Jerry999 commited on
Commit
685ede0
·
verified ·
1 Parent(s): e2ba804

Upload checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305

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 +3 -0
  2. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/chat_template.jinja +4 -0
  3. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/config.json +71 -0
  4. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/generation_config.json +12 -0
  5. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/model.safetensors +3 -0
  6. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/optimizer.pt +3 -0
  7. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/rng_state.pth +3 -0
  8. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/scheduler.pt +3 -0
  9. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokenizer.json +3 -0
  10. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokenizer_config.json +29 -0
  11. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokens_state. +1 -0
  12. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/trainer_state.json +1010 -0
  13. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/training_args.bin +3 -0
  14. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/chat_template.jinja +4 -0
  15. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/config.json +71 -0
  16. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/generation_config.json +12 -0
  17. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/model.safetensors +3 -0
  18. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/optimizer.pt +3 -0
  19. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/rng_state.pth +3 -0
  20. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/scheduler.pt +3 -0
  21. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokenizer.json +3 -0
  22. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokenizer_config.json +29 -0
  23. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokens_state. +1 -0
  24. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/trainer_state.json +1132 -0
  25. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/training_args.bin +3 -0
  26. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/chat_template.jinja +4 -0
  27. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/config.json +71 -0
  28. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/generation_config.json +12 -0
  29. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/model.safetensors +3 -0
  30. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/optimizer.pt +3 -0
  31. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/rng_state.pth +3 -0
  32. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/scheduler.pt +3 -0
  33. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokenizer.json +3 -0
  34. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokenizer_config.json +29 -0
  35. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokens_state. +1 -0
  36. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/trainer_state.json +1268 -0
  37. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/training_args.bin +3 -0
  38. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/debug.log +0 -0
  39. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/eval_results.csv +2 -0
  40. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/eval_summary.json +11 -0
  41. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/test_alpaca_converted.jsonl +0 -0
  42. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/test_alpaca_results.jsonl +0 -0
  43. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/eval_results.csv +3 -0
  44. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/eval_summary.json +11 -0
  45. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/test_alpaca_converted.jsonl +0 -0
  46. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/test_alpaca_results.jsonl +0 -0
  47. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/eval_results.csv +2 -0
  48. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/eval_summary.json +11 -0
  49. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/test_alpaca_converted.jsonl +0 -0
  50. checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/test_alpaca_results.jsonl +0 -0
.gitattributes CHANGED
@@ -56,3 +56,6 @@ checkpoints/knowledge/atomic_full_sft_50ep_t20260305/checkpoint-1274/tokenizer.j
56
  checkpoints/knowledge/atomic_full_sft_50ep_t20260305/checkpoint-1300/tokenizer.json filter=lfs diff=lfs merge=lfs -text
57
  checkpoints/knowledge/atomic_full_sft_with_phantom_wiki_compositional_data_50ep_t20260310_t20260310/checkpoint-756/tokenizer.json filter=lfs diff=lfs merge=lfs -text
58
  checkpoints/knowledge/atomic_full_sft_with_phantom_wiki_compositional_data_50ep_t20260310_t20260310/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
 
 
56
  checkpoints/knowledge/atomic_full_sft_50ep_t20260305/checkpoint-1300/tokenizer.json filter=lfs diff=lfs merge=lfs -text
57
  checkpoints/knowledge/atomic_full_sft_with_phantom_wiki_compositional_data_50ep_t20260310_t20260310/checkpoint-756/tokenizer.json filter=lfs diff=lfs merge=lfs -text
58
  checkpoints/knowledge/atomic_full_sft_with_phantom_wiki_compositional_data_50ep_t20260310_t20260310/tokenizer.json filter=lfs diff=lfs merge=lfs -text
59
+ checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokenizer.json filter=lfs diff=lfs merge=lfs -text
60
+ checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokenizer.json filter=lfs diff=lfs merge=lfs -text
61
+ checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokenizer.json filter=lfs diff=lfs merge=lfs -text
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/chat_template.jinja ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '
2
+ ' + message['content'] + '<|im_end|>' + '
3
+ '}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
4
+ ' }}{% endif %}
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "bfloat16",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2560,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 9728,
15
+ "layer_types": [
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention"
52
+ ],
53
+ "max_position_embeddings": 262144,
54
+ "max_window_layers": 36,
55
+ "model_type": "qwen3",
56
+ "num_attention_heads": 32,
57
+ "num_hidden_layers": 36,
58
+ "num_key_value_heads": 8,
59
+ "pad_token_id": 151643,
60
+ "rms_norm_eps": 1e-06,
61
+ "rope_parameters": {
62
+ "rope_theta": 5000000,
63
+ "rope_type": "default"
64
+ },
65
+ "sliding_window": null,
66
+ "tie_word_embeddings": true,
67
+ "transformers_version": "5.0.0",
68
+ "use_cache": false,
69
+ "use_sliding_window": false,
70
+ "vocab_size": 151936
71
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "temperature": 0.7,
9
+ "top_k": 20,
10
+ "top_p": 0.8,
11
+ "transformers_version": "5.0.0"
12
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4efc2c81831a80b2926165104ef4296e37674e46a4fa873b71cb49ce4777590f
3
+ size 8822894520
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e03943248822c91fcd0b0c3294043af0777b1954f1c9b93bb741076e45fdef4e
3
+ size 16090225449
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8553d3987609520495680f36a4682a760a418a930648ce46178a1399cde440b0
3
+ size 14645
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42f13ffa918c11c5393d618ff4f842bd282c5cc03a6b4927bcec919a0d5edb24
3
+ size 1465
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
3
+ size 11422650
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokenizer_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "model_max_length": 1010000,
25
+ "pad_token": "<|endoftext|>",
26
+ "split_special_tokens": false,
27
+ "tokenizer_class": "Qwen2Tokenizer",
28
+ "unk_token": null
29
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/tokens_state. ADDED
@@ -0,0 +1 @@
 
 
1
+ {"total": 4653056, "trainable": 2010981}
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/trainer_state.json ADDED
@@ -0,0 +1,1010 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 432,
3
+ "best_metric": 0.005209571216255426,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 8.0,
6
+ "eval_steps": 36,
7
+ "global_step": 568,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0,
14
+ "eval_loss": 6.340579032897949,
15
+ "eval_ppl": 567.1246,
16
+ "eval_runtime": 9.2875,
17
+ "eval_samples_per_second": 53.728,
18
+ "eval_steps_per_second": 53.728,
19
+ "memory/device_reserved (GiB)": 9.68,
20
+ "memory/max_active (GiB)": 8.96,
21
+ "memory/max_allocated (GiB)": 8.96,
22
+ "step": 0
23
+ },
24
+ {
25
+ "epoch": 0.14084507042253522,
26
+ "grad_norm": 48.75,
27
+ "learning_rate": 2.535211267605634e-06,
28
+ "loss": 6.281451797485351,
29
+ "memory/device_reserved (GiB)": 33.61,
30
+ "memory/max_active (GiB)": 32.32,
31
+ "memory/max_allocated (GiB)": 32.32,
32
+ "ppl": 534.56418,
33
+ "step": 10,
34
+ "tokens/total": 81920,
35
+ "tokens/train_per_sec_per_gpu": 9.82,
36
+ "tokens/trainable": 35649
37
+ },
38
+ {
39
+ "epoch": 0.28169014084507044,
40
+ "grad_norm": 43.0,
41
+ "learning_rate": 5.352112676056338e-06,
42
+ "loss": 5.6617687225341795,
43
+ "memory/device_reserved (GiB)": 33.61,
44
+ "memory/max_active (GiB)": 32.32,
45
+ "memory/max_allocated (GiB)": 32.32,
46
+ "ppl": 287.65698,
47
+ "step": 20,
48
+ "tokens/total": 163840,
49
+ "tokens/train_per_sec_per_gpu": 10.46,
50
+ "tokens/trainable": 71196
51
+ },
52
+ {
53
+ "epoch": 0.4225352112676056,
54
+ "grad_norm": 26.875,
55
+ "learning_rate": 8.169014084507043e-06,
56
+ "loss": 3.820334243774414,
57
+ "memory/device_reserved (GiB)": 33.61,
58
+ "memory/max_active (GiB)": 32.32,
59
+ "memory/max_allocated (GiB)": 32.32,
60
+ "ppl": 45.61945,
61
+ "step": 30,
62
+ "tokens/total": 245760,
63
+ "tokens/train_per_sec_per_gpu": 10.53,
64
+ "tokens/trainable": 106543
65
+ },
66
+ {
67
+ "epoch": 0.5070422535211268,
68
+ "eval_loss": 1.091687560081482,
69
+ "eval_ppl": 2.9793,
70
+ "eval_runtime": 8.0764,
71
+ "eval_samples_per_second": 61.785,
72
+ "eval_steps_per_second": 61.785,
73
+ "memory/device_reserved (GiB)": 33.61,
74
+ "memory/max_active (GiB)": 32.32,
75
+ "memory/max_allocated (GiB)": 32.32,
76
+ "step": 36
77
+ },
78
+ {
79
+ "epoch": 0.5633802816901409,
80
+ "grad_norm": 6.5625,
81
+ "learning_rate": 1.0985915492957748e-05,
82
+ "loss": 1.4444393157958983,
83
+ "memory/device_reserved (GiB)": 33.42,
84
+ "memory/max_active (GiB)": 32.31,
85
+ "memory/max_allocated (GiB)": 32.31,
86
+ "ppl": 4.23947,
87
+ "step": 40,
88
+ "tokens/total": 327680,
89
+ "tokens/train_per_sec_per_gpu": 10.34,
90
+ "tokens/trainable": 141719
91
+ },
92
+ {
93
+ "epoch": 0.704225352112676,
94
+ "grad_norm": 2.484375,
95
+ "learning_rate": 1.380281690140845e-05,
96
+ "loss": 0.4776052951812744,
97
+ "memory/device_reserved (GiB)": 33.42,
98
+ "memory/max_active (GiB)": 32.31,
99
+ "memory/max_allocated (GiB)": 32.31,
100
+ "ppl": 1.61221,
101
+ "step": 50,
102
+ "tokens/total": 409600,
103
+ "tokens/train_per_sec_per_gpu": 10.24,
104
+ "tokens/trainable": 177240
105
+ },
106
+ {
107
+ "epoch": 0.8450704225352113,
108
+ "grad_norm": 1.421875,
109
+ "learning_rate": 1.6619718309859155e-05,
110
+ "loss": 0.15568361282348633,
111
+ "memory/device_reserved (GiB)": 33.42,
112
+ "memory/max_active (GiB)": 32.31,
113
+ "memory/max_allocated (GiB)": 32.31,
114
+ "ppl": 1.16846,
115
+ "step": 60,
116
+ "tokens/total": 491520,
117
+ "tokens/train_per_sec_per_gpu": 10.55,
118
+ "tokens/trainable": 212815
119
+ },
120
+ {
121
+ "epoch": 0.9859154929577465,
122
+ "grad_norm": 0.95703125,
123
+ "learning_rate": 1.943661971830986e-05,
124
+ "loss": 0.06936523914337159,
125
+ "memory/device_reserved (GiB)": 33.42,
126
+ "memory/max_active (GiB)": 32.31,
127
+ "memory/max_allocated (GiB)": 32.31,
128
+ "ppl": 1.07183,
129
+ "step": 70,
130
+ "tokens/total": 573440,
131
+ "tokens/train_per_sec_per_gpu": 10.42,
132
+ "tokens/trainable": 248141
133
+ },
134
+ {
135
+ "epoch": 1.0140845070422535,
136
+ "eval_loss": 0.04077093303203583,
137
+ "eval_ppl": 1.04161,
138
+ "eval_runtime": 8.0461,
139
+ "eval_samples_per_second": 62.018,
140
+ "eval_steps_per_second": 62.018,
141
+ "memory/device_reserved (GiB)": 33.42,
142
+ "memory/max_active (GiB)": 32.31,
143
+ "memory/max_allocated (GiB)": 32.31,
144
+ "step": 72
145
+ },
146
+ {
147
+ "epoch": 1.1267605633802817,
148
+ "grad_norm": 0.859375,
149
+ "learning_rate": 1.9992266216318037e-05,
150
+ "loss": 0.03334630727767944,
151
+ "memory/device_reserved (GiB)": 33.42,
152
+ "memory/max_active (GiB)": 32.31,
153
+ "memory/max_allocated (GiB)": 32.31,
154
+ "ppl": 1.03391,
155
+ "step": 80,
156
+ "tokens/total": 655360,
157
+ "tokens/train_per_sec_per_gpu": 10.95,
158
+ "tokens/trainable": 283579
159
+ },
160
+ {
161
+ "epoch": 1.267605633802817,
162
+ "grad_norm": 0.94921875,
163
+ "learning_rate": 1.996086822074945e-05,
164
+ "loss": 0.020297910273075103,
165
+ "memory/device_reserved (GiB)": 33.42,
166
+ "memory/max_active (GiB)": 32.31,
167
+ "memory/max_allocated (GiB)": 32.31,
168
+ "ppl": 1.02051,
169
+ "step": 90,
170
+ "tokens/total": 737280,
171
+ "tokens/train_per_sec_per_gpu": 10.49,
172
+ "tokens/trainable": 319131
173
+ },
174
+ {
175
+ "epoch": 1.408450704225352,
176
+ "grad_norm": 0.482421875,
177
+ "learning_rate": 1.9905398470798207e-05,
178
+ "loss": 0.014508798718452454,
179
+ "memory/device_reserved (GiB)": 33.42,
180
+ "memory/max_active (GiB)": 32.31,
181
+ "memory/max_allocated (GiB)": 32.31,
182
+ "ppl": 1.01461,
183
+ "step": 100,
184
+ "tokens/total": 819200,
185
+ "tokens/train_per_sec_per_gpu": 10.25,
186
+ "tokens/trainable": 354528
187
+ },
188
+ {
189
+ "epoch": 1.5211267605633803,
190
+ "eval_loss": 0.011320236138999462,
191
+ "eval_ppl": 1.01138,
192
+ "eval_runtime": 8.142,
193
+ "eval_samples_per_second": 61.287,
194
+ "eval_steps_per_second": 61.287,
195
+ "memory/device_reserved (GiB)": 33.42,
196
+ "memory/max_active (GiB)": 32.31,
197
+ "memory/max_allocated (GiB)": 32.31,
198
+ "step": 108
199
+ },
200
+ {
201
+ "epoch": 1.5492957746478875,
202
+ "grad_norm": 0.58203125,
203
+ "learning_rate": 1.9825991016445387e-05,
204
+ "loss": 0.013250131905078889,
205
+ "memory/device_reserved (GiB)": 33.38,
206
+ "memory/max_active (GiB)": 32.31,
207
+ "memory/max_allocated (GiB)": 32.31,
208
+ "ppl": 1.01334,
209
+ "step": 110,
210
+ "tokens/total": 901120,
211
+ "tokens/train_per_sec_per_gpu": 10.85,
212
+ "tokens/trainable": 389882
213
+ },
214
+ {
215
+ "epoch": 1.6901408450704225,
216
+ "grad_norm": 0.23046875,
217
+ "learning_rate": 1.9722837756298112e-05,
218
+ "loss": 0.011435374617576599,
219
+ "memory/device_reserved (GiB)": 33.38,
220
+ "memory/max_active (GiB)": 32.31,
221
+ "memory/max_allocated (GiB)": 32.31,
222
+ "ppl": 1.0115,
223
+ "step": 120,
224
+ "tokens/total": 983040,
225
+ "tokens/train_per_sec_per_gpu": 10.93,
226
+ "tokens/trainable": 425400
227
+ },
228
+ {
229
+ "epoch": 1.8309859154929577,
230
+ "grad_norm": 0.419921875,
231
+ "learning_rate": 1.9596187973841218e-05,
232
+ "loss": 0.007458872348070145,
233
+ "memory/device_reserved (GiB)": 33.38,
234
+ "memory/max_active (GiB)": 32.31,
235
+ "memory/max_allocated (GiB)": 32.31,
236
+ "ppl": 1.00749,
237
+ "step": 130,
238
+ "tokens/total": 1064960,
239
+ "tokens/train_per_sec_per_gpu": 10.88,
240
+ "tokens/trainable": 460753
241
+ },
242
+ {
243
+ "epoch": 1.971830985915493,
244
+ "grad_norm": 0.74609375,
245
+ "learning_rate": 1.9446347735010765e-05,
246
+ "loss": 0.00692850798368454,
247
+ "memory/device_reserved (GiB)": 33.38,
