784,
2563
+ "eval_biology_loss": 1.2624694108963013,
2564
+ "eval_biology_mean_token_accuracy": 0.6982610473632812,
2565
+ "eval_biology_num_tokens": 35207800.0,
2566
+ "eval_biology_runtime": 48.6584,
2567
+ "eval_biology_samples_per_second": 10.276,
2568
+ "eval_biology_steps_per_second": 2.569,
2569
+ "step": 1800
2570
+ },
2571
+ {
2572
+ "epoch": 2.88,
2573
+ "eval_chemistry_entropy": 0.9439322190284729,
2574
+ "eval_chemistry_loss": 0.958177387714386,
2575
+ "eval_chemistry_mean_token_accuracy": 0.7634941415786743,
2576
+ "eval_chemistry_num_tokens": 35207800.0,
2577
+ "eval_chemistry_runtime": 60.3744,
2578
+ "eval_chemistry_samples_per_second": 8.282,
2579
+ "eval_chemistry_steps_per_second": 2.07,
2580
+ "step": 1800
2581
+ },
2582
+ {
2583
+ "epoch": 2.88,
2584
+ "eval_math_entropy": 0.8483011198043823,
2585
+ "eval_math_loss": 1.084191918373108,
2586
+ "eval_math_mean_token_accuracy": 0.749516860961914,
2587
+ "eval_math_num_tokens": 35207800.0,
2588
+ "eval_math_runtime": 61.9233,
2589
+ "eval_math_samples_per_second": 8.075,
2590
+ "eval_math_steps_per_second": 2.019,
2591
+ "step": 1800
2592
+ },
2593
+ {
2594
+ "epoch": 2.88,
2595
+ "eval_cyber_entropy": 2.233942435979843,
2596
+ "eval_cyber_loss": 2.8212552070617676,
2597
+ "eval_cyber_mean_token_accuracy": 0.49529594123363496,
2598
+ "eval_cyber_num_tokens": 35207800.0,
2599
+ "eval_cyber_runtime": 59.0321,
2600
+ "eval_cyber_samples_per_second": 6.725,
2601
+ "eval_cyber_steps_per_second": 1.694,
2602
+ "step": 1800
2603
+ },
2604
+ {
2605
+ "entropy": 1.2758983630686997,
2606
+ "epoch": 2.896,
2607
+ "grad_norm": 21.5,
2608
+ "learning_rate": 8.822222222222223e-06,
2609
+ "loss": 20.4145,
2610
+ "mean_token_accuracy": 0.6935398448258638,
2611
+ "num_tokens": 35402995.0,
2612
+ "step": 1810
2613
+ },
2614
+ {
2615
+ "entropy": 1.2687516041100024,
2616
+ "epoch": 2.912,
2617
+ "grad_norm": 21.125,
2618
+ "learning_rate": 8.74814814814815e-06,
2619
+ "loss": 20.5748,
2620
+ "mean_token_accuracy": 0.6925326142460108,
2621
+ "num_tokens": 35595456.0,
2622
+ "step": 1820
2623
+ },
2624
+ {
2625
+ "entropy": 1.2766919579356908,
2626
+ "epoch": 2.928,
2627
+ "grad_norm": 20.75,
2628
+ "learning_rate": 8.674074074074074e-06,
2629
+ "loss": 20.6155,
2630
+ "mean_token_accuracy": 0.6923059385269881,
2631
+ "num_tokens": 35787309.0,
2632
+ "step": 1830
2633
+ },
2634
+ {
2635
+ "entropy": 1.252352038025856,
2636
+ "epoch": 2.944,
2637
+ "grad_norm": 21.125,
2638
+ "learning_rate": 8.6e-06,
2639
+ "loss": 20.2043,
2640
+ "mean_token_accuracy": 0.6947959426790475,
2641
+ "num_tokens": 35982686.0,
2642
+ "step": 1840
2643
+ },
2644
+ {
2645
+ "entropy": 1.2456642746925355,
2646
+ "epoch": 2.96,
2647
+ "grad_norm": 21.375,
2648
+ "learning_rate": 8.525925925925927e-06,
2649
+ "loss": 20.1306,
2650
+ "mean_token_accuracy": 0.6962343286722898,
2651
+ "num_tokens": 36179520.0,
2652
+ "step": 1850
2653
+ },
2654
+ {
2655
+ "entropy": 1.2474573370069266,
2656
+ "epoch": 2.976,
