Hugging Face's logo

RandomDud123456
/
Mini-Project_Models

TensorBoard
Safetensors
Model card Files
xet
Metrics Community
RandomDud123456 commited on
Commit
713c540
·
verified ·
1 Parent(s): 6aab40e

Upload 33 files

Browse files
Files changed (33) hide show
  1. qwen3-150m-multilingual-checkpoints/checkpoint-108000/config.json +45 -0
  2. qwen3-150m-multilingual-checkpoints/checkpoint-108000/generation_config.json +7 -0
  3. qwen3-150m-multilingual-checkpoints/checkpoint-108000/model.safetensors +3 -0
  4. qwen3-150m-multilingual-checkpoints/checkpoint-108000/optimizer.pt +3 -0
  5. qwen3-150m-multilingual-checkpoints/checkpoint-108000/rng_state.pth +3 -0
  6. qwen3-150m-multilingual-checkpoints/checkpoint-108000/scheduler.pt +3 -0
  7. qwen3-150m-multilingual-checkpoints/checkpoint-108000/special_tokens_map.json +30 -0
  8. qwen3-150m-multilingual-checkpoints/checkpoint-108000/tokenizer.json +0 -0
  9. qwen3-150m-multilingual-checkpoints/checkpoint-108000/tokenizer_config.json +76 -0
  10. qwen3-150m-multilingual-checkpoints/checkpoint-108000/trainer_state.json +3595 -0
  11. qwen3-150m-multilingual-checkpoints/checkpoint-108000/training_args.bin +3 -0
  12. qwen3-150m-multilingual-checkpoints/checkpoint-112500/config.json +45 -0
  13. qwen3-150m-multilingual-checkpoints/checkpoint-112500/generation_config.json +7 -0
  14. qwen3-150m-multilingual-checkpoints/checkpoint-112500/model.safetensors +3 -0
  15. qwen3-150m-multilingual-checkpoints/checkpoint-112500/optimizer.pt +3 -0
  16. qwen3-150m-multilingual-checkpoints/checkpoint-112500/rng_state.pth +3 -0
  17. qwen3-150m-multilingual-checkpoints/checkpoint-112500/scheduler.pt +3 -0
  18. qwen3-150m-multilingual-checkpoints/checkpoint-112500/special_tokens_map.json +30 -0
  19. qwen3-150m-multilingual-checkpoints/checkpoint-112500/tokenizer.json +0 -0
  20. qwen3-150m-multilingual-checkpoints/checkpoint-112500/tokenizer_config.json +76 -0
  21. qwen3-150m-multilingual-checkpoints/checkpoint-112500/trainer_state.json +0 -0
  22. qwen3-150m-multilingual-checkpoints/checkpoint-112500/training_args.bin +3 -0
  23. qwen3-150m-multilingual-checkpoints/checkpoint-115888/config.json +45 -0
  24. qwen3-150m-multilingual-checkpoints/checkpoint-115888/generation_config.json +7 -0
  25. qwen3-150m-multilingual-checkpoints/checkpoint-115888/model.safetensors +3 -0
  26. qwen3-150m-multilingual-checkpoints/checkpoint-115888/optimizer.pt +3 -0
  27. qwen3-150m-multilingual-checkpoints/checkpoint-115888/rng_state.pth +3 -0
  28. qwen3-150m-multilingual-checkpoints/checkpoint-115888/scheduler.pt +3 -0
  29. qwen3-150m-multilingual-checkpoints/checkpoint-115888/special_tokens_map.json +30 -0
  30. qwen3-150m-multilingual-checkpoints/checkpoint-115888/tokenizer.json +0 -0
  31. qwen3-150m-multilingual-checkpoints/checkpoint-115888/tokenizer_config.json +76 -0
  32. qwen3-150m-multilingual-checkpoints/checkpoint-115888/trainer_state.json +0 -0
  33. qwen3-150m-multilingual-checkpoints/checkpoint-115888/training_args.bin +3 -0
qwen3-150m-multilingual-checkpoints/checkpoint-108000/config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 2,
8
+ "eos_token_id": 3,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 768,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2048,
14
+ "layer_types": [
15
+ "full_attention",
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
+ ],
28
+ "max_position_embeddings": 2048,
29
+ "max_window_layers": 28,
30
+ "model_type": "qwen3",
31
+ "num_attention_heads": 12,
32
+ "num_hidden_layers": 12,
33
+ "num_key_value_heads": 12,
34
+ "pad_token_id": 1,
35
+ "rms_norm_eps": 1e-06,
36
+ "rope_scaling": null,
37
+ "rope_theta": 10000.0,
38
+ "sliding_window": null,
39
+ "tie_word_embeddings": true,
40
+ "torch_dtype": "float32",
41
+ "transformers_version": "4.55.4",
42
+ "use_cache": true,
43
+ "use_sliding_window": false,
44
+ "vocab_size": 32000
45
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-108000/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 2,
4
+ "eos_token_id": 3,
5
+ "pad_token_id": 1,
6
+ "transformers_version": "4.55.4"
7
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-108000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8bdb4b1e5edb84f68cf03dfa6d32da72bfc35178892f849831dfcf3d4785cbe1
3
+ size 551392792
qwen3-150m-multilingual-checkpoints/checkpoint-108000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:846fea38bbc0fecb76f918fb2e38c334bfc6ad57d5c90e003bf5736150dd6c59
3
+ size 1102868538
qwen3-150m-multilingual-checkpoints/checkpoint-108000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82338f637d63f63bdaa2c79b13c19bddfff6ae35887ddee8f695a00fa1cc0f97
3
+ size 14244
qwen3-150m-multilingual-checkpoints/checkpoint-108000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:773489253c6493d88890f19da3295745d901e1b0b2301b27cc4631acd55e5c77
3
+ size 1064
qwen3-150m-multilingual-checkpoints/checkpoint-108000/special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<pad>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-108000/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3-150m-multilingual-checkpoints/checkpoint-108000/tokenizer_config.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "<unk>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<pad>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "2": {
20
+ "content": "<s>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "3": {
28
+ "content": "</s>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ },
35
+ "4": {
36
+ "content": "<en>",
37
+ "lstrip": false,
38
+ "normalized": false,
39
+ "rstrip": false,
40
+ "single_word": false,
41
+ "special": true
42
+ },
43
+ "5": {
44
+ "content": "<te>",
45
+ "lstrip": false,
46
+ "normalized": false,
47
+ "rstrip": false,
48
+ "single_word": false,
49
+ "special": true
50
+ },
51
+ "6": {
52
+ "content": "<mai>",
53
+ "lstrip": false,
54
+ "normalized": false,
55
+ "rstrip": false,
56
+ "single_word": false,
57
+ "special": true
58
+ },
59
+ "7": {
60
+ "content": "<mask>",
61
+ "lstrip": false,
62
+ "normalized": false,
63
+ "rstrip": false,
64
+ "single_word": false,
65
+ "special": true
66
+ }
67
+ },
68
+ "bos_token": "<s>",
69
+ "clean_up_tokenization_spaces": false,
70
+ "eos_token": "</s>",
71
+ "extra_special_tokens": {},
72
+ "model_max_length": 1000000000000000019884624838656,
73
+ "pad_token": "<pad>",
74
+ "tokenizer_class": "PreTrainedTokenizerFast",
75
+ "unk_token": "<unk>"
76
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-108000/trainer_state.json ADDED
@@ -0,0 +1,3595 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 24000,
3
+ "best_metric": 2.080108404159546,
4
+ "best_model_checkpoint": "./qwen3-150m-multilingual-checkpoints\\checkpoint-24000",
5
+ "epoch": 0.2070978472610231,
6
+ "eval_steps": 1000,
7
+ "global_step": 24000,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.0004314538484604648,
14
+ "grad_norm": 6.4089813232421875,
15
+ "learning_rate": 9.8e-07,
16
+ "loss": 10.257,
17
+ "step": 50
18
+ },
19
+ {
20
+ "epoch": 0.0008629076969209296,
21
+ "grad_norm": 3.5916900634765625,
22
+ "learning_rate": 1.98e-06,
23
+ "loss": 9.1849,
24
+ "step": 100
25
+ },
26
+ {
27
+ "epoch": 0.0012943615453813945,
28
+ "grad_norm": 3.186964988708496,
29
+ "learning_rate": 2.9800000000000003e-06,
30
+ "loss": 8.3015,
31
+ "step": 150
32
+ },
33
+ {
34
+ "epoch": 0.0017258153938418592,
35
+ "grad_norm": 2.7763795852661133,
36
+ "learning_rate": 3.98e-06,
37
+ "loss": 7.9443,
38
+ "step": 200
39
+ },
40
+ {
41
+ "epoch": 0.002157269242302324,
42
+ "grad_norm": 3.0714266300201416,
43
+ "learning_rate": 4.98e-06,
44
+ "loss": 7.3653,
45
+ "step": 250
46
+ },
47
+ {
48
+ "epoch": 0.002588723090762789,
49
+ "grad_norm": 2.2576541900634766,
50
+ "learning_rate": 5.98e-06,
51
+ "loss": 6.6621,
52
+ "step": 300
53
+ },
54
+ {
55
+ "epoch": 0.0030201769392232537,
56
+ "grad_norm": 2.4825663566589355,
57
+ "learning_rate": 6.98e-06,
58
+ "loss": 6.2397,
59
+ "step": 350
60
+ },
61
+ {
62
+ "epoch": 0.0034516307876837184,
63
+ "grad_norm": 1.9788450002670288,
64
+ "learning_rate": 7.98e-06,
65
+ "loss": 5.8553,
66
+ "step": 400
67
+ },
68
+ {
69
+ "epoch": 0.003883084636144183,
70
+ "grad_norm": 2.507742166519165,
71
+ "learning_rate": 8.98e-06,
72
+ "loss": 5.5195,
73
+ "step": 450
74
+ },
75
+ {
76
+ "epoch": 0.004314538484604648,
77
+ "grad_norm": 3.946438789367676,
78
+ "learning_rate": 9.980000000000001e-06,
79
+ "loss": 5.2475,
80
+ "step": 500
81
+ },
82
+ {
83
+ "epoch": 0.004745992333065113,
84
+ "grad_norm": 2.7235612869262695,
85
+ "learning_rate": 1.098e-05,
86
+ "loss": 4.9971,
87
+ "step": 550
88
+ },
89
+ {
90
+ "epoch": 0.005177446181525578,
91
+ "grad_norm": 2.88761830329895,
92
+ "learning_rate": 1.198e-05,
93
+ "loss": 4.7636,
94
+ "step": 600
95
+ },
96
+ {
97
+ "epoch": 0.005608900029986042,
98
+ "grad_norm": 2.373903512954712,
99
+ "learning_rate": 1.2980000000000001e-05,
100
+ "loss": 4.5525,
101
+ "step": 650
102
+ },
103
+ {
104
+ "epoch": 0.006040353878446507,
105
+ "grad_norm": 2.945600748062134,
106
+ "learning_rate": 1.3980000000000002e-05,
107
+ "loss": 4.447,
108
+ "step": 700
109
+ },
110
+ {
111
+ "epoch": 0.006471807726906972,
112
+ "grad_norm": 3.239373207092285,
113
+ "learning_rate": 1.4979999999999999e-05,
114
+ "loss": 4.3253,
115
+ "step": 750
116
+ },
117
+ {
118
+ "epoch": 0.006903261575367437,
119
+ "grad_norm": 3.2961957454681396,
120
+ "learning_rate": 1.598e-05,
121
+ "loss": 4.1988,
122
+ "step": 800
123
+ },
124
+ {
125
+ "epoch": 0.007334715423827902,
126
+ "grad_norm": 3.4848122596740723,
127
+ "learning_rate": 1.698e-05,
128
+ "loss": 4.1261,
129
+ "step": 850
130
+ },
131
+ {
132
+ "epoch": 0.007766169272288366,
133
+ "grad_norm": 5.849564075469971,
134
+ "learning_rate": 1.798e-05,
135
+ "loss": 4.063,
136
+ "step": 900
137
+ },
138
+ {
139
+ "epoch": 0.00819762312074883,
140
+ "grad_norm": 3.391148567199707,
141
+ "learning_rate": 1.898e-05,
142
+ "loss": 3.973,
143
+ "step": 950
144
+ },
145
+ {
146
+ "epoch": 0.008629076969209296,
147
+ "grad_norm": 2.368360757827759,
148
+ "learning_rate": 1.9980000000000002e-05,
149
+ "loss": 3.8773,
150
+ "step": 1000
151
+ },
152
+ {
153
+ "epoch": 0.008629076969209296,
154
+ "eval_loss": 3.885983943939209,
155
+ "eval_runtime": 2846.8484,
156
+ "eval_samples_per_second": 26.566,
157
+ "eval_steps_per_second": 13.283,
158
+ "step": 1000
159
+ },
160
+ {
161
+ "epoch": 0.00906053081766976,
162
+ "grad_norm": 3.211322069168091,
163
+ "learning_rate": 2.098e-05,
164
+ "loss": 3.8554,
165
+ "step": 1050
166
+ },
167
+ {
168
+ "epoch": 0.009491984666130226,
169
+ "grad_norm": 3.2583465576171875,
170
+ "learning_rate": 2.198e-05,
171
+ "loss": 3.771,
172
+ "step": 1100
173
+ },
174
+ {
175
+ "epoch": 0.009923438514590691,
176
+ "grad_norm": 3.918938636779785,
177
+ "learning_rate": 2.298e-05,
178
+ "loss": 3.7751,
179
+ "step": 1150
180
+ },
181
+ {
182
+ "epoch": 0.010354892363051156,
183
+ "grad_norm": 3.7250285148620605,
184
+ "learning_rate": 2.398e-05,
185
+ "loss": 3.6291,
186
+ "step": 1200
187
+ },
188
+ {
189
+ "epoch": 0.01078634621151162,
190
+ "grad_norm": 3.9162580966949463,
191
+ "learning_rate": 2.498e-05,
192
+ "loss": 3.6013,
193
+ "step": 1250
194
+ },
195
+ {
196
+ "epoch": 0.011217800059972085,
197
+ "grad_norm": 3.402956008911133,
198
+ "learning_rate": 2.598e-05,
199
+ "loss": 3.5911,
200
+ "step": 1300
201
+ },
202
+ {
203
+ "epoch": 0.01164925390843255,
204
+ "grad_norm": 3.117097854614258,
205
+ "learning_rate": 2.698e-05,
206
+ "loss": 3.556,
207
+ "step": 1350
208
+ },
209
+ {
210
+ "epoch": 0.012080707756893015,
211
+ "grad_norm": 2.198148012161255,
212
+ "learning_rate": 2.798e-05,
213
+ "loss": 3.4642,
214
+ "step": 1400
215
+ },
216
+ {
217
+ "epoch": 0.01251216160535348,
218
+ "grad_norm": 2.386687994003296,
219
+ "learning_rate": 2.898e-05,
220
+ "loss": 3.4827,
221
+ "step": 1450
222
+ },
223
+ {
224
+ "epoch": 0.012943615453813943,
225
+ "grad_norm": 2.13498854637146,
226
+ "learning_rate": 2.998e-05,
227
+ "loss": 3.4887,
228
+ "step": 1500
229
+ },
230
+ {
231
+ "epoch": 0.013375069302274408,
232
+ "grad_norm": 2.7259521484375,
233
+ "learning_rate": 3.0980000000000005e-05,
234
+ "loss": 3.4152,
235
+ "step": 1550
236
+ },
237
+ {
238
+ "epoch": 0.013806523150734874,
239
+ "grad_norm": 2.1376330852508545,
240
+ "learning_rate": 3.198e-05,
241
+ "loss": 3.3847,
242
+ "step": 1600
243
+ },
244
+ {
245
+ "epoch": 0.014237976999195339,
246
+ "grad_norm": 2.23085355758667,
247
+ "learning_rate": 3.298e-05,
248
+ "loss": 3.3698,
249
+ "step": 1650
250
+ },
251
+ {
252
+ "epoch": 0.014669430847655804,
253
+ "grad_norm": 2.2056055068969727,
254
+ "learning_rate": 3.398e-05,
255
+ "loss": 3.3082,
256
+ "step": 1700
257
+ },
258
+ {
259
+ "epoch": 0.015100884696116269,
260
+ "grad_norm": 2.404055595397949,
261
+ "learning_rate": 3.498e-05,
262
+ "loss": 3.218,
263
+ "step": 1750
264
+ },
265
+ {
266
+ "epoch": 0.015532338544576732,
267
+ "grad_norm": 2.5828022956848145,
268
+ "learning_rate": 3.5980000000000004e-05,
269
+ "loss": 3.2422,
270
+ "step": 1800
271
+ },
272
+ {
273
+ "epoch": 0.015963792393037197,
274
+ "grad_norm": 1.9683724641799927,
275
+ "learning_rate": 3.698e-05,
276
+ "loss": 3.1683,
277
+ "step": 1850
278
+ },
279
+ {
280
+ "epoch": 0.01639524624149766,
281
+ "grad_norm": 1.9096944332122803,
282
+ "learning_rate": 3.7980000000000006e-05,
283
+ "loss": 3.2061,
284
+ "step": 1900
285
+ },
286
+ {
287
+ "epoch": 0.016826700089958128,
288
+ "grad_norm": 1.851697325706482,
289
+ "learning_rate": 3.898e-05,
290
+ "loss": 3.1872,
291