248
+ "memory/max_active (GiB)": 32.31,
249
+ "memory/max_allocated (GiB)": 32.31,
250
+ "ppl": 1.00695,
251
+ "step": 140,
252
+ "tokens/total": 1146880,
253
+ "tokens/train_per_sec_per_gpu": 11.05,
254
+ "tokens/trainable": 496328
255
+ },
256
+ {
257
+ "epoch": 2.028169014084507,
258
+ "eval_loss": 0.006739533506333828,
259
+ "eval_ppl": 1.00676,
260
+ "eval_runtime": 8.0865,
261
+ "eval_samples_per_second": 61.708,
262
+ "eval_steps_per_second": 61.708,
263
+ "memory/device_reserved (GiB)": 33.38,
264
+ "memory/max_active (GiB)": 32.31,
265
+ "memory/max_allocated (GiB)": 32.31,
266
+ "step": 144
267
+ },
268
+ {
269
+ "epoch": 2.112676056338028,
270
+ "grad_norm": 0.0927734375,
271
+ "learning_rate": 1.9273679148545246e-05,
272
+ "loss": 0.0028733495622873307,
273
+ "memory/device_reserved (GiB)": 33.38,
274
+ "memory/max_active (GiB)": 32.31,
275
+ "memory/max_allocated (GiB)": 32.31,
276
+ "ppl": 1.00288,
277
+ "step": 150,
278
+ "tokens/total": 1228800,
279
+ "tokens/train_per_sec_per_gpu": 11.15,
280
+ "tokens/trainable": 531980
281
+ },
282
+ {
283
+ "epoch": 2.2535211267605635,
284
+ "grad_norm": 0.1025390625,
285
+ "learning_rate": 1.9078599490901984e-05,
286
+ "loss": 0.0013103673234581947,
287
+ "memory/device_reserved (GiB)": 33.38,
288
+ "memory/max_active (GiB)": 32.31,
289
+ "memory/max_allocated (GiB)": 32.31,
290
+ "ppl": 1.00131,
291
+ "step": 160,
292
+ "tokens/total": 1310720,
293
+ "tokens/train_per_sec_per_gpu": 10.97,
294
+ "tokens/trainable": 567303
295
+ },
296
+ {
297
+ "epoch": 2.3943661971830985,
298
+ "grad_norm": 0.150390625,
299
+ "learning_rate": 1.8861580197853423e-05,
300
+ "loss": 0.0011377588845789433,
301
+ "memory/device_reserved (GiB)": 33.38,
302
+ "memory/max_active (GiB)": 32.31,
303
+ "memory/max_allocated (GiB)": 32.31,
304
+ "ppl": 1.00114,
305
+ "step": 170,
306
+ "tokens/total": 1392640,
307
+ "tokens/train_per_sec_per_gpu": 10.75,
308
+ "tokens/trainable": 602482
309
+ },
310
+ {
311
+ "epoch": 2.535211267605634,
312
+ "grad_norm": 0.087890625,
313
+ "learning_rate": 1.862314572520028e-05,
314
+ "loss": 0.001653754524886608,
315
+ "memory/device_reserved (GiB)": 33.38,
316
+ "memory/max_active (GiB)": 32.31,
317
+ "memory/max_allocated (GiB)": 32.31,
318
+ "ppl": 1.00166,
319
+ "step": 180,
320
+ "tokens/total": 1474560,
321
+ "tokens/train_per_sec_per_gpu": 11.3,
322
+ "tokens/trainable": 637743
323
+ },
324
+ {
325
+ "epoch": 2.535211267605634,
326
+ "eval_loss": 0.006014068610966206,
327
+ "eval_ppl": 1.00603,
328
+ "eval_runtime": 8.1497,
329
+ "eval_samples_per_second": 61.229,
330
+ "eval_steps_per_second": 61.229,
331
+ "memory/device_reserved (GiB)": 33.38,
332
+ "memory/max_active (GiB)": 24.06,
333
+ "memory/max_allocated (GiB)": 24.06,
334
+ "step": 180
335
+ },
336
+ {
337
+ "epoch": 2.676056338028169,
338
+ "grad_norm": 0.162109375,
339
+ "learning_rate": 1.8363872281354796e-05,
340
+ "loss": 0.0019721610471606253,
341
+ "memory/device_reserved (GiB)": 33.42,
342
+ "memory/max_active (GiB)": 32.31,
343
+ "memory/max_allocated (GiB)": 32.31,
344
+ "ppl": 1.00197,
345
+ "step": 190,
346
+ "tokens/total": 1556480,
347
+ "tokens/train_per_sec_per_gpu": 10.55,
348
+ "tokens/trainable": 673120
349
+ },
350
+ {
351
+ "epoch": 2.816901408450704,
352
+ "grad_norm": 0.146484375,
353
+ "learning_rate": 1.808438643485698e-05,
354
+ "loss": 0.00105431592091918,
355
+ "memory/device_reserved (GiB)": 33.42,
356
+ "memory/max_active (GiB)": 32.31,
357
+ "memory/max_allocated (GiB)": 32.31,
358
+ "ppl": 1.00105,
359
+ "step": 200,
360
+ "tokens/total": 1638400,
361
+ "tokens/train_per_sec_per_gpu": 11.08,
362
+ "tokens/trainable": 708273
363
+ },
364
+ {
365
+ "epoch": 2.9577464788732395,
366
+ "grad_norm": 0.0830078125,
367
+ "learning_rate": 1.7785363600188894e-05,
368
+ "loss": 0.0011002192273736,
369
+ "memory/device_reserved (GiB)": 33.42,
370
+ "memory/max_active (GiB)": 32.31,
371
+ "memory/max_allocated (GiB)": 32.31,
372
+ "ppl": 1.0011,
373
+ "step": 210,
374
+ "tokens/total": 1720320,
375
+ "tokens/train_per_sec_per_gpu": 10.99,
376
+ "tokens/trainable": 743557
377
+ },
378
+ {
379
+ "epoch": 3.0422535211267605,
380
+ "eval_loss": 0.0053863488137722015,
381
+ "eval_ppl": 1.0054,
382
+ "eval_runtime": 8.0906,
383
+ "eval_samples_per_second": 61.677,
384
+ "eval_steps_per_second": 61.677,
385
+ "memory/device_reserved (GiB)": 33.42,
386
+ "memory/max_active (GiB)": 32.31,
387
+ "memory/max_allocated (GiB)": 32.31,
388
+ "step": 216
389
+ },
390
+ {
391
+ "epoch": 3.0985915492957745,
392
+ "grad_norm": 0.134765625,
393
+ "learning_rate": 1.7467526405546344e-05,
394
+ "loss": 0.001148233562707901,
395
+ "memory/device_reserved (GiB)": 33.38,
396
+ "memory/max_active (GiB)": 32.31,
397
+ "memory/max_allocated (GiB)": 32.31,
398
+ "ppl": 1.00115,
399
+ "step": 220,
400
+ "tokens/total": 1802240,
401
+ "tokens/train_per_sec_per_gpu": 10.46,
402
+ "tokens/trainable": 779210
403
+ },
404
+ {
405
+ "epoch": 3.23943661971831,
406
+ "grad_norm": 0.0791015625,
407
+ "learning_rate": 1.7131642946512314e-05,
408
+ "loss": 0.0010407534427940845,
409
+ "memory/device_reserved (GiB)": 33.38,
410
+ "memory/max_active (GiB)": 32.31,
411
+ "memory/max_allocated (GiB)": 32.31,
412
+ "ppl": 1.00104,
413
+ "step": 230,
414
+ "tokens/total": 1884160,
415
+ "tokens/train_per_sec_per_gpu": 10.43,
416
+ "tokens/trainable": 814465
417
+ },
418
+ {
419
+ "epoch": 3.380281690140845,
420
+ "grad_norm": 0.035888671875,
421
+ "learning_rate": 1.6778524929852513e-05,
422
+ "loss": 0.0008773493580520153,
423
+ "memory/device_reserved (GiB)": 33.38,
424
+ "memory/max_active (GiB)": 32.31,
425
+ "memory/max_allocated (GiB)": 32.31,
426
+ "ppl": 1.00088,
427
+ "step": 240,
428
+ "tokens/total": 1966080,
429
+ "tokens/train_per_sec_per_gpu": 10.32,
430
+ "tokens/trainable": 849679
431
+ },
432
+ {
433
+ "epoch": 3.52112676056338,
434
+ "grad_norm": 0.11328125,
435
+ "learning_rate": 1.640902571191869e-05,
436
+ "loss": 0.00072392662987113,
437
+ "memory/device_reserved (GiB)": 33.38,
438
+ "memory/max_active (GiB)": 32.31,
439
+ "memory/max_allocated (GiB)": 32.31,
440
+ "ppl": 1.00072,
441
+ "step": 250,
442
+ "tokens/total": 2048000,
443
+ "tokens/train_per_sec_per_gpu": 11.0,
444
+ "tokens/trainable": 884891
445
+ },
446
+ {
447
+ "epoch": 3.5492957746478875,
448
+ "eval_loss": 0.005309262778609991,
449
+ "eval_ppl": 1.00532,
450
+ "eval_runtime": 8.2154,
451
+ "eval_samples_per_second": 60.74,
452
+ "eval_steps_per_second": 60.74,
453
+ "memory/device_reserved (GiB)": 33.38,
454
+ "memory/max_active (GiB)": 32.31,
455
+ "memory/max_allocated (GiB)": 32.31,
456
+ "step": 252
457
+ },
458
+ {
459
+ "epoch": 3.6619718309859155,
460
+ "grad_norm": 0.037109375,
461
+ "learning_rate": 1.6024038236400246e-05,
462
+ "loss": 0.0005804049782454967,
463
+ "memory/device_reserved (GiB)": 33.42,
464
+ "memory/max_active (GiB)": 32.31,
465
+ "memory/max_allocated (GiB)": 32.31,
466
+ "ppl": 1.00058,
467
+ "step": 260,
468
+ "tokens/total": 2129920,
469
+ "tokens/train_per_sec_per_gpu": 11.12,
470
+ "tokens/trainable": 920273
471
+ },
472
+ {
473
+ "epoch": 3.802816901408451,
474
+ "grad_norm": 0.022705078125,
475
+ "learning_rate": 1.562449287640781e-05,
476
+ "loss": 0.0007765244692564011,
477
+ "memory/device_reserved (GiB)": 33.42,
478
+ "memory/max_active (GiB)": 32.31,
479
+ "memory/max_allocated (GiB)": 32.31,
480
+ "ppl": 1.00078,
481
+ "step": 270,
482
+ "tokens/total": 2211840,
483
+ "tokens/train_per_sec_per_gpu": 10.51,
484
+ "tokens/trainable": 955808
485
+ },
486
+ {
487
+ "epoch": 3.943661971830986,
488
+ "grad_norm": 0.04833984375,
489
+ "learning_rate": 1.5211355186103655e-05,
490
+ "loss": 0.0007016387302428484,
491
+ "memory/device_reserved (GiB)": 33.42,
492
+ "memory/max_active (GiB)": 32.31,
493
+ "memory/max_allocated (GiB)": 32.31,
494
+ "ppl": 1.0007,
495
+ "step": 280,
496
+ "tokens/total": 2293760,
497
+ "tokens/train_per_sec_per_gpu": 10.15,
498
+ "tokens/trainable": 991166
499
+ },
500
+ {
501
+ "epoch": 4.056338028169014,
502
+ "eval_loss": 0.005245804321020842,
503
+ "eval_ppl": 1.00526,
504
+ "eval_runtime": 8.0041,
505
+ "eval_samples_per_second": 62.343,
506
+ "eval_steps_per_second": 62.343,
507
+ "memory/device_reserved (GiB)": 33.42,
508
+ "memory/max_active (GiB)": 32.31,
509
+ "memory/max_allocated (GiB)": 32.31,
510
+ "step": 288
511
+ },
512
+ {
513
+ "epoch": 4.084507042253521,
514
+ "grad_norm": 0.047607421875,
515
+ "learning_rate": 1.4785623567312492e-05,
516
+ "loss": 0.000780455581843853,
517
+ "memory/device_reserved (GiB)": 33.42,
518
+ "memory/max_active (GiB)": 32.31,
519
+ "memory/max_allocated (GiB)": 32.31,
520
+ "ppl": 1.00078,
521
+ "step": 290,
522
+ "tokens/total": 2375680,
523
+ "tokens/train_per_sec_per_gpu": 11.01,
524
+ "tokens/trainable": 1026615
525
+ },
526
+ {
527
+ "epoch": 4.225352112676056,
528
+ "grad_norm": 0.061767578125,
529
+ "learning_rate": 1.4348326856751496e-05,
530
+ "loss": 0.0005466958507895469,
531
+ "memory/device_reserved (GiB)": 33.42,
532
+ "memory/max_active (GiB)": 32.31,
533
+ "memory/max_allocated (GiB)": 32.31,
534
+ "ppl": 1.00055,
535
+ "step": 300,
536
+ "tokens/total": 2457600,
537
+ "tokens/train_per_sec_per_gpu": 10.99,
538
+ "tokens/trainable": 1062010
539
+ },
540
+ {
541
+ "epoch": 4.366197183098592,
542
+ "grad_norm": 0.0341796875,
543
+ "learning_rate": 1.3900521839710428e-05,
544
+ "loss": 0.0006260770373046398,
545
+ "memory/device_reserved (GiB)": 33.42,
546
+ "memory/max_active (GiB)": 32.31,
547
+ "memory/max_allocated (GiB)": 32.31,
548
+ "ppl": 1.00063,
549
+ "step": 310,
550
+ "tokens/total": 2539520,
551
+ "tokens/train_per_sec_per_gpu": 11.12,
552
+ "tokens/trainable": 1097505
553
+ },
554
+ {
555
+ "epoch": 4.507042253521127,
556
+ "grad_norm": 0.03857421875,
557
+ "learning_rate": 1.3443290696190335e-05,
558
+ "loss": 0.0005137504544109106,
559
+ "memory/device_reserved (GiB)": 33.42,
560
+ "memory/max_active (GiB)": 32.31,
561
+ "memory/max_allocated (GiB)": 32.31,
562
+ "ppl": 1.00051,
563
+ "step": 320,
564
+ "tokens/total": 2621440,
565
+ "tokens/train_per_sec_per_gpu": 10.32,
566
+ "tokens/trainable": 1132758
567
+ },
568
+ {
569
+ "epoch": 4.563380281690141,
570
+ "eval_loss": 0.005256574135273695,
571
+ "eval_ppl": 1.00527,
572
+ "eval_runtime": 8.1646,
573
+ "eval_samples_per_second": 61.118,
574
+ "eval_steps_per_second": 61.118,
575
+ "memory/device_reserved (GiB)": 33.42,
576
+ "memory/max_active (GiB)": 32.31,
577
+ "memory/max_allocated (GiB)": 32.31,
578
+ "step": 324
579
+ },
580
+ {
581
+ "epoch": 4.647887323943662,
582
+ "grad_norm": 0.046142578125,
583
+ "learning_rate": 1.2977738385672558e-05,
584
+ "loss": 0.0005908971652388573,
585
+ "memory/device_reserved (GiB)": 33.42,
586
+ "memory/max_active (GiB)": 32.31,
587
+ "memory/max_allocated (GiB)": 32.31,
588
+ "ppl": 1.00059,
589
+ "step": 330,
590
+ "tokens/total": 2703360,
591
+ "tokens/train_per_sec_per_gpu": 10.25,
592
+ "tokens/trainable": 1168161
593
+ },
594
+ {
595
+ "epoch": 4.788732394366197,
596
+ "grad_norm": 0.050537109375,
597
+ "learning_rate": 1.250498997683813e-05,
598
+ "loss": 0.0006883179303258657,
599
+ "memory/device_reserved (GiB)": 33.42,
600
+ "memory/max_active (GiB)": 32.31,
601
+ "memory/max_allocated (GiB)": 32.31,
602
+ "ppl": 1.00069,
603
+ "step": 340,
604
+ "tokens/total": 2785280,
605
+ "tokens/train_per_sec_per_gpu": 10.28,
606
+ "tokens/trainable": 1203434
607
+ },
608
+ {
609
+ "epoch": 4.929577464788732,
610
+ "grad_norm": 0.04736328125,
611
+ "learning_rate": 1.202618792869063e-05,
612
+ "loss": 0.0006954259239137173,
613
+ "memory/device_reserved (GiB)": 33.42,
614
+ "memory/max_active (GiB)": 32.31,
615
+ "memory/max_allocated (GiB)": 32.31,
616
+ "ppl": 1.0007,
617
+ "step": 350,
618
+ "tokens/total": 2867200,
619
+ "tokens/train_per_sec_per_gpu": 11.09,
620
+ "tokens/trainable": 1238829
621
+ },
622
+ {
623
+ "epoch": 5.070422535211268,
624
+ "grad_norm": 0.05517578125,
625
+ "learning_rate": 1.1542489329653024e-05,
626
+ "loss": 0.000603723106905818,
627
+ "memory/device_reserved (GiB)": 33.42,
628
+ "memory/max_active (GiB)": 32.31,
629
+ "memory/max_allocated (GiB)": 32.31,
630
+ "ppl": 1.0006,
631
+ "step": 360,
632
+ "tokens/total": 2949120,
633
+ "tokens/train_per_sec_per_gpu": 11.02,
634
+ "tokens/trainable": 1274110
635
+ },
636
+ {
637
+ "epoch": 5.070422535211268,
638
+ "eval_loss": 0.005255201831459999,
639
+ "eval_ppl": 1.00527,
640
+ "eval_runtime": 8.0148,
641
+ "eval_samples_per_second": 62.26,
642
+ "eval_steps_per_second": 62.26,
643
+ "memory/device_reserved (GiB)": 33.42,
644
+ "memory/max_active (GiB)": 24.06,
645
+ "memory/max_allocated (GiB)": 24.06,
646
+ "step": 360
647
+ },
648
+ {
649
+ "epoch": 5.211267605633803,
650
+ "grad_norm": 0.0615234375,
651
+ "learning_rate": 1.1055063101310581e-05,
652
+ "loss": 0.0005278075113892555,
653
+ "memory/device_reserved (GiB)": 33.38,
654
+ "memory/max_active (GiB)": 32.31,
655
+ "memory/max_allocated (GiB)": 32.31,
656
+ "ppl": 1.00053,
657
+ "step": 370,
658
+ "tokens/total": 3031040,
659
+ "tokens/train_per_sec_per_gpu": 11.13,
660
+ "tokens/trainable": 1309814
661
+ },
662
+ {
663
+ "epoch": 5.352112676056338,
664
+ "grad_norm": 0.037841796875,
665
+ "learning_rate": 1.0565087173557396e-05,
666
+ "loss": 0.0005071659106761217,
667
+ "memory/device_reserved (GiB)": 33.38,
668
+ "memory/max_active (GiB)": 32.31,
669
+ "memory/max_allocated (GiB)": 32.31,
670
+ "ppl": 1.00051,
671
+ "step": 380,
672
+ "tokens/total": 3112960,
673
+ "tokens/train_per_sec_per_gpu": 11.21,
674
+ "tokens/trainable": 1345167
675
+ },
676
+ {
677
+ "epoch": 5.492957746478873,
678
+ "grad_norm": 0.02880859375,
679
+ "learning_rate": 1.0073745637973125e-05,
680
+ "loss": 0.0005827666725963354,
681
+ "memory/device_reserved (GiB)": 33.38,
682
+ "memory/max_active (GiB)": 32.31,
683
+ "memory/max_allocated (GiB)": 32.31,
684
+ "ppl": 1.00058,
685
+ "step": 390,
686
+ "tokens/total": 3194880,
687
+ "tokens/train_per_sec_per_gpu": 11.14,
688
+ "tokens/trainable": 1380603
689
+ },
690
+ {
691
+ "epoch": 5.577464788732394,
692
+ "eval_loss": 0.005226313602179289,
693
+ "eval_ppl": 1.00524,
694
+ "eval_runtime": 8.1606,
695
+ "eval_samples_per_second": 61.147,
696
+ "eval_steps_per_second": 61.147,
697
+ "memory/device_reserved (GiB)": 33.38,
698
+ "memory/max_active (GiB)": 32.31,
699
+ "memory/max_allocated (GiB)": 32.31,
700
+ "step": 396
701
+ },
702
+ {
703
+ "epoch": 5.633802816901408,
704
+ "grad_norm": 0.04150390625,
705
+ "learning_rate": 9.582225886309217e-06,
706
+ "loss": 0.0006697049364447593,
707
+ "memory/device_reserved (GiB)": 33.42,
708
+ "memory/max_active (GiB)": 32.31,
709
+ "memory/max_allocated (GiB)": 32.31,
710
+ "ppl": 1.00067,
711
+ "step": 400,
712
+ "tokens/total": 3276800,
713
+ "tokens/train_per_sec_per_gpu": 9.71,
714
+ "tokens/trainable": 1415822
715
+ },
716
+ {
717
+ "epoch": 5.774647887323944,
718
+ "grad_norm": 0.04833984375,
719
+ "learning_rate": 9.091715740999829e-06,
720
+ "loss": 0.0006844023242592812,
721
+ "memory/device_reserved (GiB)": 33.42,
722
+ "memory/max_active (GiB)": 32.31,
723
+ "memory/max_allocated (GiB)": 32.31,
724
+ "ppl": 1.00068,
725
+ "step": 410,
726
+ "tokens/total": 3358720,
727
+ "tokens/train_per_sec_per_gpu": 10.39,
728
+ "tokens/trainable": 1451140
729
+ },
730
+ {
731
+ "epoch": 5.915492957746479,
732
+ "grad_norm": 0.0478515625,
733
+ "learning_rate": 8.60340058463194e-06,
734
+ "loss": 0.0004917650483548641,
735