2657
+ "grad_norm": 19.875,
2658
+ "learning_rate": 8.451851851851852e-06,
2659
+ "loss": 20.1127,
2660
+ "mean_token_accuracy": 0.6960530079901218,
2661
+ "num_tokens": 36373689.0,
2662
+ "step": 1860
2663
+ },
2664
+ {
2665
+ "entropy": 1.251881342381239,
2666
+ "epoch": 2.992,
2667
+ "grad_norm": 23.25,
2668
+ "learning_rate": 8.377777777777779e-06,
2669
+ "loss": 20.2869,
2670
+ "mean_token_accuracy": 0.6959608975797892,
2671
+ "num_tokens": 36569612.0,
2672
+ "step": 1870
2673
+ },
2674
+ {
2675
+ "entropy": 1.2793707679957151,
2676
+ "epoch": 3.008,
2677
+ "grad_norm": 21.375,
2678
+ "learning_rate": 8.303703703703705e-06,
2679
+ "loss": 20.4974,
2680
+ "mean_token_accuracy": 0.6918497812002897,
2681
+ "num_tokens": 36759194.0,
2682
+ "step": 1880
2683
+ },
2684
+ {
2685
+ "entropy": 1.270806109532714,
2686
+ "epoch": 3.024,
2687
+ "grad_norm": 20.125,
2688
+ "learning_rate": 8.229629629629632e-06,
2689
+ "loss": 20.5499,
2690
+ "mean_token_accuracy": 0.6906452853232622,
2691
+ "num_tokens": 36947293.0,
2692
+ "step": 1890
2693
+ },
2694
+ {
2695
+ "entropy": 1.2329706739634276,
2696
+ "epoch": 3.04,
2697
+ "grad_norm": 21.0,
2698
+ "learning_rate": 8.155555555555556e-06,
2699
+ "loss": 19.8645,
2700
+ "mean_token_accuracy": 0.7000770628452301,
2701
+ "num_tokens": 37136901.0,
2702
+ "step": 1900
2703
+ },
2704
+ {
2705
+ "epoch": 3.04,
2706
+ "eval_biology_entropy": 1.2161930742263793,
2707
+ "eval_biology_loss": 1.2619609832763672,
2708
+ "eval_biology_mean_token_accuracy": 0.6983966584205628,
2709
+ "eval_biology_num_tokens": 37136901.0,
2710
+ "eval_biology_runtime": 48.5676,
2711
+ "eval_biology_samples_per_second": 10.295,
2712
+ "eval_biology_steps_per_second": 2.574,
2713
+ "step": 1900
2714
+ },
2715
+ {
2716
+ "epoch": 3.04,
2717
+ "eval_chemistry_entropy": 0.9266538290977478,
2718
+ "eval_chemistry_loss": 0.9588263034820557,
2719
+ "eval_chemistry_mean_token_accuracy": 0.763595272064209,
2720
+ "eval_chemistry_num_tokens": 37136901.0,
2721
+ "eval_chemistry_runtime": 60.2182,
2722
+ "eval_chemistry_samples_per_second": 8.303,
2723
+ "eval_chemistry_steps_per_second": 2.076,
2724
+ "step": 1900
2725
+ },
2726
+ {
2727
+ "epoch": 3.04,
2728
+ "eval_math_entropy": 0.8339179096221924,
2729
+ "eval_math_loss": 1.0852128267288208,
2730
+ "eval_math_mean_token_accuracy": 0.7496648592948914,
2731
+ "eval_math_num_tokens": 37136901.0,
2732
+ "eval_math_runtime": 61.793,
2733
+ "eval_math_samples_per_second": 8.092,
2734
+ "eval_math_steps_per_second": 2.023,
2735
+ "step": 1900
2736
+ },
2737
+ {
2738
+ "epoch": 3.04,
2739
+ "eval_cyber_entropy": 2.1908029651641847,
2740
+ "eval_cyber_loss": 2.824298143386841,
2741
+ "eval_cyber_mean_token_accuracy": 0.4978000572323799,
2742
+ "eval_cyber_num_tokens": 37136901.0,
2743
+ "eval_cyber_runtime": 58.9512,
2744
+ "eval_cyber_samples_per_second": 6.734,
2745
+ "eval_cyber_steps_per_second": 1.696,
2746
+ "step": 1900
2747
+ },
2748
+ {
2749
+ "entropy": 1.246962970867753,