+ "step": 1950
292
+ },
293
+ {
294
+ "epoch": 0.01725815393841859,
295
+ "grad_norm": 1.6863641738891602,
296
+ "learning_rate": 3.998e-05,
297
+ "loss": 3.1296,
298
+ "step": 2000
299
+ },
300
+ {
301
+ "epoch": 0.01725815393841859,
302
+ "eval_loss": 3.1464343070983887,
303
+ "eval_runtime": 2765.1571,
304
+ "eval_samples_per_second": 27.35,
305
+ "eval_steps_per_second": 13.675,
306
+ "step": 2000
307
+ },
308
+ {
309
+ "epoch": 0.017689607786879058,
310
+ "grad_norm": 1.947570562362671,
311
+ "learning_rate": 4.0980000000000004e-05,
312
+ "loss": 3.1589,
313
+ "step": 2050
314
+ },
315
+ {
316
+ "epoch": 0.01812106163533952,
317
+ "grad_norm": 1.683140516281128,
318
+ "learning_rate": 4.198e-05,
319
+ "loss": 3.1179,
320
+ "step": 2100
321
+ },
322
+ {
323
+ "epoch": 0.018552515483799988,
324
+ "grad_norm": 1.7471286058425903,
325
+ "learning_rate": 4.2980000000000005e-05,
326
+ "loss": 3.1115,
327
+ "step": 2150
328
+ },
329
+ {
330
+ "epoch": 0.01898396933226045,
331
+ "grad_norm": 1.4706114530563354,
332
+ "learning_rate": 4.398e-05,
333
+ "loss": 3.061,
334
+ "step": 2200
335
+ },
336
+ {
337
+ "epoch": 0.019415423180720915,
338
+ "grad_norm": 1.6069599390029907,
339
+ "learning_rate": 4.498e-05,
340
+ "loss": 3.0233,
341
+ "step": 2250
342
+ },
343
+ {
344
+ "epoch": 0.019846877029181382,
345
+ "grad_norm": 1.2190749645233154,
346
+ "learning_rate": 4.5980000000000004e-05,
347
+ "loss": 3.0544,
348
+ "step": 2300
349
+ },
350
+ {
351
+ "epoch": 0.020278330877641845,
352
+ "grad_norm": 1.2313108444213867,
353
+ "learning_rate": 4.698e-05,
354
+ "loss": 3.0468,
355
+ "step": 2350
356
+ },
357
+ {
358
+ "epoch": 0.020709784726102312,
359
+ "grad_norm": 1.1253961324691772,
360
+ "learning_rate": 4.7980000000000005e-05,
361
+ "loss": 2.9753,
362
+ "step": 2400
363
+ },
364
+ {
365
+ "epoch": 0.021141238574562776,
366
+ "grad_norm": 1.1460264921188354,
367
+ "learning_rate": 4.898e-05,
368
+ "loss": 2.955,
369
+ "step": 2450
370
+ },
371
+ {
372
+ "epoch": 0.02157269242302324,
373
+ "grad_norm": 1.2542543411254883,
374
+ "learning_rate": 4.9980000000000006e-05,
375
+ "loss": 2.9418,
376
+ "step": 2500
377
+ },
378
+ {
379
+ "epoch": 0.022004146271483706,
380
+ "grad_norm": 1.0879753828048706,
381
+ "learning_rate": 5.098e-05,
382
+ "loss": 2.9779,
383
+ "step": 2550
384
+ },
385
+ {
386
+ "epoch": 0.02243560011994417,
387
+ "grad_norm": 1.0592762231826782,
388
+ "learning_rate": 5.198000000000001e-05,
389
+ "loss": 2.9218,
390
+ "step": 2600
391
+ },
392
+ {
393
+ "epoch": 0.022867053968404636,
394
+ "grad_norm": 1.1562455892562866,
395
+ "learning_rate": 5.2980000000000004e-05,
396
+ "loss": 2.9204,
397
+ "step": 2650
398
+ },
399
+ {
400
+ "epoch": 0.0232985078168651,
401
+ "grad_norm": 1.013615369796753,
402
+ "learning_rate": 5.3979999999999995e-05,
403
+ "loss": 2.9179,
404
+ "step": 2700
405
+ },
406
+ {
407
+ "epoch": 0.023729961665325563,
408
+ "grad_norm": 1.0790969133377075,
409
+ "learning_rate": 5.498e-05,
410
+ "loss": 2.9417,
411
+ "step": 2750
412
+ },
413
+ {
414
+ "epoch": 0.02416141551378603,
415
+ "grad_norm": 0.9028209447860718,
416
+ "learning_rate": 5.5979999999999996e-05,
417
+ "loss": 2.8743,
418
+ "step": 2800
419
+ },
420
+ {
421
+ "epoch": 0.024592869362246493,
422
+ "grad_norm": 0.8891520500183105,
423
+ "learning_rate": 5.698e-05,
424
+ "loss": 2.8934,
425
+ "step": 2850
426
+ },
427
+ {
428
+ "epoch": 0.02502432321070696,
429
+ "grad_norm": 1.0406967401504517,
430
+ "learning_rate": 5.7980000000000004e-05,
431
+ "loss": 2.8679,
432
+ "step": 2900
433
+ },
434
+ {
435
+ "epoch": 0.025455777059167423,
436
+ "grad_norm": 0.938779354095459,
437
+ "learning_rate": 5.898e-05,
438
+ "loss": 2.8523,
439
+ "step": 2950
440
+ },
441
+ {
442
+ "epoch": 0.025887230907627887,
443
+ "grad_norm": 0.8158487677574158,
444
+ "learning_rate": 5.9980000000000005e-05,
445
+ "loss": 2.827,
446
+ "step": 3000
447
+ },
448
+ {
449
+ "epoch": 0.025887230907627887,
450
+ "eval_loss": 2.8239262104034424,
451
+ "eval_runtime": 2782.2115,
452
+ "eval_samples_per_second": 27.183,
453
+ "eval_steps_per_second": 13.591,
454
+ "step": 3000
455
+ },
456
+ {
457
+ "epoch": 0.026318684756088354,
458
+ "grad_norm": 0.8945357203483582,
459
+ "learning_rate": 6.098e-05,
460
+ "loss": 2.8226,
461
+ "step": 3050
462
+ },
463
+ {
464
+ "epoch": 0.026750138604548817,
465
+ "grad_norm": 0.8776882886886597,
466
+ "learning_rate": 6.198e-05,
467
+ "loss": 2.7725,
468
+ "step": 3100
469
+ },
470
+ {
471
+ "epoch": 0.027181592453009284,
472
+ "grad_norm": 0.8972533941268921,
473
+ "learning_rate": 6.298000000000001e-05,
474
+ "loss": 2.795,
475
+ "step": 3150
476
+ },
477
+ {
478
+ "epoch": 0.027613046301469747,
479
+ "grad_norm": 0.7877374887466431,
480
+ "learning_rate": 6.398000000000001e-05,
481
+ "loss": 2.8148,
482
+ "step": 3200
483
+ },
484
+ {
485
+ "epoch": 0.028044500149930214,
486
+ "grad_norm": 0.7548953890800476,
487
+ "learning_rate": 6.498e-05,
488
+ "loss": 2.7789,
489
+ "step": 3250
490
+ },
491
+ {
492
+ "epoch": 0.028475953998390677,
493
+ "grad_norm": 0.7239817380905151,
494
+ "learning_rate": 6.598e-05,
495
+ "loss": 2.7506,
496
+ "step": 3300
497
+ },
498
+ {
499
+ "epoch": 0.02890740784685114,
500
+ "grad_norm": 0.8170995116233826,
501
+ "learning_rate": 6.698e-05,
502
+ "loss": 2.7415,
503
+ "step": 3350
504
+ },
505
+ {
506
+ "epoch": 0.029338861695311608,
507
+ "grad_norm": 0.7359878420829773,
508
+ "learning_rate": 6.798e-05,
509
+ "loss": 2.7643,
510
+ "step": 3400
511
+ },
512
+ {
513
+ "epoch": 0.02977031554377207,
514
+ "grad_norm": 0.7392879128456116,
515
+ "learning_rate": 6.898e-05,
516
+ "loss": 2.7541,
517
+ "step": 3450
518
+ },
519
+ {
520
+ "epoch": 0.030201769392232538,
521
+ "grad_norm": 0.7366235852241516,
522
+ "learning_rate": 6.998e-05,
523
+ "loss": 2.7391,
524
+ "step": 3500
525
+ },
526
+ {
527
+ "epoch": 0.030633223240693,
528
+ "grad_norm": 0.6895750164985657,
529
+ "learning_rate": 7.098e-05,
530
+ "loss": 2.7625,
531
+ "step": 3550
532
+ },
533
+ {
534
+ "epoch": 0.031064677089153465,
535
+ "grad_norm": 0.6986274719238281,
536
+ "learning_rate": 7.198e-05,
537
+ "loss": 2.714,
538
+ "step": 3600
539
+ },
540
+ {
541
+ "epoch": 0.03149613093761393,
542
+ "grad_norm": 0.7208825945854187,
543
+ "learning_rate": 7.298000000000001e-05,
544
+ "loss": 2.6966,
545
+ "step": 3650
546
+ },
547
+ {
548
+ "epoch": 0.031927584786074395,
549
+ "grad_norm": 0.7059122323989868,
550
+ "learning_rate": 7.398e-05,
551
+ "loss": 2.7436,
552
+ "step": 3700
553
+ },
554
+ {
555
+ "epoch": 0.03235903863453486,
556
+ "grad_norm": 0.6853424310684204,
557
+ "learning_rate": 7.498e-05,
558
+ "loss": 2.696,
559
+ "step": 3750
560
+ },
561
+ {
562
+ "epoch": 0.03279049248299532,
563
+ "grad_norm": 0.6768801212310791,
564
+ "learning_rate": 7.598e-05,
565
+ "loss": 2.7349,
566
+ "step": 3800
567
+ },
568
+ {
569
+ "epoch": 0.03322194633145579,
570
+ "grad_norm": 0.7065775990486145,
571
+ "learning_rate": 7.698000000000001e-05,
572
+ "loss": 2.6729,
573
+ "step": 3850
574
+ },
575
+ {
576
+ "epoch": 0.033653400179916255,
577
+ "grad_norm": 0.6956036686897278,
578
+ "learning_rate": 7.798000000000001e-05,
579
+ "loss": 2.6356,
580
+ "step": 3900
581
+ },
582
+ {
583
+ "epoch": 0.03408485402837672,
584
+ "grad_norm": 0.6357026696205139,
585
+ "learning_rate": 7.897999999999999e-05,
586
+ "loss": 2.5888,
587
+ "step": 3950
588
+ },
589
+ {
590
+ "epoch": 0.03451630787683718,
591
+ "grad_norm": 0.6445960998535156,
592
+ "learning_rate": 7.998e-05,
593
+ "loss": 2.6392,
594
+ "step": 4000
595
+ },
596
+ {
597
+ "epoch": 0.03451630787683718,
598
+ "eval_loss": 2.6430516242980957,
599
+ "eval_runtime": 2751.3719,
600
+ "eval_samples_per_second": 27.487,
601
+ "eval_steps_per_second": 13.744,
602
+ "step": 4000
603
+ },
604
+ {
605
+ "epoch": 0.03494776172529765,
606
+ "grad_norm": 0.6698920130729675,
607
+ "learning_rate": 8.098e-05,
608
+ "loss": 2.6621,
609
+ "step": 4050
610
+ },
611
+ {
612
+ "epoch": 0.035379215573758116,
613
+ "grad_norm": 0.6329808235168457,
614
+ "learning_rate": 8.198e-05,
615
+ "loss": 2.6276,
616
+ "step": 4100
617
+ },
618
+ {
619
+ "epoch": 0.035810669422218576,
620
+ "grad_norm": 0.6281711459159851,
621
+ "learning_rate": 8.298000000000001e-05,
622
+ "loss": 2.6093,
623
+ "step": 4150
624
+ },
625
+ {
626
+ "epoch": 0.03624212327067904,
627
+ "grad_norm": 0.6180399060249329,
628
+ "learning_rate": 8.398e-05,
629
+ "loss": 2.645,
630
+ "step": 4200
631
+ },
632
+ {
633
+ "epoch": 0.03667357711913951,
634
+ "grad_norm": 0.620257556438446,
635
+ "learning_rate": 8.498e-05,
636
+ "loss": 2.6221,
637
+ "step": 4250
638
+ },
639
+ {
640
+ "epoch": 0.037105030967599976,
641
+ "grad_norm": 0.6216643452644348,
642
+ "learning_rate": 8.598e-05,
643
+ "loss": 2.5944,
644
+ "step": 4300
645
+ },
646
+ {
647
+ "epoch": 0.037536484816060436,
648
+ "grad_norm": 0.6922804117202759,
649
+ "learning_rate": 8.698000000000001e-05,
650
+ "loss": 2.5989,
651
+ "step": 4350
652
+ },
653
+ {
654
+ "epoch": 0.0379679386645209,
655
+ "grad_norm": 0.5947348475456238,
656
+ "learning_rate": 8.798000000000001e-05,
657
+ "loss": 2.5872,
658
+ "step": 4400
659
+ },
660
+ {
661
+ "epoch": 0.03839939251298137,
662
+ "grad_norm": 0.6008750796318054,
663
+ "learning_rate": 8.898e-05,
664
+ "loss": 2.5942,
665
+ "step": 4450
666
+ },
667
+ {
668
+ "epoch": 0.03883084636144183,
669
+ "grad_norm": 0.604884922504425,
670
+ "learning_rate": 8.998e-05,
671
+ "loss": 2.5764,
672
+ "step": 4500
673
+ },
674
+ {
675
+ "epoch": 0.0392623002099023,
676
+ "grad_norm": 0.6107255816459656,
677
+ "learning_rate": 9.098000000000001e-05,
678
+ "loss": 2.5419,
679
+ "step": 4550
680
+ },
681
+ {
682
+ "epoch": 0.039693754058362764,
683
+ "grad_norm": 0.5564243793487549,
684
+ "learning_rate": 9.198e-05,
685
+ "loss": 2.565,
686
+ "step": 4600
687
+ },
688
+ {
689
+ "epoch": 0.040125207906823224,
690
+ "grad_norm": 0.5812260508537292,
691
+ "learning_rate": 9.298e-05,
692
+ "loss": 2.5734,
693
+ "step": 4650
694
+ },
695
+ {
696
+ "epoch": 0.04055666175528369,
697
+ "grad_norm": 0.5526607632637024,
698
+ "learning_rate": 9.398e-05,
699
+ "loss": 2.586,
700
+ "step": 4700
701
+ },
702
+ {
703
+ "epoch": 0.04098811560374416,
704
+ "grad_norm": 0.6254459023475647,
705
+ "learning_rate": 9.498e-05,
706
+ "loss": 2.4797,
707
+ "step": 4750
708
+ },
709
+ {
710
+ "epoch": 0.041419569452204624,
711
+ "grad_norm": 0.5473371148109436,
712
+ "learning_rate": 9.598e-05,
713
+ "loss": 2.5454,
714
+ "step": 4800
715
+ },
716
+ {
717
+ "epoch": 0.041851023300665084,
718
+ "grad_norm": 0.5498418807983398,
719
+ "learning_rate": 9.698000000000001e-05,
720
+ "loss": 2.5581,
721
+ "step": 4850
722
+ },
723
+ {
724
+ "epoch": 0.04228247714912555,
725
+ "grad_norm": 0.565069317817688,
726
+ "learning_rate": 9.798000000000001e-05,
727
+ "loss": 2.5014,
728
+ "step": 4900
729
+ },
730
+ {
731
+ "epoch": 0.04271393099758602,
732
+ "grad_norm": 0.5747342705726624,
733
+ "learning_rate": 9.898e-05,
734
+ "loss": 2.5312,
735
+ "step": 4950
736
+ },
737
+ {
738
+ "epoch": 0.04314538484604648,
739
+ "grad_norm": 0.5452004075050354,
740
+ "learning_rate": 9.998e-05,
741
+ "loss": 2.5331,
742
+ "step": 5000
743
+ },
744
+ {
745
+ "epoch": 0.04314538484604648,
746
+ "eval_loss": 2.526031494140625,
747
+ "eval_runtime": 3108.7919,
748
+ "eval_samples_per_second": 24.327,
749
+ "eval_steps_per_second": 12.164,
750
+ "step": 5000
751
+ },
752
+ {
753
+ "epoch": 0.043576838694506945,
754
+ "grad_norm": 0.5333559513092041,
755
+ "learning_rate": 9.999995182044957e-05,
756
+ "loss": 2.5314,
757
+ "step": 5050
758
+ },
759
+ {
760
+ "epoch": 0.04400829254296741,
761
+ "grad_norm": 0.5360190272331238,
762
+ "learning_rate": 9.99998033288046e-05,
763
+ "loss": 2.5069,
764
+ "step": 5100
765
+ },
766
+ {
767
+ "epoch": 0.04443974639142787,
768
+ "grad_norm": 0.5400934219360352,
769
+ "learning_rate": 9.999955450529664e-05,
770
+ "loss": 2.5327,
771
+ "step": 5150
772
+ },
773
+ {
774
+ "epoch": 0.04487120023988834,
775
+ "grad_norm": 0.5095759630203247,
776
+ "learning_rate": 9.999920535042497e-05,
777
+ "loss": 2.496,
778
+ "step": 5200
779
+ },
780
+ {
781
+ "epoch": 0.045302654088348805,
782
+ "grad_norm": 0.5121382474899292,
783
+ "learning_rate": 9.999875586489024e-05,
784
+ "loss": 2.4837,
785
+ "step": 5250
786
+ },
787
+ {
788
+ "epoch": 0.04573410793680927,
789
+ "grad_norm": 0.5363543033599854,
790
+ "learning_rate": 9.999820604959439e-05,
791
+ "loss": 2.513,
792
+ "step": 5300
793
+ },
794
+ {
795
+ "epoch": 0.04616556178526973,
796
+ "grad_norm": 0.5149393081665039,
797
+ "learning_rate": 9.99975559056407e-05,
798
+ "loss": 2.5031,
799
+ "step": 5350
800
+ },
801
+ {
802
+ "epoch": 0.0465970156337302,
803
+ "grad_norm": 0.5039998888969421,
804
+ "learning_rate": 9.999680543433381e-05,
805
+ "loss": 2.488,
806
+ "step": 5400
807
+ },
808
+ {
809
+ "epoch": 0.047028469482190666,
810
+ "grad_norm": 0.5018443465232849,
811
+ "learning_rate": 9.999595463717962e-05,
812
+ "loss": 2.4613,
813
+ "step": 5450
814
+ },
815
+ {
816
+ "epoch": 0.047459923330651126,
817
+ "grad_norm": 0.49239611625671387,
818
+ "learning_rate": 9.999500351588537e-05,
819
+ "loss": 2.4413,
820
+ "step": 5500
821
+ },
822
+ {
823
+ "epoch": 0.04789137717911159,
824
+ "grad_norm": 0.4872280955314636,
825
+ "learning_rate": 9.999395207235966e-05,
826
+ "loss": 2.454,
827
+ "step": 5550
828
+ },
829
+ {
830
+ "epoch": 0.04832283102757206,
831
+ "grad_norm": 0.47833895683288574,
832
+ "learning_rate": 9.999280030871234e-05,
833
+ "loss": 2.4903,
834
+ "step": 5600
835
+ },
836
+ {
837
+ "epoch": 0.048754284876032526,
838
+ "grad_norm": 0.472711980342865,
839
+ "learning_rate": 9.99915482272546e-05,
840
+ "loss": 2.4238,
841
+ "step": 5650
842
+ },
843
+ {
844
+ "epoch": 0.049185738724492986,
845
+ "grad_norm": 0.47533369064331055,
846
+ "learning_rate": 9.99901958304989e-05,
847
+ "loss": 2.4662,
848
+ "step": 5700
849
+ },
850
+ {
851
+ "epoch": 0.04961719257295345,
852
+ "grad_norm": 0.47553250193595886,
853
+ "learning_rate": 9.998874312115904e-05,
854
+ "loss": 2.4653,
855
+ "step": 5750
856
+ },
857
+ {
858
+ "epoch": 0.05004864642141392,
859
+ "grad_norm": 0.5022785663604736,
860
+ "learning_rate": 9.99871901021501e-05,
861
+ "loss": 2.4652,
862
+ "step": 5800
863
+ },
864
+ {
865
+ "epoch": 0.05048010026987438,
866
+ "grad_norm": 0.4645737409591675,
867
+ "learning_rate": 9.998553677658843e-05,
868
+ "loss": 2.4522,
869
+ "step": 5850
870
+ },
871
+ {
872
+ "epoch": 0.05091155411833485,
873
+ "grad_norm": 0.47363319993019104,
874
+ "learning_rate": 9.998378314779166e-05,
875
+ "loss": 2.4609,
876
+ "step": 5900
877
+ },
878
+ {
879
+ "epoch": 0.051343007966795313,
880
+ "grad_norm": 0.4770340323448181,
881
+ "learning_rate": 9.998192921927872e-05,
882
+ "loss": 2.3911,
883
+ "step": 5950
884
+ },
885
+ {
886
+ "epoch": 0.05177446181525577,
887
+ "grad_norm": 0.48888033628463745,
888
+ "learning_rate": 9.997997499476975e-05,
889
+ "loss": 2.4673,
890
+ "step": 6000
891
+ },
892
+ {
893
+ "epoch": 0.05177446181525577,
894
+ "eval_loss": 2.4369618892669678,
895
+ "eval_runtime": 2769.3132,
896
+ "eval_samples_per_second": 27.309,
897
+ "eval_steps_per_second": 13.655,
898
+ "step": 6000
899
+ },
900
+ {
901
+ "epoch": 0.05220591566371624,
902
+ "grad_norm": 0.4705820083618164,
903
+ "learning_rate": 9.997792047818622e-05,
904
+ "loss": 2.465,
905
+ "step": 6050
906
+ },
907
+ {
908
+ "epoch": 0.05263736951217671,
909
+ "grad_norm": 0.47672560811042786,
910
+ "learning_rate": 9.99757656736508e-05,
911
+ "loss": 2.4048,
912
+ "step": 6100
913
+ },
914
+ {
915
+ "epoch": 0.053068823360637174,
916
+ "grad_norm": 0.4655727744102478,
917
+ "learning_rate": 9.997351058548742e-05,
918
+ "loss": 2.4552,
919
+ "step": 6150
920
+ },
921
+ {
922
+ "epoch": 0.053500277209097634,
923
+ "grad_norm": 0.47453373670578003,
924
+ "learning_rate": 9.997115521822125e-05,
925
+ "loss": 2.4435,
926
+ "step": 6200
927
+ },
928
+ {
929
+ "epoch": 0.0539317310575581,
930
+ "grad_norm": 0.4548121392726898,
931
+ "learning_rate": 9.996869957657866e-05,
932
+ "loss": 2.3989,
933
+ "step": 6250
934
+ },
935
+ {
936
+ "epoch": 0.05436318490601857,
937
+ "grad_norm": 0.4468235373497009,
938
+ "learning_rate": 9.996614366548727e-05,
939
+ "loss": 2.3674,
940
+ "step": 6300
941
+ },
942
+ {
943
+ "epoch": 0.05479463875447903,
944
+ "grad_norm": 0.4576756954193115,
945
+ "learning_rate": 9.996348749007589e-05,
946
+ "loss": 2.3689,
947
+ "step": 6350
948
+ },
949
+ {
950
+ "epoch": 0.055226092602939494,
951
+ "grad_norm": 0.4505023956298828,
952
+ "learning_rate": 9.996073105567449e-05,
953
+ "loss": 2.4009,
954
+ "step": 6400
955
+ },
956
+ {
957
+ "epoch": 0.05565754645139996,
958
+ "grad_norm": 0.4463840425014496,
959
+ "learning_rate": 9.99578743678143e-05,
960
+ "loss": 2.3846,
961
+ "step": 6450
962
+ },
963
+ {
964
+ "epoch": 0.05608900029986043,
965
+ "grad_norm": 0.44599780440330505,
966
+ "learning_rate": 9.995491743222763e-05,
967
+ "loss": 2.4243,
968
+ "step": 6500
969
+ },
970
+ {
971
+ "epoch": 0.05652045414832089,
972
+ "grad_norm": 0.44260427355766296,
973
+ "learning_rate": 9.995186025484803e-05,
974
+ "loss": 2.366,
975
+ "step": 6550
976
+ },
977
+ {
978
+ "epoch": 0.056951907996781355,
979
+ "grad_norm": 0.4404599368572235,
980
+ "learning_rate": 9.994870284181017e-05,
981
+ "loss": 2.3635,
982
+ "step": 6600
983
+ },
984
+ {
985
+ "epoch": 0.05738336184524182,
986
+ "grad_norm": 0.43710505962371826,
987
+ "learning_rate": 9.994544519944984e-05,
988
+ "loss": 2.3563,
989
+ "step": 6650
990
+ },
991
+ {
992
+ "epoch": 0.05781481569370228,
993
+ "grad_norm": 0.42433786392211914,
994
+ "learning_rate": 9.9942087334304e-05,
995
+ "loss": 2.3752,
996
+ "step": 6700
997
+ },
998
+ {
999
+ "epoch": 0.05824626954216275,
1000
+ "grad_norm": 0.4220280945301056,
1001
+ "learning_rate": 9.993862925311067e-05,
1002
+ "loss": 2.394,
1003
+ "step": 6750
1004
+ },
1005
+ {
1006
+ "epoch": 0.058677723390623215,
1007
+ "grad_norm": 0.4359777271747589,
1008
+ "learning_rate": 9.9935070962809e-05,
1009
+ "loss": 2.3852,
1010
+ "step": 6800
1011
+ },
1012
+ {
1013
+ "epoch": 0.059109177239083675,
1014
+ "grad_norm": 0.4458872973918915,
1015
+ "learning_rate": 9.993141247053923e-05,
1016
+ "loss": 2.3866,
1017
+ "step": 6850
1018
+ },
1019
+ {
1020
+ "epoch": 0.05954063108754414,
1021
+ "grad_norm": 0.44301262497901917,
1022
+ "learning_rate": 9.992765378364263e-05,
1023
+ "loss": 2.3885,
1024
+ "step": 6900
1025
+ },
1026
+ {
1027
+ "epoch": 0.05997208493600461,
1028
+ "grad_norm": 0.4382476508617401,
1029
+ "learning_rate": 9.992379490966156e-05,
1030
+ "loss": 2.3537,
1031
+ "step": 6950
1032
+ },
1033
+ {
1034
+ "epoch": 0.060403538784465076,
1035
+ "grad_norm": 0.435742050409317,
1036
+ "learning_rate": 9.991983585633941e-05,
1037
+ "loss": 2.3634,
1038
+ "step": 7000
1039
+ },
1040
+ {
1041
+ "epoch": 0.060403538784465076,
1042
+ "eval_loss": 2.374647617340088,
1043
+ "eval_runtime": 2732.6747,
1044
+ "eval_samples_per_second": 27.675,
1045
+ "eval_steps_per_second": 13.838,
1046
+ "step": 7000
1047
+ },
1048
+ {
1049
+ "epoch": 0.060834992632925536,
1050
+ "grad_norm": 0.46364474296569824,
1051
+ "learning_rate": 9.991577663162062e-05,
1052
+ "loss": 2.3667,
1053
+ "step": 7050
1054
+ },
1055
+ {
1056
+ "epoch": 0.061266446481386,
1057
+ "grad_norm": 0.43860843777656555,
1058
+ "learning_rate": 9.991161724365057e-05,
1059
+ "loss": 2.3723,
1060
+ "step": 7100
1061
+ },
1062
+ {
1063
+ "epoch": 0.06169790032984647,
1064
+ "grad_norm": 0.42228421568870544,
1065
+ "learning_rate": 9.990735770077571e-05,
1066
+ "loss": 2.3963,
1067
+ "step": 7150
1068
+ },
1069
+ {
1070
+ "epoch": 0.06212935417830693,
1071
+ "grad_norm": 0.41791942715644836,
1072
+ "learning_rate": 9.990299801154342e-05,
1073
+ "loss": 2.3626,
1074
+ "step": 7200
1075
+ },
1076
+ {
1077
+ "epoch": 0.0625608080267674,
1078
+ "grad_norm": 0.4274737238883972,
1079
+ "learning_rate": 9.989853818470206e-05,
1080
+ "loss": 2.3904,
1081
+ "step": 7250
1082
+ },
1083
+ {
1084
+ "epoch": 0.06299226187522786,
1085
+ "grad_norm": 0.4228742718696594,
1086
+ "learning_rate": 9.989397822920088e-05,
1087
+ "loss": 2.3824,
1088
+ "step": 7300
1089
+ },
1090
+ {
1091
+ "epoch": 0.06342371572368832,
1092
+ "grad_norm": 0.4259440004825592,
1093
+ "learning_rate": 9.988931815419013e-05,
1094
+ "loss": 2.3413,
1095
+ "step": 7350
1096
+ },
1097
+ {
1098
+ "epoch": 0.06385516957214879,
1099
+ "grad_norm": 0.40943819284439087,
1100
+ "learning_rate": 9.988455796902094e-05,
1101
+ "loss": 2.3766,
1102
+ "step": 7400
1103
+ },
1104
+ {
1105
+ "epoch": 0.06428662342060926,
1106
+ "grad_norm": 0.4232550859451294,
1107
+ "learning_rate": 9.987969768324528e-05,
1108
+ "loss": 2.3293,
1109
+ "step": 7450
1110
+ },
1111
+ {
1112
+ "epoch": 0.06471807726906972,
1113
+ "grad_norm": 0.41405293345451355,
1114
+ "learning_rate": 9.987473730661602e-05,
1115
+ "loss": 2.3586,
1116
+ "step": 7500
1117
+ },
1118
+ {
1119
+ "epoch": 0.06514953111753019,
1120
+ "grad_norm": 0.4351080060005188,
1121
+ "learning_rate": 9.986967684908688e-05,
1122
+ "loss": 2.3015,
1123
+ "step": 7550
1124
+ },
1125
+ {
1126
+ "epoch": 0.06558098496599064,
1127
+ "grad_norm": 0.4127940237522125,
1128
+ "learning_rate": 9.986451632081242e-05,
1129
+ "loss": 2.3478,
1130
+ "step": 7600
1131
+ },
1132
+ {
1133
+ "epoch": 0.06601243881445111,
1134
+ "grad_norm": 0.4163227379322052,
1135
+ "learning_rate": 9.985925573214798e-05,
1136
+ "loss": 2.3294,
1137
+ "step": 7650
1138
+ },
1139
+ {
1140
+ "epoch": 0.06644389266291158,
1141
+ "grad_norm": 0.4200190603733063,
1142
+ "learning_rate": 9.985389509364968e-05,
1143
+ "loss": 2.3291,
1144
+ "step": 7700
1145
+ },
1146
+ {
1147
+ "epoch": 0.06687534651137204,
1148
+ "grad_norm": 0.4104471206665039,
1149
+ "learning_rate": 9.984843441607443e-05,
1150
+ "loss": 2.3712,
1151
+ "step": 7750
1152
+ },
1153
+ {
1154
+ "epoch": 0.06730680035983251,
1155
+ "grad_norm": 0.4211075007915497,
1156
+ "learning_rate": 9.984287371037988e-05,
1157
+ "loss": 2.2668,
1158
+ "step": 7800
1159
+ },
1160
+ {
1161
+ "epoch": 0.06773825420829298,
1162
+ "grad_norm": 0.42335718870162964,
1163
+ "learning_rate": 9.98372129877244e-05,
1164
+ "loss": 2.329,
1165
+ "step": 7850
1166
+ },
1167
+ {
1168
+ "epoch": 0.06816970805675344,
1169
+ "grad_norm": 0.4369964301586151,
1170
+ "learning_rate": 9.983145225946702e-05,
1171
+ "loss": 2.3559,
1172
+ "step": 7900
1173
+ },
1174
+ {
1175
+ "epoch": 0.0686011619052139,
1176
+ "grad_norm": 0.4052545130252838,
1177
+ "learning_rate": 9.982559153716749e-05,
1178
+ "loss": 2.3116,
1179
+ "step": 7950
1180
+ },
1181
+ {
1182
+ "epoch": 0.06903261575367436,
1183
+ "grad_norm": 0.4120001196861267,
1184
+ "learning_rate": 9.981963083258622e-05,
1185
+ "loss": 2.3408,
1186
+ "step": 8000
1187
+ },
1188
+ {
1189
+ "epoch": 0.06903261575367436,
1190
+ "eval_loss": 2.3286283016204834,
1191
+ "eval_runtime": 2709.736,
1192
+ "eval_samples_per_second": 27.91,
1193
+ "eval_steps_per_second": 13.955,
1194
+ "step": 8000
1195
+ },
1196
+ {
1197
+ "epoch": 0.06946406960213483,
1198
+ "grad_norm": 0.4071601629257202,
1199
+ "learning_rate": 9.98135701576842e-05,
1200
+ "loss": 2.3101,
1201
+ "step": 8050
1202
+ },
1203
+ {
1204
+ "epoch": 0.0698955234505953,
1205
+ "grad_norm": 0.4207800626754761,
1206
+ "learning_rate": 9.980740952462309e-05,
1207
+ "loss": 2.3192,
1208
+ "step": 8100
1209
+ },
1210
+ {
1211
+ "epoch": 0.07032697729905577,
1212
+ "grad_norm": 0.43044227361679077,
1213
+ "learning_rate": 9.980114894576504e-05,
1214
+ "loss": 2.3054,
1215
+ "step": 8150
1216
+ },
1217
+ {
1218
+ "epoch": 0.07075843114751623,
1219
+ "grad_norm": 0.3968140184879303,
1220
+ "learning_rate": 9.979478843367287e-05,
1221
+ "loss": 2.2884,
1222
+ "step": 8200
1223
+ },
1224
+ {
1225
+ "epoch": 0.0711898849959767,
1226
+ "grad_norm": 0.39424625039100647,
1227
+ "learning_rate": 9.978832800110983e-05,
1228
+ "loss": 2.336,
1229
+ "step": 8250
1230
+ },
1231
+ {
1232
+ "epoch": 0.07162133884443715,
1233
+ "grad_norm": 0.40985167026519775,
1234
+ "learning_rate": 9.978176766103974e-05,
1235
+ "loss": 2.3301,
1236
+ "step": 8300
1237
+ },
1238
+ {
1239
+ "epoch": 0.07205279269289762,
1240
+ "grad_norm": 0.3952298164367676,
1241
+ "learning_rate": 9.977510742662685e-05,
1242
+ "loss": 2.3406,
1243
+ "step": 8350
1244
+ },
1245
+ {
1246
+ "epoch": 0.07248424654135809,
1247
+ "grad_norm": 0.4053609073162079,
1248
+ "learning_rate": 9.976834731123591e-05,
1249
+ "loss": 2.2933,
1250
+ "step": 8400
1251
+ },
1252
+ {
1253
+ "epoch": 0.07291570038981855,
1254
+ "grad_norm": 0.4119470417499542,
1255
+ "learning_rate": 9.976148732843207e-05,
1256
+ "loss": 2.2927,
1257
+ "step": 8450
1258
+ },
1259
+ {
1260
+ "epoch": 0.07334715423827902,
1261
+ "grad_norm": 0.4225163459777832,
1262
+ "learning_rate": 9.975452749198086e-05,
1263
+ "loss": 2.3243,
1264
+ "step": 8500
1265
+ },
1266
+ {
1267
+ "epoch": 0.07377860808673949,
1268
+ "grad_norm": 0.410567969083786,
1269
+ "learning_rate": 9.974746781584823e-05,
1270
+ "loss": 2.2995,
1271
+ "step": 8550
1272
+ },
1273
+ {
1274
+ "epoch": 0.07421006193519995,
1275
+ "grad_norm": 0.3905579447746277,
1276
+ "learning_rate": 9.974030831420043e-05,
1277
+ "loss": 2.2875,
1278
+ "step": 8600
1279
+ },
1280
+ {
1281
+ "epoch": 0.0746415157836604,
1282
+ "grad_norm": 0.4086805284023285,
1283
+ "learning_rate": 9.973304900140404e-05,
1284
+ "loss": 2.3132,
1285
+ "step": 8650
1286
+ },
1287
+ {
1288
+ "epoch": 0.07507296963212087,
1289
+ "grad_norm": 0.4010264575481415,
1290
+ "learning_rate": 9.972568989202591e-05,
1291
+ "loss": 2.2886,
1292
+ "step": 8700
1293
+ },
1294
+ {
1295
+ "epoch": 0.07550442348058134,
1296
+ "grad_norm": 0.41073790192604065,
1297
+ "learning_rate": 9.971823100083319e-05,
1298
+ "loss": 2.3267,
1299
+ "step": 8750
1300
+ },
1301
+ {
1302
+ "epoch": 0.0759358773290418,
1303
+ "grad_norm": 0.39557185769081116,
1304
+ "learning_rate": 9.971067234279322e-05,
1305
+ "loss": 2.3053,
1306
+ "step": 8800
1307
+ },
1308
+ {
1309
+ "epoch": 0.07636733117750227,
1310
+ "grad_norm": 0.39963647723197937,
1311
+ "learning_rate": 9.970301393307353e-05,
1312
+ "loss": 2.3193,
1313
+ "step": 8850
1314
+ },
1315
+ {
1316
+ "epoch": 0.07679878502596274,
1317
+ "grad_norm": 0.38299834728240967,
1318
+ "learning_rate": 9.969525578704186e-05,
1319
+ "loss": 2.2845,
1320
+ "step": 8900
1321
+ },
1322
+ {
1323
+ "epoch": 0.0772302388744232,
1324
+ "grad_norm": 0.3826388418674469,
1325
+ "learning_rate": 9.968739792026602e-05,
1326
+ "loss": 2.2879,
1327
+ "step": 8950
1328
+ },
1329
+ {
1330
+ "epoch": 0.07766169272288366,
1331
+ "grad_norm": 0.39117395877838135,
1332
+ "learning_rate": 9.967944034851398e-05,
1333
+ "loss": 2.2922,
1334
+ "step": 9000
1335
+ },
1336
+ {
1337
+ "epoch": 0.07766169272288366,
1338
+ "eval_loss": 2.290743589401245,
1339
+ "eval_runtime": 2787.9306,
1340
+ "eval_samples_per_second": 27.127,
1341
+ "eval_steps_per_second": 13.563,
1342
+ "step": 9000
1343
+ },
1344
+ {
1345
+ "epoch": 0.07809314657134413,
1346
+ "grad_norm": 0.4105989933013916,
1347
+ "learning_rate": 9.967138308775374e-05,
1348
+ "loss": 2.3296,
1349
+ "step": 9050
1350
+ },
1351
+ {
1352
+ "epoch": 0.0785246004198046,
1353
+ "grad_norm": 0.3829767107963562,
1354
+ "learning_rate": 9.966322615415341e-05,
1355
+ "loss": 2.2909,
1356
+ "step": 9100
1357
+ },
1358
+ {
1359
+ "epoch": 0.07895605426826506,
1360
+ "grad_norm": 0.38462504744529724,
1361
+ "learning_rate": 9.965496956408102e-05,
1362
+ "loss": 2.3014,
1363
+ "step": 9150
1364
+ },
1365
+ {
1366
+ "epoch": 0.07938750811672553,
1367
+ "grad_norm": 0.3984980583190918,
1368
+ "learning_rate": 9.964661333410464e-05,
1369
+ "loss": 2.2933,
1370
+ "step": 9200
1371
+ },
1372
+ {
1373
+ "epoch": 0.079818961965186,
1374
+ "grad_norm": 0.41218262910842896,
1375
+ "learning_rate": 9.963815748099225e-05,
1376
+ "loss": 2.3321,
1377
+ "step": 9250
1378
+ },
1379
+ {
1380
+ "epoch": 0.08025041581364645,
1381
+ "grad_norm": 0.39859455823898315,
1382
+ "learning_rate": 9.962960202171174e-05,
1383
+ "loss": 2.2745,
1384
+ "step": 9300
1385
+ },
1386
+ {
1387
+ "epoch": 0.08068186966210691,
1388
+ "grad_norm": 0.3806203305721283,
1389
+ "learning_rate": 9.962094697343089e-05,
1390
+ "loss": 2.2691,
1391
+ "step": 9350
1392
+ },
1393
+ {
1394
+ "epoch": 0.08111332351056738,
1395
+ "grad_norm": 0.3818638324737549,
1396
+ "learning_rate": 9.961219235351729e-05,
1397
+ "loss": 2.3255,
1398
+ "step": 9400
1399
+ },
1400
+ {
1401
+ "epoch": 0.08154477735902785,
1402
+ "grad_norm": 0.390399307012558,
1403
+ "learning_rate": 9.96033381795384e-05,
1404
+ "loss": 2.2906,
1405
+ "step": 9450
1406
+ },
1407
+ {
1408
+ "epoch": 0.08197623120748831,
1409
+ "grad_norm": 0.37718382477760315,
1410
+ "learning_rate": 9.959438446926135e-05,
1411
+ "loss": 2.2903,
1412
+ "step": 9500
1413
+ },
1414
+ {
1415
+ "epoch": 0.08240768505594878,
1416
+ "grad_norm": 0.3930068016052246,
1417
+ "learning_rate": 9.95853312406531e-05,
1418
+ "loss": 2.2832,
1419
+ "step": 9550
1420
+ },
1421
+ {
1422
+ "epoch": 0.08283913890440925,
1423
+ "grad_norm": 0.3778354227542877,
1424
+ "learning_rate": 9.957617851188026e-05,
1425
+ "loss": 2.2438,
1426
+ "step": 9600
1427
+ },
1428
+ {
1429
+ "epoch": 0.0832705927528697,
1430
+ "grad_norm": 0.39089974761009216,
1431
+ "learning_rate": 9.956692630130911e-05,
1432
+ "loss": 2.3055,
1433
+ "step": 9650
1434
+ },
1435
+ {
1436
+ "epoch": 0.08370204660133017,
1437
+ "grad_norm": 0.3801150619983673,
1438
+ "learning_rate": 9.955757462750554e-05,
1439
+ "loss": 2.2887,
1440
+ "step": 9700
1441
+ },
1442
+ {
1443
+ "epoch": 0.08413350044979064,
1444
+ "grad_norm": 0.37772729992866516,
1445
+ "learning_rate": 9.954812350923504e-05,
1446
+ "loss": 2.2617,
1447
+ "step": 9750
1448
+ },
1449
+ {
1450
+ "epoch": 0.0845649542982511,
1451
+ "grad_norm": 0.4012361168861389,
1452
+ "learning_rate": 9.953857296546267e-05,
1453
+ "loss": 2.2583,
1454
+ "step": 9800
1455
+ },
1456
+ {
1457
+ "epoch": 0.08499640814671157,
1458
+ "grad_norm": 0.37904253602027893,
1459
+ "learning_rate": 9.952892301535295e-05,
1460
+ "loss": 2.2582,
1461
+ "step": 9850
1462
+ },
1463
+ {
1464
+ "epoch": 0.08542786199517204,
1465
+ "grad_norm": 0.38929301500320435,
1466
+ "learning_rate": 9.951917367826995e-05,
1467
+ "loss": 2.2581,
1468
+ "step": 9900
1469
+ },
1470
+ {
1471
+ "epoch": 0.0858593158436325,
1472
+ "grad_norm": 0.37145188450813293,
1473
+ "learning_rate": 9.95093249737771e-05,
1474
+ "loss": 2.2422,
1475
+ "step": 9950
1476
+ },
1477
+ {
1478
+ "epoch": 0.08629076969209296,
1479
+ "grad_norm": 0.3830346465110779,
1480
+ "learning_rate": 9.949937692163726e-05,
1481
+ "loss": 2.2527,
1482
+ "step": 10000
1483
+ },
1484
+ {
1485
+ "epoch": 0.08629076969209296,
1486
+ "eval_loss": 2.2615647315979004,
1487
+ "eval_runtime": 2822.4337,
1488
+ "eval_samples_per_second": 26.795,
1489
+ "eval_steps_per_second": 13.398,
1490
+ "step": 10000
1491
+ },
1492
+ {
1493
+ "epoch": 0.08672222354055342,
1494
+ "grad_norm": 0.37199145555496216,
1495
+ "learning_rate": 9.948932954181264e-05,
1496
+ "loss": 2.2656,
1497
+ "step": 10050
1498
+ },
1499
+ {
1500
+ "epoch": 0.08715367738901389,
1501
+ "grad_norm": 0.385570764541626,
1502
+ "learning_rate": 9.947918285446475e-05,
1503
+ "loss": 2.2667,
1504
+ "step": 10100
1505
+ },
1506
+ {
1507
+ "epoch": 0.08758513123747436,
1508
+ "grad_norm": 0.38618603348731995,
1509
+ "learning_rate": 9.946893687995444e-05,
1510
+ "loss": 2.2459,
1511
+ "step": 10150
1512
+ },
1513
+ {
1514
+ "epoch": 0.08801658508593482,
1515
+ "grad_norm": 0.37942782044410706,
1516
+ "learning_rate": 9.94585916388417e-05,
1517
+ "loss": 2.2473,
1518
+ "step": 10200
1519
+ },
1520
+ {
1521
+ "epoch": 0.08844803893439529,
1522
+ "grad_norm": 0.37803226709365845,
1523
+ "learning_rate": 9.944814715188578e-05,
1524
+ "loss": 2.2623,
1525
+ "step": 10250
1526
+ },
1527
+ {
1528
+ "epoch": 0.08887949278285574,
1529
+ "grad_norm": 0.37103068828582764,
1530
+ "learning_rate": 9.943760344004505e-05,
1531
+ "loss": 2.2822,
1532
+ "step": 10300
1533
+ },
1534
+ {
1535
+ "epoch": 0.08931094663131621,
1536
+ "grad_norm": 0.3779771327972412,
1537
+ "learning_rate": 9.942696052447699e-05,
1538
+ "loss": 2.2473,
1539
+ "step": 10350
1540
+ },
1541
+ {
1542
+ "epoch": 0.08974240047977668,
1543
+ "grad_norm": 0.38384005427360535,
1544
+ "learning_rate": 9.94162184265382e-05,
1545
+ "loss": 2.2615,
1546
+ "step": 10400
1547
+ },
1548
+ {
1549
+ "epoch": 0.09017385432823714,
1550
+ "grad_norm": 0.36723431944847107,
1551
+ "learning_rate": 9.940537716778417e-05,
1552
+ "loss": 2.2656,
1553
+ "step": 10450
1554
+ },
1555
+ {
1556
+ "epoch": 0.09060530817669761,
1557
+ "grad_norm": 0.39431023597717285,
1558
+ "learning_rate": 9.939443676996955e-05,
1559
+ "loss": 2.2645,
1560
+ "step": 10500
1561
+ },
1562
+ {
1563
+ "epoch": 0.09103676202515808,
1564
+ "grad_norm": 0.3861570954322815,
1565
+ "learning_rate": 9.938339725504778e-05,
1566
+ "loss": 2.2158,
1567
+ "step": 10550
1568
+ },
1569
+ {
1570
+ "epoch": 0.09146821587361854,
1571
+ "grad_norm": 0.38060426712036133,
1572
+ "learning_rate": 9.937225864517128e-05,
1573
+ "loss": 2.2341,
1574
+ "step": 10600
1575
+ },
1576
+ {
1577
+ "epoch": 0.091899669722079,
1578
+ "grad_norm": 0.3649984300136566,
1579
+ "learning_rate": 9.936102096269126e-05,
1580
+ "loss": 2.2333,
1581
+ "step": 10650
1582
+ },
1583
+ {
1584
+ "epoch": 0.09233112357053946,
1585
+ "grad_norm": 0.37162846326828003,
1586
+ "learning_rate": 9.93496842301578e-05,
1587
+ "loss": 2.262,
1588
+ "step": 10700
1589
+ },
1590
+ {
1591
+ "epoch": 0.09276257741899993,
1592
+ "grad_norm": 0.3689543604850769,
1593
+ "learning_rate": 9.933824847031967e-05,
1594
+ "loss": 2.2342,
1595
+ "step": 10750
1596
+ },
1597
+ {
1598
+ "epoch": 0.0931940312674604,
1599
+ "grad_norm": 0.3751949071884155,
1600
+ "learning_rate": 9.932671370612437e-05,
1601
+ "loss": 2.2452,
1602
+ "step": 10800
1603
+ },
1604
+ {
1605
+ "epoch": 0.09362548511592086,
1606
+ "grad_norm": 0.3793388605117798,
1607
+ "learning_rate": 9.931507996071812e-05,
1608
+ "loss": 2.216,
1609
+ "step": 10850
1610
+ },
1611
+ {
1612
+ "epoch": 0.09405693896438133,
1613
+ "grad_norm": 0.37986430525779724,
1614
+ "learning_rate": 9.930334725744568e-05,
1615
+ "loss": 2.2414,
1616
+ "step": 10900
1617
+ },
1618
+ {
1619
+ "epoch": 0.0944883928128418,
1620
+ "grad_norm": 0.37280675768852234,
1621
+ "learning_rate": 9.929151561985044e-05,
1622
+ "loss": 2.2244,
1623
+ "step": 10950
1624
+ },
1625
+ {
1626
+ "epoch": 0.09491984666130225,
1627
+ "grad_norm": 0.39744243025779724,
1628
+ "learning_rate": 9.927958507167429e-05,
1629
+ "loss": 2.2515,
1630
+ "step": 11000
1631
+ },
1632
+ {
1633
+ "epoch": 0.09491984666130225,
1634
+ "eval_loss": 2.235520362854004,
1635
+ "eval_runtime": 2908.2681,
1636
+ "eval_samples_per_second": 26.004,
1637
+ "eval_steps_per_second": 13.002,
1638
+ "step": 11000
1639
+ },
1640
+ {
1641
+ "epoch": 0.09535130050976272,
1642
+ "grad_norm": 0.3687732517719269,
1643
+ "learning_rate": 9.926755563685763e-05,
1644
+ "loss": 2.2522,
1645
+ "step": 11050
1646
+ },
1647
+ {
1648
+ "epoch": 0.09578275435822318,
1649
+ "grad_norm": 0.36527740955352783,
1650
+ "learning_rate": 9.925542733953923e-05,
1651
+ "loss": 2.2731,
1652
+ "step": 11100
1653
+ },
1654
+ {
1655
+ "epoch": 0.09621420820668365,
1656
+ "grad_norm": 0.3645661175251007,
1657
+ "learning_rate": 9.924320020405632e-05,
1658
+ "loss": 2.2084,
1659
+ "step": 11150
1660
+ },
1661
+ {
1662
+ "epoch": 0.09664566205514412,
1663
+ "grad_norm": 0.36427396535873413,
1664
+ "learning_rate": 9.92308742549444e-05,
1665
+ "loss": 2.2074,
1666
+ "step": 11200
1667
+ },
1668
+ {
1669
+ "epoch": 0.09707711590360459,
1670
+ "grad_norm": 0.35815566778182983,
1671
+ "learning_rate": 9.921844951693726e-05,
1672
+ "loss": 2.212,
1673
+ "step": 11250
1674
+ },
1675
+ {
1676
+ "epoch": 0.09750856975206505,
1677
+ "grad_norm": 0.3574189841747284,
1678
+ "learning_rate": 9.920592601496696e-05,
1679
+ "loss": 2.2262,
1680
+ "step": 11300
1681
+ },
1682
+ {
1683
+ "epoch": 0.0979400236005255,
1684
+ "grad_norm": 0.3735974133014679,
1685
+ "learning_rate": 9.919330377416373e-05,
1686
+ "loss": 2.23,
1687
+ "step": 11350
1688
+ },
1689
+ {
1690
+ "epoch": 0.09837147744898597,
1691
+ "grad_norm": 0.38068506121635437,
1692
+ "learning_rate": 9.918058281985592e-05,
1693
+ "loss": 2.1978,
1694
+ "step": 11400
1695
+ },
1696
+ {
1697
+ "epoch": 0.09880293129744644,
1698
+ "grad_norm": 0.36570560932159424,
1699
+ "learning_rate": 9.916776317756996e-05,
1700
+ "loss": 2.2228,
1701
+ "step": 11450
1702
+ },
1703
+ {
1704
+ "epoch": 0.0992343851459069,
1705
+ "grad_norm": 0.3642862141132355,
1706
+ "learning_rate": 9.915484487303035e-05,
1707
+ "loss": 2.2383,
1708
+ "step": 11500
1709
+ },
1710
+ {
1711
+ "epoch": 0.09966583899436737,
1712
+ "grad_norm": 0.36596134305000305,
1713
+ "learning_rate": 9.914182793215951e-05,
1714
+ "loss": 2.2302,
1715
+ "step": 11550
1716
+ },
1717
+ {
1718
+ "epoch": 0.10009729284282784,
1719
+ "grad_norm": 0.36998990178108215,
1720
+ "learning_rate": 9.912871238107786e-05,
1721
+ "loss": 2.2384,
1722
+ "step": 11600
1723
+ },
1724
+ {
1725
+ "epoch": 0.1005287466912883,
1726
+ "grad_norm": 0.36373093724250793,
1727
+ "learning_rate": 9.911549824610362e-05,
1728
+ "loss": 2.1848,
1729
+ "step": 11650
1730
+ },
1731
+ {
1732
+ "epoch": 0.10096020053974876,
1733
+ "grad_norm": 0.36697497963905334,
1734
+ "learning_rate": 9.910218555375289e-05,
1735
+ "loss": 2.2126,
1736
+ "step": 11700
1737
+ },
1738
+ {
1739
+ "epoch": 0.10139165438820923,
1740
+ "grad_norm": 0.36996662616729736,
1741
+ "learning_rate": 9.908877433073953e-05,
1742
+ "loss": 2.2276,
1743
+ "step": 11750
1744
+ },
1745
+ {
1746
+ "epoch": 0.1018231082366697,
1747
+ "grad_norm": 0.36146995425224304,
1748
+ "learning_rate": 9.907526460397507e-05,
1749
+ "loss": 2.2043,
1750
+ "step": 11800
1751
+ },
1752
+ {
1753
+ "epoch": 0.10225456208513016,
1754
+ "grad_norm": 0.36770108342170715,
1755
+ "learning_rate": 9.906165640056877e-05,
1756
+ "loss": 2.2223,
1757
+ "step": 11850
1758
+ },
1759
+ {
1760
+ "epoch": 0.10268601593359063,
1761
+ "grad_norm": 0.3661516606807709,
1762
+ "learning_rate": 9.904794974782745e-05,
1763
+ "loss": 2.1971,
1764
+ "step": 11900
1765
+ },
1766
+ {
1767
+ "epoch": 0.1031174697820511,
1768
+ "grad_norm": 0.36954957246780396,
1769
+ "learning_rate": 9.90341446732555e-05,
1770
+ "loss": 2.2063,
1771
+ "step": 11950
1772
+ },
1773
+ {
1774
+ "epoch": 0.10354892363051155,
1775
+ "grad_norm": 0.3615468740463257,
1776
+ "learning_rate": 9.902024120455482e-05,
1777
+ "loss": 2.2116,
1778
+ "step": 12000
1779
+ },
1780
+ {
1781
+ "epoch": 0.10354892363051155,
1782
+ "eval_loss": 2.214376926422119,
1783
+ "eval_runtime": 2985.5808,
1784
+ "eval_samples_per_second": 25.331,
1785
+ "eval_steps_per_second": 12.666,
1786
+ "step": 12000
1787
+ },
1788
+ {
1789
+ "epoch": 0.10398037747897201,
1790
+ "grad_norm": 0.35979902744293213,
1791
+ "learning_rate": 9.900623936962471e-05,
1792
+ "loss": 2.1718,
1793
+ "step": 12050
1794
+ },
1795
+ {
1796
+ "epoch": 0.10441183132743248,
1797
+ "grad_norm": 0.37038519978523254,
1798
+ "learning_rate": 9.899213919656191e-05,
1799
+ "loss": 2.2364,
1800
+ "step": 12100
1801
+ },
1802
+ {
1803
+ "epoch": 0.10484328517589295,
1804
+ "grad_norm": 0.349633127450943,
1805
+ "learning_rate": 9.897794071366045e-05,
1806
+ "loss": 2.2055,
1807
+ "step": 12150
1808
+ },
1809
+ {
1810
+ "epoch": 0.10527473902435341,
1811
+ "grad_norm": 0.365032821893692,
1812
+ "learning_rate": 9.896364394941166e-05,
1813
+ "loss": 2.2218,
1814
+ "step": 12200
1815
+ },
1816
+ {
1817
+ "epoch": 0.10570619287281388,
1818
+ "grad_norm": 0.3733338415622711,
1819
+ "learning_rate": 9.894924893250406e-05,
1820
+ "loss": 2.2368,
1821
+ "step": 12250
1822
+ },
1823
+ {
1824
+ "epoch": 0.10613764672127435,
1825
+ "grad_norm": 0.3741872310638428,
1826
+ "learning_rate": 9.893475569182335e-05,
1827
+ "loss": 2.2012,
1828
+ "step": 12300
1829
+ },
1830
+ {
1831
+ "epoch": 0.1065691005697348,
1832
+ "grad_norm": 0.35796815156936646,
1833
+ "learning_rate": 9.892016425645232e-05,
1834
+ "loss": 2.2186,
1835
+ "step": 12350
1836
+ },
1837
+ {
1838
+ "epoch": 0.10700055441819527,
1839
+ "grad_norm": 0.35955625772476196,
1840
+ "learning_rate": 9.89054746556708e-05,
1841
+ "loss": 2.2257,
1842
+ "step": 12400
1843
+ },
1844
+ {
1845
+ "epoch": 0.10743200826665573,
1846
+ "grad_norm": 0.36324945092201233,
1847
+ "learning_rate": 9.889068691895558e-05,
1848
+ "loss": 2.2069,
1849
+ "step": 12450
1850
+ },
1851
+ {
1852
+ "epoch": 0.1078634621151162,
1853
+ "grad_norm": 0.3579570949077606,
1854
+ "learning_rate": 9.887580107598044e-05,
1855
+ "loss": 2.2368,
1856
+ "step": 12500
1857
+ },
1858
+ {
1859
+ "epoch": 0.10829491596357667,
1860
+ "grad_norm": 0.3664620518684387,
1861
+ "learning_rate": 9.886081715661599e-05,
1862
+ "loss": 2.2064,
1863
+ "step": 12550
1864
+ },
1865
+ {
1866
+ "epoch": 0.10872636981203714,
1867
+ "grad_norm": 0.3607580065727234,
1868
+ "learning_rate": 9.884573519092959e-05,
1869
+ "loss": 2.2008,
1870
+ "step": 12600
1871
+ },
1872
+ {
1873
+ "epoch": 0.1091578236604976,
1874
+ "grad_norm": 0.3472786247730255,
1875
+ "learning_rate": 9.883055520918544e-05,
1876
+ "loss": 2.1839,
1877
+ "step": 12650
1878
+ },
1879
+ {
1880
+ "epoch": 0.10958927750895806,
1881
+ "grad_norm": 0.34541594982147217,
1882
+ "learning_rate": 9.881527724184436e-05,
1883
+ "loss": 2.2265,
1884
+ "step": 12700
1885
+ },
1886
+ {
1887
+ "epoch": 0.11002073135741852,
1888
+ "grad_norm": 0.360601544380188,
1889
+ "learning_rate": 9.87999013195638e-05,
1890
+ "loss": 2.2101,
1891
+ "step": 12750
1892
+ },
1893
+ {
1894
+ "epoch": 0.11045218520587899,
1895
+ "grad_norm": 0.3593735992908478,
1896
+ "learning_rate": 9.87844274731978e-05,
1897
+ "loss": 2.2014,
1898
+ "step": 12800
1899
+ },
1900
+ {
1901
+ "epoch": 0.11088363905433946,
1902
+ "grad_norm": 0.3615984320640564,
1903
+ "learning_rate": 9.876885573379687e-05,
1904
+ "loss": 2.2163,
1905
+ "step": 12850
1906
+ },
1907
+ {
1908
+ "epoch": 0.11131509290279992,
1909
+ "grad_norm": 0.3485114574432373,
1910
+ "learning_rate": 9.875318613260794e-05,
1911
+ "loss": 2.2154,
1912
+ "step": 12900
1913
+ },
1914
+ {
1915
+ "epoch": 0.11174654675126039,
1916
+ "grad_norm": 0.3505745232105255,
1917
+ "learning_rate": 9.873741870107438e-05,
1918
+ "loss": 2.1613,
1919
+ "step": 12950
1920
+ },
1921
+ {
1922
+ "epoch": 0.11217800059972086,
1923
+ "grad_norm": 0.3599075376987457,
1924
+ "learning_rate": 9.872155347083582e-05,
1925
+ "loss": 2.1743,
1926
+ "step": 13000
1927
+ },
1928
+ {
1929
+ "epoch": 0.11217800059972086,
1930
+ "eval_loss": 2.195662021636963,
1931
+ "eval_runtime": 2862.6512,
1932
+ "eval_samples_per_second": 26.419,
1933
+ "eval_steps_per_second": 13.209,
1934
+ "step": 13000
1935
+ },
1936
+ {
1937
+ "epoch": 0.11260945444818131,
1938
+ "grad_norm": 0.35765302181243896,
1939
+ "learning_rate": 9.870559047372811e-05,
1940
+ "loss": 2.2033,
1941
+ "step": 13050
1942
+ },
1943
+ {
1944
+ "epoch": 0.11304090829664178,
1945
+ "grad_norm": 0.3540419340133667,
1946
+ "learning_rate": 9.868952974178339e-05,
1947
+ "loss": 2.153,
1948
+ "step": 13100
1949
+ },
1950
+ {
1951
+ "epoch": 0.11347236214510224,
1952
+ "grad_norm": 0.3633355498313904,
1953
+ "learning_rate": 9.867337130722978e-05,
1954
+ "loss": 2.1969,
1955
+ "step": 13150
1956
+ },
1957
+ {
1958
+ "epoch": 0.11390381599356271,
1959
+ "grad_norm": 0.35366296768188477,
1960
+ "learning_rate": 9.865711520249156e-05,
1961
+ "loss": 2.1863,
1962
+ "step": 13200
1963
+ },
1964
+ {
1965
+ "epoch": 0.11433526984202318,
1966
+ "grad_norm": 0.3593711853027344,
1967
+ "learning_rate": 9.864076146018897e-05,
1968
+ "loss": 2.1632,
1969
+ "step": 13250
1970
+ },
1971
+ {
1972
+ "epoch": 0.11476672369048364,
1973
+ "grad_norm": 0.3670215904712677,
1974
+ "learning_rate": 9.862431011313815e-05,
1975
+ "loss": 2.1884,
1976
+ "step": 13300
1977
+ },
1978
+ {
1979
+ "epoch": 0.1151981775389441,
1980
+ "grad_norm": 0.3451824486255646,
1981
+ "learning_rate": 9.860776119435114e-05,
1982
+ "loss": 2.1869,
1983
+ "step": 13350
1984
+ },
1985
+ {
1986
+ "epoch": 0.11562963138740456,
1987
+ "grad_norm": 0.36133304238319397,
1988
+ "learning_rate": 9.85911147370357e-05,
1989
+ "loss": 2.1586,
1990
+ "step": 13400
1991
+ },
1992
+ {
1993
+ "epoch": 0.11606108523586503,
1994
+ "grad_norm": 0.35326626896858215,
1995
+ "learning_rate": 9.85743707745954e-05,
1996
+ "loss": 2.2124,
1997
+ "step": 13450
1998
+ },
1999
+ {
2000
+ "epoch": 0.1164925390843255,
2001
+ "grad_norm": 0.3657006025314331,
2002
+ "learning_rate": 9.855752934062943e-05,
2003
+ "loss": 2.1942,
2004
+ "step": 13500
2005
+ },
2006
+ {
2007
+ "epoch": 0.11692399293278596,
2008
+ "grad_norm": 0.3521730601787567,
2009
+ "learning_rate": 9.854059046893257e-05,
2010
+ "loss": 2.1945,
2011
+ "step": 13550
2012
+ },
2013
+ {
2014
+ "epoch": 0.11735544678124643,
2015
+ "grad_norm": 0.3588502109050751,
2016
+ "learning_rate": 9.852355419349511e-05,
2017
+ "loss": 2.2061,
2018
+ "step": 13600
2019
+ },
2020
+ {
2021
+ "epoch": 0.1177869006297069,
2022
+ "grad_norm": 0.36119502782821655,
2023
+ "learning_rate": 9.850642054850284e-05,
2024
+ "loss": 2.1927,
2025
+ "step": 13650
2026
+ },
2027
+ {
2028
+ "epoch": 0.11821835447816735,
2029
+ "grad_norm": 0.35996031761169434,
2030
+ "learning_rate": 9.848918956833686e-05,
2031
+ "loss": 2.1971,
2032
+ "step": 13700
2033
+ },
2034
+ {
2035
+ "epoch": 0.11864980832662782,
2036
+ "grad_norm": 0.34750646352767944,
2037
+ "learning_rate": 9.847186128757366e-05,
2038
+ "loss": 2.1965,
2039
+ "step": 13750
2040
+ },
2041
+ {
2042
+ "epoch": 0.11908126217508828,
2043
+ "grad_norm": 0.3519524335861206,
2044
+ "learning_rate": 9.845443574098496e-05,
2045
+ "loss": 2.1557,
2046
+ "step": 13800
2047
+ },
2048
+ {
2049
+ "epoch": 0.11951271602354875,
2050
+ "grad_norm": 0.35633113980293274,
2051
+ "learning_rate": 9.843691296353765e-05,
2052
+ "loss": 2.1867,
2053
+ "step": 13850
2054
+ },
2055
+ {
2056
+ "epoch": 0.11994416987200922,
2057
+ "grad_norm": 0.37132638692855835,
2058
+ "learning_rate": 9.84192929903937e-05,
2059
+ "loss": 2.1871,
2060
+ "step": 13900
2061
+ },
2062
+ {
2063
+ "epoch": 0.12037562372046968,
2064
+ "grad_norm": 0.3432469367980957,
2065
+ "learning_rate": 9.840157585691014e-05,
2066
+ "loss": 2.1724,
2067
+ "step": 13950
2068
+ },
2069
+ {
2070
+ "epoch": 0.12080707756893015,
2071
+ "grad_norm": 0.34275302290916443,
2072
+ "learning_rate": 9.838376159863902e-05,
2073
+ "loss": 2.1507,
2074
+ "step": 14000
2075
+ },
2076
+ {
2077
+ "epoch": 0.12080707756893015,
2078
+ "eval_loss": 2.179051637649536,
2079
+ "eval_runtime": 2834.8461,
2080
+ "eval_samples_per_second": 26.678,
2081
+ "eval_steps_per_second": 13.339,
2082
+ "step": 14000
2083
+ },
2084
+ {
2085
+ "epoch": 0.1212385314173906,
2086
+ "grad_norm": 0.34372302889823914,
2087
+ "learning_rate": 9.836585025132719e-05,
2088
+ "loss": 2.1675,
2089
+ "step": 14050
2090
+ },
2091
+ {
2092
+ "epoch": 0.12166998526585107,
2093
+ "grad_norm": 0.34981855750083923,
2094
+ "learning_rate": 9.834784185091638e-05,
2095
+ "loss": 2.2123,
2096
+ "step": 14100
2097
+ },
2098
+ {
2099
+ "epoch": 0.12210143911431154,
2100
+ "grad_norm": 0.3538338840007782,
2101
+ "learning_rate": 9.832973643354307e-05,
2102
+ "loss": 2.1979,
2103
+ "step": 14150
2104
+ },
2105
+ {
2106
+ "epoch": 0.122532892962772,
2107
+ "grad_norm": 0.3431382477283478,
2108
+ "learning_rate": 9.831153403553838e-05,
2109
+ "loss": 2.2037,
2110
+ "step": 14200
2111
+ },
2112
+ {
2113
+ "epoch": 0.12296434681123247,
2114
+ "grad_norm": 0.35622259974479675,
2115
+ "learning_rate": 9.829323469342808e-05,
2116
+ "loss": 2.159,
2117
+ "step": 14250
2118
+ },
2119
+ {
2120
+ "epoch": 0.12339580065969294,
2121
+ "grad_norm": 0.35167625546455383,
2122
+ "learning_rate": 9.827483844393249e-05,
2123
+ "loss": 2.1838,
2124
+ "step": 14300
2125
+ },
2126
+ {
2127
+ "epoch": 0.1238272545081534,
2128
+ "grad_norm": 0.3605762720108032,
2129
+ "learning_rate": 9.825634532396628e-05,
2130
+ "loss": 2.1947,
2131
+ "step": 14350
2132
+ },
2133
+ {
2134
+ "epoch": 0.12425870835661386,
2135
+ "grad_norm": 0.36447280645370483,
2136
+ "learning_rate": 9.823775537063867e-05,
2137
+ "loss": 2.1854,
2138
+ "step": 14400
2139
+ },
2140
+ {
2141
+ "epoch": 0.12469016220507433,
2142
+ "grad_norm": 0.3465559780597687,
2143
+ "learning_rate": 9.821906862125301e-05,
2144
+ "loss": 2.1467,
2145
+ "step": 14450
2146
+ },
2147
+ {
2148
+ "epoch": 0.1251216160535348,
2149
+ "grad_norm": 0.34778252243995667,
2150
+ "learning_rate": 9.820028511330703e-05,
2151
+ "loss": 2.1531,
2152
+ "step": 14500
2153
+ },
2154
+ {
2155
+ "epoch": 0.12555306990199525,
2156
+ "grad_norm": 0.34816524386405945,
2157
+ "learning_rate": 9.818140488449257e-05,
2158
+ "loss": 2.168,
2159
+ "step": 14550
2160
+ },
2161
+ {
2162
+ "epoch": 0.1259845237504557,
2163
+ "grad_norm": 0.3492778241634369,
2164
+ "learning_rate": 9.816242797269553e-05,
2165
+ "loss": 2.1599,
2166
+ "step": 14600
2167
+ },
2168
+ {
2169
+ "epoch": 0.12641597759891618,
2170
+ "grad_norm": 0.3482818901538849,
2171
+ "learning_rate": 9.814335441599583e-05,
2172
+ "loss": 2.1515,
2173
+ "step": 14650
2174
+ },
2175
+ {
2176
+ "epoch": 0.12684743144737665,
2177
+ "grad_norm": 0.3657529950141907,
2178
+ "learning_rate": 9.812418425266733e-05,
2179
+ "loss": 2.1656,
2180
+ "step": 14700
2181
+ },
2182
+ {
2183
+ "epoch": 0.1272788852958371,
2184
+ "grad_norm": 0.3417568802833557,
2185
+ "learning_rate": 9.810491752117776e-05,
2186
+ "loss": 2.156,
2187
+ "step": 14750
2188
+ },
2189
+ {
2190
+ "epoch": 0.12771033914429758,
2191
+ "grad_norm": 0.35225942730903625,
2192
+ "learning_rate": 9.808555426018863e-05,
2193
+ "loss": 2.1734,
2194
+ "step": 14800
2195
+ },
2196
+ {
2197
+ "epoch": 0.12814179299275805,
2198
+ "grad_norm": 0.3499196767807007,
2199
+ "learning_rate": 9.806609450855511e-05,
2200
+ "loss": 2.1257,
2201
+ "step": 14850
2202
+ },
2203
+ {
2204
+ "epoch": 0.1285732468412185,
2205
+ "grad_norm": 0.3475897014141083,
2206
+ "learning_rate": 9.804653830532599e-05,
2207
+ "loss": 2.1718,
2208
+ "step": 14900
2209
+ },
2210
+ {
2211
+ "epoch": 0.12900470068967898,
2212
+ "grad_norm": 0.3465918302536011,
2213
+ "learning_rate": 9.80268856897437e-05,
2214
+ "loss": 2.1536,
2215
+ "step": 14950
2216
+ },
2217
+ {
2218
+ "epoch": 0.12943615453813945,
2219
+ "grad_norm": 0.3532946705818176,
2220
+ "learning_rate": 9.800713670124399e-05,
2221
+ "loss": 2.1877,
2222
+ "step": 15000
2223
+ },
2224
+ {
2225
+ "epoch": 0.12943615453813945,
2226
+ "eval_loss": 2.16441011428833,
2227
+ "eval_runtime": 2772.359,
2228
+ "eval_samples_per_second": 27.279,
2229
+ "eval_steps_per_second": 13.64,
2230
+ "step": 15000
2231
+ },
2232
+ {
2233
+ "epoch": 0.12986760838659991,
2234
+ "grad_norm": 0.3611823618412018,
2235
+ "learning_rate": 9.798729137945612e-05,
2236
+ "loss": 2.1417,
2237
+ "step": 15050
2238
+ },
2239
+ {
2240
+ "epoch": 0.13029906223506038,
2241
+ "grad_norm": 0.35656318068504333,
2242
+ "learning_rate": 9.79673497642026e-05,
2243
+ "loss": 2.1631,
2244
+ "step": 15100
2245
+ },
2246
+ {
2247
+ "epoch": 0.13073051608352085,
2248
+ "grad_norm": 0.3490156829357147,
2249
+ "learning_rate": 9.794731189549918e-05,
2250
+ "loss": 2.1591,
2251
+ "step": 15150
2252
+ },
2253
+ {
2254
+ "epoch": 0.1311619699319813,
2255
+ "grad_norm": 0.34025728702545166,
2256
+ "learning_rate": 9.792717781355473e-05,
2257
+ "loss": 2.1742,
2258
+ "step": 15200
2259
+ },
2260
+ {
2261
+ "epoch": 0.13159342378044175,
2262
+ "grad_norm": 0.33115556836128235,
2263
+ "learning_rate": 9.790694755877123e-05,
2264
+ "loss": 2.1254,
2265
+ "step": 15250
2266
+ },
2267
+ {
2268
+ "epoch": 0.13202487762890222,
2269
+ "grad_norm": 0.3736386001110077,
2270
+ "learning_rate": 9.788662117174362e-05,
2271
+ "loss": 2.1415,
2272
+ "step": 15300
2273
+ },
2274
+ {
2275
+ "epoch": 0.1324563314773627,
2276
+ "grad_norm": 0.3499143719673157,
2277
+ "learning_rate": 9.786619869325972e-05,
2278
+ "loss": 2.1464,
2279
+ "step": 15350
2280
+ },
2281
+ {
2282
+ "epoch": 0.13288778532582315,
2283
+ "grad_norm": 0.33793404698371887,
2284
+ "learning_rate": 9.784568016430024e-05,
2285
+ "loss": 2.1691,
2286
+ "step": 15400
2287
+ },
2288
+ {
2289
+ "epoch": 0.13331923917428362,
2290
+ "grad_norm": 0.3370670974254608,
2291
+ "learning_rate": 9.782506562603857e-05,
2292
+ "loss": 2.178,
2293
+ "step": 15450
2294
+ },
2295
+ {
2296
+ "epoch": 0.1337506930227441,
2297
+ "grad_norm": 0.34184974431991577,
2298
+ "learning_rate": 9.780435511984073e-05,
2299
+ "loss": 2.1577,
2300
+ "step": 15500
2301
+ },
2302
+ {
2303
+ "epoch": 0.13418214687120456,
2304
+ "grad_norm": 0.36029738187789917,
2305
+ "learning_rate": 9.778354868726544e-05,
2306
+ "loss": 2.1941,
2307
+ "step": 15550
2308
+ },
2309
+ {
2310
+ "epoch": 0.13461360071966502,
2311
+ "grad_norm": 0.342674195766449,
2312
+ "learning_rate": 9.776264637006376e-05,
2313
+ "loss": 2.1351,
2314
+ "step": 15600
2315
+ },
2316
+ {
2317
+ "epoch": 0.1350450545681255,
2318
+ "grad_norm": 0.35412395000457764,
2319
+ "learning_rate": 9.774164821017924e-05,
2320
+ "loss": 2.1266,
2321
+ "step": 15650
2322
+ },
2323
+ {
2324
+ "epoch": 0.13547650841658596,
2325
+ "grad_norm": 0.3524283766746521,
2326
+ "learning_rate": 9.772055424974775e-05,
2327
+ "loss": 2.1708,
2328
+ "step": 15700
2329
+ },
2330
+ {
2331
+ "epoch": 0.13590796226504642,
2332
+ "grad_norm": 0.33916088938713074,
2333
+ "learning_rate": 9.769936453109738e-05,
2334
+ "loss": 2.1399,
2335
+ "step": 15750
2336
+ },
2337
+ {
2338
+ "epoch": 0.1363394161135069,
2339
+ "grad_norm": 0.3382519483566284,
2340
+ "learning_rate": 9.767807909674838e-05,
2341
+ "loss": 2.1421,
2342
+ "step": 15800
2343
+ },
2344
+ {
2345
+ "epoch": 0.13677086996196736,
2346
+ "grad_norm": 0.3560401499271393,
2347
+ "learning_rate": 9.765669798941304e-05,
2348
+ "loss": 2.1618,
2349
+ "step": 15850
2350
+ },
2351
+ {
2352
+ "epoch": 0.1372023238104278,
2353
+ "grad_norm": 0.34324225783348083,
2354
+ "learning_rate": 9.763522125199567e-05,
2355
+ "loss": 2.1811,
2356
+ "step": 15900
2357
+ },
2358
+ {
2359
+ "epoch": 0.13763377765888826,
2360
+ "grad_norm": 0.3366933763027191,
2361
+ "learning_rate": 9.761364892759246e-05,
2362
+ "loss": 2.1512,
2363
+ "step": 15950
2364
+ },
2365
+ {
2366
+ "epoch": 0.13806523150734873,
2367
+ "grad_norm": 0.3429983854293823,
2368
+ "learning_rate": 9.759198105949141e-05,
2369
+ "loss": 2.155,
2370
+ "step": 16000
2371
+ },
2372
+ {
2373
+ "epoch": 0.13806523150734873,
2374
+ "eval_loss": 2.151211977005005,
2375
+ "eval_runtime": 2914.79,
2376
+ "eval_samples_per_second": 25.946,
2377
+ "eval_steps_per_second": 12.973,
2378
+ "step": 16000
2379
+ },
2380
+ {
2381
+ "epoch": 0.1384966853558092,
2382
+ "grad_norm": 0.349167138338089,
2383
+ "learning_rate": 9.757021769117225e-05,
2384
+ "loss": 2.1388,
2385
+ "step": 16050
2386
+ },
2387
+ {
2388
+ "epoch": 0.13892813920426966,
2389
+ "grad_norm": 0.3502015471458435,
2390
+ "learning_rate": 9.754835886630634e-05,
2391
+ "loss": 2.1562,
2392
+ "step": 16100
2393
+ },
2394
+ {
2395
+ "epoch": 0.13935959305273013,
2396
+ "grad_norm": 0.3419562876224518,
2397
+ "learning_rate": 9.752640462875655e-05,
2398
+ "loss": 2.1739,
2399
+ "step": 16150
2400
+ },
2401
+ {
2402
+ "epoch": 0.1397910469011906,
2403
+ "grad_norm": 0.3341248631477356,
2404
+ "learning_rate": 9.750435502257729e-05,
2405
+ "loss": 2.1397,
2406
+ "step": 16200
2407
+ },
2408
+ {
2409
+ "epoch": 0.14022250074965106,
2410
+ "grad_norm": 0.3449030816555023,
2411
+ "learning_rate": 9.748221009201429e-05,
2412
+ "loss": 2.1816,
2413
+ "step": 16250
2414
+ },
2415
+ {
2416
+ "epoch": 0.14065395459811153,
2417
+ "grad_norm": 0.33365193009376526,
2418
+ "learning_rate": 9.745996988150453e-05,
2419
+ "loss": 2.1348,
2420
+ "step": 16300
2421
+ },
2422
+ {
2423
+ "epoch": 0.141085408446572,
2424
+ "grad_norm": 0.34570497274398804,
2425
+ "learning_rate": 9.743763443567624e-05,
2426
+ "loss": 2.1359,
2427
+ "step": 16350
2428
+ },
2429
+ {
2430
+ "epoch": 0.14151686229503246,
2431
+ "grad_norm": 0.34485459327697754,
2432
+ "learning_rate": 9.741520379934876e-05,
2433
+ "loss": 2.1368,
2434
+ "step": 16400
2435
+ },
2436
+ {
2437
+ "epoch": 0.14194831614349293,
2438
+ "grad_norm": 0.3529675304889679,
2439
+ "learning_rate": 9.739267801753238e-05,
2440
+ "loss": 2.1268,
2441
+ "step": 16450
2442
+ },
2443
+ {
2444
+ "epoch": 0.1423797699919534,
2445
+ "grad_norm": 0.34084296226501465,
2446
+ "learning_rate": 9.737005713542837e-05,
2447
+ "loss": 2.1367,
2448
+ "step": 16500
2449
+ },
2450
+ {
2451
+ "epoch": 0.14281122384041384,
2452
+ "grad_norm": 0.33580780029296875,
2453
+ "learning_rate": 9.734734119842882e-05,
2454
+ "loss": 2.154,
2455
+ "step": 16550
2456
+ },
2457
+ {
2458
+ "epoch": 0.1432426776888743,
2459
+ "grad_norm": 0.3536083996295929,
2460
+ "learning_rate": 9.732453025211654e-05,
2461
+ "loss": 2.1375,
2462
+ "step": 16600
2463
+ },
2464
+ {
2465
+ "epoch": 0.14367413153733477,
2466
+ "grad_norm": 0.34920045733451843,
2467
+ "learning_rate": 9.730162434226502e-05,
2468
+ "loss": 2.1264,
2469
+ "step": 16650
2470
+ },
2471
+ {
2472
+ "epoch": 0.14410558538579524,
2473
+ "grad_norm": 0.34508806467056274,
2474
+ "learning_rate": 9.727862351483827e-05,
2475
+ "loss": 2.1525,
2476
+ "step": 16700
2477
+ },
2478
+ {
2479
+ "epoch": 0.1445370392342557,
2480
+ "grad_norm": 0.3422926664352417,
2481
+ "learning_rate": 9.72555278159908e-05,
2482
+ "loss": 2.1303,
2483
+ "step": 16750
2484
+ },
2485
+ {
2486
+ "epoch": 0.14496849308271617,
2487
+ "grad_norm": 0.3599984645843506,
2488
+ "learning_rate": 9.723233729206749e-05,
2489
+ "loss": 2.0945,
2490
+ "step": 16800
2491
+ },
2492
+ {
2493
+ "epoch": 0.14539994693117664,
2494
+ "grad_norm": 0.33985811471939087,
2495
+ "learning_rate": 9.720905198960348e-05,
2496
+ "loss": 2.1518,
2497
+ "step": 16850
2498
+ },
2499
+ {
2500
+ "epoch": 0.1458314007796371,
2501
+ "grad_norm": 0.3494996726512909,
2502
+ "learning_rate": 9.718567195532413e-05,
2503
+ "loss": 2.1526,
2504
+ "step": 16900
2505
+ },
2506
+ {
2507
+ "epoch": 0.14626285462809757,
2508
+ "grad_norm": 0.34371593594551086,
2509
+ "learning_rate": 9.716219723614486e-05,
2510
+ "loss": 2.1285,
2511
+ "step": 16950
2512
+ },
2513
+ {
2514
+ "epoch": 0.14669430847655804,
2515
+ "grad_norm": 0.34658172726631165,
2516
+ "learning_rate": 9.713862787917109e-05,
2517
+ "loss": 2.1572,
2518
+ "step": 17000
2519
+ },
2520
+ {
2521
+ "epoch": 0.14669430847655804,
2522
+ "eval_loss": 2.1390128135681152,
2523
+ "eval_runtime": 3020.0517,
2524
+ "eval_samples_per_second": 25.042,
2525
+ "eval_steps_per_second": 12.521,
2526
+ "step": 17000
2527
+ },
2528
+ {
2529
+ "epoch": 0.1471257623250185,
2530
+ "grad_norm": 0.3454444408416748,
2531
+ "learning_rate": 9.711496393169817e-05,
2532
+ "loss": 2.1443,
2533
+ "step": 17050
2534
+ },
2535
+ {
2536
+ "epoch": 0.14755721617347897,
2537
+ "grad_norm": 0.3368396759033203,
2538
+ "learning_rate": 9.709120544121123e-05,
2539
+ "loss": 2.1328,
2540
+ "step": 17100
2541
+ },
2542
+ {
2543
+ "epoch": 0.14798867002193944,
2544
+ "grad_norm": 0.35348692536354065,
2545
+ "learning_rate": 9.706735245538515e-05,
2546
+ "loss": 2.16,
2547
+ "step": 17150
2548
+ },
2549
+ {
2550
+ "epoch": 0.1484201238703999,
2551
+ "grad_norm": 0.35535895824432373,
2552
+ "learning_rate": 9.70434050220844e-05,
2553
+ "loss": 2.1613,
2554
+ "step": 17200
2555
+ },
2556
+ {
2557
+ "epoch": 0.14885157771886035,
2558
+ "grad_norm": 0.3357240557670593,
2559
+ "learning_rate": 9.701936318936297e-05,
2560
+ "loss": 2.1302,
2561
+ "step": 17250
2562
+ },
2563
+ {
2564
+ "epoch": 0.1492830315673208,
2565
+ "grad_norm": 0.33816105127334595,
2566
+ "learning_rate": 9.69952270054643e-05,
2567
+ "loss": 2.1315,
2568
+ "step": 17300
2569
+ },
2570
+ {
2571
+ "epoch": 0.14971448541578128,
2572
+ "grad_norm": 0.35641545057296753,
2573
+ "learning_rate": 9.697099651882116e-05,
2574
+ "loss": 2.121,
2575
+ "step": 17350
2576
+ },
2577
+ {
2578
+ "epoch": 0.15014593926424175,
2579
+ "grad_norm": 0.33782002329826355,
2580
+ "learning_rate": 9.694667177805552e-05,
2581
+ "loss": 2.1376,
2582
+ "step": 17400
2583
+ },
2584
+ {
2585
+ "epoch": 0.1505773931127022,
2586
+ "grad_norm": 0.33257773518562317,
2587
+ "learning_rate": 9.692225283197849e-05,
2588
+ "loss": 2.1159,
2589
+ "step": 17450
2590
+ },
2591
+ {
2592
+ "epoch": 0.15100884696116268,
2593
+ "grad_norm": 0.3327760100364685,
2594
+ "learning_rate": 9.689773972959026e-05,
2595
+ "loss": 2.1422,
2596
+ "step": 17500
2597
+ },
2598
+ {
2599
+ "epoch": 0.15144030080962315,
2600
+ "grad_norm": 0.33078470826148987,
2601
+ "learning_rate": 9.68731325200799e-05,
2602
+ "loss": 2.1279,
2603
+ "step": 17550
2604
+ },
2605
+ {
2606
+ "epoch": 0.1518717546580836,
2607
+ "grad_norm": 0.35639920830726624,
2608
+ "learning_rate": 9.684843125282538e-05,
2609
+ "loss": 2.1226,
2610
+ "step": 17600
2611
+ },
2612
+ {
2613
+ "epoch": 0.15230320850654408,
2614
+ "grad_norm": 0.33569735288619995,
2615
+ "learning_rate": 9.682363597739334e-05,
2616
+ "loss": 2.1263,
2617
+ "step": 17650
2618
+ },
2619
+ {
2620
+ "epoch": 0.15273466235500455,
2621
+ "grad_norm": 0.35261237621307373,
2622
+ "learning_rate": 9.679874674353913e-05,
2623
+ "loss": 2.1135,
2624
+ "step": 17700
2625
+ },
2626
+ {
2627
+ "epoch": 0.153166116203465,
2628
+ "grad_norm": 0.3444455564022064,
2629
+ "learning_rate": 9.67737636012066e-05,
2630
+ "loss": 2.0787,
2631
+ "step": 17750
2632
+ },
2633
+ {
2634
+ "epoch": 0.15359757005192548,
2635
+ "grad_norm": 0.32956936955451965,
2636
+ "learning_rate": 9.674868660052805e-05,
2637
+ "loss": 2.1283,
2638
+ "step": 17800
2639
+ },
2640
+ {
2641
+ "epoch": 0.15402902390038595,
2642
+ "grad_norm": 0.3369726836681366,
2643
+ "learning_rate": 9.672351579182413e-05,
2644
+ "loss": 2.1337,
2645
+ "step": 17850
2646
+ },
2647
+ {
2648
+ "epoch": 0.1544604777488464,
2649
+ "grad_norm": 0.3625190854072571,
2650
+ "learning_rate": 9.66982512256037e-05,
2651
+ "loss": 2.1165,
2652
+ "step": 17900
2653
+ },
2654
+ {
2655
+ "epoch": 0.15489193159730685,
2656
+ "grad_norm": 0.3413086235523224,
2657
+ "learning_rate": 9.667289295256379e-05,
2658
+ "loss": 2.1242,
2659
+ "step": 17950
2660
+ },
2661
+ {
2662
+ "epoch": 0.15532338544576732,
2663
+ "grad_norm": 0.34174641966819763,
2664
+ "learning_rate": 9.664744102358947e-05,
2665
+ "loss": 2.1325,
2666
+ "step": 18000
2667
+ },
2668
+ {
2669
+ "epoch": 0.15532338544576732,
2670
+ "eval_loss": 2.128389835357666,
2671
+ "eval_runtime": 2954.9582,
2672
+ "eval_samples_per_second": 25.594,
2673
+ "eval_steps_per_second": 12.797,
2674
+ "step": 18000
2675
+ },
2676
+ {
2677
+ "epoch": 0.1557548392942278,
2678
+ "grad_norm": 0.3447898328304291,
2679
+ "learning_rate": 9.662189548975373e-05,
2680
+ "loss": 2.1418,
2681
+ "step": 18050
2682
+ },
2683
+ {
2684
+ "epoch": 0.15618629314268825,
2685
+ "grad_norm": 0.3458497226238251,
2686
+ "learning_rate": 9.659625640231735e-05,
2687
+ "loss": 2.1415,
2688
+ "step": 18100
2689
+ },
2690
+ {
2691
+ "epoch": 0.15661774699114872,
2692
+ "grad_norm": 0.3496282994747162,
2693
+ "learning_rate": 9.657052381272893e-05,
2694
+ "loss": 2.1318,
2695
+ "step": 18150
2696
+ },
2697
+ {
2698
+ "epoch": 0.1570492008396092,
2699
+ "grad_norm": 0.3375896215438843,
2700
+ "learning_rate": 9.654469777262462e-05,
2701
+ "loss": 2.1342,
2702
+ "step": 18200
2703
+ },
2704
+ {
2705
+ "epoch": 0.15748065468806965,
2706
+ "grad_norm": 0.3378855288028717,
2707
+ "learning_rate": 9.651877833382815e-05,
2708
+ "loss": 2.126,
2709
+ "step": 18250
2710
+ },
2711
+ {
2712
+ "epoch": 0.15791210853653012,
2713
+ "grad_norm": 0.3357175588607788,
2714
+ "learning_rate": 9.649276554835058e-05,
2715
+ "loss": 2.14,
2716
+ "step": 18300
2717
+ },
2718
+ {
2719
+ "epoch": 0.1583435623849906,
2720
+ "grad_norm": 0.3392798602581024,
2721
+ "learning_rate": 9.64666594683904e-05,
2722
+ "loss": 2.1071,
2723
+ "step": 18350
2724
+ },
2725
+ {
2726
+ "epoch": 0.15877501623345106,
2727
+ "grad_norm": 0.33450230956077576,
2728
+ "learning_rate": 9.644046014633319e-05,
2729
+ "loss": 2.1219,
2730
+ "step": 18400
2731
+ },
2732
+ {
2733
+ "epoch": 0.15920647008191152,
2734
+ "grad_norm": 0.33854368329048157,
2735
+ "learning_rate": 9.641416763475174e-05,
2736
+ "loss": 2.1206,
2737
+ "step": 18450
2738
+ },
2739
+ {
2740
+ "epoch": 0.159637923930372,
2741
+ "grad_norm": 0.3468214273452759,
2742
+ "learning_rate": 9.638778198640576e-05,
2743
+ "loss": 2.1227,
2744
+ "step": 18500
2745
+ },
2746
+ {
2747
+ "epoch": 0.16006937777883246,
2748
+ "grad_norm": 0.3371368944644928,
2749
+ "learning_rate": 9.636130325424192e-05,
2750
+ "loss": 2.1259,
2751
+ "step": 18550
2752
+ },
2753
+ {
2754
+ "epoch": 0.1605008316272929,
2755
+ "grad_norm": 0.3576333224773407,
2756
+ "learning_rate": 9.633473149139361e-05,
2757
+ "loss": 2.1255,
2758
+ "step": 18600
2759
+ },
2760
+ {
2761
+ "epoch": 0.16093228547575336,
2762
+ "grad_norm": 0.34628191590309143,
2763
+ "learning_rate": 9.630806675118093e-05,
2764
+ "loss": 2.1616,
2765
+ "step": 18650
2766
+ },
2767
+ {
2768
+ "epoch": 0.16136373932421383,
2769
+ "grad_norm": 0.3629646599292755,
2770
+ "learning_rate": 9.628130908711056e-05,
2771
+ "loss": 2.1303,
2772
+ "step": 18700
2773
+ },
2774
+ {
2775
+ "epoch": 0.1617951931726743,
2776
+ "grad_norm": 0.3423612117767334,
2777
+ "learning_rate": 9.625445855287562e-05,
2778
+ "loss": 2.1291,
2779
+ "step": 18750
2780
+ },
2781
+ {
2782
+ "epoch": 0.16222664702113476,
2783
+ "grad_norm": 0.3248550295829773,
2784
+ "learning_rate": 9.622751520235564e-05,
2785
+ "loss": 2.1231,
2786
+ "step": 18800
2787
+ },
2788
+ {
2789
+ "epoch": 0.16265810086959523,
2790
+ "grad_norm": 0.3261340260505676,
2791
+ "learning_rate": 9.620047908961633e-05,
2792
+ "loss": 2.1206,
2793
+ "step": 18850
2794
+ },
2795
+ {
2796
+ "epoch": 0.1630895547180557,
2797
+ "grad_norm": 0.33387574553489685,
2798
+ "learning_rate": 9.61733502689096e-05,
2799
+ "loss": 2.1371,
2800
+ "step": 18900
2801
+ },
2802
+ {
2803
+ "epoch": 0.16352100856651616,
2804
+ "grad_norm": 0.3315752446651459,
2805
+ "learning_rate": 9.614612879467336e-05,
2806
+ "loss": 2.1134,
2807
+ "step": 18950
2808
+ },
2809
+ {
2810
+ "epoch": 0.16395246241497663,
2811
+ "grad_norm": 0.340299129486084,
2812
+ "learning_rate": 9.611881472153141e-05,
2813
+ "loss": 2.0927,
2814
+ "step": 19000
2815
+ },
2816
+ {
2817
+ "epoch": 0.16395246241497663,
2818
+ "eval_loss": 2.118671417236328,
2819
+ "eval_runtime": 3059.0098,
2820
+ "eval_samples_per_second": 24.723,
2821
+ "eval_steps_per_second": 12.362,
2822
+ "step": 19000
2823
+ },
2824
+ {
2825
+ "epoch": 0.1643839162634371,
2826
+ "grad_norm": 0.3347771465778351,
2827
+ "learning_rate": 9.609140810429347e-05,
2828
+ "loss": 2.072,
2829
+ "step": 19050
2830
+ },
2831
+ {
2832
+ "epoch": 0.16481537011189756,
2833
+ "grad_norm": 0.3293216824531555,
2834
+ "learning_rate": 9.606390899795483e-05,
2835
+ "loss": 2.0933,
2836
+ "step": 19100
2837
+ },
2838
+ {
2839
+ "epoch": 0.16524682396035803,
2840
+ "grad_norm": 0.3466013967990875,
2841
+ "learning_rate": 9.60363174576965e-05,
2842
+ "loss": 2.0871,
2843
+ "step": 19150
2844
+ },
2845
+ {
2846
+ "epoch": 0.1656782778088185,
2847
+ "grad_norm": 0.34989696741104126,
2848
+ "learning_rate": 9.600863353888485e-05,
2849
+ "loss": 2.1029,
2850
+ "step": 19200
2851
+ },
2852
+ {
2853
+ "epoch": 0.16610973165727894,
2854
+ "grad_norm": 0.3422117829322815,
2855
+ "learning_rate": 9.598085729707174e-05,
2856
+ "loss": 2.1076,
2857
+ "step": 19250
2858
+ },
2859
+ {
2860
+ "epoch": 0.1665411855057394,
2861
+ "grad_norm": 0.32813316583633423,
2862
+ "learning_rate": 9.59529887879942e-05,
2863
+ "loss": 2.089,
2864
+ "step": 19300
2865
+ },
2866
+ {
2867
+ "epoch": 0.16697263935419987,
2868
+ "grad_norm": 0.3397763967514038,
2869
+ "learning_rate": 9.592502806757442e-05,
2870
+ "loss": 2.0949,
2871
+ "step": 19350
2872
+ },
2873
+ {
2874
+ "epoch": 0.16740409320266034,
2875
+ "grad_norm": 0.3325379490852356,
2876
+ "learning_rate": 9.589697519191967e-05,
2877
+ "loss": 2.1023,
2878
+ "step": 19400
2879
+ },
2880
+ {
2881
+ "epoch": 0.1678355470511208,
2882
+ "grad_norm": 0.3426936864852905,
2883
+ "learning_rate": 9.586883021732213e-05,
2884
+ "loss": 2.1364,
2885
+ "step": 19450
2886
+ },
2887
+ {
2888
+ "epoch": 0.16826700089958127,
2889
+ "grad_norm": 0.3614189922809601,
2890
+ "learning_rate": 9.584059320025876e-05,
2891
+ "loss": 2.13,
2892
+ "step": 19500
2893
+ },
2894
+ {
2895
+ "epoch": 0.16869845474804174,
2896
+ "grad_norm": 0.3224940001964569,
2897
+ "learning_rate": 9.581226419739122e-05,
2898
+ "loss": 2.094,
2899
+ "step": 19550
2900
+ },
2901
+ {
2902
+ "epoch": 0.1691299085965022,
2903
+ "grad_norm": 0.34935611486434937,
2904
+ "learning_rate": 9.578384326556577e-05,
2905
+ "loss": 2.1284,
2906
+ "step": 19600
2907
+ },
2908
+ {
2909
+ "epoch": 0.16956136244496267,
2910
+ "grad_norm": 0.33929896354675293,
2911
+ "learning_rate": 9.575533046181315e-05,
2912
+ "loss": 2.1261,
2913
+ "step": 19650
2914
+ },
2915
+ {
2916
+ "epoch": 0.16999281629342314,
2917
+ "grad_norm": 0.31563815474510193,
2918
+ "learning_rate": 9.572672584334845e-05,
2919
+ "loss": 2.1374,
2920
+ "step": 19700
2921
+ },
2922
+ {
2923
+ "epoch": 0.1704242701418836,
2924
+ "grad_norm": 0.33976227045059204,
2925
+ "learning_rate": 9.569802946757097e-05,
2926
+ "loss": 2.0851,
2927
+ "step": 19750
2928
+ },
2929
+ {
2930
+ "epoch": 0.17085572399034407,
2931
+ "grad_norm": 0.35277485847473145,
2932
+ "learning_rate": 9.566924139206413e-05,
2933
+ "loss": 2.1082,
2934
+ "step": 19800
2935
+ },
2936
+ {
2937
+ "epoch": 0.17128717783880454,
2938
+ "grad_norm": 0.34546715021133423,
2939
+ "learning_rate": 9.564036167459544e-05,
2940
+ "loss": 2.1219,
2941
+ "step": 19850
2942
+ },
2943
+ {
2944
+ "epoch": 0.171718631687265,
2945
+ "grad_norm": 0.32824838161468506,
2946
+ "learning_rate": 9.56113903731162e-05,
2947
+ "loss": 2.1053,
2948
+ "step": 19900
2949
+ },
2950
+ {
2951
+ "epoch": 0.17215008553572544,
2952
+ "grad_norm": 0.35051465034484863,
2953
+ "learning_rate": 9.558232754576154e-05,
2954
+ "loss": 2.0876,
2955
+ "step": 19950
2956
+ },
2957
+ {
2958
+ "epoch": 0.1725815393841859,
2959
+ "grad_norm": 0.3445969223976135,
2960
+ "learning_rate": 9.555317325085024e-05,
2961
+ "loss": 2.116,
2962
+ "step": 20000
2963
+ },
2964
+ {
2965
+ "epoch": 0.1725815393841859,
2966
+ "eval_loss": 2.1093990802764893,
2967
+ "eval_runtime": 2936.2197,
2968
+ "eval_samples_per_second": 25.757,
2969
+ "eval_steps_per_second": 12.878,
2970
+ "step": 20000
2971
+ },
2972
+ {
2973
+ "epoch": 0.17301299323264638,
2974
+ "grad_norm": 0.35169127583503723,
2975
+ "learning_rate": 9.552392754688465e-05,
2976
+ "loss": 2.075,
2977
+ "step": 20050
2978
+ },
2979
+ {
2980
+ "epoch": 0.17344444708110685,
2981
+ "grad_norm": 0.34275829792022705,
2982
+ "learning_rate": 9.549459049255049e-05,
2983
+ "loss": 2.1033,
2984
+ "step": 20100
2985
+ },
2986
+ {
2987
+ "epoch": 0.1738759009295673,
2988
+ "grad_norm": 0.3412991464138031,
2989
+ "learning_rate": 9.546516214671684e-05,
2990
+ "loss": 2.1315,
2991
+ "step": 20150
2992
+ },
2993
+ {
2994
+ "epoch": 0.17430735477802778,
2995
+ "grad_norm": 0.32284075021743774,
2996
+ "learning_rate": 9.543564256843593e-05,
2997
+ "loss": 2.121,
2998
+ "step": 20200
2999
+ },
3000
+ {
3001
+ "epoch": 0.17473880862648825,
3002
+ "grad_norm": 0.33116912841796875,
3003
+ "learning_rate": 9.540603181694307e-05,
3004
+ "loss": 2.0749,
3005
+ "step": 20250
3006
+ },
3007
+ {
3008
+ "epoch": 0.1751702624749487,
3009
+ "grad_norm": 0.34137117862701416,
3010
+ "learning_rate": 9.537632995165658e-05,
3011
+ "loss": 2.1103,
3012
+ "step": 20300
3013
+ },
3014
+ {
3015
+ "epoch": 0.17560171632340918,
3016
+ "grad_norm": 0.3352222442626953,
3017
+ "learning_rate": 9.534653703217751e-05,
3018
+ "loss": 2.0891,
3019
+ "step": 20350
3020
+ },
3021
+ {
3022
+ "epoch": 0.17603317017186965,
3023
+ "grad_norm": 0.3311406373977661,
3024
+ "learning_rate": 9.53166531182897e-05,
3025
+ "loss": 2.1171,
3026
+ "step": 20400
3027
+ },
3028
+ {
3029
+ "epoch": 0.1764646240203301,
3030
+ "grad_norm": 0.33530110120773315,
3031
+ "learning_rate": 9.528667826995955e-05,
3032
+ "loss": 2.1015,
3033
+ "step": 20450
3034
+ },
3035
+ {
3036
+ "epoch": 0.17689607786879058,
3037
+ "grad_norm": 0.32751473784446716,
3038
+ "learning_rate": 9.525661254733596e-05,
3039
+ "loss": 2.0727,
3040
+ "step": 20500
3041
+ },
3042
+ {
3043
+ "epoch": 0.17732753171725105,
3044
+ "grad_norm": 0.3411867320537567,
3045
+ "learning_rate": 9.522645601075016e-05,
3046
+ "loss": 2.1006,
3047
+ "step": 20550
3048
+ },
3049
+ {
3050
+ "epoch": 0.17775898556571149,
3051
+ "grad_norm": 0.3302497863769531,
3052
+ "learning_rate": 9.519620872071561e-05,
3053
+ "loss": 2.1083,
3054
+ "step": 20600
3055
+ },
3056
+ {
3057
+ "epoch": 0.17819043941417195,
3058
+ "grad_norm": 0.3380363881587982,
3059
+ "learning_rate": 9.516587073792792e-05,
3060
+ "loss": 2.115,
3061
+ "step": 20650
3062
+ },
3063
+ {
3064
+ "epoch": 0.17862189326263242,
3065
+ "grad_norm": 0.3405621349811554,
3066
+ "learning_rate": 9.513544212326461e-05,
3067
+ "loss": 2.1424,
3068
+ "step": 20700
3069
+ },
3070
+ {
3071
+ "epoch": 0.1790533471110929,
3072
+ "grad_norm": 0.3401827812194824,
3073
+ "learning_rate": 9.510492293778513e-05,
3074
+ "loss": 2.0871,
3075
+ "step": 20750
3076
+ },
3077
+ {
3078
+ "epoch": 0.17948480095955335,
3079
+ "grad_norm": 0.34303560853004456,
3080
+ "learning_rate": 9.507431324273067e-05,
3081
+ "loss": 2.1215,
3082
+ "step": 20800
3083
+ },
3084
+ {
3085
+ "epoch": 0.17991625480801382,
3086
+ "grad_norm": 0.34137266874313354,
3087
+ "learning_rate": 9.504361309952401e-05,
3088
+ "loss": 2.1171,
3089
+ "step": 20850
3090
+ },
3091
+ {
3092
+ "epoch": 0.1803477086564743,
3093
+ "grad_norm": 0.32752570509910583,
3094
+ "learning_rate": 9.501282256976945e-05,
3095
+ "loss": 2.1038,
3096
+ "step": 20900
3097
+ },
3098
+ {
3099
+ "epoch": 0.18077916250493475,
3100
+ "grad_norm": 0.33904024958610535,
3101
+ "learning_rate": 9.498194171525267e-05,
3102
+ "loss": 2.1143,
3103
+ "step": 20950
3104
+ },
3105
+ {
3106
+ "epoch": 0.18121061635339522,
3107
+ "grad_norm": 0.3322898745536804,
3108
+ "learning_rate": 9.495097059794057e-05,
3109
+ "loss": 2.1101,
3110
+ "step": 21000
3111
+ },
3112
+ {
3113
+ "epoch": 0.18121061635339522,
3114
+ "eval_loss": 2.1013271808624268,
3115
+ "eval_runtime": 2858.1247,
3116
+ "eval_samples_per_second": 26.461,
3117
+ "eval_steps_per_second": 13.23,
3118
+ "step": 21000
3119
+ },
3120
+ {
3121
+ "epoch": 0.1816420702018557,
3122
+ "grad_norm": 0.3301975131034851,
3123
+ "learning_rate": 9.491990927998117e-05,
3124
+ "loss": 2.0876,
3125
+ "step": 21050
3126
+ },
3127
+ {
3128
+ "epoch": 0.18207352405031615,
3129
+ "grad_norm": 0.3501399755477905,
3130
+ "learning_rate": 9.488875782370356e-05,
3131
+ "loss": 2.1131,
3132
+ "step": 21100
3133
+ },
3134
+ {
3135
+ "epoch": 0.18250497789877662,
3136
+ "grad_norm": 0.33008062839508057,
3137
+ "learning_rate": 9.485751629161763e-05,
3138
+ "loss": 2.1086,
3139
+ "step": 21150
3140
+ },
3141
+ {
3142
+ "epoch": 0.1829364317472371,
3143
+ "grad_norm": 0.3411719501018524,
3144
+ "learning_rate": 9.482618474641407e-05,
3145
+ "loss": 2.0703,
3146
+ "step": 21200
3147
+ },
3148
+ {
3149
+ "epoch": 0.18336788559569756,
3150
+ "grad_norm": 0.33804383873939514,
3151
+ "learning_rate": 9.479476325096414e-05,
3152
+ "loss": 2.1167,
3153
+ "step": 21250
3154
+ },
3155
+ {
3156
+ "epoch": 0.183799339444158,
3157
+ "grad_norm": 0.34236985445022583,
3158
+ "learning_rate": 9.476325186831967e-05,
3159
+ "loss": 2.0685,
3160
+ "step": 21300
3161
+ },
3162
+ {
3163
+ "epoch": 0.18423079329261846,
3164
+ "grad_norm": 0.3437114953994751,
3165
+ "learning_rate": 9.47316506617128e-05,
3166
+ "loss": 2.0929,
3167
+ "step": 21350
3168
+ },
3169
+ {
3170
+ "epoch": 0.18466224714107893,
3171
+ "grad_norm": 0.33738139271736145,
3172
+ "learning_rate": 9.469995969455593e-05,
3173
+ "loss": 2.0855,
3174
+ "step": 21400
3175
+ },
3176
+ {
3177
+ "epoch": 0.1850937009895394,
3178
+ "grad_norm": 0.3287017345428467,
3179
+ "learning_rate": 9.46681790304416e-05,
3180
+ "loss": 2.1089,
3181
+ "step": 21450
3182
+ },
3183
+ {
3184
+ "epoch": 0.18552515483799986,
3185
+ "grad_norm": 0.3263178765773773,
3186
+ "learning_rate": 9.463630873314233e-05,
3187
+ "loss": 2.0775,
3188
+ "step": 21500
3189
+ },
3190
+ {
3191
+ "epoch": 0.18595660868646033,
3192
+ "grad_norm": 0.3348768353462219,
3193
+ "learning_rate": 9.460434886661049e-05,
3194
+ "loss": 2.1013,
3195
+ "step": 21550
3196
+ },
3197
+ {
3198
+ "epoch": 0.1863880625349208,
3199
+ "grad_norm": 0.3428316116333008,
3200
+ "learning_rate": 9.457229949497822e-05,
3201
+ "loss": 2.0863,
3202
+ "step": 21600
3203
+ },
3204
+ {
3205
+ "epoch": 0.18681951638338126,
3206
+ "grad_norm": 0.3480794429779053,
3207
+ "learning_rate": 9.454016068255717e-05,
3208
+ "loss": 2.1077,
3209
+ "step": 21650
3210
+ },
3211
+ {
3212
+ "epoch": 0.18725097023184173,
3213
+ "grad_norm": 0.3354591131210327,
3214
+ "learning_rate": 9.450793249383858e-05,
3215
+ "loss": 2.0962,
3216
+ "step": 21700
3217
+ },
3218
+ {
3219
+ "epoch": 0.1876824240803022,
3220
+ "grad_norm": 0.33805784583091736,
3221
+ "learning_rate": 9.4475614993493e-05,
3222
+ "loss": 2.1287,
3223
+ "step": 21750
3224
+ },
3225
+ {
3226
+ "epoch": 0.18811387792876266,
3227
+ "grad_norm": 0.32974615693092346,
3228
+ "learning_rate": 9.444320824637014e-05,
3229
+ "loss": 2.0657,
3230
+ "step": 21800
3231
+ },
3232
+ {
3233
+ "epoch": 0.18854533177722313,
3234
+ "grad_norm": 0.33754146099090576,
3235
+ "learning_rate": 9.441071231749889e-05,
3236
+ "loss": 2.0863,
3237
+ "step": 21850
3238
+ },
3239
+ {
3240
+ "epoch": 0.1889767856256836,
3241
+ "grad_norm": 0.3371632993221283,
3242
+ "learning_rate": 9.437812727208702e-05,
3243
+ "loss": 2.1099,
3244
+ "step": 21900
3245
+ },
3246
+ {
3247
+ "epoch": 0.18940823947414406,
3248
+ "grad_norm": 0.3343505561351776,
3249
+ "learning_rate": 9.434545317552115e-05,
3250
+ "loss": 2.0833,
3251
+ "step": 21950
3252
+ },
3253
+ {
3254
+ "epoch": 0.1898396933226045,
3255
+ "grad_norm": 0.33940064907073975,
3256
+ "learning_rate": 9.431269009336661e-05,
3257
+ "loss": 2.1066,
3258
+ "step": 22000
3259
+ },
3260
+ {
3261
+ "epoch": 0.1898396933226045,
3262
+ "eval_loss": 2.093938112258911,
3263
+ "eval_runtime": 2820.9238,
3264
+ "eval_samples_per_second": 26.81,
3265
+ "eval_steps_per_second": 13.405,
3266
+ "step": 22000
3267
+ },
3268
+ {
3269
+ "epoch": 0.19027114717106497,
3270
+ "grad_norm": 0.32727915048599243,
3271
+ "learning_rate": 9.427983809136727e-05,
3272
+ "loss": 2.0494,
3273
+ "step": 22050
3274
+ },
3275
+ {
3276
+ "epoch": 0.19070260101952544,
3277
+ "grad_norm": 0.3239200711250305,
3278
+ "learning_rate": 9.424689723544545e-05,
3279
+ "loss": 2.0716,
3280
+ "step": 22100
3281
+ },
3282
+ {
3283
+ "epoch": 0.1911340548679859,
3284
+ "grad_norm": 0.3169387876987457,
3285
+ "learning_rate": 9.421386759170177e-05,
3286
+ "loss": 2.0903,
3287
+ "step": 22150
3288
+ },
3289
+ {
3290
+ "epoch": 0.19156550871644637,
3291
+ "grad_norm": 0.3533649146556854,
3292
+ "learning_rate": 9.418074922641498e-05,
3293
+ "loss": 2.1177,
3294
+ "step": 22200
3295
+ },
3296
+ {
3297
+ "epoch": 0.19199696256490684,
3298
+ "grad_norm": 0.3385666608810425,
3299
+ "learning_rate": 9.41475422060419e-05,
3300
+ "loss": 2.0788,
3301
+ "step": 22250
3302
+ },
3303
+ {
3304
+ "epoch": 0.1924284164133673,
3305
+ "grad_norm": 0.3421185612678528,
3306
+ "learning_rate": 9.411424659721722e-05,
3307
+ "loss": 2.1098,
3308
+ "step": 22300
3309
+ },
3310
+ {
3311
+ "epoch": 0.19285987026182777,
3312
+ "grad_norm": 0.3351210653781891,
3313
+ "learning_rate": 9.408086246675345e-05,
3314
+ "loss": 2.0829,
3315
+ "step": 22350
3316
+ },
3317
+ {
3318
+ "epoch": 0.19329132411028824,
3319
+ "grad_norm": 0.32542884349823,
3320
+ "learning_rate": 9.404738988164065e-05,
3321
+ "loss": 2.056,
3322
+ "step": 22400
3323
+ },
3324
+ {
3325
+ "epoch": 0.1937227779587487,
3326
+ "grad_norm": 0.32976922392845154,
3327
+ "learning_rate": 9.401382890904645e-05,
3328
+ "loss": 2.1095,
3329
+ "step": 22450
3330
+ },
3331
+ {
3332
+ "epoch": 0.19415423180720917,
3333
+ "grad_norm": 0.3366084098815918,
3334
+ "learning_rate": 9.39801796163158e-05,
3335
+ "loss": 2.051,
3336
+ "step": 22500
3337
+ },
3338
+ {
3339
+ "epoch": 0.19458568565566964,
3340
+ "grad_norm": 0.32825997471809387,
3341
+ "learning_rate": 9.394644207097089e-05,
3342
+ "loss": 2.0851,
3343
+ "step": 22550
3344
+ },
3345
+ {
3346
+ "epoch": 0.1950171395041301,
3347
+ "grad_norm": 0.33241936564445496,
3348
+ "learning_rate": 9.391261634071102e-05,
3349
+ "loss": 2.1218,
3350
+ "step": 22600
3351
+ },
3352
+ {
3353
+ "epoch": 0.19544859335259054,
3354
+ "grad_norm": 0.33546289801597595,
3355
+ "learning_rate": 9.38787024934124e-05,
3356
+ "loss": 2.0866,
3357
+ "step": 22650
3358
+ },
3359
+ {
3360
+ "epoch": 0.195880047201051,
3361
+ "grad_norm": 0.3313877284526825,
3362
+ "learning_rate": 9.384470059712811e-05,
3363
+ "loss": 2.0722,
3364
+ "step": 22700
3365
+ },
3366
+ {
3367
+ "epoch": 0.19631150104951148,
3368
+ "grad_norm": 0.3336232900619507,
3369
+ "learning_rate": 9.381061072008786e-05,
3370
+ "loss": 2.0831,
3371
+ "step": 22750
3372
+ },
3373
+ {
3374
+ "epoch": 0.19674295489797194,
3375
+ "grad_norm": 0.3445178270339966,
3376
+ "learning_rate": 9.377643293069798e-05,
3377
+ "loss": 2.1004,
3378
+ "step": 22800
3379
+ },
3380
+ {
3381
+ "epoch": 0.1971744087464324,
3382
+ "grad_norm": 0.33248141407966614,
3383
+ "learning_rate": 9.374216729754113e-05,
3384
+ "loss": 2.071,
3385
+ "step": 22850
3386
+ },
3387
+ {
3388
+ "epoch": 0.19760586259489288,
3389
+ "grad_norm": 0.3260776996612549,
3390
+ "learning_rate": 9.370781388937631e-05,
3391
+ "loss": 2.1116,
3392
+ "step": 22900
3393
+ },
3394
+ {
3395
+ "epoch": 0.19803731644335335,
3396
+ "grad_norm": 0.3381562829017639,
3397
+ "learning_rate": 9.367337277513858e-05,
3398
+ "loss": 2.0624,
3399
+ "step": 22950
3400
+ },
3401
+ {
3402
+ "epoch": 0.1984687702918138,
3403
+ "grad_norm": 0.3296578824520111,
3404
+ "learning_rate": 9.363884402393905e-05,
3405
+ "loss": 2.0885,
3406
+ "step": 23000
3407
+ },
3408
+ {
3409
+ "epoch": 0.1984687702918138,
3410
+ "eval_loss": 2.086437463760376,
3411
+ "eval_runtime": 2791.6728,
3412
+ "eval_samples_per_second": 27.091,
3413
+ "eval_steps_per_second": 13.545,
3414
+ "step": 23000
3415
+ },
3416
+ {
3417
+ "epoch": 0.19890022414027428,
3418
+ "grad_norm": 0.3351718485355377,
3419
+ "learning_rate": 9.360422770506469e-05,
3420
+ "loss": 2.1152,
3421
+ "step": 23050
3422
+ },
3423
+ {
3424
+ "epoch": 0.19933167798873475,
3425
+ "grad_norm": 0.344475120306015,
3426
+ "learning_rate": 9.356952388797815e-05,
3427
+ "loss": 2.0992,
3428
+ "step": 23100
3429
+ },
3430
+ {
3431
+ "epoch": 0.1997631318371952,
3432
+ "grad_norm": 0.32130861282348633,
3433
+ "learning_rate": 9.353473264231768e-05,
3434
+ "loss": 2.1008,
3435
+ "step": 23150
3436
+ },
3437
+ {
3438
+ "epoch": 0.20019458568565568,
3439
+ "grad_norm": 0.3410738408565521,
3440
+ "learning_rate": 9.349985403789696e-05,
3441
+ "loss": 2.0973,
3442
+ "step": 23200
3443
+ },
3444
+ {
3445
+ "epoch": 0.20062603953411615,
3446
+ "grad_norm": 0.3305729627609253,
3447
+ "learning_rate": 9.346488814470498e-05,
3448
+ "loss": 2.0821,
3449
+ "step": 23250
3450
+ },
3451
+ {
3452
+ "epoch": 0.2010574933825766,
3453
+ "grad_norm": 0.3449229598045349,
3454
+ "learning_rate": 9.342983503290587e-05,
3455
+ "loss": 2.1085,
3456
+ "step": 23300
3457
+ },
3458
+ {
3459
+ "epoch": 0.20148894723103705,
3460
+ "grad_norm": 0.3349471092224121,
3461
+ "learning_rate": 9.339469477283882e-05,
3462
+ "loss": 2.0927,
3463
+ "step": 23350
3464
+ },
3465
+ {
3466
+ "epoch": 0.20192040107949752,
3467
+ "grad_norm": 0.35080793499946594,
3468
+ "learning_rate": 9.335946743501783e-05,
3469
+ "loss": 2.083,
3470
+ "step": 23400
3471
+ },
3472
+ {
3473
+ "epoch": 0.20235185492795799,
3474
+ "grad_norm": 0.33375632762908936,
3475
+ "learning_rate": 9.332415309013165e-05,
3476
+ "loss": 2.1197,
3477
+ "step": 23450
3478
+ },
3479
+ {
3480
+ "epoch": 0.20278330877641845,
3481
+ "grad_norm": 0.3398565351963043,
3482
+ "learning_rate": 9.328875180904368e-05,
3483
+ "loss": 2.0895,
3484
+ "step": 23500
3485
+ },
3486
+ {
3487
+ "epoch": 0.20321476262487892,
3488
+ "grad_norm": 0.3695748746395111,
3489
+ "learning_rate": 9.32532636627917e-05,
3490
+ "loss": 2.0526,
3491
+ "step": 23550
3492
+ },
3493
+ {
3494
+ "epoch": 0.2036462164733394,
3495
+ "grad_norm": 0.3366464078426361,
3496
+ "learning_rate": 9.321768872258784e-05,
3497
+ "loss": 2.0781,
3498
+ "step": 23600
3499
+ },
3500
+ {
3501
+ "epoch": 0.20407767032179985,
3502
+ "grad_norm": 0.32691192626953125,
3503
+ "learning_rate": 9.31820270598184e-05,
3504
+ "loss": 2.0826,
3505
+ "step": 23650
3506
+ },
3507
+ {
3508
+ "epoch": 0.20450912417026032,
3509
+ "grad_norm": 0.33307206630706787,
3510
+ "learning_rate": 9.314627874604364e-05,
3511
+ "loss": 2.0769,
3512
+ "step": 23700
3513
+ },
3514
+ {
3515
+ "epoch": 0.2049405780187208,
3516
+ "grad_norm": 0.32544660568237305,
3517
+ "learning_rate": 9.311044385299778e-05,
3518
+ "loss": 2.0789,
3519
+ "step": 23750
3520
+ },
3521
+ {
3522
+ "epoch": 0.20537203186718125,
3523
+ "grad_norm": 0.3445671796798706,
3524
+ "learning_rate": 9.307452245258872e-05,
3525
+ "loss": 2.063,
3526
+ "step": 23800
3527
+ },
3528
+ {
3529
+ "epoch": 0.20580348571564172,
3530
+ "grad_norm": 0.3327982723712921,
3531
+ "learning_rate": 9.303851461689794e-05,
3532
+ "loss": 2.0549,
3533
+ "step": 23850
3534
+ },
3535
+ {
3536
+ "epoch": 0.2062349395641022,
3537
+ "grad_norm": 0.34181317687034607,
3538
+ "learning_rate": 9.300242041818044e-05,
3539
+ "loss": 2.1197,
3540
+ "step": 23900
3541
+ },
3542
+ {
3543
+ "epoch": 0.20666639341256265,
3544
+ "grad_norm": 0.3366542160511017,
3545
+ "learning_rate": 9.296623992886442e-05,
3546
+ "loss": 2.0681,
3547
+ "step": 23950
3548
+ },
3549
+ {
3550
+ "epoch": 0.2070978472610231,
3551
+ "grad_norm": 0.33302879333496094,
3552
+ "learning_rate": 9.292997322155129e-05,
3553
+ "loss": 2.096,
3554
+ "step": 24000
3555
+ },
3556
+ {
3557
+ "epoch": 0.2070978472610231,
3558
+ "eval_loss": 2.080108404159546,
3559
+ "eval_runtime": 2808.0771,
3560
+ "eval_samples_per_second": 26.932,
3561
+ "eval_steps_per_second": 13.466,
3562
+ "step": 24000
3563
+ }
3564
+ ],
3565
+ "logging_steps": 50,
3566
+ "max_steps": 115888,
3567
+ "num_input_tokens_seen": 0,
3568
+ "num_train_epochs": 1,
3569
+ "save_steps": 1000,
3570
+ "stateful_callbacks": {
3571
+ "EarlyStoppingCallback": {
3572
+ "args": {
3573
+ "early_stopping_patience": 3,
3574
+ "early_stopping_threshold": 0.0
3575
+ },
3576
+ "attributes": {
3577
+ "early_stopping_patience_counter": 0
3578
+ }
3579
+ },
3580
+ "TrainerControl": {
3581
+ "args": {
3582
+ "should_epoch_stop": false,
3583
+ "should_evaluate": false,
3584
+ "should_log": false,
3585
+ "should_save": true,
3586
+ "should_training_stop": false
3587
+ },
3588
+ "attributes": {}
3589
+ }
3590
+ },
3591
+ "total_flos": 1.06893548716032e+18,
3592
+ "train_batch_size": 2,
3593
+ "trial_name": null,
3594
+ "trial_params": null
3595
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-108000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45400e5719e350eae239c5ac1f19604c105bbfee7f1eb21d9e7308c70716fc5f
3
+ size 5304
qwen3-150m-multilingual-checkpoints/checkpoint-112500/config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 2,
8
+ "eos_token_id": 3,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 768,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2048,
14
+ "layer_types": [
15
+ "full_attention",
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
+ ],
28
+ "max_position_embeddings": 2048,
29
+ "max_window_layers": 28,
30
+ "model_type": "qwen3",
31
+ "num_attention_heads": 12,
32
+ "num_hidden_layers": 12,
33
+ "num_key_value_heads": 12,
34
+ "pad_token_id": 1,
35
+ "rms_norm_eps": 1e-06,
36
+ "rope_scaling": null,
37
+ "rope_theta": 10000.0,
38
+ "sliding_window": null,
39
+ "tie_word_embeddings": true,
40
+ "torch_dtype": "float32",
41
+ "transformers_version": "4.55.4",
42
+ "use_cache": true,
43
+ "use_sliding_window": false,
44
+ "vocab_size": 32000
45
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-112500/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 2,
4
+ "eos_token_id": 3,
5
+ "pad_token_id": 1,
6
+ "transformers_version": "4.55.4"
7
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-112500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00a6c13d93e27205d4aa9171ad80cf2ebbc16173e87ee166796e0e8434fa9a27
3
+ size 551392792
qwen3-150m-multilingual-checkpoints/checkpoint-112500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2203e587cb5abf8b23997dd59db38ff0f57de67d13290cbaf471259c182b28b7
3
+ size 1102868538
qwen3-150m-multilingual-checkpoints/checkpoint-112500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ac5c921b80607bc6b542eec40c2324594db2b30c39cd6ead0c5c2d4ac8e298e
3
+ size 14244
qwen3-150m-multilingual-checkpoints/checkpoint-112500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3bfd960a056ca52df0d037a7a79891806b97a1edcec1152025bd4f7f1cec5001
3
+ size 1064
qwen3-150m-multilingual-checkpoints/checkpoint-112500/special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<pad>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-112500/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3-150m-multilingual-checkpoints/checkpoint-112500/tokenizer_config.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "<unk>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<pad>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "2": {
20
+ "content": "<s>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "3": {
28
+ "content": "</s>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ },
35
+ "4": {
36
+ "content": "<en>",
37
+ "lstrip": false,
38
+ "normalized": false,
39
+ "rstrip": false,
40
+ "single_word": false,
41
+ "special": true
42
+ },
43
+ "5": {
44
+ "content": "<te>",
45
+ "lstrip": false,
46
+ "normalized": false,
47
+ "rstrip": false,
48
+ "single_word": false,
49
+ "special": true
50
+ },
51
+ "6": {
52
+ "content": "<mai>",
53
+ "lstrip": false,
54
+ "normalized": false,
55
+ "rstrip": false,
56
+ "single_word": false,
57
+ "special": true
58
+ },
59
+ "7": {
60
+ "content": "<mask>",
61
+ "lstrip": false,
62
+ "normalized": false,
63
+ "rstrip": false,
64
+ "single_word": false,
65
+ "special": true
66
+ }
67
+ },
68
+ "bos_token": "<s>",
69
+ "clean_up_tokenization_spaces": false,
70
+ "eos_token": "</s>",
71
+ "extra_special_tokens": {},
72
+ "model_max_length": 1000000000000000019884624838656,
73
+ "pad_token": "<pad>",
74
+ "tokenizer_class": "PreTrainedTokenizerFast",
75
+ "unk_token": "<unk>"
76
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-112500/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3-150m-multilingual-checkpoints/checkpoint-112500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45400e5719e350eae239c5ac1f19604c105bbfee7f1eb21d9e7308c70716fc5f
3
+ size 5304
qwen3-150m-multilingual-checkpoints/checkpoint-115888/config.json ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 2,
8
+ "eos_token_id": 3,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 768,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2048,
14
+ "layer_types": [
15
+ "full_attention",
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
+ ],
28
+ "max_position_embeddings": 2048,
29
+ "max_window_layers": 28,
30
+ "model_type": "qwen3",
31
+ "num_attention_heads": 12,
32
+ "num_hidden_layers": 12,
33
+ "num_key_value_heads": 12,
34
+ "pad_token_id": 1,
35
+ "rms_norm_eps": 1e-06,
36
+ "rope_scaling": null,
37
+ "rope_theta": 10000.0,
38
+ "sliding_window": null,
39
+ "tie_word_embeddings": true,
40
+ "torch_dtype": "float32",
41
+ "transformers_version": "4.55.4",
42
+ "use_cache": true,
43
+ "use_sliding_window": false,
44
+ "vocab_size": 32000
45
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-115888/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 2,
4
+ "eos_token_id": 3,
5
+ "pad_token_id": 1,
6
+ "transformers_version": "4.55.4"
7
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-115888/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09dd28fbead93244946e8d251b97fd238ef611d47245122514f2f5d568863f40
3
+ size 551392792
qwen3-150m-multilingual-checkpoints/checkpoint-115888/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f5a4114f1d6443c8e0360db2006fe8f4002ce8844c42e1a097fe43842f785a92
3
+ size 1102868538
qwen3-150m-multilingual-checkpoints/checkpoint-115888/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec47496b3888b8aa93cd4b19682ea4c5daddf9fe0b5f1ccb067dbfcb9f814628
3
+ size 14244
qwen3-150m-multilingual-checkpoints/checkpoint-115888/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c69aebfdbbadf76e645b9d4bb873e795d7760902add78b6ab52d068dbcadbde1
3
+ size 1064
qwen3-150m-multilingual-checkpoints/checkpoint-115888/special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "<pad>",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-115888/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3-150m-multilingual-checkpoints/checkpoint-115888/tokenizer_config.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "<unk>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<pad>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "2": {
20
+ "content": "<s>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ },
27
+ "3": {
28
+ "content": "</s>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false,
33
+ "special": true
34
+ },
35
+ "4": {
36
+ "content": "<en>",
37
+ "lstrip": false,
38
+ "normalized": false,
39
+ "rstrip": false,
40
+ "single_word": false,
41
+ "special": true
42
+ },
43
+ "5": {
44
+ "content": "<te>",
45
+ "lstrip": false,
46
+ "normalized": false,
47
+ "rstrip": false,
48
+ "single_word": false,
49
+ "special": true
50
+ },
51
+ "6": {
52
+ "content": "<mai>",
53
+ "lstrip": false,
54
+ "normalized": false,
55
+ "rstrip": false,
56
+ "single_word": false,
57
+ "special": true
58
+ },
59
+ "7": {
60
+ "content": "<mask>",
61
+ "lstrip": false,
62
+ "normalized": false,
63
+ "rstrip": false,
64
+ "single_word": false,
65
+ "special": true
66
+ }
67
+ },
68
+ "bos_token": "<s>",
69
+ "clean_up_tokenization_spaces": false,
70
+ "eos_token": "</s>",
71
+ "extra_special_tokens": {},
72
+ "model_max_length": 1000000000000000019884624838656,
73
+ "pad_token": "<pad>",
74
+ "tokenizer_class": "PreTrainedTokenizerFast",
75
+ "unk_token": "<unk>"
76
+ }
qwen3-150m-multilingual-checkpoints/checkpoint-115888/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
qwen3-150m-multilingual-checkpoints/checkpoint-115888/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45400e5719e350eae239c5ac1f19604c105bbfee7f1eb21d9e7308c70716fc5f
3
+ size 5304