+ "memory/device_reserved (GiB)": 33.42,
736
+ "memory/max_active (GiB)": 32.31,
737
+ "memory/max_allocated (GiB)": 32.31,
738
+ "ppl": 1.00049,
739
+ "step": 420,
740
+ "tokens/total": 3440640,
741
+ "tokens/train_per_sec_per_gpu": 11.04,
742
+ "tokens/trainable": 1486616
743
+ },
744
+ {
745
+ "epoch": 6.056338028169014,
746
+ "grad_norm": 0.047607421875,
747
+ "learning_rate": 8.118460495311687e-06,
748
+ "loss": 0.0005019189789891243,
749
+ "memory/device_reserved (GiB)": 33.42,
750
+ "memory/max_active (GiB)": 32.31,
751
+ "memory/max_allocated (GiB)": 32.31,
752
+ "ppl": 1.0005,
753
+ "step": 430,
754
+ "tokens/total": 3522560,
755
+ "tokens/train_per_sec_per_gpu": 9.55,
756
+ "tokens/trainable": 1522001
757
+ },
758
+ {
759
+ "epoch": 6.084507042253521,
760
+ "eval_loss": 0.005209571216255426,
761
+ "eval_ppl": 1.00522,
762
+ "eval_runtime": 8.0091,
763
+ "eval_samples_per_second": 62.304,
764
+ "eval_steps_per_second": 62.304,
765
+ "memory/device_reserved (GiB)": 33.42,
766
+ "memory/max_active (GiB)": 32.31,
767
+ "memory/max_allocated (GiB)": 32.31,
768
+ "step": 432
769
+ },
770
+ {
771
+ "epoch": 6.197183098591549,
772
+ "grad_norm": 0.134765625,
773
+ "learning_rate": 7.638067394849672e-06,
774
+ "loss": 0.000784461572766304,
775
+ "memory/device_reserved (GiB)": 33.38,
776
+ "memory/max_active (GiB)": 32.31,
777
+ "memory/max_allocated (GiB)": 32.31,
778
+ "ppl": 1.00078,
779
+ "step": 440,
780
+ "tokens/total": 3604480,
781
+ "tokens/train_per_sec_per_gpu": 11.09,
782
+ "tokens/trainable": 1557368
783
+ },
784
+ {
785
+ "epoch": 6.338028169014084,
786
+ "grad_norm": 0.05712890625,
787
+ "learning_rate": 7.163382216657033e-06,
788
+ "loss": 0.0005743904504925013,
789
+ "memory/device_reserved (GiB)": 33.38,
790
+ "memory/max_active (GiB)": 32.31,
791
+ "memory/max_allocated (GiB)": 32.31,
792
+ "ppl": 1.00057,
793
+ "step": 450,
794
+ "tokens/total": 3686400,
795
+ "tokens/train_per_sec_per_gpu": 10.46,
796
+ "tokens/trainable": 1592826
797
+ },
798
+ {
799
+ "epoch": 6.47887323943662,
800
+ "grad_norm": 0.03076171875,
801
+ "learning_rate": 6.6955521001964526e-06,
802
+ "loss": 0.0004704087972640991,
803
+ "memory/device_reserved (GiB)": 33.38,
804
+ "memory/max_active (GiB)": 32.31,
805
+ "memory/max_allocated (GiB)": 32.31,
806
+ "ppl": 1.00047,
807
+ "step": 460,
808
+ "tokens/total": 3768320,
809
+ "tokens/train_per_sec_per_gpu": 10.41,
810
+ "tokens/trainable": 1628143
811
+ },
812
+ {
813
+ "epoch": 6.591549295774648,
814
+ "eval_loss": 0.005213646683841944,
815
+ "eval_ppl": 1.00523,
816
+ "eval_runtime": 8.1767,
817
+ "eval_samples_per_second": 61.027,
818
+ "eval_steps_per_second": 61.027,
819
+ "memory/device_reserved (GiB)": 33.38,
820
+ "memory/max_active (GiB)": 32.31,
821
+ "memory/max_allocated (GiB)": 32.31,
822
+ "step": 468
823
+ },
824
+ {
825
+ "epoch": 6.619718309859155,
826
+ "grad_norm": 0.04736328125,
827
+ "learning_rate": 6.2357076187680325e-06,
828
+ "loss": 0.0004482423886656761,
829
+ "memory/device_reserved (GiB)": 33.42,
830
+ "memory/max_active (GiB)": 32.31,
831
+ "memory/max_allocated (GiB)": 32.31,
832
+ "ppl": 1.00045,
833
+ "step": 470,
834
+ "tokens/total": 3850240,
835
+ "tokens/train_per_sec_per_gpu": 10.41,
836
+ "tokens/trainable": 1663545
837
+ },
838
+ {
839
+ "epoch": 6.76056338028169,
840
+ "grad_norm": 0.0380859375,
841
+ "learning_rate": 5.78496004732952e-06,
842
+ "loss": 0.0004957062192261219,
843
+ "memory/device_reserved (GiB)": 33.42,
844
+ "memory/max_active (GiB)": 32.31,
845
+ "memory/max_allocated (GiB)": 32.31,
846
+ "ppl": 1.0005,
847
+ "step": 480,
848
+ "tokens/total": 3932160,
849
+ "tokens/train_per_sec_per_gpu": 10.4,
850
+ "tokens/trainable": 1699181
851
+ },
852
+ {
853
+ "epoch": 6.901408450704225,
854
+ "grad_norm": 0.0361328125,
855
+ "learning_rate": 5.344398676953526e-06,
856
+ "loss": 0.000544667523354292,
857
+ "memory/device_reserved (GiB)": 33.42,
858
+ "memory/max_active (GiB)": 32.31,
859
+ "memory/max_allocated (GiB)": 32.31,
860
+ "ppl": 1.00054,
861
+ "step": 490,
862
+ "tokens/total": 4014080,
863
+ "tokens/train_per_sec_per_gpu": 10.21,
864
+ "tokens/trainable": 1734683
865
+ },
866
+ {
867
+ "epoch": 7.042253521126761,
868
+ "grad_norm": 0.03759765625,
869
+ "learning_rate": 4.915088182411674e-06,
870
+ "loss": 0.0004880569409579039,
871
+ "memory/device_reserved (GiB)": 33.42,
872
+ "memory/max_active (GiB)": 32.31,
873
+ "memory/max_allocated (GiB)": 32.31,
874
+ "ppl": 1.00049,
875
+ "step": 500,
876
+ "tokens/total": 4096000,
877
+ "tokens/train_per_sec_per_gpu": 11.44,
878
+ "tokens/trainable": 1770105
879
+ },
880
+ {
881
+ "epoch": 7.098591549295775,
882
+ "eval_loss": 0.00523661682382226,
883
+ "eval_ppl": 1.00525,
884
+ "eval_runtime": 8.0317,
885
+ "eval_samples_per_second": 62.129,
886
+ "eval_steps_per_second": 62.129,
887
+ "memory/device_reserved (GiB)": 33.42,
888
+ "memory/max_active (GiB)": 32.31,
889
+ "memory/max_allocated (GiB)": 32.31,
890
+ "step": 504
891
+ },
892
+ {
893
+ "epoch": 7.183098591549296,
894
+ "grad_norm": 0.0257568359375,
895
+ "learning_rate": 4.498066049247344e-06,
896
+ "loss": 0.0005014274269342422,
897
+ "memory/device_reserved (GiB)": 33.38,
898
+ "memory/max_active (GiB)": 32.31,
899
+ "memory/max_allocated (GiB)": 32.31,
900
+ "ppl": 1.0005,
901
+ "step": 510,
902
+ "tokens/total": 4177920,
903
+ "tokens/train_per_sec_per_gpu": 10.53,
904
+ "tokens/trainable": 1805486
905
+ },
906
+ {
907
+ "epoch": 7.323943661971831,
908
+ "grad_norm": 0.03466796875,
909
+ "learning_rate": 4.094340066554742e-06,
910
+ "loss": 0.0006260806694626808,
911
+ "memory/device_reserved (GiB)": 33.38,
912
+ "memory/max_active (GiB)": 32.31,
913
+ "memory/max_allocated (GiB)": 32.31,
914
+ "ppl": 1.00063,
915
+ "step": 520,
916
+ "tokens/total": 4259840,
917
+ "tokens/train_per_sec_per_gpu": 10.8,
918
+ "tokens/trainable": 1840987
919
+ },
920
+ {
921
+ "epoch": 7.464788732394366,
922
+ "grad_norm": 0.036865234375,
923
+ "learning_rate": 3.7048858915233665e-06,
924
+ "loss": 0.0005529518239200115,
925
+ "memory/device_reserved (GiB)": 33.38,
926
+ "memory/max_active (GiB)": 32.31,
927
+ "memory/max_allocated (GiB)": 32.31,
928
+ "ppl": 1.00055,
929
+ "step": 530,
930
+ "tokens/total": 4341760,
931
+ "tokens/train_per_sec_per_gpu": 10.79,
932
+ "tokens/trainable": 1876364
933
+ },
934
+ {
935
+ "epoch": 7.605633802816901,
936
+ "grad_norm": 0.034912109375,
937
+ "learning_rate": 3.330644691633492e-06,
938
+ "loss": 0.0005182296968996525,
939
+ "memory/device_reserved (GiB)": 33.38,
940
+ "memory/max_active (GiB)": 32.31,
941
+ "memory/max_allocated (GiB)": 32.31,
942
+ "ppl": 1.00052,
943
+ "step": 540,
944
+ "tokens/total": 4423680,
945
+ "tokens/train_per_sec_per_gpu": 10.9,
946
+ "tokens/trainable": 1911823
947
+ },
948
+ {
949
+ "epoch": 7.605633802816901,
950
+ "eval_loss": 0.005228586029261351,
951
+ "eval_ppl": 1.00524,
952
+ "eval_runtime": 8.2216,
953
+ "eval_samples_per_second": 60.694,
954
+ "eval_steps_per_second": 60.694,
955
+ "memory/device_reserved (GiB)": 33.38,
956
+ "memory/max_active (GiB)": 24.06,
957
+ "memory/max_allocated (GiB)": 24.06,
958
+ "step": 540
959
+ },
960
+ {
961
+ "epoch": 7.746478873239437,
962
+ "grad_norm": 0.025634765625,
963
+ "learning_rate": 2.9725208702005736e-06,
964
+ "loss": 0.0005175352562218904,
965
+ "memory/device_reserved (GiB)": 33.42,
966
+ "memory/max_active (GiB)": 32.31,
967
+ "memory/max_allocated (GiB)": 32.31,
968
+ "ppl": 1.00052,
969
+ "step": 550,
970
+ "tokens/total": 4505600,
971
+ "tokens/train_per_sec_per_gpu": 11.35,
972
+ "tokens/trainable": 1947212
973
+ },
974
+ {
975
+ "epoch": 7.887323943661972,
976
+ "grad_norm": 0.0341796875,
977
+ "learning_rate": 2.631379880765107e-06,
978
+ "loss": 0.0005252756644040346,
979
+ "memory/device_reserved (GiB)": 33.42,
980
+ "memory/max_active (GiB)": 32.31,
981
+ "memory/max_allocated (GiB)": 32.31,
982
+ "ppl": 1.00053,
983
+ "step": 560,
984
+ "tokens/total": 4587520,
985
+ "tokens/train_per_sec_per_gpu": 11.28,
986
+ "tokens/trainable": 1982716
987
+ }
988
+ ],
989
+ "logging_steps": 10,
990
+ "max_steps": 710,
991
+ "num_input_tokens_seen": 0,
992
+ "num_train_epochs": 10,
993
+ "save_steps": 71,
994
+ "stateful_callbacks": {
995
+ "TrainerControl": {
996
+ "args": {
997
+ "should_epoch_stop": false,
998
+ "should_evaluate": false,
999
+ "should_log": false,
1000
+ "should_save": true,
1001
+ "should_training_stop": false
1002
+ },
1003
+ "attributes": {}
1004
+ }
1005
+ },
1006
+ "total_flos": 1.014416070892585e+17,
1007
+ "train_batch_size": 1,
1008
+ "trial_name": null,
1009
+ "trial_params": null
1010
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-568/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c56b5e3616cf3d2a8bf8aabf031ff524a587484ec751048484926648eb0ad82
3
+ size 7121
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/chat_template.jinja ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '
2
+ ' + message['content'] + '<|im_end|>' + '
3
+ '}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
4
+ ' }}{% endif %}
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "bfloat16",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2560,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 9728,
15
+ "layer_types": [
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention"
52
+ ],
53
+ "max_position_embeddings": 262144,
54
+ "max_window_layers": 36,
55
+ "model_type": "qwen3",
56
+ "num_attention_heads": 32,
57
+ "num_hidden_layers": 36,
58
+ "num_key_value_heads": 8,
59
+ "pad_token_id": 151643,
60
+ "rms_norm_eps": 1e-06,
61
+ "rope_parameters": {
62
+ "rope_theta": 5000000,
63
+ "rope_type": "default"
64
+ },
65
+ "sliding_window": null,
66
+ "tie_word_embeddings": true,
67
+ "transformers_version": "5.0.0",
68
+ "use_cache": false,
69
+ "use_sliding_window": false,
70
+ "vocab_size": 151936
71
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "temperature": 0.7,
9
+ "top_k": 20,
10
+ "top_p": 0.8,
11
+ "transformers_version": "5.0.0"
12
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5bea0c4b1228c29f5d660021b0d1fb69f3b261de76f875901b07a563922bbeaa
3
+ size 8822894520
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a1e0db2a84b488e8b75beb8d41bc4aef7605f1dcbcff30ef971590fed07cf5d
3
+ size 16090225449
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8bd75f3b287b15c92e6927ee3d2a9e3e9a7c2ea768e141eb8d1ab87cfbf9392a
3
+ size 14645
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb76ee2909d16ed6f82f442fd3c4f64612245c92ced7758b90e37ec342c67a9d
3
+ size 1465
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
3
+ size 11422650
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokenizer_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "model_max_length": 1010000,
25
+ "pad_token": "<|endoftext|>",
26
+ "split_special_tokens": false,
27
+ "tokenizer_class": "Qwen2Tokenizer",
28
+ "unk_token": null
29
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/tokens_state. ADDED
@@ -0,0 +1 @@
 
 
1
+ {"total": 5234688, "trainable": 2262545}
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/trainer_state.json ADDED
@@ -0,0 +1,1132 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 432,
3
+ "best_metric": 0.005209571216255426,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 9.0,
6
+ "eval_steps": 36,
7
+ "global_step": 639,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0,
14
+ "eval_loss": 6.340579032897949,
15
+ "eval_ppl": 567.1246,
16
+ "eval_runtime": 9.2875,
17
+ "eval_samples_per_second": 53.728,
18
+ "eval_steps_per_second": 53.728,
19
+ "memory/device_reserved (GiB)": 9.68,
20
+ "memory/max_active (GiB)": 8.96,
21
+ "memory/max_allocated (GiB)": 8.96,
22
+ "step": 0
23
+ },
24
+ {
25
+ "epoch": 0.14084507042253522,
26
+ "grad_norm": 48.75,
27
+ "learning_rate": 2.535211267605634e-06,
28
+ "loss": 6.281451797485351,
29
+ "memory/device_reserved (GiB)": 33.61,
30
+ "memory/max_active (GiB)": 32.32,
31
+ "memory/max_allocated (GiB)": 32.32,
32
+ "ppl": 534.56418,
33
+ "step": 10,
34
+ "tokens/total": 81920,
35
+ "tokens/train_per_sec_per_gpu": 9.82,
36
+ "tokens/trainable": 35649
37
+ },
38
+ {
39
+ "epoch": 0.28169014084507044,
40
+ "grad_norm": 43.0,
41
+ "learning_rate": 5.352112676056338e-06,
42
+ "loss": 5.6617687225341795,
43
+ "memory/device_reserved (GiB)": 33.61,
44
+ "memory/max_active (GiB)": 32.32,
45
+ "memory/max_allocated (GiB)": 32.32,
46
+ "ppl": 287.65698,
47
+ "step": 20,
48
+ "tokens/total": 163840,
49
+ "tokens/train_per_sec_per_gpu": 10.46,
50
+ "tokens/trainable": 71196
51
+ },
52
+ {
53
+ "epoch": 0.4225352112676056,
54
+ "grad_norm": 26.875,
55
+ "learning_rate": 8.169014084507043e-06,
56
+ "loss": 3.820334243774414,
57
+ "memory/device_reserved (GiB)": 33.61,
58
+ "memory/max_active (GiB)": 32.32,
59
+ "memory/max_allocated (GiB)": 32.32,
60
+ "ppl": 45.61945,
61
+ "step": 30,
62
+ "tokens/total": 245760,
63
+ "tokens/train_per_sec_per_gpu": 10.53,
64
+ "tokens/trainable": 106543
65
+ },
66
+ {
67
+ "epoch": 0.5070422535211268,
68
+ "eval_loss": 1.091687560081482,
69
+ "eval_ppl": 2.9793,
70
+ "eval_runtime": 8.0764,
71
+ "eval_samples_per_second": 61.785,
72
+ "eval_steps_per_second": 61.785,
73
+ "memory/device_reserved (GiB)": 33.61,
74
+ "memory/max_active (GiB)": 32.32,
75
+ "memory/max_allocated (GiB)": 32.32,
76
+ "step": 36
77
+ },
78
+ {
79
+ "epoch": 0.5633802816901409,
80
+ "grad_norm": 6.5625,
81
+ "learning_rate": 1.0985915492957748e-05,
82
+ "loss": 1.4444393157958983,
83
+ "memory/device_reserved (GiB)": 33.42,
84
+ "memory/max_active (GiB)": 32.31,
85
+ "memory/max_allocated (GiB)": 32.31,
86
+ "ppl": 4.23947,
87
+ "step": 40,
88
+ "tokens/total": 327680,
89
+ "tokens/train_per_sec_per_gpu": 10.34,
90
+ "tokens/trainable": 141719
91
+ },
92
+ {
93
+ "epoch": 0.704225352112676,
94
+ "grad_norm": 2.484375,
95
+ "learning_rate": 1.380281690140845e-05,
96
+ "loss": 0.4776052951812744,
97
+ "memory/device_reserved (GiB)": 33.42,
98
+ "memory/max_active (GiB)": 32.31,
99
+ "memory/max_allocated (GiB)": 32.31,
100
+ "ppl": 1.61221,
101
+ "step": 50,
102
+ "tokens/total": 409600,
103
+ "tokens/train_per_sec_per_gpu": 10.24,
104
+ "tokens/trainable": 177240
105
+ },
106
+ {
107
+ "epoch": 0.8450704225352113,
108
+ "grad_norm": 1.421875,
109
+ "learning_rate": 1.6619718309859155e-05,
110
+ "loss": 0.15568361282348633,
111
+ "memory/device_reserved (GiB)": 33.42,
112
+ "memory/max_active (GiB)": 32.31,
113
+ "memory/max_allocated (GiB)": 32.31,
114
+ "ppl": 1.16846,
115
+ "step": 60,
116
+ "tokens/total": 491520,
117
+ "tokens/train_per_sec_per_gpu": 10.55,
118
+ "tokens/trainable": 212815
119
+ },
120
+ {
121
+ "epoch": 0.9859154929577465,
122
+ "grad_norm": 0.95703125,
123
+ "learning_rate": 1.943661971830986e-05,
124
+ "loss": 0.06936523914337159,
125
+ "memory/device_reserved (GiB)": 33.42,
126
+ "memory/max_active (GiB)": 32.31,
127
+ "memory/max_allocated (GiB)": 32.31,
128
+ "ppl": 1.07183,
129
+ "step": 70,
130
+ "tokens/total": 573440,
131
+ "tokens/train_per_sec_per_gpu": 10.42,
132
+ "tokens/trainable": 248141
133
+ },
134
+ {
135
+ "epoch": 1.0140845070422535,
136
+ "eval_loss": 0.04077093303203583,
137
+ "eval_ppl": 1.04161,
138
+ "eval_runtime": 8.0461,
139
+ "eval_samples_per_second": 62.018,
140
+ "eval_steps_per_second": 62.018,
141
+ "memory/device_reserved (GiB)": 33.42,
142
+ "memory/max_active (GiB)": 32.31,
143
+ "memory/max_allocated (GiB)": 32.31,
144
+ "step": 72
145
+ },
146
+ {
147
+ "epoch": 1.1267605633802817,
148
+ "grad_norm": 0.859375,
149
+ "learning_rate": 1.9992266216318037e-05,
150
+ "loss": 0.03334630727767944,
151
+ "memory/device_reserved (GiB)": 33.42,
152
+ "memory/max_active (GiB)": 32.31,
153
+ "memory/max_allocated (GiB)": 32.31,
154
+ "ppl": 1.03391,
155
+ "step": 80,
156
+ "tokens/total": 655360,
157
+ "tokens/train_per_sec_per_gpu": 10.95,
158
+ "tokens/trainable": 283579
159
+ },
160
+ {
161
+ "epoch": 1.267605633802817,
162
+ "grad_norm": 0.94921875,
163
+ "learning_rate": 1.996086822074945e-05,
164
+ "loss": 0.020297910273075103,
165
+ "memory/device_reserved (GiB)": 33.42,
166
+ "memory/max_active (GiB)": 32.31,
167
+ "memory/max_allocated (GiB)": 32.31,
168
+ "ppl": 1.02051,
169
+ "step": 90,
170
+ "tokens/total": 737280,
171
+ "tokens/train_per_sec_per_gpu": 10.49,
172
+ "tokens/trainable": 319131
173
+ },
174
+ {
175
+ "epoch": 1.408450704225352,
176
+ "grad_norm": 0.482421875,
177
+ "learning_rate": 1.9905398470798207e-05,
178
+ "loss": 0.014508798718452454,
179
+ "memory/device_reserved (GiB)": 33.42,
180
+ "memory/max_active (GiB)": 32.31,
181
+ "memory/max_allocated (GiB)": 32.31,
182
+ "ppl": 1.01461,
183
+ "step": 100,
184
+ "tokens/total": 819200,
185
+ "tokens/train_per_sec_per_gpu": 10.25,
186
+ "tokens/trainable": 354528
187
+ },
188
+ {
189
+ "epoch": 1.5211267605633803,
190
+ "eval_loss": 0.011320236138999462,
191
+ "eval_ppl": 1.01138,
192
+ "eval_runtime": 8.142,
193
+ "eval_samples_per_second": 61.287,
194
+ "eval_steps_per_second": 61.287,
195
+ "memory/device_reserved (GiB)": 33.42,
196
+ "memory/max_active (GiB)": 32.31,
197
+ "memory/max_allocated (GiB)": 32.31,
198
+ "step": 108
199
+ },
200
+ {
201
+ "epoch": 1.5492957746478875,
202
+ "grad_norm": 0.58203125,
203
+ "learning_rate": 1.9825991016445387e-05,
204
+ "loss": 0.013250131905078889,
205
+ "memory/device_reserved (GiB)": 33.38,
206
+ "memory/max_active (GiB)": 32.31,
207
+ "memory/max_allocated (GiB)": 32.31,
208
+ "ppl": 1.01334,
209
+ "step": 110,
210
+ "tokens/total": 901120,
211
+ "tokens/train_per_sec_per_gpu": 10.85,
212
+ "tokens/trainable": 389882
213
+ },
214
+ {
215
+ "epoch": 1.6901408450704225,
216
+ "grad_norm": 0.23046875,
217
+ "learning_rate": 1.9722837756298112e-05,
218
+ "loss": 0.011435374617576599,
219
+ "memory/device_reserved (GiB)": 33.38,
220
+ "memory/max_active (GiB)": 32.31,
221
+ "memory/max_allocated (GiB)": 32.31,
222
+ "ppl": 1.0115,
223
+ "step": 120,
224
+ "tokens/total": 983040,
225
+ "tokens/train_per_sec_per_gpu": 10.93,
226
+ "tokens/trainable": 425400
227
+ },
228
+ {
229
+ "epoch": 1.8309859154929577,
230
+ "grad_norm": 0.419921875,
231
+ "learning_rate": 1.9596187973841218e-05,
232
+ "loss": 0.007458872348070145,
233
+ "memory/device_reserved (GiB)": 33.38,
234
+ "memory/max_active (GiB)": 32.31,
235
+ "memory/max_allocated (GiB)": 32.31,
236
+ "ppl": 1.00749,
237
+ "step": 130,
238
+ "tokens/total": 1064960,
239
+ "tokens/train_per_sec_per_gpu": 10.88,
240
+ "tokens/trainable": 460753
241
+ },
242
+ {
243
+ "epoch": 1.971830985915493,
244
+ "grad_norm": 0.74609375,
245
+ "learning_rate": 1.9446347735010765e-05,
246
+ "loss": 0.00692850798368454,
247
+ "memory/device_reserved (GiB)": 33.38,
248
+ "memory/max_active (GiB)": 32.31,
249
+ "memory/max_allocated (GiB)": 32.31,
250
+ "ppl": 1.00695,
251
+ "step": 140,
252
+ "tokens/total": 1146880,
253
+ "tokens/train_per_sec_per_gpu": 11.05,
254
+ "tokens/trainable": 496328
255
+ },
256
+ {
257
+ "epoch": 2.028169014084507,
258
+ "eval_loss": 0.006739533506333828,
259
+ "eval_ppl": 1.00676,
260
+ "eval_runtime": 8.0865,
261
+ "eval_samples_per_second": 61.708,
262
+ "eval_steps_per_second": 61.708,
263
+ "memory/device_reserved (GiB)": 33.38,
264
+ "memory/max_active (GiB)": 32.31,
265
+ "memory/max_allocated (GiB)": 32.31,
266
+ "step": 144
267
+ },
268
+ {
269
+ "epoch": 2.112676056338028,
270
+ "grad_norm": 0.0927734375,
271
+ "learning_rate": 1.9273679148545246e-05,
272
+ "loss": 0.0028733495622873307,
273
+ "memory/device_reserved (GiB)": 33.38,
274
+ "memory/max_active (GiB)": 32.31,
275
+ "memory/max_allocated (GiB)": 32.31,
276
+ "ppl": 1.00288,
277
+ "step": 150,
278
+ "tokens/total": 1228800,
279
+ "tokens/train_per_sec_per_gpu": 11.15,
280
+ "tokens/trainable": 531980
281
+ },
282
+ {
283
+ "epoch": 2.2535211267605635,
284
+ "grad_norm": 0.1025390625,
285
+ "learning_rate": 1.9078599490901984e-05,
286
+ "loss": 0.0013103673234581947,
287
+ "memory/device_reserved (GiB)": 33.38,
288
+ "memory/max_active (GiB)": 32.31,
289
+ "memory/max_allocated (GiB)": 32.31,
290
+ "ppl": 1.00131,
291
+ "step": 160,
292
+ "tokens/total": 1310720,
293
+ "tokens/train_per_sec_per_gpu": 10.97,
294
+ "tokens/trainable": 567303
295
+ },
296
+ {
297
+ "epoch": 2.3943661971830985,
298
+ "grad_norm": 0.150390625,
299
+ "learning_rate": 1.8861580197853423e-05,
300
+ "loss": 0.0011377588845789433,
301
+ "memory/device_reserved (GiB)": 33.38,
302
+ "memory/max_active (GiB)": 32.31,
303
+ "memory/max_allocated (GiB)": 32.31,
304
+ "ppl": 1.00114,
305
+ "step": 170,
306
+ "tokens/total": 1392640,
307
+ "tokens/train_per_sec_per_gpu": 10.75,
308
+ "tokens/trainable": 602482
309
+ },
310
+ {
311
+ "epoch": 2.535211267605634,
312
+ "grad_norm": 0.087890625,
313
+ "learning_rate": 1.862314572520028e-05,
314
+ "loss": 0.001653754524886608,
315
+ "memory/device_reserved (GiB)": 33.38,
316
+ "memory/max_active (GiB)": 32.31,
317
+ "memory/max_allocated (GiB)": 32.31,
318
+ "ppl": 1.00166,
319
+ "step": 180,
320
+ "tokens/total": 1474560,
321
+ "tokens/train_per_sec_per_gpu": 11.3,
322
+ "tokens/trainable": 637743
323
+ },
324
+ {
325
+ "epoch": 2.535211267605634,
326
+ "eval_loss": 0.006014068610966206,
327
+ "eval_ppl": 1.00603,
328
+ "eval_runtime": 8.1497,
329
+ "eval_samples_per_second": 61.229,
330
+ "eval_steps_per_second": 61.229,
331
+ "memory/device_reserved (GiB)": 33.38,
332
+ "memory/max_active (GiB)": 24.06,
333
+ "memory/max_allocated (GiB)": 24.06,
334
+ "step": 180
335
+ },
336
+ {
337
+ "epoch": 2.676056338028169,
338
+ "grad_norm": 0.162109375,
339
+ "learning_rate": 1.8363872281354796e-05,
340
+ "loss": 0.0019721610471606253,
341
+ "memory/device_reserved (GiB)": 33.42,
342
+ "memory/max_active (GiB)": 32.31,
343
+ "memory/max_allocated (GiB)": 32.31,
344
+ "ppl": 1.00197,
345
+ "step": 190,
346
+ "tokens/total": 1556480,
347
+ "tokens/train_per_sec_per_gpu": 10.55,
348
+ "tokens/trainable": 673120
349
+ },
350
+ {
351
+ "epoch": 2.816901408450704,
352
+ "grad_norm": 0.146484375,
353
+ "learning_rate": 1.808438643485698e-05,
354
+ "loss": 0.00105431592091918,
355
+ "memory/device_reserved (GiB)": 33.42,
356
+ "memory/max_active (GiB)": 32.31,
357
+ "memory/max_allocated (GiB)": 32.31,
358
+ "ppl": 1.00105,
359
+ "step": 200,
360
+ "tokens/total": 1638400,
361
+ "tokens/train_per_sec_per_gpu": 11.08,
362
+ "tokens/trainable": 708273
363
+ },
364
+ {
365
+ "epoch": 2.9577464788732395,
366
+ "grad_norm": 0.0830078125,
367
+ "learning_rate": 1.7785363600188894e-05,
368
+ "loss": 0.0011002192273736,
369
+ "memory/device_reserved (GiB)": 33.42,
370
+ "memory/max_active (GiB)": 32.31,
371
+ "memory/max_allocated (GiB)": 32.31,
372
+ "ppl": 1.0011,
373
+ "step": 210,
374
+ "tokens/total": 1720320,
375
+ "tokens/train_per_sec_per_gpu": 10.99,
376
+ "tokens/trainable": 743557
377
+ },
378
+ {
379
+ "epoch": 3.0422535211267605,
380
+ "eval_loss": 0.0053863488137722015,
381
+ "eval_ppl": 1.0054,
382
+ "eval_runtime": 8.0906,
383
+ "eval_samples_per_second": 61.677,
384
+ "eval_steps_per_second": 61.677,
385
+ "memory/device_reserved (GiB)": 33.42,
386
+ "memory/max_active (GiB)": 32.31,
387
+ "memory/max_allocated (GiB)": 32.31,
388
+ "step": 216
389
+ },
390
+ {
391
+ "epoch": 3.0985915492957745,
392
+ "grad_norm": 0.134765625,
393
+ "learning_rate": 1.7467526405546344e-05,
394
+ "loss": 0.001148233562707901,
395
+ "memory/device_reserved (GiB)": 33.38,
396
+ "memory/max_active (GiB)": 32.31,
397
+ "memory/max_allocated (GiB)": 32.31,
398
+ "ppl": 1.00115,
399
+ "step": 220,
400
+ "tokens/total": 1802240,
401
+ "tokens/train_per_sec_per_gpu": 10.46,
402
+ "tokens/trainable": 779210
403
+ },
404
+ {
405
+ "epoch": 3.23943661971831,
406
+ "grad_norm": 0.0791015625,
407
+ "learning_rate": 1.7131642946512314e-05,
408
+ "loss": 0.0010407534427940845,
409
+ "memory/device_reserved (GiB)": 33.38,
410
+ "memory/max_active (GiB)": 32.31,
411
+ "memory/max_allocated (GiB)": 32.31,
412
+ "ppl": 1.00104,
413
+ "step": 230,
414
+ "tokens/total": 1884160,
415
+ "tokens/train_per_sec_per_gpu": 10.43,
416
+ "tokens/trainable": 814465
417
+ },
418
+ {
419
+ "epoch": 3.380281690140845,
420
+ "grad_norm": 0.035888671875,
421
+ "learning_rate": 1.6778524929852513e-05,
422
+ "loss": 0.0008773493580520153,
423
+ "memory/device_reserved (GiB)": 33.38,
424
+ "memory/max_active (GiB)": 32.31,
425
+ "memory/max_allocated (GiB)": 32.31,
426
+ "ppl": 1.00088,
427
+ "step": 240,
428
+ "tokens/total": 1966080,
429
+ "tokens/train_per_sec_per_gpu": 10.32,
430
+ "tokens/trainable": 849679
431
+ },
432
+ {
433
+ "epoch": 3.52112676056338,
434
+ "grad_norm": 0.11328125,
435
+ "learning_rate": 1.640902571191869e-05,
436
+ "loss": 0.00072392662987113,
437
+ "memory/device_reserved (GiB)": 33.38,
438
+ "memory/max_active (GiB)": 32.31,
439
+ "memory/max_allocated (GiB)": 32.31,
440
+ "ppl": 1.00072,
441
+ "step": 250,
442
+ "tokens/total": 2048000,
443
+ "tokens/train_per_sec_per_gpu": 11.0,
444
+ "tokens/trainable": 884891
445
+ },
446
+ {
447
+ "epoch": 3.5492957746478875,
448
+ "eval_loss": 0.005309262778609991,
449
+ "eval_ppl": 1.00532,
450
+ "eval_runtime": 8.2154,
451
+ "eval_samples_per_second": 60.74,
452
+ "eval_steps_per_second": 60.74,
453
+ "memory/device_reserved (GiB)": 33.38,
454
+ "memory/max_active (GiB)": 32.31,
455
+ "memory/max_allocated (GiB)": 32.31,
456
+ "step": 252
457
+ },
458
+ {
459
+ "epoch": 3.6619718309859155,
460
+ "grad_norm": 0.037109375,
461
+ "learning_rate": 1.6024038236400246e-05,
462
+ "loss": 0.0005804049782454967,
463
+ "memory/device_reserved (GiB)": 33.42,
464
+ "memory/max_active (GiB)": 32.31,
465
+ "memory/max_allocated (GiB)": 32.31,
466
+ "ppl": 1.00058,
467
+ "step": 260,
468
+ "tokens/total": 2129920,
469
+ "tokens/train_per_sec_per_gpu": 11.12,
470
+ "tokens/trainable": 920273
471
+ },
472
+ {
473
+ "epoch": 3.802816901408451,
474
+ "grad_norm": 0.022705078125,
475
+ "learning_rate": 1.562449287640781e-05,
476
+ "loss": 0.0007765244692564011,
477
+ "memory/device_reserved (GiB)": 33.42,
478
+ "memory/max_active (GiB)": 32.31,
479
+ "memory/max_allocated (GiB)": 32.31,
480
+ "ppl": 1.00078,
481
+ "step": 270,
482
+ "tokens/total": 2211840,
483
+ "tokens/train_per_sec_per_gpu": 10.51,
484
+ "tokens/trainable": 955808
485
+ },
486
+ {
487
+ "epoch": 3.943661971830986,
488
+ "grad_norm": 0.04833984375,
489
+ "learning_rate": 1.5211355186103655e-05,
490
+ "loss": 0.0007016387302428484,
491
+ "memory/device_reserved (GiB)": 33.42,
492
+ "memory/max_active (GiB)": 32.31,
493
+ "memory/max_allocated (GiB)": 32.31,
494
+ "ppl": 1.0007,
495
+ "step": 280,
496
+ "tokens/total": 2293760,
497
+ "tokens/train_per_sec_per_gpu": 10.15,
498
+ "tokens/trainable": 991166
499
+ },
500
+ {
501
+ "epoch": 4.056338028169014,
502
+ "eval_loss": 0.005245804321020842,
503
+ "eval_ppl": 1.00526,
504
+ "eval_runtime": 8.0041,
505
+ "eval_samples_per_second": 62.343,
506
+ "eval_steps_per_second": 62.343,
507
+ "memory/device_reserved (GiB)": 33.42,
508
+ "memory/max_active (GiB)": 32.31,
509
+ "memory/max_allocated (GiB)": 32.31,
510
+ "step": 288
511
+ },
512
+ {
513
+ "epoch": 4.084507042253521,
514
+ "grad_norm": 0.047607421875,
515
+ "learning_rate": 1.4785623567312492e-05,
516
+ "loss": 0.000780455581843853,
517
+ "memory/device_reserved (GiB)": 33.42,
518
+ "memory/max_active (GiB)": 32.31,
519
+ "memory/max_allocated (GiB)": 32.31,
520
+ "ppl": 1.00078,
521
+ "step": 290,
522
+ "tokens/total": 2375680,
523
+ "tokens/train_per_sec_per_gpu": 11.01,
524
+ "tokens/trainable": 1026615
525
+ },
526
+ {
527
+ "epoch": 4.225352112676056,
528
+ "grad_norm": 0.061767578125,
529
+ "learning_rate": 1.4348326856751496e-05,
530
+ "loss": 0.0005466958507895469,
531
+ "memory/device_reserved (GiB)": 33.42,
532
+ "memory/max_active (GiB)": 32.31,
533
+ "memory/max_allocated (GiB)": 32.31,
534
+ "ppl": 1.00055,
535
+ "step": 300,
536
+ "tokens/total": 2457600,
537
+ "tokens/train_per_sec_per_gpu": 10.99,
538
+ "tokens/trainable": 1062010
539
+ },
540
+ {
541
+ "epoch": 4.366197183098592,
542
+ "grad_norm": 0.0341796875,
543
+ "learning_rate": 1.3900521839710428e-05,
544
+ "loss": 0.0006260770373046398,
545
+ "memory/device_reserved (GiB)": 33.42,
546
+ "memory/max_active (GiB)": 32.31,
547
+ "memory/max_allocated (GiB)": 32.31,
548
+ "ppl": 1.00063,
549
+ "step": 310,
550
+ "tokens/total": 2539520,
551
+ "tokens/train_per_sec_per_gpu": 11.12,
552
+ "tokens/trainable": 1097505
553
+ },
554
+ {
555
+ "epoch": 4.507042253521127,
556
+ "grad_norm": 0.03857421875,
557
+ "learning_rate": 1.3443290696190335e-05,
558
+ "loss": 0.0005137504544109106,
559
+ "memory/device_reserved (GiB)": 33.42,
560
+ "memory/max_active (GiB)": 32.31,
561
+ "memory/max_allocated (GiB)": 32.31,
562
+ "ppl": 1.00051,
563
+ "step": 320,
564
+ "tokens/total": 2621440,
565
+ "tokens/train_per_sec_per_gpu": 10.32,
566
+ "tokens/trainable": 1132758
567
+ },
568
+ {
569
+ "epoch": 4.563380281690141,
570
+ "eval_loss": 0.005256574135273695,
571
+ "eval_ppl": 1.00527,
572
+ "eval_runtime": 8.1646,
573
+ "eval_samples_per_second": 61.118,
574
+ "eval_steps_per_second": 61.118,
575
+ "memory/device_reserved (GiB)": 33.42,
576
+ "memory/max_active (GiB)": 32.31,
577
+ "memory/max_allocated (GiB)": 32.31,
578
+ "step": 324
579
+ },
580
+ {
581
+ "epoch": 4.647887323943662,
582
+ "grad_norm": 0.046142578125,
583
+ "learning_rate": 1.2977738385672558e-05,
584
+ "loss": 0.0005908971652388573,
585
+ "memory/device_reserved (GiB)": 33.42,
586
+ "memory/max_active (GiB)": 32.31,
587
+ "memory/max_allocated (GiB)": 32.31,
588
+ "ppl": 1.00059,
589
+ "step": 330,
590
+ "tokens/total": 2703360,
591
+ "tokens/train_per_sec_per_gpu": 10.25,
592
+ "tokens/trainable": 1168161
593
+ },
594
+ {
595
+ "epoch": 4.788732394366197,
596
+ "grad_norm": 0.050537109375,
597
+ "learning_rate": 1.250498997683813e-05,
598
+ "loss": 0.0006883179303258657,
599
+ "memory/device_reserved (GiB)": 33.42,
600
+ "memory/max_active (GiB)": 32.31,
601
+ "memory/max_allocated (GiB)": 32.31,
602
+ "ppl": 1.00069,
603
+ "step": 340,
604
+ "tokens/total": 2785280,
605
+ "tokens/train_per_sec_per_gpu": 10.28,
606
+ "tokens/trainable": 1203434
607
+ },
608
+ {
609
+ "epoch": 4.929577464788732,
610
+ "grad_norm": 0.04736328125,
611
+ "learning_rate": 1.202618792869063e-05,
612
+ "loss": 0.0006954259239137173,
613
+ "memory/device_reserved (GiB)": 33.42,
614
+ "memory/max_active (GiB)": 32.31,
615
+ "memory/max_allocated (GiB)": 32.31,
616
+ "ppl": 1.0007,
617
+ "step": 350,
618
+ "tokens/total": 2867200,
619
+ "tokens/train_per_sec_per_gpu": 11.09,
620
+ "tokens/trainable": 1238829
621
+ },
622
+ {
623
+ "epoch": 5.070422535211268,
624
+ "grad_norm": 0.05517578125,
625
+ "learning_rate": 1.1542489329653024e-05,
626
+ "loss": 0.000603723106905818,
627
+ "memory/device_reserved (GiB)": 33.42,
628
+ "memory/max_active (GiB)": 32.31,
629
+ "memory/max_allocated (GiB)": 32.31,
630
+ "ppl": 1.0006,
631
+ "step": 360,
632
+ "tokens/total": 2949120,
633
+ "tokens/train_per_sec_per_gpu": 11.02,
634
+ "tokens/trainable": 1274110
635
+ },
636
+ {
637
+ "epoch": 5.070422535211268,
638
+ "eval_loss": 0.005255201831459999,
639
+ "eval_ppl": 1.00527,
640
+ "eval_runtime": 8.0148,
641
+ "eval_samples_per_second": 62.26,
642
+ "eval_steps_per_second": 62.26,
643
+ "memory/device_reserved (GiB)": 33.42,
644
+ "memory/max_active (GiB)": 24.06,
645
+ "memory/max_allocated (GiB)": 24.06,
646
+ "step": 360
647
+ },
648
+ {
649
+ "epoch": 5.211267605633803,
650
+ "grad_norm": 0.0615234375,
651
+ "learning_rate": 1.1055063101310581e-05,
652
+ "loss": 0.0005278075113892555,
653
+ "memory/device_reserved (GiB)": 33.38,
654
+ "memory/max_active (GiB)": 32.31,
655
+ "memory/max_allocated (GiB)": 32.31,
656
+ "ppl": 1.00053,
657
+ "step": 370,
658
+ "tokens/total": 3031040,
659
+ "tokens/train_per_sec_per_gpu": 11.13,
660
+ "tokens/trainable": 1309814
661
+ },
662
+ {
663
+ "epoch": 5.352112676056338,
664
+ "grad_norm": 0.037841796875,
665
+ "learning_rate": 1.0565087173557396e-05,
666
+ "loss": 0.0005071659106761217,
667
+ "memory/device_reserved (GiB)": 33.38,
668
+ "memory/max_active (GiB)": 32.31,
669
+ "memory/max_allocated (GiB)": 32.31,
670
+ "ppl": 1.00051,
671
+ "step": 380,
672
+ "tokens/total": 3112960,
673
+ "tokens/train_per_sec_per_gpu": 11.21,
674
+ "tokens/trainable": 1345167
675
+ },
676
+ {
677
+ "epoch": 5.492957746478873,
678
+ "grad_norm": 0.02880859375,
679
+ "learning_rate": 1.0073745637973125e-05,
680
+ "loss": 0.0005827666725963354,
681
+ "memory/device_reserved (GiB)": 33.38,
682
+ "memory/max_active (GiB)": 32.31,
683
+ "memory/max_allocated (GiB)": 32.31,
684
+ "ppl": 1.00058,
685
+ "step": 390,
686
+ "tokens/total": 3194880,
687
+ "tokens/train_per_sec_per_gpu": 11.14,
688
+ "tokens/trainable": 1380603
689
+ },
690
+ {
691
+ "epoch": 5.577464788732394,
692
+ "eval_loss": 0.005226313602179289,
693
+ "eval_ppl": 1.00524,
694
+ "eval_runtime": 8.1606,
695
+ "eval_samples_per_second": 61.147,
696
+ "eval_steps_per_second": 61.147,
697
+ "memory/device_reserved (GiB)": 33.38,
698
+ "memory/max_active (GiB)": 32.31,
699
+ "memory/max_allocated (GiB)": 32.31,
700
+ "step": 396
701
+ },
702
+ {
703
+ "epoch": 5.633802816901408,
704
+ "grad_norm": 0.04150390625,
705
+ "learning_rate": 9.582225886309217e-06,
706
+ "loss": 0.0006697049364447593,
707
+ "memory/device_reserved (GiB)": 33.42,
708
+ "memory/max_active (GiB)": 32.31,
709
+ "memory/max_allocated (GiB)": 32.31,
710
+ "ppl": 1.00067,
711
+ "step": 400,
712
+ "tokens/total": 3276800,
713
+ "tokens/train_per_sec_per_gpu": 9.71,
714
+ "tokens/trainable": 1415822
715
+ },
716
+ {
717
+ "epoch": 5.774647887323944,
718
+ "grad_norm": 0.04833984375,
719
+ "learning_rate": 9.091715740999829e-06,
720
+ "loss": 0.0006844023242592812,
721
+ "memory/device_reserved (GiB)": 33.42,
722
+ "memory/max_active (GiB)": 32.31,
723
+ "memory/max_allocated (GiB)": 32.31,
724
+ "ppl": 1.00068,
725
+ "step": 410,
726
+ "tokens/total": 3358720,
727
+ "tokens/train_per_sec_per_gpu": 10.39,
728
+ "tokens/trainable": 1451140
729
+ },
730
+ {
731
+ "epoch": 5.915492957746479,
732
+ "grad_norm": 0.0478515625,
733
+ "learning_rate": 8.60340058463194e-06,
734
+ "loss": 0.0004917650483548641,
735
+ "memory/device_reserved (GiB)": 33.42,
736
+ "memory/max_active (GiB)": 32.31,
737
+ "memory/max_allocated (GiB)": 32.31,
738
+ "ppl": 1.00049,
739
+ "step": 420,
740
+ "tokens/total": 3440640,
741
+ "tokens/train_per_sec_per_gpu": 11.04,
742
+ "tokens/trainable": 1486616
743
+ },
744
+ {
745
+ "epoch": 6.056338028169014,
746
+ "grad_norm": 0.047607421875,
747
+ "learning_rate": 8.118460495311687e-06,
748
+ "loss": 0.0005019189789891243,
749
+ "memory/device_reserved (GiB)": 33.42,
750
+ "memory/max_active (GiB)": 32.31,
751
+ "memory/max_allocated (GiB)": 32.31,
752
+ "ppl": 1.0005,
753
+ "step": 430,
754
+ "tokens/total": 3522560,
755
+ "tokens/train_per_sec_per_gpu": 9.55,
756
+ "tokens/trainable": 1522001
757
+ },
758
+ {
759
+ "epoch": 6.084507042253521,
760
+ "eval_loss": 0.005209571216255426,
761
+ "eval_ppl": 1.00522,
762
+ "eval_runtime": 8.0091,
763
+ "eval_samples_per_second": 62.304,
764
+ "eval_steps_per_second": 62.304,
765
+ "memory/device_reserved (GiB)": 33.42,
766
+ "memory/max_active (GiB)": 32.31,
767
+ "memory/max_allocated (GiB)": 32.31,
768
+ "step": 432
769
+ },
770
+ {
771
+ "epoch": 6.197183098591549,
772
+ "grad_norm": 0.134765625,
773
+ "learning_rate": 7.638067394849672e-06,
774
+ "loss": 0.000784461572766304,
775
+ "memory/device_reserved (GiB)": 33.38,
776
+ "memory/max_active (GiB)": 32.31,
777
+ "memory/max_allocated (GiB)": 32.31,
778
+ "ppl": 1.00078,
779
+ "step": 440,
780
+ "tokens/total": 3604480,
781
+ "tokens/train_per_sec_per_gpu": 11.09,
782
+ "tokens/trainable": 1557368
783
+ },
784
+ {
785
+ "epoch": 6.338028169014084,
786
+ "grad_norm": 0.05712890625,
787
+ "learning_rate": 7.163382216657033e-06,
788
+ "loss": 0.0005743904504925013,
789
+ "memory/device_reserved (GiB)": 33.38,
790
+ "memory/max_active (GiB)": 32.31,
791
+ "memory/max_allocated (GiB)": 32.31,
792
+ "ppl": 1.00057,
793
+ "step": 450,
794
+ "tokens/total": 3686400,
795
+ "tokens/train_per_sec_per_gpu": 10.46,
796
+ "tokens/trainable": 1592826
797
+ },
798
+ {
799
+ "epoch": 6.47887323943662,
800
+ "grad_norm": 0.03076171875,
801
+ "learning_rate": 6.6955521001964526e-06,
802
+ "loss": 0.0004704087972640991,
803
+ "memory/device_reserved (GiB)": 33.38,
804
+ "memory/max_active (GiB)": 32.31,
805
+ "memory/max_allocated (GiB)": 32.31,
806
+ "ppl": 1.00047,
807
+ "step": 460,
808
+ "tokens/total": 3768320,
809
+ "tokens/train_per_sec_per_gpu": 10.41,
810
+ "tokens/trainable": 1628143
811
+ },
812
+ {
813
+ "epoch": 6.591549295774648,
814
+ "eval_loss": 0.005213646683841944,
815
+ "eval_ppl": 1.00523,
816
+ "eval_runtime": 8.1767,
817
+ "eval_samples_per_second": 61.027,
818
+ "eval_steps_per_second": 61.027,
819
+ "memory/device_reserved (GiB)": 33.38,
820
+ "memory/max_active (GiB)": 32.31,
821
+ "memory/max_allocated (GiB)": 32.31,
822
+ "step": 468
823
+ },
824
+ {
825
+ "epoch": 6.619718309859155,
826
+ "grad_norm": 0.04736328125,
827
+ "learning_rate": 6.2357076187680325e-06,
828
+ "loss": 0.0004482423886656761,
829
+ "memory/device_reserved (GiB)": 33.42,
830
+ "memory/max_active (GiB)": 32.31,
831
+ "memory/max_allocated (GiB)": 32.31,
832
+ "ppl": 1.00045,
833
+ "step": 470,
834
+ "tokens/total": 3850240,
835
+ "tokens/train_per_sec_per_gpu": 10.41,
836
+ "tokens/trainable": 1663545
837
+ },
838
+ {
839
+ "epoch": 6.76056338028169,
840
+ "grad_norm": 0.0380859375,
841
+ "learning_rate": 5.78496004732952e-06,
842
+ "loss": 0.0004957062192261219,
843
+ "memory/device_reserved (GiB)": 33.42,
844
+ "memory/max_active (GiB)": 32.31,
845
+ "memory/max_allocated (GiB)": 32.31,
846
+ "ppl": 1.0005,
847
+ "step": 480,
848
+ "tokens/total": 3932160,
849
+ "tokens/train_per_sec_per_gpu": 10.4,
850
+ "tokens/trainable": 1699181
851
+ },
852
+ {
853
+ "epoch": 6.901408450704225,
854
+ "grad_norm": 0.0361328125,
855
+ "learning_rate": 5.344398676953526e-06,
856
+ "loss": 0.000544667523354292,
857
+ "memory/device_reserved (GiB)": 33.42,
858
+ "memory/max_active (GiB)": 32.31,
859
+ "memory/max_allocated (GiB)": 32.31,
860
+ "ppl": 1.00054,
861
+ "step": 490,
862
+ "tokens/total": 4014080,
863
+ "tokens/train_per_sec_per_gpu": 10.21,
864
+ "tokens/trainable": 1734683
865
+ },
866
+ {
867
+ "epoch": 7.042253521126761,
868
+ "grad_norm": 0.03759765625,
869
+ "learning_rate": 4.915088182411674e-06,
870
+ "loss": 0.0004880569409579039,
871
+ "memory/device_reserved (GiB)": 33.42,
872
+ "memory/max_active (GiB)": 32.31,
873
+ "memory/max_allocated (GiB)": 32.31,
874
+ "ppl": 1.00049,
875
+ "step": 500,
876
+ "tokens/total": 4096000,
877
+ "tokens/train_per_sec_per_gpu": 11.44,
878
+ "tokens/trainable": 1770105
879
+ },
880
+ {
881
+ "epoch": 7.098591549295775,
882
+ "eval_loss": 0.00523661682382226,
883
+ "eval_ppl": 1.00525,
884
+ "eval_runtime": 8.0317,
885
+ "eval_samples_per_second": 62.129,
886
+ "eval_steps_per_second": 62.129,
887
+ "memory/device_reserved (GiB)": 33.42,
888
+ "memory/max_active (GiB)": 32.31,
889
+ "memory/max_allocated (GiB)": 32.31,
890
+ "step": 504
891
+ },
892
+ {
893
+ "epoch": 7.183098591549296,
894
+ "grad_norm": 0.0257568359375,
895
+ "learning_rate": 4.498066049247344e-06,
896
+ "loss": 0.0005014274269342422,
897
+ "memory/device_reserved (GiB)": 33.38,
898
+ "memory/max_active (GiB)": 32.31,
899
+ "memory/max_allocated (GiB)": 32.31,
900
+ "ppl": 1.0005,
901
+ "step": 510,
902
+ "tokens/total": 4177920,
903
+ "tokens/train_per_sec_per_gpu": 10.53,
904
+ "tokens/trainable": 1805486
905
+ },
906
+ {
907
+ "epoch": 7.323943661971831,
908
+ "grad_norm": 0.03466796875,
909
+ "learning_rate": 4.094340066554742e-06,
910
+ "loss": 0.0006260806694626808,
911
+ "memory/device_reserved (GiB)": 33.38,
912
+ "memory/max_active (GiB)": 32.31,
913
+ "memory/max_allocated (GiB)": 32.31,
914
+ "ppl": 1.00063,
915
+ "step": 520,
916
+ "tokens/total": 4259840,
917
+ "tokens/train_per_sec_per_gpu": 10.8,
918
+ "tokens/trainable": 1840987
919
+ },
920
+ {
921
+ "epoch": 7.464788732394366,
922
+ "grad_norm": 0.036865234375,
923
+ "learning_rate": 3.7048858915233665e-06,
924
+ "loss": 0.0005529518239200115,
925
+ "memory/device_reserved (GiB)": 33.38,
926
+ "memory/max_active (GiB)": 32.31,
927
+ "memory/max_allocated (GiB)": 32.31,
928
+ "ppl": 1.00055,
929
+ "step": 530,
930
+ "tokens/total": 4341760,
931
+ "tokens/train_per_sec_per_gpu": 10.79,
932
+ "tokens/trainable": 1876364
933
+ },
934
+ {
935
+ "epoch": 7.605633802816901,
936
+ "grad_norm": 0.034912109375,
937
+ "learning_rate": 3.330644691633492e-06,
938
+ "loss": 0.0005182296968996525,
939
+ "memory/device_reserved (GiB)": 33.38,
940
+ "memory/max_active (GiB)": 32.31,
941
+ "memory/max_allocated (GiB)": 32.31,
942
+ "ppl": 1.00052,
943
+ "step": 540,
944
+ "tokens/total": 4423680,
945
+ "tokens/train_per_sec_per_gpu": 10.9,
946
+ "tokens/trainable": 1911823
947
+ },
948
+ {
949
+ "epoch": 7.605633802816901,
950
+ "eval_loss": 0.005228586029261351,
951
+ "eval_ppl": 1.00524,
952
+ "eval_runtime": 8.2216,
953
+ "eval_samples_per_second": 60.694,
954
+ "eval_steps_per_second": 60.694,
955
+ "memory/device_reserved (GiB)": 33.38,
956
+ "memory/max_active (GiB)": 24.06,
957
+ "memory/max_allocated (GiB)": 24.06,
958
+ "step": 540
959
+ },
960
+ {
961
+ "epoch": 7.746478873239437,
962
+ "grad_norm": 0.025634765625,
963
+ "learning_rate": 2.9725208702005736e-06,
964
+ "loss": 0.0005175352562218904,
965
+ "memory/device_reserved (GiB)": 33.42,
966
+ "memory/max_active (GiB)": 32.31,
967
+ "memory/max_allocated (GiB)": 32.31,
968
+ "ppl": 1.00052,
969
+ "step": 550,
970
+ "tokens/total": 4505600,
971
+ "tokens/train_per_sec_per_gpu": 11.35,
972
+ "tokens/trainable": 1947212
973
+ },
974
+ {
975
+ "epoch": 7.887323943661972,
976
+ "grad_norm": 0.0341796875,
977
+ "learning_rate": 2.631379880765107e-06,
978
+ "loss": 0.0005252756644040346,
979
+ "memory/device_reserved (GiB)": 33.42,
980
+ "memory/max_active (GiB)": 32.31,
981
+ "memory/max_allocated (GiB)": 32.31,
982
+ "ppl": 1.00053,
983
+ "step": 560,
984
+ "tokens/total": 4587520,
985
+ "tokens/train_per_sec_per_gpu": 11.28,
986
+ "tokens/trainable": 1982716
987
+ },
988
+ {
989
+ "epoch": 8.028169014084508,
990
+ "grad_norm": 0.046142578125,
991
+ "learning_rate": 2.3080461356097938e-06,
992
+ "loss": 0.0004542431328445673,
993
+ "memory/device_reserved (GiB)": 33.42,
994
+ "memory/max_active (GiB)": 32.31,
995
+ "memory/max_allocated (GiB)": 32.31,
996
+ "ppl": 1.00045,
997
+ "step": 570,
998
+ "tokens/total": 4669440,
999
+ "tokens/train_per_sec_per_gpu": 10.28,
1000
+ "tokens/trainable": 2018112
1001
+ },
1002
+ {
1003
+ "epoch": 8.112676056338028,
1004
+ "eval_loss": 0.005249009933322668,
1005
+ "eval_ppl": 1.00526,
1006
+ "eval_runtime": 8.1416,
1007
+ "eval_samples_per_second": 61.29,
1008
+ "eval_steps_per_second": 61.29,
1009
+ "memory/device_reserved (GiB)": 33.42,
1010
+ "memory/max_active (GiB)": 32.31,
1011
+ "memory/max_allocated (GiB)": 32.31,
1012
+ "step": 576
1013
+ },
1014
+ {
1015
+ "epoch": 8.169014084507042,
1016
+ "grad_norm": 0.043212890625,
1017
+ "learning_rate": 2.0033010134583085e-06,
1018
+ "loss": 0.000499055441468954,
1019
+ "memory/device_reserved (GiB)": 33.42,
1020
+ "memory/max_active (GiB)": 32.31,
1021
+ "memory/max_allocated (GiB)": 32.31,
1022
+ "ppl": 1.0005,
1023
+ "step": 580,
1024
+ "tokens/total": 4751360,
1025
+ "tokens/train_per_sec_per_gpu": 10.13,
1026
+ "tokens/trainable": 2053569
1027
+ },
1028
+ {
1029
+ "epoch": 8.309859154929578,
1030
+ "grad_norm": 0.029052734375,
1031
+ "learning_rate": 1.7178809711703525e-06,
1032
+ "loss": 0.0006432968657463789,
1033
+ "memory/device_reserved (GiB)": 33.42,
1034
+ "memory/max_active (GiB)": 32.31,
1035
+ "memory/max_allocated (GiB)": 32.31,
1036
+ "ppl": 1.00064,
1037
+ "step": 590,
1038
+ "tokens/total": 4833280,
1039
+ "tokens/train_per_sec_per_gpu": 10.98,
1040
+ "tokens/trainable": 2089056
1041
+ },
1042
+ {
1043
+ "epoch": 8.450704225352112,
1044
+ "grad_norm": 0.04443359375,
1045
+ "learning_rate": 1.452475763996326e-06,
1046
+ "loss": 0.0005107030272483825,
1047
+ "memory/device_reserved (GiB)": 33.42,
1048
+ "memory/max_active (GiB)": 32.31,
1049
+ "memory/max_allocated (GiB)": 32.31,
1050
+ "ppl": 1.00051,
1051
+ "step": 600,
1052
+ "tokens/total": 4915200,
1053
+ "tokens/train_per_sec_per_gpu": 11.03,
1054
+ "tokens/trainable": 2124556
1055
+ },
1056
+ {
1057
+ "epoch": 8.591549295774648,
1058
+ "grad_norm": 0.04150390625,
1059
+ "learning_rate": 1.207726778692625e-06,
1060
+ "loss": 0.00045921015553176405,
1061
+ "memory/device_reserved (GiB)": 33.42,
1062
+ "memory/max_active (GiB)": 32.31,
1063
+ "memory/max_allocated (GiB)": 32.31,
1064
+ "ppl": 1.00046,
1065
+ "step": 610,
1066
+ "tokens/total": 4997120,
1067
+ "tokens/train_per_sec_per_gpu": 11.16,
1068
+ "tokens/trainable": 2159952
1069
+ },
1070
+ {
1071
+ "epoch": 8.619718309859154,
1072
+ "eval_loss": 0.0052222758531570435,
1073
+ "eval_ppl": 1.00524,
1074
+ "eval_runtime": 8.2144,
1075
+ "eval_samples_per_second": 60.747,
1076
+ "eval_steps_per_second": 60.747,
1077
+ "memory/device_reserved (GiB)": 33.42,
1078
+ "memory/max_active (GiB)": 32.31,
1079
+ "memory/max_allocated (GiB)": 32.31,
1080
+ "step": 612
1081
+ },
1082
+ {
1083
+ "epoch": 8.732394366197184,
1084
+ "grad_norm": 0.03369140625,
1085
+ "learning_rate": 9.842254835257792e-07,
1086
+ "loss": 0.0005835144780576229,
1087
+ "memory/device_reserved (GiB)": 33.38,
1088
+ "memory/max_active (GiB)": 32.31,
1089
+ "memory/max_allocated (GiB)": 32.31,
1090
+ "ppl": 1.00058,
1091
+ "step": 620,
1092
+ "tokens/total": 5079040,
1093
+ "tokens/train_per_sec_per_gpu": 10.43,
1094
+ "tokens/trainable": 2195458
1095
+ },
1096
+ {
1097
+ "epoch": 8.873239436619718,
1098
+ "grad_norm": 0.035400390625,
1099
+ "learning_rate": 7.825119989112173e-07,
1100
+ "loss": 0.00045958077535033224,
1101
+ "memory/device_reserved (GiB)": 33.38,
1102
+ "memory/max_active (GiB)": 32.31,
1103
+ "memory/max_allocated (GiB)": 32.31,
1104
+ "ppl": 1.00046,
1105
+ "step": 630,
1106
+ "tokens/total": 5160960,
1107
+ "tokens/train_per_sec_per_gpu": 10.35,
1108
+ "tokens/trainable": 2230834
1109
+ }
1110
+ ],
1111
+ "logging_steps": 10,
1112
+ "max_steps": 710,
1113
+ "num_input_tokens_seen": 0,
1114
+ "num_train_epochs": 10,
1115
+ "save_steps": 71,
1116
+ "stateful_callbacks": {
1117
+ "TrainerControl": {
1118
+ "args": {
1119
+ "should_epoch_stop": false,
1120
+ "should_evaluate": false,
1121
+ "should_log": false,
1122
+ "should_save": true,
1123
+ "should_training_stop": false
1124
+ },
1125
+ "attributes": {}
1126
+ }
1127
+ },
1128
+ "total_flos": 1.141218079754158e+17,
1129
+ "train_batch_size": 1,
1130
+ "trial_name": null,
1131
+ "trial_params": null
1132
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-639/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c56b5e3616cf3d2a8bf8aabf031ff524a587484ec751048484926648eb0ad82
3
+ size 7121
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/chat_template.jinja ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '
2
+ ' + message['content'] + '<|im_end|>' + '
3
+ '}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
4
+ ' }}{% endif %}
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "dtype": "bfloat16",
9
+ "eos_token_id": 151645,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2560,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 9728,
15
+ "layer_types": [
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention"
52
+ ],
53
+ "max_position_embeddings": 262144,
54
+ "max_window_layers": 36,
55
+ "model_type": "qwen3",
56
+ "num_attention_heads": 32,
57
+ "num_hidden_layers": 36,
58
+ "num_key_value_heads": 8,
59
+ "pad_token_id": 151643,
60
+ "rms_norm_eps": 1e-06,
61
+ "rope_parameters": {
62
+ "rope_theta": 5000000,
63
+ "rope_type": "default"
64
+ },
65
+ "sliding_window": null,
66
+ "tie_word_embeddings": true,
67
+ "transformers_version": "5.0.0",
68
+ "use_cache": false,
69
+ "use_sliding_window": false,
70
+ "vocab_size": 151936
71
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_sample": true,
3
+ "eos_token_id": [
4
+ 151645,
5
+ 151643
6
+ ],
7
+ "pad_token_id": 151643,
8
+ "temperature": 0.7,
9
+ "top_k": 20,
10
+ "top_p": 0.8,
11
+ "transformers_version": "5.0.0"
12
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e23f6344d15f9d4e240f524ca8d68d9d092a7ee823e86a4c1737957794c545ba
3
+ size 8822894520
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0259a41ca1f650cc1741ffdfc70e09c0459ee40279a9291bb7685917b879a35
3
+ size 16090225449
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:63aab5945fa6a4f19ef11eca5b6add2fab56216a2af2a42b2f4db10b37425e8e
3
+ size 14645
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9fab60981798c54e021a9281fcfa4c4519ffd9bdd8a217bc9242ded887653d0d
3
+ size 1465
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
3
+ size 11422650
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokenizer_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|im_start|>",
10
+ "<|im_end|>",
11
+ "<|object_ref_start|>",
12
+ "<|object_ref_end|>",
13
+ "<|box_start|>",
14
+ "<|box_end|>",
15
+ "<|quad_start|>",
16
+ "<|quad_end|>",
17
+ "<|vision_start|>",
18
+ "<|vision_end|>",
19
+ "<|vision_pad|>",
20
+ "<|image_pad|>",
21
+ "<|video_pad|>"
22
+ ],
23
+ "is_local": true,
24
+ "model_max_length": 1010000,
25
+ "pad_token": "<|endoftext|>",
26
+ "split_special_tokens": false,
27
+ "tokenizer_class": "Qwen2Tokenizer",
28
+ "unk_token": null
29
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/tokens_state. ADDED
@@ -0,0 +1 @@
 
 
1
+ {"total": 5816320, "trainable": 2513851}
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/trainer_state.json ADDED
@@ -0,0 +1,1268 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 684,
3
+ "best_metric": 0.005193324759602547,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 10.0,
6
+ "eval_steps": 36,
7
+ "global_step": 710,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0,
14
+ "eval_loss": 6.340579032897949,
15
+ "eval_ppl": 567.1246,
16
+ "eval_runtime": 9.2875,
17
+ "eval_samples_per_second": 53.728,
18
+ "eval_steps_per_second": 53.728,
19
+ "memory/device_reserved (GiB)": 9.68,
20
+ "memory/max_active (GiB)": 8.96,
21
+ "memory/max_allocated (GiB)": 8.96,
22
+ "step": 0
23
+ },
24
+ {
25
+ "epoch": 0.14084507042253522,
26
+ "grad_norm": 48.75,
27
+ "learning_rate": 2.535211267605634e-06,
28
+ "loss": 6.281451797485351,
29
+ "memory/device_reserved (GiB)": 33.61,
30
+ "memory/max_active (GiB)": 32.32,
31
+ "memory/max_allocated (GiB)": 32.32,
32
+ "ppl": 534.56418,
33
+ "step": 10,
34
+ "tokens/total": 81920,
35
+ "tokens/train_per_sec_per_gpu": 9.82,
36
+ "tokens/trainable": 35649
37
+ },
38
+ {
39
+ "epoch": 0.28169014084507044,
40
+ "grad_norm": 43.0,
41
+ "learning_rate": 5.352112676056338e-06,
42
+ "loss": 5.6617687225341795,
43
+ "memory/device_reserved (GiB)": 33.61,
44
+ "memory/max_active (GiB)": 32.32,
45
+ "memory/max_allocated (GiB)": 32.32,
46
+ "ppl": 287.65698,
47
+ "step": 20,
48
+ "tokens/total": 163840,
49
+ "tokens/train_per_sec_per_gpu": 10.46,
50
+ "tokens/trainable": 71196
51
+ },
52
+ {
53
+ "epoch": 0.4225352112676056,
54
+ "grad_norm": 26.875,
55
+ "learning_rate": 8.169014084507043e-06,
56
+ "loss": 3.820334243774414,
57
+ "memory/device_reserved (GiB)": 33.61,
58
+ "memory/max_active (GiB)": 32.32,
59
+ "memory/max_allocated (GiB)": 32.32,
60
+ "ppl": 45.61945,
61
+ "step": 30,
62
+ "tokens/total": 245760,
63
+ "tokens/train_per_sec_per_gpu": 10.53,
64
+ "tokens/trainable": 106543
65
+ },
66
+ {
67
+ "epoch": 0.5070422535211268,
68
+ "eval_loss": 1.091687560081482,
69
+ "eval_ppl": 2.9793,
70
+ "eval_runtime": 8.0764,
71
+ "eval_samples_per_second": 61.785,
72
+ "eval_steps_per_second": 61.785,
73
+ "memory/device_reserved (GiB)": 33.61,
74
+ "memory/max_active (GiB)": 32.32,
75
+ "memory/max_allocated (GiB)": 32.32,
76
+ "step": 36
77
+ },
78
+ {
79
+ "epoch": 0.5633802816901409,
80
+ "grad_norm": 6.5625,
81
+ "learning_rate": 1.0985915492957748e-05,
82
+ "loss": 1.4444393157958983,
83
+ "memory/device_reserved (GiB)": 33.42,
84
+ "memory/max_active (GiB)": 32.31,
85
+ "memory/max_allocated (GiB)": 32.31,
86
+ "ppl": 4.23947,
87
+ "step": 40,
88
+ "tokens/total": 327680,
89
+ "tokens/train_per_sec_per_gpu": 10.34,
90
+ "tokens/trainable": 141719
91
+ },
92
+ {
93
+ "epoch": 0.704225352112676,
94
+ "grad_norm": 2.484375,
95
+ "learning_rate": 1.380281690140845e-05,
96
+ "loss": 0.4776052951812744,
97
+ "memory/device_reserved (GiB)": 33.42,
98
+ "memory/max_active (GiB)": 32.31,
99
+ "memory/max_allocated (GiB)": 32.31,
100
+ "ppl": 1.61221,
101
+ "step": 50,
102
+ "tokens/total": 409600,
103
+ "tokens/train_per_sec_per_gpu": 10.24,
104
+ "tokens/trainable": 177240
105
+ },
106
+ {
107
+ "epoch": 0.8450704225352113,
108
+ "grad_norm": 1.421875,
109
+ "learning_rate": 1.6619718309859155e-05,
110
+ "loss": 0.15568361282348633,
111
+ "memory/device_reserved (GiB)": 33.42,
112
+ "memory/max_active (GiB)": 32.31,
113
+ "memory/max_allocated (GiB)": 32.31,
114
+ "ppl": 1.16846,
115
+ "step": 60,
116
+ "tokens/total": 491520,
117
+ "tokens/train_per_sec_per_gpu": 10.55,
118
+ "tokens/trainable": 212815
119
+ },
120
+ {
121
+ "epoch": 0.9859154929577465,
122
+ "grad_norm": 0.95703125,
123
+ "learning_rate": 1.943661971830986e-05,
124
+ "loss": 0.06936523914337159,
125
+ "memory/device_reserved (GiB)": 33.42,
126
+ "memory/max_active (GiB)": 32.31,
127
+ "memory/max_allocated (GiB)": 32.31,
128
+ "ppl": 1.07183,
129
+ "step": 70,
130
+ "tokens/total": 573440,
131
+ "tokens/train_per_sec_per_gpu": 10.42,
132
+ "tokens/trainable": 248141
133
+ },
134
+ {
135
+ "epoch": 1.0140845070422535,
136
+ "eval_loss": 0.04077093303203583,
137
+ "eval_ppl": 1.04161,
138
+ "eval_runtime": 8.0461,
139
+ "eval_samples_per_second": 62.018,
140
+ "eval_steps_per_second": 62.018,
141
+ "memory/device_reserved (GiB)": 33.42,
142
+ "memory/max_active (GiB)": 32.31,
143
+ "memory/max_allocated (GiB)": 32.31,
144
+ "step": 72
145
+ },
146
+ {
147
+ "epoch": 1.1267605633802817,
148
+ "grad_norm": 0.859375,
149
+ "learning_rate": 1.9992266216318037e-05,
150
+ "loss": 0.03334630727767944,
151
+ "memory/device_reserved (GiB)": 33.42,
152
+ "memory/max_active (GiB)": 32.31,
153
+ "memory/max_allocated (GiB)": 32.31,
154
+ "ppl": 1.03391,
155
+ "step": 80,
156
+ "tokens/total": 655360,
157
+ "tokens/train_per_sec_per_gpu": 10.95,
158
+ "tokens/trainable": 283579
159
+ },
160
+ {
161
+ "epoch": 1.267605633802817,
162
+ "grad_norm": 0.94921875,
163
+ "learning_rate": 1.996086822074945e-05,
164
+ "loss": 0.020297910273075103,
165
+ "memory/device_reserved (GiB)": 33.42,
166
+ "memory/max_active (GiB)": 32.31,
167
+ "memory/max_allocated (GiB)": 32.31,
168
+ "ppl": 1.02051,
169
+ "step": 90,
170
+ "tokens/total": 737280,
171
+ "tokens/train_per_sec_per_gpu": 10.49,
172
+ "tokens/trainable": 319131
173
+ },
174
+ {
175
+ "epoch": 1.408450704225352,
176
+ "grad_norm": 0.482421875,
177
+ "learning_rate": 1.9905398470798207e-05,
178
+ "loss": 0.014508798718452454,
179
+ "memory/device_reserved (GiB)": 33.42,
180
+ "memory/max_active (GiB)": 32.31,
181
+ "memory/max_allocated (GiB)": 32.31,
182
+ "ppl": 1.01461,
183
+ "step": 100,
184
+ "tokens/total": 819200,
185
+ "tokens/train_per_sec_per_gpu": 10.25,
186
+ "tokens/trainable": 354528
187
+ },
188
+ {
189
+ "epoch": 1.5211267605633803,
190
+ "eval_loss": 0.011320236138999462,
191
+ "eval_ppl": 1.01138,
192
+ "eval_runtime": 8.142,
193
+ "eval_samples_per_second": 61.287,
194
+ "eval_steps_per_second": 61.287,
195
+ "memory/device_reserved (GiB)": 33.42,
196
+ "memory/max_active (GiB)": 32.31,
197
+ "memory/max_allocated (GiB)": 32.31,
198
+ "step": 108
199
+ },
200
+ {
201
+ "epoch": 1.5492957746478875,
202
+ "grad_norm": 0.58203125,
203
+ "learning_rate": 1.9825991016445387e-05,
204
+ "loss": 0.013250131905078889,
205
+ "memory/device_reserved (GiB)": 33.38,
206
+ "memory/max_active (GiB)": 32.31,
207
+ "memory/max_allocated (GiB)": 32.31,
208
+ "ppl": 1.01334,
209
+ "step": 110,
210
+ "tokens/total": 901120,
211
+ "tokens/train_per_sec_per_gpu": 10.85,
212
+ "tokens/trainable": 389882
213
+ },
214
+ {
215
+ "epoch": 1.6901408450704225,
216
+ "grad_norm": 0.23046875,
217
+ "learning_rate": 1.9722837756298112e-05,
218
+ "loss": 0.011435374617576599,
219
+ "memory/device_reserved (GiB)": 33.38,
220
+ "memory/max_active (GiB)": 32.31,
221
+ "memory/max_allocated (GiB)": 32.31,
222
+ "ppl": 1.0115,
223
+ "step": 120,
224
+ "tokens/total": 983040,
225
+ "tokens/train_per_sec_per_gpu": 10.93,
226
+ "tokens/trainable": 425400
227
+ },
228
+ {
229
+ "epoch": 1.8309859154929577,
230
+ "grad_norm": 0.419921875,
231
+ "learning_rate": 1.9596187973841218e-05,
232
+ "loss": 0.007458872348070145,
233
+ "memory/device_reserved (GiB)": 33.38,
234
+ "memory/max_active (GiB)": 32.31,
235
+ "memory/max_allocated (GiB)": 32.31,
236
+ "ppl": 1.00749,
237
+ "step": 130,
238
+ "tokens/total": 1064960,
239
+ "tokens/train_per_sec_per_gpu": 10.88,
240
+ "tokens/trainable": 460753
241
+ },
242
+ {
243
+ "epoch": 1.971830985915493,
244
+ "grad_norm": 0.74609375,
245
+ "learning_rate": 1.9446347735010765e-05,
246
+ "loss": 0.00692850798368454,
247
+ "memory/device_reserved (GiB)": 33.38,
248
+ "memory/max_active (GiB)": 32.31,
249
+ "memory/max_allocated (GiB)": 32.31,
250
+ "ppl": 1.00695,
251
+ "step": 140,
252
+ "tokens/total": 1146880,
253
+ "tokens/train_per_sec_per_gpu": 11.05,
254
+ "tokens/trainable": 496328
255
+ },
256
+ {
257
+ "epoch": 2.028169014084507,
258
+ "eval_loss": 0.006739533506333828,
259
+ "eval_ppl": 1.00676,
260
+ "eval_runtime": 8.0865,
261
+ "eval_samples_per_second": 61.708,
262
+ "eval_steps_per_second": 61.708,
263
+ "memory/device_reserved (GiB)": 33.38,
264
+ "memory/max_active (GiB)": 32.31,
265
+ "memory/max_allocated (GiB)": 32.31,
266
+ "step": 144
267
+ },
268
+ {
269
+ "epoch": 2.112676056338028,
270
+ "grad_norm": 0.0927734375,
271
+ "learning_rate": 1.9273679148545246e-05,
272
+ "loss": 0.0028733495622873307,
273
+ "memory/device_reserved (GiB)": 33.38,
274
+ "memory/max_active (GiB)": 32.31,
275
+ "memory/max_allocated (GiB)": 32.31,
276
+ "ppl": 1.00288,
277
+ "step": 150,
278
+ "tokens/total": 1228800,
279
+ "tokens/train_per_sec_per_gpu": 11.15,
280
+ "tokens/trainable": 531980
281
+ },
282
+ {
283
+ "epoch": 2.2535211267605635,
284
+ "grad_norm": 0.1025390625,
285
+ "learning_rate": 1.9078599490901984e-05,
286
+ "loss": 0.0013103673234581947,
287
+ "memory/device_reserved (GiB)": 33.38,
288
+ "memory/max_active (GiB)": 32.31,
289
+ "memory/max_allocated (GiB)": 32.31,
290
+ "ppl": 1.00131,
291
+ "step": 160,
292
+ "tokens/total": 1310720,
293
+ "tokens/train_per_sec_per_gpu": 10.97,
294
+ "tokens/trainable": 567303
295
+ },
296
+ {
297
+ "epoch": 2.3943661971830985,
298
+ "grad_norm": 0.150390625,
299
+ "learning_rate": 1.8861580197853423e-05,
300
+ "loss": 0.0011377588845789433,
301
+ "memory/device_reserved (GiB)": 33.38,
302
+ "memory/max_active (GiB)": 32.31,
303
+ "memory/max_allocated (GiB)": 32.31,
304
+ "ppl": 1.00114,
305
+ "step": 170,
306
+ "tokens/total": 1392640,
307
+ "tokens/train_per_sec_per_gpu": 10.75,
308
+ "tokens/trainable": 602482
309
+ },
310
+ {
311
+ "epoch": 2.535211267605634,
312
+ "grad_norm": 0.087890625,
313
+ "learning_rate": 1.862314572520028e-05,
314
+ "loss": 0.001653754524886608,
315
+ "memory/device_reserved (GiB)": 33.38,
316
+ "memory/max_active (GiB)": 32.31,
317
+ "memory/max_allocated (GiB)": 32.31,
318
+ "ppl": 1.00166,
319
+ "step": 180,
320
+ "tokens/total": 1474560,
321
+ "tokens/train_per_sec_per_gpu": 11.3,
322
+ "tokens/trainable": 637743
323
+ },
324
+ {
325
+ "epoch": 2.535211267605634,
326
+ "eval_loss": 0.006014068610966206,
327
+ "eval_ppl": 1.00603,
328
+ "eval_runtime": 8.1497,
329
+ "eval_samples_per_second": 61.229,
330
+ "eval_steps_per_second": 61.229,
331
+ "memory/device_reserved (GiB)": 33.38,
332
+ "memory/max_active (GiB)": 24.06,
333
+ "memory/max_allocated (GiB)": 24.06,
334
+ "step": 180
335
+ },
336
+ {
337
+ "epoch": 2.676056338028169,
338
+ "grad_norm": 0.162109375,
339
+ "learning_rate": 1.8363872281354796e-05,
340
+ "loss": 0.0019721610471606253,
341
+ "memory/device_reserved (GiB)": 33.42,
342
+ "memory/max_active (GiB)": 32.31,
343
+ "memory/max_allocated (GiB)": 32.31,
344
+ "ppl": 1.00197,
345
+ "step": 190,
346
+ "tokens/total": 1556480,
347
+ "tokens/train_per_sec_per_gpu": 10.55,
348
+ "tokens/trainable": 673120
349
+ },
350
+ {
351
+ "epoch": 2.816901408450704,
352
+ "grad_norm": 0.146484375,
353
+ "learning_rate": 1.808438643485698e-05,
354
+ "loss": 0.00105431592091918,
355
+ "memory/device_reserved (GiB)": 33.42,
356
+ "memory/max_active (GiB)": 32.31,
357
+ "memory/max_allocated (GiB)": 32.31,
358
+ "ppl": 1.00105,
359
+ "step": 200,
360
+ "tokens/total": 1638400,
361
+ "tokens/train_per_sec_per_gpu": 11.08,
362
+ "tokens/trainable": 708273
363
+ },
364
+ {
365
+ "epoch": 2.9577464788732395,
366
+ "grad_norm": 0.0830078125,
367
+ "learning_rate": 1.7785363600188894e-05,
368
+ "loss": 0.0011002192273736,
369
+ "memory/device_reserved (GiB)": 33.42,
370
+ "memory/max_active (GiB)": 32.31,
371
+ "memory/max_allocated (GiB)": 32.31,
372
+ "ppl": 1.0011,
373
+ "step": 210,
374
+ "tokens/total": 1720320,
375
+ "tokens/train_per_sec_per_gpu": 10.99,
376
+ "tokens/trainable": 743557
377
+ },
378
+ {
379
+ "epoch": 3.0422535211267605,
380
+ "eval_loss": 0.0053863488137722015,
381
+ "eval_ppl": 1.0054,
382
+ "eval_runtime": 8.0906,
383
+ "eval_samples_per_second": 61.677,
384
+ "eval_steps_per_second": 61.677,
385
+ "memory/device_reserved (GiB)": 33.42,
386
+ "memory/max_active (GiB)": 32.31,
387
+ "memory/max_allocated (GiB)": 32.31,
388
+ "step": 216
389
+ },
390
+ {
391
+ "epoch": 3.0985915492957745,
392
+ "grad_norm": 0.134765625,
393
+ "learning_rate": 1.7467526405546344e-05,
394
+ "loss": 0.001148233562707901,
395
+ "memory/device_reserved (GiB)": 33.38,
396
+ "memory/max_active (GiB)": 32.31,
397
+ "memory/max_allocated (GiB)": 32.31,
398
+ "ppl": 1.00115,
399
+ "step": 220,
400
+ "tokens/total": 1802240,
401
+ "tokens/train_per_sec_per_gpu": 10.46,
402
+ "tokens/trainable": 779210
403
+ },
404
+ {
405
+ "epoch": 3.23943661971831,
406
+ "grad_norm": 0.0791015625,
407
+ "learning_rate": 1.7131642946512314e-05,
408
+ "loss": 0.0010407534427940845,
409
+ "memory/device_reserved (GiB)": 33.38,
410
+ "memory/max_active (GiB)": 32.31,
411
+ "memory/max_allocated (GiB)": 32.31,
412
+ "ppl": 1.00104,
413
+ "step": 230,
414
+ "tokens/total": 1884160,
415
+ "tokens/train_per_sec_per_gpu": 10.43,
416
+ "tokens/trainable": 814465
417
+ },
418
+ {
419
+ "epoch": 3.380281690140845,
420
+ "grad_norm": 0.035888671875,
421
+ "learning_rate": 1.6778524929852513e-05,
422
+ "loss": 0.0008773493580520153,
423
+ "memory/device_reserved (GiB)": 33.38,
424
+ "memory/max_active (GiB)": 32.31,
425
+ "memory/max_allocated (GiB)": 32.31,
426
+ "ppl": 1.00088,
427
+ "step": 240,
428
+ "tokens/total": 1966080,
429
+ "tokens/train_per_sec_per_gpu": 10.32,
430
+ "tokens/trainable": 849679
431
+ },
432
+ {
433
+ "epoch": 3.52112676056338,
434
+ "grad_norm": 0.11328125,
435
+ "learning_rate": 1.640902571191869e-05,
436
+ "loss": 0.00072392662987113,
437
+ "memory/device_reserved (GiB)": 33.38,
438
+ "memory/max_active (GiB)": 32.31,
439
+ "memory/max_allocated (GiB)": 32.31,
440
+ "ppl": 1.00072,
441
+ "step": 250,
442
+ "tokens/total": 2048000,
443
+ "tokens/train_per_sec_per_gpu": 11.0,
444
+ "tokens/trainable": 884891
445
+ },
446
+ {
447
+ "epoch": 3.5492957746478875,
448
+ "eval_loss": 0.005309262778609991,
449
+ "eval_ppl": 1.00532,
450
+ "eval_runtime": 8.2154,
451
+ "eval_samples_per_second": 60.74,
452
+ "eval_steps_per_second": 60.74,
453
+ "memory/device_reserved (GiB)": 33.38,
454
+ "memory/max_active (GiB)": 32.31,
455
+ "memory/max_allocated (GiB)": 32.31,
456
+ "step": 252
457
+ },
458
+ {
459
+ "epoch": 3.6619718309859155,
460
+ "grad_norm": 0.037109375,
461
+ "learning_rate": 1.6024038236400246e-05,
462
+ "loss": 0.0005804049782454967,
463
+ "memory/device_reserved (GiB)": 33.42,
464
+ "memory/max_active (GiB)": 32.31,
465
+ "memory/max_allocated (GiB)": 32.31,
466
+ "ppl": 1.00058,
467
+ "step": 260,
468
+ "tokens/total": 2129920,
469
+ "tokens/train_per_sec_per_gpu": 11.12,
470
+ "tokens/trainable": 920273
471
+ },
472
+ {
473
+ "epoch": 3.802816901408451,
474
+ "grad_norm": 0.022705078125,
475
+ "learning_rate": 1.562449287640781e-05,
476
+ "loss": 0.0007765244692564011,
477
+ "memory/device_reserved (GiB)": 33.42,
478
+ "memory/max_active (GiB)": 32.31,
479
+ "memory/max_allocated (GiB)": 32.31,
480
+ "ppl": 1.00078,
481
+ "step": 270,
482
+ "tokens/total": 2211840,
483
+ "tokens/train_per_sec_per_gpu": 10.51,
484
+ "tokens/trainable": 955808
485
+ },
486
+ {
487
+ "epoch": 3.943661971830986,
488
+ "grad_norm": 0.04833984375,
489
+ "learning_rate": 1.5211355186103655e-05,
490
+ "loss": 0.0007016387302428484,
491
+ "memory/device_reserved (GiB)": 33.42,
492
+ "memory/max_active (GiB)": 32.31,
493
+ "memory/max_allocated (GiB)": 32.31,
494
+ "ppl": 1.0007,
495
+ "step": 280,
496
+ "tokens/total": 2293760,
497
+ "tokens/train_per_sec_per_gpu": 10.15,
498
+ "tokens/trainable": 991166
499
+ },
500
+ {
501
+ "epoch": 4.056338028169014,
502
+ "eval_loss": 0.005245804321020842,
503
+ "eval_ppl": 1.00526,
504
+ "eval_runtime": 8.0041,
505
+ "eval_samples_per_second": 62.343,
506
+ "eval_steps_per_second": 62.343,
507
+ "memory/device_reserved (GiB)": 33.42,
508
+ "memory/max_active (GiB)": 32.31,
509
+ "memory/max_allocated (GiB)": 32.31,
510
+ "step": 288
511
+ },
512
+ {
513
+ "epoch": 4.084507042253521,
514
+ "grad_norm": 0.047607421875,
515
+ "learning_rate": 1.4785623567312492e-05,
516
+ "loss": 0.000780455581843853,
517
+ "memory/device_reserved (GiB)": 33.42,
518
+ "memory/max_active (GiB)": 32.31,
519
+ "memory/max_allocated (GiB)": 32.31,
520
+ "ppl": 1.00078,
521
+ "step": 290,
522
+ "tokens/total": 2375680,
523
+ "tokens/train_per_sec_per_gpu": 11.01,
524
+ "tokens/trainable": 1026615
525
+ },
526
+ {
527
+ "epoch": 4.225352112676056,
528
+ "grad_norm": 0.061767578125,
529
+ "learning_rate": 1.4348326856751496e-05,
530
+ "loss": 0.0005466958507895469,
531
+ "memory/device_reserved (GiB)": 33.42,
532
+ "memory/max_active (GiB)": 32.31,
533
+ "memory/max_allocated (GiB)": 32.31,
534
+ "ppl": 1.00055,
535
+ "step": 300,
536
+ "tokens/total": 2457600,
537
+ "tokens/train_per_sec_per_gpu": 10.99,
538
+ "tokens/trainable": 1062010
539
+ },
540
+ {
541
+ "epoch": 4.366197183098592,
542
+ "grad_norm": 0.0341796875,
543
+ "learning_rate": 1.3900521839710428e-05,
544
+ "loss": 0.0006260770373046398,
545
+ "memory/device_reserved (GiB)": 33.42,
546
+ "memory/max_active (GiB)": 32.31,
547
+ "memory/max_allocated (GiB)": 32.31,
548
+ "ppl": 1.00063,
549
+ "step": 310,
550
+ "tokens/total": 2539520,
551
+ "tokens/train_per_sec_per_gpu": 11.12,
552
+ "tokens/trainable": 1097505
553
+ },
554
+ {
555
+ "epoch": 4.507042253521127,
556
+ "grad_norm": 0.03857421875,
557
+ "learning_rate": 1.3443290696190335e-05,
558
+ "loss": 0.0005137504544109106,
559
+ "memory/device_reserved (GiB)": 33.42,
560
+ "memory/max_active (GiB)": 32.31,
561
+ "memory/max_allocated (GiB)": 32.31,
562
+ "ppl": 1.00051,
563
+ "step": 320,
564
+ "tokens/total": 2621440,
565
+ "tokens/train_per_sec_per_gpu": 10.32,
566
+ "tokens/trainable": 1132758
567
+ },
568
+ {
569
+ "epoch": 4.563380281690141,
570
+ "eval_loss": 0.005256574135273695,
571
+ "eval_ppl": 1.00527,
572
+ "eval_runtime": 8.1646,
573
+ "eval_samples_per_second": 61.118,
574
+ "eval_steps_per_second": 61.118,
575
+ "memory/device_reserved (GiB)": 33.42,
576
+ "memory/max_active (GiB)": 32.31,
577
+ "memory/max_allocated (GiB)": 32.31,
578
+ "step": 324
579
+ },
580
+ {
581
+ "epoch": 4.647887323943662,
582
+ "grad_norm": 0.046142578125,
583
+ "learning_rate": 1.2977738385672558e-05,
584
+ "loss": 0.0005908971652388573,
585
+ "memory/device_reserved (GiB)": 33.42,
586
+ "memory/max_active (GiB)": 32.31,
587
+ "memory/max_allocated (GiB)": 32.31,
588
+ "ppl": 1.00059,
589
+ "step": 330,
590
+ "tokens/total": 2703360,
591
+ "tokens/train_per_sec_per_gpu": 10.25,
592
+ "tokens/trainable": 1168161
593
+ },
594
+ {
595
+ "epoch": 4.788732394366197,
596
+ "grad_norm": 0.050537109375,
597
+ "learning_rate": 1.250498997683813e-05,
598
+ "loss": 0.0006883179303258657,
599
+ "memory/device_reserved (GiB)": 33.42,
600
+ "memory/max_active (GiB)": 32.31,
601
+ "memory/max_allocated (GiB)": 32.31,
602
+ "ppl": 1.00069,
603
+ "step": 340,
604
+ "tokens/total": 2785280,
605
+ "tokens/train_per_sec_per_gpu": 10.28,
606
+ "tokens/trainable": 1203434
607
+ },
608
+ {
609
+ "epoch": 4.929577464788732,
610
+ "grad_norm": 0.04736328125,
611
+ "learning_rate": 1.202618792869063e-05,
612
+ "loss": 0.0006954259239137173,
613
+ "memory/device_reserved (GiB)": 33.42,
614
+ "memory/max_active (GiB)": 32.31,
615
+ "memory/max_allocated (GiB)": 32.31,
616
+ "ppl": 1.0007,
617
+ "step": 350,
618
+ "tokens/total": 2867200,
619
+ "tokens/train_per_sec_per_gpu": 11.09,
620
+ "tokens/trainable": 1238829
621
+ },
622
+ {
623
+ "epoch": 5.070422535211268,
624
+ "grad_norm": 0.05517578125,
625
+ "learning_rate": 1.1542489329653024e-05,
626
+ "loss": 0.000603723106905818,
627
+ "memory/device_reserved (GiB)": 33.42,
628
+ "memory/max_active (GiB)": 32.31,
629
+ "memory/max_allocated (GiB)": 32.31,
630
+ "ppl": 1.0006,
631
+ "step": 360,
632
+ "tokens/total": 2949120,
633
+ "tokens/train_per_sec_per_gpu": 11.02,
634
+ "tokens/trainable": 1274110
635
+ },
636
+ {
637
+ "epoch": 5.070422535211268,
638
+ "eval_loss": 0.005255201831459999,
639
+ "eval_ppl": 1.00527,
640
+ "eval_runtime": 8.0148,
641
+ "eval_samples_per_second": 62.26,
642
+ "eval_steps_per_second": 62.26,
643
+ "memory/device_reserved (GiB)": 33.42,
644
+ "memory/max_active (GiB)": 24.06,
645
+ "memory/max_allocated (GiB)": 24.06,
646
+ "step": 360
647
+ },
648
+ {
649
+ "epoch": 5.211267605633803,
650
+ "grad_norm": 0.0615234375,
651
+ "learning_rate": 1.1055063101310581e-05,
652
+ "loss": 0.0005278075113892555,
653
+ "memory/device_reserved (GiB)": 33.38,
654
+ "memory/max_active (GiB)": 32.31,
655
+ "memory/max_allocated (GiB)": 32.31,
656
+ "ppl": 1.00053,
657
+ "step": 370,
658
+ "tokens/total": 3031040,
659
+ "tokens/train_per_sec_per_gpu": 11.13,
660
+ "tokens/trainable": 1309814
661
+ },
662
+ {
663
+ "epoch": 5.352112676056338,
664
+ "grad_norm": 0.037841796875,
665
+ "learning_rate": 1.0565087173557396e-05,
666
+ "loss": 0.0005071659106761217,
667
+ "memory/device_reserved (GiB)": 33.38,
668
+ "memory/max_active (GiB)": 32.31,
669
+ "memory/max_allocated (GiB)": 32.31,
670
+ "ppl": 1.00051,
671
+ "step": 380,
672
+ "tokens/total": 3112960,
673
+ "tokens/train_per_sec_per_gpu": 11.21,
674
+ "tokens/trainable": 1345167
675
+ },
676
+ {
677
+ "epoch": 5.492957746478873,
678
+ "grad_norm": 0.02880859375,
679
+ "learning_rate": 1.0073745637973125e-05,
680
+ "loss": 0.0005827666725963354,
681
+ "memory/device_reserved (GiB)": 33.38,
682
+ "memory/max_active (GiB)": 32.31,
683
+ "memory/max_allocated (GiB)": 32.31,
684
+ "ppl": 1.00058,
685
+ "step": 390,
686
+ "tokens/total": 3194880,
687
+ "tokens/train_per_sec_per_gpu": 11.14,
688
+ "tokens/trainable": 1380603
689
+ },
690
+ {
691
+ "epoch": 5.577464788732394,
692
+ "eval_loss": 0.005226313602179289,
693
+ "eval_ppl": 1.00524,
694
+ "eval_runtime": 8.1606,
695
+ "eval_samples_per_second": 61.147,
696
+ "eval_steps_per_second": 61.147,
697
+ "memory/device_reserved (GiB)": 33.38,
698
+ "memory/max_active (GiB)": 32.31,
699
+ "memory/max_allocated (GiB)": 32.31,
700
+ "step": 396
701
+ },
702
+ {
703
+ "epoch": 5.633802816901408,
704
+ "grad_norm": 0.04150390625,
705
+ "learning_rate": 9.582225886309217e-06,
706
+ "loss": 0.0006697049364447593,
707
+ "memory/device_reserved (GiB)": 33.42,
708
+ "memory/max_active (GiB)": 32.31,
709
+ "memory/max_allocated (GiB)": 32.31,
710
+ "ppl": 1.00067,
711
+ "step": 400,
712
+ "tokens/total": 3276800,
713
+ "tokens/train_per_sec_per_gpu": 9.71,
714
+ "tokens/trainable": 1415822
715
+ },
716
+ {
717
+ "epoch": 5.774647887323944,
718
+ "grad_norm": 0.04833984375,
719
+ "learning_rate": 9.091715740999829e-06,
720
+ "loss": 0.0006844023242592812,
721
+ "memory/device_reserved (GiB)": 33.42,
722
+ "memory/max_active (GiB)": 32.31,
723
+ "memory/max_allocated (GiB)": 32.31,
724
+ "ppl": 1.00068,
725
+ "step": 410,
726
+ "tokens/total": 3358720,
727
+ "tokens/train_per_sec_per_gpu": 10.39,
728
+ "tokens/trainable": 1451140
729
+ },
730
+ {
731
+ "epoch": 5.915492957746479,
732
+ "grad_norm": 0.0478515625,
733
+ "learning_rate": 8.60340058463194e-06,
734
+ "loss": 0.0004917650483548641,
735
+ "memory/device_reserved (GiB)": 33.42,
736
+ "memory/max_active (GiB)": 32.31,
737
+ "memory/max_allocated (GiB)": 32.31,
738
+ "ppl": 1.00049,
739
+ "step": 420,
740
+ "tokens/total": 3440640,
741
+ "tokens/train_per_sec_per_gpu": 11.04,
742
+ "tokens/trainable": 1486616
743
+ },
744
+ {
745
+ "epoch": 6.056338028169014,
746
+ "grad_norm": 0.047607421875,
747
+ "learning_rate": 8.118460495311687e-06,
748
+ "loss": 0.0005019189789891243,
749
+ "memory/device_reserved (GiB)": 33.42,
750
+ "memory/max_active (GiB)": 32.31,
751
+ "memory/max_allocated (GiB)": 32.31,
752
+ "ppl": 1.0005,
753
+ "step": 430,
754
+ "tokens/total": 3522560,
755
+ "tokens/train_per_sec_per_gpu": 9.55,
756
+ "tokens/trainable": 1522001
757
+ },
758
+ {
759
+ "epoch": 6.084507042253521,
760
+ "eval_loss": 0.005209571216255426,
761
+ "eval_ppl": 1.00522,
762
+ "eval_runtime": 8.0091,
763
+ "eval_samples_per_second": 62.304,
764
+ "eval_steps_per_second": 62.304,
765
+ "memory/device_reserved (GiB)": 33.42,
766
+ "memory/max_active (GiB)": 32.31,
767
+ "memory/max_allocated (GiB)": 32.31,
768
+ "step": 432
769
+ },
770
+ {
771
+ "epoch": 6.197183098591549,
772
+ "grad_norm": 0.134765625,
773
+ "learning_rate": 7.638067394849672e-06,
774
+ "loss": 0.000784461572766304,
775
+ "memory/device_reserved (GiB)": 33.38,
776
+ "memory/max_active (GiB)": 32.31,
777
+ "memory/max_allocated (GiB)": 32.31,
778
+ "ppl": 1.00078,
779
+ "step": 440,
780
+ "tokens/total": 3604480,
781
+ "tokens/train_per_sec_per_gpu": 11.09,
782
+ "tokens/trainable": 1557368
783
+ },
784
+ {
785
+ "epoch": 6.338028169014084,
786
+ "grad_norm": 0.05712890625,
787
+ "learning_rate": 7.163382216657033e-06,
788
+ "loss": 0.0005743904504925013,
789
+ "memory/device_reserved (GiB)": 33.38,
790
+ "memory/max_active (GiB)": 32.31,
791
+ "memory/max_allocated (GiB)": 32.31,
792
+ "ppl": 1.00057,
793
+ "step": 450,
794
+ "tokens/total": 3686400,
795
+ "tokens/train_per_sec_per_gpu": 10.46,
796
+ "tokens/trainable": 1592826
797
+ },
798
+ {
799
+ "epoch": 6.47887323943662,
800
+ "grad_norm": 0.03076171875,
801
+ "learning_rate": 6.6955521001964526e-06,
802
+ "loss": 0.0004704087972640991,
803
+ "memory/device_reserved (GiB)": 33.38,
804
+ "memory/max_active (GiB)": 32.31,
805
+ "memory/max_allocated (GiB)": 32.31,
806
+ "ppl": 1.00047,
807
+ "step": 460,
808
+ "tokens/total": 3768320,
809
+ "tokens/train_per_sec_per_gpu": 10.41,
810
+ "tokens/trainable": 1628143
811
+ },
812
+ {
813
+ "epoch": 6.591549295774648,
814
+ "eval_loss": 0.005213646683841944,
815
+ "eval_ppl": 1.00523,
816
+ "eval_runtime": 8.1767,
817
+ "eval_samples_per_second": 61.027,
818
+ "eval_steps_per_second": 61.027,
819
+ "memory/device_reserved (GiB)": 33.38,
820
+ "memory/max_active (GiB)": 32.31,
821
+ "memory/max_allocated (GiB)": 32.31,
822
+ "step": 468
823
+ },
824
+ {
825
+ "epoch": 6.619718309859155,
826
+ "grad_norm": 0.04736328125,
827
+ "learning_rate": 6.2357076187680325e-06,
828
+ "loss": 0.0004482423886656761,
829
+ "memory/device_reserved (GiB)": 33.42,
830
+ "memory/max_active (GiB)": 32.31,
831
+ "memory/max_allocated (GiB)": 32.31,
832
+ "ppl": 1.00045,
833
+ "step": 470,
834
+ "tokens/total": 3850240,
835
+ "tokens/train_per_sec_per_gpu": 10.41,
836
+ "tokens/trainable": 1663545
837
+ },
838
+ {
839
+ "epoch": 6.76056338028169,
840
+ "grad_norm": 0.0380859375,
841
+ "learning_rate": 5.78496004732952e-06,
842
+ "loss": 0.0004957062192261219,
843
+ "memory/device_reserved (GiB)": 33.42,
844
+ "memory/max_active (GiB)": 32.31,
845
+ "memory/max_allocated (GiB)": 32.31,
846
+ "ppl": 1.0005,
847
+ "step": 480,
848
+ "tokens/total": 3932160,
849
+ "tokens/train_per_sec_per_gpu": 10.4,
850
+ "tokens/trainable": 1699181
851
+ },
852
+ {
853
+ "epoch": 6.901408450704225,
854
+ "grad_norm": 0.0361328125,
855
+ "learning_rate": 5.344398676953526e-06,
856
+ "loss": 0.000544667523354292,
857
+ "memory/device_reserved (GiB)": 33.42,
858
+ "memory/max_active (GiB)": 32.31,
859
+ "memory/max_allocated (GiB)": 32.31,
860
+ "ppl": 1.00054,
861
+ "step": 490,
862
+ "tokens/total": 4014080,
863
+ "tokens/train_per_sec_per_gpu": 10.21,
864
+ "tokens/trainable": 1734683
865
+ },
866
+ {
867
+ "epoch": 7.042253521126761,
868
+ "grad_norm": 0.03759765625,
869
+ "learning_rate": 4.915088182411674e-06,
870
+ "loss": 0.0004880569409579039,
871
+ "memory/device_reserved (GiB)": 33.42,
872
+ "memory/max_active (GiB)": 32.31,
873
+ "memory/max_allocated (GiB)": 32.31,
874
+ "ppl": 1.00049,
875
+ "step": 500,
876
+ "tokens/total": 4096000,
877
+ "tokens/train_per_sec_per_gpu": 11.44,
878
+ "tokens/trainable": 1770105
879
+ },
880
+ {
881
+ "epoch": 7.098591549295775,
882
+ "eval_loss": 0.00523661682382226,
883
+ "eval_ppl": 1.00525,
884
+ "eval_runtime": 8.0317,
885
+ "eval_samples_per_second": 62.129,
886
+ "eval_steps_per_second": 62.129,
887
+ "memory/device_reserved (GiB)": 33.42,
888
+ "memory/max_active (GiB)": 32.31,
889
+ "memory/max_allocated (GiB)": 32.31,
890
+ "step": 504
891
+ },
892
+ {
893
+ "epoch": 7.183098591549296,
894
+ "grad_norm": 0.0257568359375,
895
+ "learning_rate": 4.498066049247344e-06,
896
+ "loss": 0.0005014274269342422,
897
+ "memory/device_reserved (GiB)": 33.38,
898
+ "memory/max_active (GiB)": 32.31,
899
+ "memory/max_allocated (GiB)": 32.31,
900
+ "ppl": 1.0005,
901
+ "step": 510,
902
+ "tokens/total": 4177920,
903
+ "tokens/train_per_sec_per_gpu": 10.53,
904
+ "tokens/trainable": 1805486
905
+ },
906
+ {
907
+ "epoch": 7.323943661971831,
908
+ "grad_norm": 0.03466796875,
909
+ "learning_rate": 4.094340066554742e-06,
910
+ "loss": 0.0006260806694626808,
911
+ "memory/device_reserved (GiB)": 33.38,
912
+ "memory/max_active (GiB)": 32.31,
913
+ "memory/max_allocated (GiB)": 32.31,
914
+ "ppl": 1.00063,
915
+ "step": 520,
916
+ "tokens/total": 4259840,
917
+ "tokens/train_per_sec_per_gpu": 10.8,
918
+ "tokens/trainable": 1840987
919
+ },
920
+ {
921
+ "epoch": 7.464788732394366,
922
+ "grad_norm": 0.036865234375,
923
+ "learning_rate": 3.7048858915233665e-06,
924
+ "loss": 0.0005529518239200115,
925
+ "memory/device_reserved (GiB)": 33.38,
926
+ "memory/max_active (GiB)": 32.31,
927
+ "memory/max_allocated (GiB)": 32.31,
928
+ "ppl": 1.00055,
929
+ "step": 530,
930
+ "tokens/total": 4341760,
931
+ "tokens/train_per_sec_per_gpu": 10.79,
932
+ "tokens/trainable": 1876364
933
+ },
934
+ {
935
+ "epoch": 7.605633802816901,
936
+ "grad_norm": 0.034912109375,
937
+ "learning_rate": 3.330644691633492e-06,
938
+ "loss": 0.0005182296968996525,
939
+ "memory/device_reserved (GiB)": 33.38,
940
+ "memory/max_active (GiB)": 32.31,
941
+ "memory/max_allocated (GiB)": 32.31,
942
+ "ppl": 1.00052,
943
+ "step": 540,
944
+ "tokens/total": 4423680,
945
+ "tokens/train_per_sec_per_gpu": 10.9,
946
+ "tokens/trainable": 1911823
947
+ },
948
+ {
949
+ "epoch": 7.605633802816901,
950
+ "eval_loss": 0.005228586029261351,
951
+ "eval_ppl": 1.00524,
952
+ "eval_runtime": 8.2216,
953
+ "eval_samples_per_second": 60.694,
954
+ "eval_steps_per_second": 60.694,
955
+ "memory/device_reserved (GiB)": 33.38,
956
+ "memory/max_active (GiB)": 24.06,
957
+ "memory/max_allocated (GiB)": 24.06,
958
+ "step": 540
959
+ },
960
+ {
961
+ "epoch": 7.746478873239437,
962
+ "grad_norm": 0.025634765625,
963
+ "learning_rate": 2.9725208702005736e-06,
964
+ "loss": 0.0005175352562218904,
965
+ "memory/device_reserved (GiB)": 33.42,
966
+ "memory/max_active (GiB)": 32.31,
967
+ "memory/max_allocated (GiB)": 32.31,
968
+ "ppl": 1.00052,
969
+ "step": 550,
970
+ "tokens/total": 4505600,
971
+ "tokens/train_per_sec_per_gpu": 11.35,
972
+ "tokens/trainable": 1947212
973
+ },
974
+ {
975
+ "epoch": 7.887323943661972,
976
+ "grad_norm": 0.0341796875,
977
+ "learning_rate": 2.631379880765107e-06,
978
+ "loss": 0.0005252756644040346,
979
+ "memory/device_reserved (GiB)": 33.42,
980
+ "memory/max_active (GiB)": 32.31,
981
+ "memory/max_allocated (GiB)": 32.31,
982
+ "ppl": 1.00053,
983
+ "step": 560,
984
+ "tokens/total": 4587520,
985
+ "tokens/train_per_sec_per_gpu": 11.28,
986
+ "tokens/trainable": 1982716
987
+ },
988
+ {
989
+ "epoch": 8.028169014084508,
990
+ "grad_norm": 0.046142578125,
991
+ "learning_rate": 2.3080461356097938e-06,
992
+ "loss": 0.0004542431328445673,
993
+ "memory/device_reserved (GiB)": 33.42,
994
+ "memory/max_active (GiB)": 32.31,
995
+ "memory/max_allocated (GiB)": 32.31,
996
+ "ppl": 1.00045,
997
+ "step": 570,
998
+ "tokens/total": 4669440,
999
+ "tokens/train_per_sec_per_gpu": 10.28,
1000
+ "tokens/trainable": 2018112
1001
+ },
1002
+ {
1003
+ "epoch": 8.112676056338028,
1004
+ "eval_loss": 0.005249009933322668,
1005
+ "eval_ppl": 1.00526,
1006
+ "eval_runtime": 8.1416,
1007
+ "eval_samples_per_second": 61.29,
1008
+ "eval_steps_per_second": 61.29,
1009
+ "memory/device_reserved (GiB)": 33.42,
1010
+ "memory/max_active (GiB)": 32.31,
1011
+ "memory/max_allocated (GiB)": 32.31,
1012
+ "step": 576
1013
+ },
1014
+ {
1015
+ "epoch": 8.169014084507042,
1016
+ "grad_norm": 0.043212890625,
1017
+ "learning_rate": 2.0033010134583085e-06,
1018
+ "loss": 0.000499055441468954,
1019
+ "memory/device_reserved (GiB)": 33.42,
1020
+ "memory/max_active (GiB)": 32.31,
1021
+ "memory/max_allocated (GiB)": 32.31,
1022
+ "ppl": 1.0005,
1023
+ "step": 580,
1024
+ "tokens/total": 4751360,
1025
+ "tokens/train_per_sec_per_gpu": 10.13,
1026
+ "tokens/trainable": 2053569
1027
+ },
1028
+ {
1029
+ "epoch": 8.309859154929578,
1030
+ "grad_norm": 0.029052734375,
1031
+ "learning_rate": 1.7178809711703525e-06,
1032
+ "loss": 0.0006432968657463789,
1033
+ "memory/device_reserved (GiB)": 33.42,
1034
+ "memory/max_active (GiB)": 32.31,
1035
+ "memory/max_allocated (GiB)": 32.31,
1036
+ "ppl": 1.00064,
1037
+ "step": 590,
1038
+ "tokens/total": 4833280,
1039
+ "tokens/train_per_sec_per_gpu": 10.98,
1040
+ "tokens/trainable": 2089056
1041
+ },
1042
+ {
1043
+ "epoch": 8.450704225352112,
1044
+ "grad_norm": 0.04443359375,
1045
+ "learning_rate": 1.452475763996326e-06,
1046
+ "loss": 0.0005107030272483825,
1047
+ "memory/device_reserved (GiB)": 33.42,
1048
+ "memory/max_active (GiB)": 32.31,
1049
+ "memory/max_allocated (GiB)": 32.31,
1050
+ "ppl": 1.00051,
1051
+ "step": 600,
1052
+ "tokens/total": 4915200,
1053
+ "tokens/train_per_sec_per_gpu": 11.03,
1054
+ "tokens/trainable": 2124556
1055
+ },
1056
+ {
1057
+ "epoch": 8.591549295774648,
1058
+ "grad_norm": 0.04150390625,
1059
+ "learning_rate": 1.207726778692625e-06,
1060
+ "loss": 0.00045921015553176405,
1061
+ "memory/device_reserved (GiB)": 33.42,
1062
+ "memory/max_active (GiB)": 32.31,
1063
+ "memory/max_allocated (GiB)": 32.31,
1064
+ "ppl": 1.00046,
1065
+ "step": 610,
1066
+ "tokens/total": 4997120,
1067
+ "tokens/train_per_sec_per_gpu": 11.16,
1068
+ "tokens/trainable": 2159952
1069
+ },
1070
+ {
1071
+ "epoch": 8.619718309859154,
1072
+ "eval_loss": 0.0052222758531570435,
1073
+ "eval_ppl": 1.00524,
1074
+ "eval_runtime": 8.2144,
1075
+ "eval_samples_per_second": 60.747,
1076
+ "eval_steps_per_second": 60.747,
1077
+ "memory/device_reserved (GiB)": 33.42,
1078
+ "memory/max_active (GiB)": 32.31,
1079
+ "memory/max_allocated (GiB)": 32.31,
1080
+ "step": 612
1081
+ },
1082
+ {
1083
+ "epoch": 8.732394366197184,
1084
+ "grad_norm": 0.03369140625,
1085
+ "learning_rate": 9.842254835257792e-07,
1086
+ "loss": 0.0005835144780576229,
1087
+ "memory/device_reserved (GiB)": 33.38,
1088
+ "memory/max_active (GiB)": 32.31,
1089
+ "memory/max_allocated (GiB)": 32.31,
1090
+ "ppl": 1.00058,
1091
+ "step": 620,
1092
+ "tokens/total": 5079040,
1093
+ "tokens/train_per_sec_per_gpu": 10.43,
1094
+ "tokens/trainable": 2195458
1095
+ },
1096
+ {
1097
+ "epoch": 8.873239436619718,
1098
+ "grad_norm": 0.035400390625,
1099
+ "learning_rate": 7.825119989112173e-07,
1100
+ "loss": 0.00045958077535033224,
1101
+ "memory/device_reserved (GiB)": 33.38,
1102
+ "memory/max_active (GiB)": 32.31,
1103
+ "memory/max_allocated (GiB)": 32.31,
1104
+ "ppl": 1.00046,
1105
+ "step": 630,
1106
+ "tokens/total": 5160960,
1107
+ "tokens/train_per_sec_per_gpu": 10.35,
1108
+ "tokens/trainable": 2230834
1109
+ },
1110
+ {
1111
+ "epoch": 9.014084507042254,
1112
+ "grad_norm": 0.04248046875,
1113
+ "learning_rate": 6.030737921409169e-07,
1114
+ "loss": 0.0005321558564901352,
1115
+ "memory/device_reserved (GiB)": 33.38,
1116
+ "memory/max_active (GiB)": 32.31,
1117
+ "memory/max_allocated (GiB)": 32.31,
1118
+ "ppl": 1.00053,
1119
+ "step": 640,
1120
+ "tokens/total": 5242880,
1121
+ "tokens/train_per_sec_per_gpu": 10.91,
1122
+ "tokens/trainable": 2266102
1123
+ },
1124
+ {
1125
+ "epoch": 9.126760563380282,
1126
+ "eval_loss": 0.005234830547124147,
1127
+ "eval_ppl": 1.00525,
1128
+ "eval_runtime": 8.0713,
1129
+ "eval_samples_per_second": 61.824,
1130
+ "eval_steps_per_second": 61.824,
1131
+ "memory/device_reserved (GiB)": 33.38,
1132
+ "memory/max_active (GiB)": 32.31,
1133
+ "memory/max_allocated (GiB)": 32.31,
1134
+ "step": 648
1135
+ },
1136
+ {
1137
+ "epoch": 9.154929577464788,
1138
+ "grad_norm": 0.06103515625,
1139
+ "learning_rate": 4.4634449935427203e-07,
1140
+ "loss": 0.0005488845054060221,
1141
+ "memory/device_reserved (GiB)": 33.42,
1142
+ "memory/max_active (GiB)": 32.31,
1143
+ "memory/max_allocated (GiB)": 32.31,
1144
+ "ppl": 1.00055,
1145
+ "step": 650,
1146
+ "tokens/total": 5324800,
1147
+ "tokens/train_per_sec_per_gpu": 10.42,
1148
+ "tokens/trainable": 2301492
1149
+ },
1150
+ {
1151
+ "epoch": 9.295774647887324,
1152
+ "grad_norm": 0.049072265625,
1153
+ "learning_rate": 3.127028775990515e-07,
1154
+ "loss": 0.0004435606300830841,
1155
+ "memory/device_reserved (GiB)": 33.42,
1156
+ "memory/max_active (GiB)": 32.31,
1157
+ "memory/max_allocated (GiB)": 32.31,
1158
+ "ppl": 1.00044,
1159
+ "step": 660,
1160
+ "tokens/total": 5406720,
1161
+ "tokens/train_per_sec_per_gpu": 10.97,
1162
+ "tokens/trainable": 2337078
1163
+ },
1164
+ {
1165
+ "epoch": 9.43661971830986,
1166
+ "grad_norm": 0.032958984375,
1167
+ "learning_rate": 2.02471889514948e-07,
1168
+ "loss": 0.0005032818298786879,
1169
+ "memory/device_reserved (GiB)": 33.42,
1170
+ "memory/max_active (GiB)": 32.31,
1171
+ "memory/max_allocated (GiB)": 32.31,
1172
+ "ppl": 1.0005,
1173
+ "step": 670,
1174
+ "tokens/total": 5488640,
1175
+ "tokens/train_per_sec_per_gpu": 9.53,
1176
+ "tokens/trainable": 2372502
1177
+ },
1178
+ {
1179
+ "epoch": 9.577464788732394,
1180
+ "grad_norm": 0.062255859375,
1181
+ "learning_rate": 1.1591792285167603e-07,
1182
+ "loss": 0.0005544357933104038,
1183
+ "memory/device_reserved (GiB)": 33.42,
1184
+ "memory/max_active (GiB)": 32.31,
1185
+ "memory/max_allocated (GiB)": 32.31,
1186
+ "ppl": 1.00055,
1187
+ "step": 680,
1188
+ "tokens/total": 5570560,
1189
+ "tokens/train_per_sec_per_gpu": 10.35,
1190
+ "tokens/trainable": 2407771
1191
+ },
1192
+ {
1193
+ "epoch": 9.633802816901408,
1194
+ "eval_loss": 0.005193324759602547,
1195
+ "eval_ppl": 1.00521,
1196
+ "eval_runtime": 8.0758,
1197
+ "eval_samples_per_second": 61.789,
1198
+ "eval_steps_per_second": 61.789,
1199
+ "memory/device_reserved (GiB)": 33.42,
1200
+ "memory/max_active (GiB)": 32.31,
1201
+ "memory/max_allocated (GiB)": 32.31,
1202
+ "step": 684
1203
+ },
1204
+ {
1205
+ "epoch": 9.71830985915493,
1206
+ "grad_norm": 0.0302734375,
1207
+ "learning_rate": 5.325014670776951e-08,
1208
+ "loss": 0.0006243317387998104,
1209
+ "memory/device_reserved (GiB)": 33.42,
1210
+ "memory/max_active (GiB)": 32.31,
1211
+ "memory/max_allocated (GiB)": 32.31,
1212
+ "ppl": 1.00062,
1213
+ "step": 690,
1214
+ "tokens/total": 5652480,
1215
+ "tokens/train_per_sec_per_gpu": 11.15,
1216
+ "tokens/trainable": 2443031
1217
+ },
1218
+ {
1219
+ "epoch": 9.859154929577464,
1220
+ "grad_norm": 0.04248046875,
1221
+ "learning_rate": 1.4620006045816814e-08,
1222
+ "loss": 0.000509471632540226,
1223
+ "memory/device_reserved (GiB)": 33.42,
1224
+ "memory/max_active (GiB)": 32.31,
1225
+ "memory/max_allocated (GiB)": 32.31,
1226
+ "ppl": 1.00051,
1227
+ "step": 700,
1228
+ "tokens/total": 5734400,
1229
+ "tokens/train_per_sec_per_gpu": 10.7,
1230
+ "tokens/trainable": 2478484
1231
+ },
1232
+ {
1233
+ "epoch": 10.0,
1234
+ "grad_norm": 0.032958984375,
1235
+ "learning_rate": 1.2085570569642102e-10,
1236
+ "loss": 0.00048337155021727085,
1237
+ "memory/device_reserved (GiB)": 33.42,
1238
+ "memory/max_active (GiB)": 32.31,
1239
+ "memory/max_allocated (GiB)": 32.31,
1240
+ "ppl": 1.00048,
1241
+ "step": 710,
1242
+ "tokens/total": 5816320,
1243
+ "tokens/train_per_sec_per_gpu": 10.67,
1244
+ "tokens/trainable": 2513851
1245
+ }
1246
+ ],
1247
+ "logging_steps": 10,
1248
+ "max_steps": 710,
1249
+ "num_input_tokens_seen": 0,
1250
+ "num_train_epochs": 10,
1251
+ "save_steps": 71,
1252
+ "stateful_callbacks": {
1253
+ "TrainerControl": {
1254
+ "args": {
1255
+ "should_epoch_stop": false,
1256
+ "should_evaluate": false,
1257
+ "should_log": false,
1258
+ "should_save": true,
1259
+ "should_training_stop": true
1260
+ },
1261
+ "attributes": {}
1262
+ }
1263
+ },
1264
+ "total_flos": 1.2680200886157312e+17,
1265
+ "train_batch_size": 1,
1266
+ "trial_name": null,
1267
+ "trial_params": null
1268
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/checkpoint-710/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c56b5e3616cf3d2a8bf8aabf031ff524a587484ec751048484926648eb0ad82
3
+ size 7121
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/debug.log ADDED
The diff for this file is too large to render. See raw diff
 
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/eval_results.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ category,filename,total,correct,accuracy,format_found,format_accuracy,errors_count
2
+ knowledge,test_alpaca_results,450,156,34.67,0,0.00,294
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/eval_summary.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "overall": {
3
+ "total": 450,
4
+ "correct": 156,
5
+ "accuracy": 34.67,
6
+ "format_found": 0,
7
+ "format_accuracy": 0.0
8
+ },
9
+ "n_errors": 294,
10
+ "results_file": "/home/jiaruil5/math_rl/mix_teachers/r3lit_rl/mix_teachers/checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/test_alpaca_results.jsonl"
11
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/test_alpaca_converted.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/atomic/test_alpaca_results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/eval_results.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ category,filename,total,correct,accuracy,format_found,format_accuracy,errors_count
2
+ knowledge,test_alpaca_results,450,155,34.44,0,0.00,295
3
+ knowledge,test_alpaca_results,499,143,28.66,420,84.17,356
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/eval_summary.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "overall": {
3
+ "total": 499,
4
+ "correct": 143,
5
+ "accuracy": 28.66,
6
+ "format_found": 420,
7
+ "format_accuracy": 84.17
8
+ },
9
+ "n_errors": 356,
10
+ "results_file": "/home/jiaruil5/math_rl/mix_teachers/r3lit_rl/mix_teachers/checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/test_alpaca_results.jsonl"
11
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/test_alpaca_converted.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context/compositional_2step/test_alpaca_results.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/eval_results.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ category,filename,total,correct,accuracy,format_found,format_accuracy,errors_count
2
+ knowledge,test_alpaca_results,450,55,12.22,426,94.67,395
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/eval_summary.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "overall": {
3
+ "total": 450,
4
+ "correct": 55,
5
+ "accuracy": 12.22,
6
+ "format_found": 426,
7
+ "format_accuracy": 94.67
8
+ },
9
+ "n_errors": 395,
10
+ "results_file": "/home/jiaruil5/math_rl/mix_teachers/r3lit_rl/mix_teachers/checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/test_alpaca_results.jsonl"
11
+ }
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/test_alpaca_converted.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
checkpoints/knowledge/atomic_full_then_2step_full_sft_t20260305/eval_results_with_context_cot/atomic/test_alpaca_results.jsonl ADDED
The diff for this file is too large to render. See raw diff