2750
+ "epoch": 3.056,
2751
+ "grad_norm": 21.5,
2752
+ "learning_rate": 8.081481481481483e-06,
2753
+ "loss": 20.0058,
2754
+ "mean_token_accuracy": 0.6961042139679193,
2755
+ "num_tokens": 37331588.0,
2756
+ "step": 1910
2757
+ },
2758
+ {
2759
+ "entropy": 1.2139686714857816,
2760
+ "epoch": 3.072,
2761
+ "grad_norm": 21.25,
2762
+ "learning_rate": 8.007407407407408e-06,
2763
+ "loss": 19.5736,
2764
+ "mean_token_accuracy": 0.7032940357923507,
2765
+ "num_tokens": 37521211.0,
2766
+ "step": 1920
2767
+ },
2768
+ {
2769
+ "entropy": 1.23736840672791,
2770
+ "epoch": 3.088,
2771
+ "grad_norm": 20.5,
2772
+ "learning_rate": 7.933333333333334e-06,
2773
+ "loss": 20.0982,
2774
+ "mean_token_accuracy": 0.6960875768214464,
2775
+ "num_tokens": 37713328.0,
2776
+ "step": 1930
2777
+ },
2778
+ {
2779
+ "entropy": 1.2229486364871263,
2780
+ "epoch": 3.104,
2781
+ "grad_norm": 19.125,
2782
+ "learning_rate": 7.859259259259259e-06,
2783
+ "loss": 19.6047,
2784
+ "mean_token_accuracy": 0.7018660567700863,
2785
+ "num_tokens": 37913470.0,
2786
+ "step": 1940
2787
+ },
2788
+ {
2789
+ "entropy": 1.223341765627265,
2790
+ "epoch": 3.12,
2791
+ "grad_norm": 20.5,
2792
+ "learning_rate": 7.785185185185185e-06,
2793
+ "loss": 19.92,
2794
+ "mean_token_accuracy": 0.6998291172087192,
2795
+ "num_tokens": 38106718.0,
2796
+ "step": 1950
2797
+ },
2798
+ {
2799
+ "entropy": 1.2402896065264941,
2800
+ "epoch": 3.136,
2801
+ "grad_norm": 22.5,
2802
+ "learning_rate": 7.711111111111112e-06,
2803
+ "loss": 19.879,
2804
+ "mean_token_accuracy": 0.6992235485464334,
2805
+ "num_tokens": 38295701.0,
2806
+ "step": 1960
2807
+ },
2808
+ {
2809
+ "entropy": 1.2329155914485455,
2810
+ "epoch": 3.152,
2811
+ "grad_norm": 20.875,
2812
+ "learning_rate": 7.637037037037037e-06,
2813
+ "loss": 19.8162,
2814
+ "mean_token_accuracy": 0.7000128466635942,
2815
+ "num_tokens": 38493951.0,
2816
+ "step": 1970
2817
+ },
2818
+ {
2819
+ "entropy": 1.2596189703792333,
2820
+ "epoch": 3.168,
2821
+ "grad_norm": 21.625,
2822
+ "learning_rate": 7.562962962962963e-06,
2823
+ "loss": 20.4478,
2824
+ "mean_token_accuracy": 0.6925735887140035,
2825
+ "num_tokens": 38688390.0,
2826
+ "step": 1980
2827
+ },
2828
+ {
2829
+ "entropy": 1.2458199262619019,
2830
+ "epoch": 3.184,
2831
+ "grad_norm": 19.875,
2832
+ "learning_rate": 7.48888888888889e-06,
2833
+ "loss": 20.0605,
2834
+ "mean_token_accuracy": 0.6980946302413941,
2835
+ "num_tokens": 38880157.0,
2836
+ "step": 1990
2837
+ },
2838
+ {
2839
+ "entropy": 1.2279883187264204,
2840
+ "epoch": 3.2,
2841
+ "grad_norm": 19.875,
2842
+ "learning_rate": 7.4148148148148155e-06,
2843
+ "loss": 19.8355,
2844
+ "mean_token_accuracy": 0.7001625929027796,
2845
+ "num_tokens": 39081782.0,
2846
+ "step": 2000
2847
+ },
2848
+ {
2849
+ "epoch": 3.2,
2850
+ "eval_biology_entropy": 1.203757492542267,
2851
+ "eval_biology_loss": 1.2618530988693237,
2852
+ "eval_biology_mean_token_accuracy": 0.6981693663597107,
2853
+ "eval_biology_num_tokens": 39081782.0,
2854
+ "eval_biology_runtime": 48.5462,
2855
+ "eval_biology_samples_per_second": 10.299,
2856
+ "eval_biology_steps_per_second": 2.575,
2857
+ "step": 2000
2858
+ },
2859
+ {
2860
+ "epoch": 3.2,
2861
+ "eval_chemistry_entropy": 0.9181569867134094,
2862
+ "eval_chemistry_loss": 0.9586373567581177,
2863
+ "eval_chemistry_mean_token_accuracy": 0.7633922476768493,
2864
+ "eval_chemistry_num_tokens": 39081782.0,
2865
+ "eval_chemistry_runtime": 59.9544,
2866
+ "eval_chemistry_samples_per_second": 8.34,
2867
+ "eval_chemistry_steps_per_second": 2.085,
2868
+ "step": 2000
2869
+ },
2870
+ {
2871
+ "epoch": 3.2,
2872
+ "eval_math_entropy": 0.827507004737854,
2873
+ "eval_math_loss": 1.0850911140441895,
2874
+ "eval_math_mean_token_accuracy": 0.7499541993141174,
2875
+ "eval_math_num_tokens": 39081782.0,
2876
+ "eval_math_runtime": 61.4817,
2877
+ "eval_math_samples_per_second": 8.133,
2878
+ "eval_math_steps_per_second": 2.033,
2879
+ "step": 2000
2880
+ },
2881
+ {
2882
+ "epoch": 3.2,
2883
+ "eval_cyber_entropy": 2.1836040806770325,
2884
+ "eval_cyber_loss": 2.829263925552368,
2885
+ "eval_cyber_mean_token_accuracy": 0.4958575034141541,
2886
+ "eval_cyber_num_tokens": 39081782.0,
2887
+ "eval_cyber_runtime": 58.8995,
2888
+ "eval_cyber_samples_per_second": 6.74,
2889
+ "eval_cyber_steps_per_second": 1.698,
2890
+ "step": 2000
2891
+ }
2892
+ ],
2893
+ "logging_steps": 10,
2894
+ "max_steps": 3000,
2895
+ "num_input_tokens_seen": 0,
2896
+ "num_train_epochs": 5,
2897
+ "save_steps": 1000,
2898
+ "stateful_callbacks": {
2899
+ "TrainerControl": {
2900
+ "args": {
2901
+ "should_epoch_stop": false,
2902
+ "should_evaluate": false,
2903
+ "should_log": false,
2904
+ "should_save": true,
2905
+ "should_training_stop": false
2906
+ },
2907
+ "attributes": {}
2908
+ }
2909
+ },
2910
+ "total_flos": 4.4082567723215933e+18,
2911
+ "train_batch_size": 4,
2912
+ "trial_name": null,
2913
+ "trial_params": null
2914
+ }
recover/README.md ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: google/gemma-3-12b-it
3
+ library_name: transformers
4
+ model_name: recover
5
+ tags:
6
+ - generated_from_trainer
7
+ - sft
8
+ - trl
9
+ licence: license
10
+ ---
11
+
12
+ # Model Card for recover
13
+
14
+ This model is a fine-tuned version of [google/gemma-3-12b-it](https://huggingface.co/google/gemma-3-12b-it).
15
+ It has been trained using [TRL](https://github.com/huggingface/trl).
16
+
17
+ ## Quick start
18
+
19
+ ```python
20
+ from transformers import pipeline
21
+
22
+ question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
23
+ generator = pipeline("text-generation", model="None", device="cuda")
24
+ output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
25
+ print(output["generated_text"])
26
+ ```
27
+
28
+ ## Training procedure
29
+
30
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/arunasank/sae-scoping-stemqa-biology/runs/124xpcap)
31
+
32
+
33
+ This model was trained with SFT.
34
+
35
+ ### Framework versions
36
+
37
+ - TRL: 0.22.2
38
+ - Transformers: 4.56.1
39
+ - Pytorch: 2.7.1+cu128
40
+ - Datasets: 4.0.0
41
+ - Tokenizers: 0.22.2
42
+
43
+ ## Citations
44
+
45
+
46
+
47
+ Cite TRL as:
48
+
49
+ ```bibtex
50
+ @misc{vonwerra2022trl,
51
+ title = {{TRL: Transformer Reinforcement Learning}},
52
+ author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
53
+ year = 2020,
54
+ journal = {GitHub repository},
55
+ publisher = {GitHub},
56
+ howpublished = {\url{https://github.com/huggingface/trl}}
57
+ }
58
+ ```
recover/checkpoint-3000/added_tokens.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "<image_soft_token>": 262144
3
+ }
recover/checkpoint-3000/chat_template.jinja ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'image' -%}
33
+ {{ '<start_of_image>' }}
34
+ {%- elif item['type'] == 'text' -%}
35
+ {{ item['text'] | trim }}
36
+ {%- endif -%}
37
+ {%- endfor -%}
38
+ {%- else -%}
39
+ {{ raise_exception("Invalid content type") }}
40
+ {%- endif -%}
41
+ {{ '<end_of_turn>
42
+ ' }}
43
+ {%- endfor -%}
44
+ {%- if add_generation_prompt -%}
45
+ {{'<start_of_turn>model
46
+ '}}
47
+ {%- endif -%}
recover/checkpoint-3000/generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 1,
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "top_k": 64,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.56.1"
13
+ }
recover/checkpoint-3000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
recover/checkpoint-3000/special_tokens_map.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "boi_token": "<start_of_image>",
3
+ "bos_token": {
4
+ "content": "<bos>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "eoi_token": "<end_of_image>",
11
+ "eos_token": {
12
+ "content": "<eos>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false
17
+ },
18
+ "image_token": "<image_soft_token>",
19
+ "pad_token": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "unk_token": {
27
+ "content": "<unk>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ }
33
+ }
recover/checkpoint-3000/tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff
 
recover/checkpoint-3000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
recover/final/added_tokens.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "<image_soft_token>": 262144
3
+ }
recover/final/chat_template.jinja ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'image' -%}
33
+ {{ '<start_of_image>' }}
34
+ {%- elif item['type'] == 'text' -%}
35
+ {{ item['text'] | trim }}
36
+ {%- endif -%}
37
+ {%- endfor -%}
38
+ {%- else -%}
39
+ {{ raise_exception("Invalid content type") }}
40
+ {%- endif -%}
41
+ {{ '<end_of_turn>
42
+ ' }}
43
+ {%- endfor -%}
44
+ {%- if add_generation_prompt -%}
45
+ {{'<start_of_turn>model
46
+ '}}
47
+ {%- endif -%}
recover/final/config.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "dtype": "bfloat16",
21
+ "final_logit_softcapping": null,
22
+ "head_dim": 256,
23
+ "hidden_activation": "gelu_pytorch_tanh",
24
+ "hidden_size": 3840,
25
+ "initializer_range": 0.02,
26
+ "intermediate_size": 15360,
27
+ "layer_types": [
28
+ "sliding_attention",
29
+ "sliding_attention",
30
+ "sliding_attention",
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "full_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "sliding_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "full_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "sliding_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "full_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "sliding_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "full_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "sliding_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "full_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "sliding_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "full_attention",
64
+ "sliding_attention",
65
+ "sliding_attention",
66
+ "sliding_attention",
67
+ "sliding_attention",
68
+ "sliding_attention",
69
+ "full_attention",
70
+ "sliding_attention",
71
+ "sliding_attention",
72
+ "sliding_attention",
73
+ "sliding_attention",
74
+ "sliding_attention",
75
+ "full_attention"
76
+ ],
77
+ "max_position_embeddings": 131072,
78
+ "model_type": "gemma3_text",
79
+ "num_attention_heads": 16,
80
+ "num_hidden_layers": 48,
81
+ "num_key_value_heads": 8,
82
+ "query_pre_attn_scalar": 256,
83
+ "rms_norm_eps": 1e-06,
84
+ "rope_local_base_freq": 10000.0,
85
+ "rope_scaling": {
86
+ "factor": 8.0,
87
+ "rope_type": "linear"
88
+ },
89
+ "rope_theta": 1000000.0,
90
+ "sliding_window": 1024,
91
+ "use_cache": true,
92
+ "vocab_size": 262208
93
+ },
94
+ "transformers_version": "4.56.1",
95
+ "vision_config": {
96
+ "attention_dropout": 0.0,
97
+ "dtype": "bfloat16",
98
+ "hidden_act": "gelu_pytorch_tanh",
99
+ "hidden_size": 1152,
100
+ "image_size": 896,
101
+ "intermediate_size": 4304,
102
+ "layer_norm_eps": 1e-06,
103
+ "model_type": "siglip_vision_model",
104
+ "num_attention_heads": 16,
105
+ "num_channels": 3,
106
+ "num_hidden_layers": 27,
107
+ "patch_size": 14,
108
+ "vision_use_head": false
109
+ }
110
+ }
recover/final/generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 1,
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "top_k": 64,
11
+ "top_p": 0.95,
12
+ "transformers_version": "4.56.1"
13
+ }
recover/final/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff