Hugging Face's logo

ewre324
/
ewre324-R1-SmolLM2-135M-Distill

TensorBoard
Safetensors
llama
Model card Files Metrics Community
ewre324 commited on
Commit
5b1a49d
·
verified ·
1 Parent(s): c279528

Upload folder using huggingface_hub

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. checkpoint-1000/config.json +39 -0
  2. checkpoint-1000/generation_config.json +7 -0
  3. checkpoint-1000/merges.txt +0 -0
  4. checkpoint-1000/model.safetensors +3 -0
  5. checkpoint-1000/optimizer.pt +3 -0
  6. checkpoint-1000/rng_state.pth +3 -0
  7. checkpoint-1000/scheduler.pt +3 -0
  8. checkpoint-1000/special_tokens_map.json +28 -0
  9. checkpoint-1000/tokenizer.json +0 -0
  10. checkpoint-1000/tokenizer_config.json +162 -0
  11. checkpoint-1000/trainer_state.json +1473 -0
  12. checkpoint-1000/training_args.bin +3 -0
  13. checkpoint-1000/vocab.json +0 -0
  14. checkpoint-1500/config.json +39 -0
  15. checkpoint-1500/generation_config.json +7 -0
  16. checkpoint-1500/merges.txt +0 -0
  17. checkpoint-1500/model.safetensors +3 -0
  18. checkpoint-1500/optimizer.pt +3 -0
  19. checkpoint-1500/rng_state.pth +3 -0
  20. checkpoint-1500/scheduler.pt +3 -0
  21. checkpoint-1500/special_tokens_map.json +28 -0
  22. checkpoint-1500/tokenizer.json +0 -0
  23. checkpoint-1500/tokenizer_config.json +162 -0
  24. checkpoint-1500/trainer_state.json +2189 -0
  25. checkpoint-1500/training_args.bin +3 -0
  26. checkpoint-1500/vocab.json +0 -0
  27. checkpoint-2000/config.json +39 -0
  28. checkpoint-2000/generation_config.json +7 -0
  29. checkpoint-2000/merges.txt +0 -0
  30. checkpoint-2000/model.safetensors +3 -0
  31. checkpoint-2000/optimizer.pt +3 -0
  32. checkpoint-2000/rng_state.pth +3 -0
  33. checkpoint-2000/scheduler.pt +3 -0
  34. checkpoint-2000/special_tokens_map.json +28 -0
  35. checkpoint-2000/tokenizer.json +0 -0
  36. checkpoint-2000/tokenizer_config.json +162 -0
  37. checkpoint-2000/trainer_state.json +2913 -0
  38. checkpoint-2000/training_args.bin +3 -0
  39. checkpoint-2000/vocab.json +0 -0
  40. checkpoint-2500/config.json +39 -0
  41. checkpoint-2500/generation_config.json +7 -0
  42. checkpoint-2500/merges.txt +0 -0
  43. checkpoint-2500/model.safetensors +3 -0
  44. checkpoint-2500/optimizer.pt +3 -0
  45. checkpoint-2500/rng_state.pth +3 -0
  46. checkpoint-2500/scheduler.pt +3 -0
  47. checkpoint-2500/special_tokens_map.json +28 -0
  48. checkpoint-2500/tokenizer.json +0 -0
  49. checkpoint-2500/tokenizer_config.json +162 -0
  50. checkpoint-2500/trainer_state.json +3629 -0
checkpoint-1000/config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "ewre324/ewre324-Thinker-SmolLM2-135M-Instruct-Reasoning",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "head_dim": 64,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 576,
13
+ "initializer_range": 0.041666666666666664,
14
+ "intermediate_size": 1536,
15
+ "is_llama_config": true,
16
+ "max_position_embeddings": 8192,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "num_attention_heads": 9,
20
+ "num_hidden_layers": 30,
21
+ "num_key_value_heads": 3,
22
+ "pad_token_id": 2,
23
+ "pretraining_tp": 1,
24
+ "rms_norm_eps": 1e-05,
25
+ "rope_interleaved": false,
26
+ "rope_scaling": null,
27
+ "rope_theta": 100000,
28
+ "tie_word_embeddings": true,
29
+ "torch_dtype": "float32",
30
+ "transformers.js_config": {
31
+ "kv_cache_dtype": {
32
+ "fp16": "float16",
33
+ "q4f16": "float16"
34
+ }
35
+ },
36
+ "transformers_version": "4.49.0.dev0",
37
+ "use_cache": false,
38
+ "vocab_size": 49152
39
+ }
checkpoint-1000/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 2,
6
+ "transformers_version": "4.49.0.dev0"
7
+ }
checkpoint-1000/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-1000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9af5d173e7564cdda57bbecc1ee1d1373fe60eea67539113ad1a474bbdf36ffe
3
+ size 538090408
checkpoint-1000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e399725b7cd89d340045b61df439892bc6485fe485d7419b3333b659fce20330
3
+ size 1076349050
checkpoint-1000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c062f7f375beded48b5337f5a3f3a5cb38807fa3e85dbf3e294c0ab6b627bfc2
3
+ size 14244
checkpoint-1000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:753b60ceb48341188690448f803646e87b0db50bdbf0da140937469ffbf4f610
3
+ size 1064
checkpoint-1000/special_tokens_map.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>"
5
+ ],
6
+ "bos_token": {
7
+ "content": "<|im_start|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false
12
+ },
13
+ "eos_token": {
14
+ "content": "<|im_end|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ },
20
+ "pad_token": "<|im_end|>",
21
+ "unk_token": {
22
+ "content": "<|endoftext|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false
27
+ }
28
+ }
checkpoint-1000/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-1000/tokenizer_config.json ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "0": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "1": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "2": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "3": {
29
+ "content": "<repo_name>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "4": {
37
+ "content": "<reponame>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "5": {
45
+ "content": "<file_sep>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "6": {
53
+ "content": "<filename>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "7": {
61
+ "content": "<gh_stars>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "8": {
69
+ "content": "<issue_start>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "9": {
77
+ "content": "<issue_comment>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "10": {
85
+ "content": "<issue_closed>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "11": {
93
+ "content": "<jupyter_start>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "12": {
101
+ "content": "<jupyter_text>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "13": {
109
+ "content": "<jupyter_code>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "14": {
117
+ "content": "<jupyter_output>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": true
123
+ },
124
+ "15": {
125
+ "content": "<jupyter_script>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": true
131
+ },
132
+ "16": {
133
+ "content": "<empty_output>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": true
139
+ }
140
+ },
141
+ "additional_special_tokens": [
142
+ "<|im_start|>",
143
+ "<|im_end|>"
144
+ ],
145
+ "bos_token": "<|im_start|>",
146
+ "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
147
+ "clean_up_tokenization_spaces": false,
148
+ "eos_token": "<|im_end|>",
149
+ "extra_special_tokens": {},
150
+ "max_length": 2048,
151
+ "model_max_length": 2048,
152
+ "pad_to_multiple_of": null,
153
+ "pad_token": "<|im_end|>",
154
+ "pad_token_type_id": 0,
155
+ "padding_side": "right",
156
+ "stride": 0,
157
+ "tokenizer_class": "GPT2Tokenizer",
158
+ "truncation_side": "right",
159
+ "truncation_strategy": "longest_first",
160
+ "unk_token": "<|endoftext|>",
161
+ "vocab_size": 49152
162
+ }
checkpoint-1000/trainer_state.json ADDED
@@ -0,0 +1,1473 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.2408187838651415,
5
+ "eval_steps": 200,
6
+ "global_step": 1000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0012040939193257074,
13
+ "grad_norm": 1.2087849378585815,
14
+ "learning_rate": 1.9975915221579964e-05,
15
+ "loss": 1.7501,
16
+ "step": 5
17
+ },
18
+ {
19
+ "epoch": 0.002408187838651415,
20
+ "grad_norm": 1.0137544870376587,
21
+ "learning_rate": 1.9951830443159926e-05,
22
+ "loss": 1.3294,
23
+ "step": 10
24
+ },
25
+ {
26
+ "epoch": 0.003612281757977122,
27
+ "grad_norm": 0.8906710147857666,
28
+ "learning_rate": 1.9927745664739885e-05,
29
+ "loss": 1.4793,
30
+ "step": 15
31
+ },
32
+ {
33
+ "epoch": 0.00481637567730283,
34
+ "grad_norm": 0.8175792098045349,
35
+ "learning_rate": 1.9903660886319848e-05,
36
+ "loss": 1.4537,
37
+ "step": 20
38
+ },
39
+ {
40
+ "epoch": 0.006020469596628537,
41
+ "grad_norm": 0.8326422572135925,
42
+ "learning_rate": 1.987957610789981e-05,
43
+ "loss": 1.5579,
44
+ "step": 25
45
+ },
46
+ {
47
+ "epoch": 0.007224563515954244,
48
+ "grad_norm": 0.8402993083000183,
49
+ "learning_rate": 1.985549132947977e-05,
50
+ "loss": 1.4146,
51
+ "step": 30
52
+ },
53
+ {
54
+ "epoch": 0.008428657435279952,
55
+ "grad_norm": 0.9211377501487732,
56
+ "learning_rate": 1.983140655105973e-05,
57
+ "loss": 1.4986,
58
+ "step": 35
59
+ },
60
+ {
61
+ "epoch": 0.00963275135460566,
62
+ "grad_norm": 0.7729477286338806,
63
+ "learning_rate": 1.9807321772639694e-05,
64
+ "loss": 1.3949,
65
+ "step": 40
66
+ },
67
+ {
68
+ "epoch": 0.010836845273931367,
69
+ "grad_norm": 0.8256559371948242,
70
+ "learning_rate": 1.9783236994219656e-05,
71
+ "loss": 1.2517,
72
+ "step": 45
73
+ },
74
+ {
75
+ "epoch": 0.012040939193257074,
76
+ "grad_norm": 0.7404232025146484,
77
+ "learning_rate": 1.9759152215799615e-05,
78
+ "loss": 1.4761,
79
+ "step": 50
80
+ },
81
+ {
82
+ "epoch": 0.013245033112582781,
83
+ "grad_norm": 0.773309588432312,
84
+ "learning_rate": 1.9735067437379577e-05,
85
+ "loss": 1.227,
86
+ "step": 55
87
+ },
88
+ {
89
+ "epoch": 0.014449127031908489,
90
+ "grad_norm": 0.7535562515258789,
91
+ "learning_rate": 1.971098265895954e-05,
92
+ "loss": 1.3443,
93
+ "step": 60
94
+ },
95
+ {
96
+ "epoch": 0.015653220951234198,
97
+ "grad_norm": 0.8561655282974243,
98
+ "learning_rate": 1.9686897880539502e-05,
99
+ "loss": 1.3229,
100
+ "step": 65
101
+ },
102
+ {
103
+ "epoch": 0.016857314870559904,
104
+ "grad_norm": 0.7888396978378296,
105
+ "learning_rate": 1.966281310211946e-05,
106
+ "loss": 1.4274,
107
+ "step": 70
108
+ },
109
+ {
110
+ "epoch": 0.018061408789885613,
111
+ "grad_norm": 0.7886033654212952,
112
+ "learning_rate": 1.9638728323699423e-05,
113
+ "loss": 1.4469,
114
+ "step": 75
115
+ },
116
+ {
117
+ "epoch": 0.01926550270921132,
118
+ "grad_norm": 0.8100034594535828,
119
+ "learning_rate": 1.9614643545279385e-05,
120
+ "loss": 1.314,
121
+ "step": 80
122
+ },
123
+ {
124
+ "epoch": 0.020469596628537028,
125
+ "grad_norm": 0.7786366939544678,
126
+ "learning_rate": 1.9590558766859348e-05,
127
+ "loss": 1.2631,
128
+ "step": 85
129
+ },
130
+ {
131
+ "epoch": 0.021673690547862733,
132
+ "grad_norm": 0.8259329795837402,
133
+ "learning_rate": 1.9566473988439307e-05,
134
+ "loss": 1.4006,
135
+ "step": 90
136
+ },
137
+ {
138
+ "epoch": 0.022877784467188442,
139
+ "grad_norm": 0.7402159571647644,
140
+ "learning_rate": 1.954238921001927e-05,
141
+ "loss": 1.416,
142
+ "step": 95
143
+ },
144
+ {
145
+ "epoch": 0.024081878386514148,
146
+ "grad_norm": 0.8619882464408875,
147
+ "learning_rate": 1.951830443159923e-05,
148
+ "loss": 1.2583,
149
+ "step": 100
150
+ },
151
+ {
152
+ "epoch": 0.025285972305839857,
153
+ "grad_norm": 0.7436647415161133,
154
+ "learning_rate": 1.949421965317919e-05,
155
+ "loss": 1.3527,
156
+ "step": 105
157
+ },
158
+ {
159
+ "epoch": 0.026490066225165563,
160
+ "grad_norm": 0.7261799573898315,
161
+ "learning_rate": 1.9470134874759156e-05,
162
+ "loss": 1.4234,
163
+ "step": 110
164
+ },
165
+ {
166
+ "epoch": 0.027694160144491272,
167
+ "grad_norm": 0.8648309707641602,
168
+ "learning_rate": 1.9446050096339115e-05,
169
+ "loss": 1.2541,
170
+ "step": 115
171
+ },
172
+ {
173
+ "epoch": 0.028898254063816978,
174
+ "grad_norm": 0.7091889381408691,
175
+ "learning_rate": 1.9421965317919077e-05,
176
+ "loss": 1.2473,
177
+ "step": 120
178
+ },
179
+ {
180
+ "epoch": 0.030102347983142687,
181
+ "grad_norm": 0.7450091242790222,
182
+ "learning_rate": 1.939788053949904e-05,
183
+ "loss": 1.2555,
184
+ "step": 125
185
+ },
186
+ {
187
+ "epoch": 0.031306441902468396,
188
+ "grad_norm": 0.8190895318984985,
189
+ "learning_rate": 1.9373795761079e-05,
190
+ "loss": 1.2874,
191
+ "step": 130
192
+ },
193
+ {
194
+ "epoch": 0.0325105358217941,
195
+ "grad_norm": 0.7458487749099731,
196
+ "learning_rate": 1.934971098265896e-05,
197
+ "loss": 1.4139,
198
+ "step": 135
199
+ },
200
+ {
201
+ "epoch": 0.03371462974111981,
202
+ "grad_norm": 0.806698203086853,
203
+ "learning_rate": 1.9325626204238923e-05,
204
+ "loss": 1.4061,
205
+ "step": 140
206
+ },
207
+ {
208
+ "epoch": 0.034918723660445516,
209
+ "grad_norm": 0.7941407561302185,
210
+ "learning_rate": 1.9301541425818882e-05,
211
+ "loss": 1.2819,
212
+ "step": 145
213
+ },
214
+ {
215
+ "epoch": 0.036122817579771226,
216
+ "grad_norm": 0.6824349164962769,
217
+ "learning_rate": 1.9277456647398845e-05,
218
+ "loss": 1.3025,
219
+ "step": 150
220
+ },
221
+ {
222
+ "epoch": 0.03732691149909693,
223
+ "grad_norm": 0.8056986927986145,
224
+ "learning_rate": 1.9253371868978807e-05,
225
+ "loss": 1.1617,
226
+ "step": 155
227
+ },
228
+ {
229
+ "epoch": 0.03853100541842264,
230
+ "grad_norm": 0.7486716508865356,
231
+ "learning_rate": 1.922928709055877e-05,
232
+ "loss": 1.3523,
233
+ "step": 160
234
+ },
235
+ {
236
+ "epoch": 0.039735099337748346,
237
+ "grad_norm": 0.7773532867431641,
238
+ "learning_rate": 1.920520231213873e-05,
239
+ "loss": 1.3013,
240
+ "step": 165
241
+ },
242
+ {
243
+ "epoch": 0.040939193257074055,
244
+ "grad_norm": 0.7498481869697571,
245
+ "learning_rate": 1.918111753371869e-05,
246
+ "loss": 1.2563,
247
+ "step": 170
248
+ },
249
+ {
250
+ "epoch": 0.04214328717639976,
251
+ "grad_norm": 0.7634468078613281,
252
+ "learning_rate": 1.9157032755298653e-05,
253
+ "loss": 1.2365,
254
+ "step": 175
255
+ },
256
+ {
257
+ "epoch": 0.04334738109572547,
258
+ "grad_norm": 0.6561328172683716,
259
+ "learning_rate": 1.9132947976878615e-05,
260
+ "loss": 1.3058,
261
+ "step": 180
262
+ },
263
+ {
264
+ "epoch": 0.044551475015051176,
265
+ "grad_norm": 0.7443628907203674,
266
+ "learning_rate": 1.9108863198458578e-05,
267
+ "loss": 1.3264,
268
+ "step": 185
269
+ },
270
+ {
271
+ "epoch": 0.045755568934376885,
272
+ "grad_norm": 0.6697310209274292,
273
+ "learning_rate": 1.9084778420038536e-05,
274
+ "loss": 1.1029,
275
+ "step": 190
276
+ },
277
+ {
278
+ "epoch": 0.04695966285370259,
279
+ "grad_norm": 0.7119889259338379,
280
+ "learning_rate": 1.90606936416185e-05,
281
+ "loss": 1.2503,
282
+ "step": 195
283
+ },
284
+ {
285
+ "epoch": 0.048163756773028296,
286
+ "grad_norm": 0.8120700120925903,
287
+ "learning_rate": 1.903660886319846e-05,
288
+ "loss": 1.246,
289
+ "step": 200
290
+ },
291
+ {
292
+ "epoch": 0.048163756773028296,
293
+ "eval_loss": 1.2944953441619873,
294
+ "eval_runtime": 10.5151,
295
+ "eval_samples_per_second": 9.51,
296
+ "eval_steps_per_second": 1.236,
297
+ "step": 200
298
+ },
299
+ {
300
+ "epoch": 0.049367850692354005,
301
+ "grad_norm": 0.7876719832420349,
302
+ "learning_rate": 1.901252408477842e-05,
303
+ "loss": 1.2097,
304
+ "step": 205
305
+ },
306
+ {
307
+ "epoch": 0.050571944611679714,
308
+ "grad_norm": 0.7367354035377502,
309
+ "learning_rate": 1.8988439306358382e-05,
310
+ "loss": 1.3332,
311
+ "step": 210
312
+ },
313
+ {
314
+ "epoch": 0.05177603853100542,
315
+ "grad_norm": 0.7384558320045471,
316
+ "learning_rate": 1.8964354527938345e-05,
317
+ "loss": 1.3897,
318
+ "step": 215
319
+ },
320
+ {
321
+ "epoch": 0.052980132450331126,
322
+ "grad_norm": 0.8261380195617676,
323
+ "learning_rate": 1.8940269749518304e-05,
324
+ "loss": 1.2677,
325
+ "step": 220
326
+ },
327
+ {
328
+ "epoch": 0.054184226369656835,
329
+ "grad_norm": 0.7237825989723206,
330
+ "learning_rate": 1.891618497109827e-05,
331
+ "loss": 1.2919,
332
+ "step": 225
333
+ },
334
+ {
335
+ "epoch": 0.055388320288982544,
336
+ "grad_norm": 0.7296762466430664,
337
+ "learning_rate": 1.889210019267823e-05,
338
+ "loss": 1.2977,
339
+ "step": 230
340
+ },
341
+ {
342
+ "epoch": 0.056592414208308246,
343
+ "grad_norm": 0.8224279880523682,
344
+ "learning_rate": 1.886801541425819e-05,
345
+ "loss": 1.3751,
346
+ "step": 235
347
+ },
348
+ {
349
+ "epoch": 0.057796508127633955,
350
+ "grad_norm": 0.7233418822288513,
351
+ "learning_rate": 1.8843930635838153e-05,
352
+ "loss": 1.293,
353
+ "step": 240
354
+ },
355
+ {
356
+ "epoch": 0.059000602046959665,
357
+ "grad_norm": 0.7015657424926758,
358
+ "learning_rate": 1.8819845857418112e-05,
359
+ "loss": 1.3093,
360
+ "step": 245
361
+ },
362
+ {
363
+ "epoch": 0.060204695966285374,
364
+ "grad_norm": 0.8122280240058899,
365
+ "learning_rate": 1.8795761078998074e-05,
366
+ "loss": 1.3979,
367
+ "step": 250
368
+ },
369
+ {
370
+ "epoch": 0.061408789885611076,
371
+ "grad_norm": 0.7582533955574036,
372
+ "learning_rate": 1.8771676300578037e-05,
373
+ "loss": 1.3147,
374
+ "step": 255
375
+ },
376
+ {
377
+ "epoch": 0.06261288380493679,
378
+ "grad_norm": 0.746967077255249,
379
+ "learning_rate": 1.8747591522158e-05,
380
+ "loss": 1.1785,
381
+ "step": 260
382
+ },
383
+ {
384
+ "epoch": 0.0638169777242625,
385
+ "grad_norm": 0.7588161826133728,
386
+ "learning_rate": 1.8723506743737958e-05,
387
+ "loss": 1.3648,
388
+ "step": 265
389
+ },
390
+ {
391
+ "epoch": 0.0650210716435882,
392
+ "grad_norm": 0.696869432926178,
393
+ "learning_rate": 1.869942196531792e-05,
394
+ "loss": 1.2854,
395
+ "step": 270
396
+ },
397
+ {
398
+ "epoch": 0.06622516556291391,
399
+ "grad_norm": 0.8297455906867981,
400
+ "learning_rate": 1.8675337186897883e-05,
401
+ "loss": 1.2011,
402
+ "step": 275
403
+ },
404
+ {
405
+ "epoch": 0.06742925948223961,
406
+ "grad_norm": 0.7183229327201843,
407
+ "learning_rate": 1.8651252408477845e-05,
408
+ "loss": 1.1944,
409
+ "step": 280
410
+ },
411
+ {
412
+ "epoch": 0.06863335340156532,
413
+ "grad_norm": 0.7092033624649048,
414
+ "learning_rate": 1.8627167630057804e-05,
415
+ "loss": 1.3296,
416
+ "step": 285
417
+ },
418
+ {
419
+ "epoch": 0.06983744732089103,
420
+ "grad_norm": 0.7024344801902771,
421
+ "learning_rate": 1.8603082851637766e-05,
422
+ "loss": 1.2349,
423
+ "step": 290
424
+ },
425
+ {
426
+ "epoch": 0.07104154124021674,
427
+ "grad_norm": 0.7504032254219055,
428
+ "learning_rate": 1.857899807321773e-05,
429
+ "loss": 1.1986,
430
+ "step": 295
431
+ },
432
+ {
433
+ "epoch": 0.07224563515954245,
434
+ "grad_norm": 0.7584034204483032,
435
+ "learning_rate": 1.855491329479769e-05,
436
+ "loss": 1.2998,
437
+ "step": 300
438
+ },
439
+ {
440
+ "epoch": 0.07344972907886815,
441
+ "grad_norm": 0.7563251852989197,
442
+ "learning_rate": 1.853082851637765e-05,
443
+ "loss": 1.3703,
444
+ "step": 305
445
+ },
446
+ {
447
+ "epoch": 0.07465382299819386,
448
+ "grad_norm": 0.728660523891449,
449
+ "learning_rate": 1.8506743737957612e-05,
450
+ "loss": 1.2916,
451
+ "step": 310
452
+ },
453
+ {
454
+ "epoch": 0.07585791691751957,
455
+ "grad_norm": 0.6596492528915405,
456
+ "learning_rate": 1.8482658959537575e-05,
457
+ "loss": 1.1821,
458
+ "step": 315
459
+ },
460
+ {
461
+ "epoch": 0.07706201083684527,
462
+ "grad_norm": 0.7137461304664612,
463
+ "learning_rate": 1.8458574181117533e-05,
464
+ "loss": 1.1982,
465
+ "step": 320
466
+ },
467
+ {
468
+ "epoch": 0.07826610475617098,
469
+ "grad_norm": 0.7411230206489563,
470
+ "learning_rate": 1.8434489402697496e-05,
471
+ "loss": 1.2711,
472
+ "step": 325
473
+ },
474
+ {
475
+ "epoch": 0.07947019867549669,
476
+ "grad_norm": 0.7566368579864502,
477
+ "learning_rate": 1.8410404624277458e-05,
478
+ "loss": 1.1603,
479
+ "step": 330
480
+ },
481
+ {
482
+ "epoch": 0.0806742925948224,
483
+ "grad_norm": 0.7235690355300903,
484
+ "learning_rate": 1.838631984585742e-05,
485
+ "loss": 1.1993,
486
+ "step": 335
487
+ },
488
+ {
489
+ "epoch": 0.08187838651414811,
490
+ "grad_norm": 0.7648451328277588,
491
+ "learning_rate": 1.8362235067437383e-05,
492
+ "loss": 1.134,
493
+ "step": 340
494
+ },
495
+ {
496
+ "epoch": 0.08308248043347381,
497
+ "grad_norm": 0.740710973739624,
498
+ "learning_rate": 1.8338150289017342e-05,
499
+ "loss": 1.1533,
500
+ "step": 345
501
+ },
502
+ {
503
+ "epoch": 0.08428657435279951,
504
+ "grad_norm": 0.7610028982162476,
505
+ "learning_rate": 1.8314065510597304e-05,
506
+ "loss": 1.3126,
507
+ "step": 350
508
+ },
509
+ {
510
+ "epoch": 0.08549066827212523,
511
+ "grad_norm": 0.6956422924995422,
512
+ "learning_rate": 1.8289980732177266e-05,
513
+ "loss": 1.2272,
514
+ "step": 355
515
+ },
516
+ {
517
+ "epoch": 0.08669476219145093,
518
+ "grad_norm": 0.6914700269699097,
519
+ "learning_rate": 1.8265895953757225e-05,
520
+ "loss": 1.3049,
521
+ "step": 360
522
+ },
523
+ {
524
+ "epoch": 0.08789885611077664,
525
+ "grad_norm": 0.7015173435211182,
526
+ "learning_rate": 1.8241811175337188e-05,
527
+ "loss": 1.3569,
528
+ "step": 365
529
+ },
530
+ {
531
+ "epoch": 0.08910295003010235,
532
+ "grad_norm": 0.6985088586807251,
533
+ "learning_rate": 1.821772639691715e-05,
534
+ "loss": 1.2307,
535
+ "step": 370
536
+ },
537
+ {
538
+ "epoch": 0.09030704394942805,
539
+ "grad_norm": 0.7313199043273926,
540
+ "learning_rate": 1.8193641618497112e-05,
541
+ "loss": 1.1783,
542
+ "step": 375
543
+ },
544
+ {
545
+ "epoch": 0.09151113786875377,
546
+ "grad_norm": 0.7657830715179443,
547
+ "learning_rate": 1.8169556840077075e-05,
548
+ "loss": 1.2228,
549
+ "step": 380
550
+ },
551
+ {
552
+ "epoch": 0.09271523178807947,
553
+ "grad_norm": 0.8051377534866333,
554
+ "learning_rate": 1.8145472061657034e-05,
555
+ "loss": 1.25,
556
+ "step": 385
557
+ },
558
+ {
559
+ "epoch": 0.09391932570740517,
560
+ "grad_norm": 0.8114392161369324,
561
+ "learning_rate": 1.8121387283236996e-05,
562
+ "loss": 1.2347,
563
+ "step": 390
564
+ },
565
+ {
566
+ "epoch": 0.09512341962673089,
567
+ "grad_norm": 0.7304120659828186,
568
+ "learning_rate": 1.809730250481696e-05,
569
+ "loss": 1.2437,
570
+ "step": 395
571
+ },
572
+ {
573
+ "epoch": 0.09632751354605659,
574
+ "grad_norm": 0.8418581485748291,
575
+ "learning_rate": 1.8073217726396917e-05,
576
+ "loss": 1.3255,
577
+ "step": 400
578
+ },
579
+ {
580
+ "epoch": 0.09632751354605659,
581
+ "eval_loss": 1.255526065826416,
582
+ "eval_runtime": 10.4675,
583
+ "eval_samples_per_second": 9.553,
584
+ "eval_steps_per_second": 1.242,
585
+ "step": 400
586
+ },
587
+ {
588
+ "epoch": 0.0975316074653823,
589
+ "grad_norm": 0.7186573147773743,
590
+ "learning_rate": 1.804913294797688e-05,
591
+ "loss": 1.2859,
592
+ "step": 405
593
+ },
594
+ {
595
+ "epoch": 0.09873570138470801,
596
+ "grad_norm": 0.7544155716896057,
597
+ "learning_rate": 1.8025048169556842e-05,
598
+ "loss": 1.2771,
599
+ "step": 410
600
+ },
601
+ {
602
+ "epoch": 0.09993979530403371,
603
+ "grad_norm": 0.700529158115387,
604
+ "learning_rate": 1.8000963391136804e-05,
605
+ "loss": 1.2696,
606
+ "step": 415
607
+ },
608
+ {
609
+ "epoch": 0.10114388922335943,
610
+ "grad_norm": 0.6894243955612183,
611
+ "learning_rate": 1.7976878612716763e-05,
612
+ "loss": 1.2556,
613
+ "step": 420
614
+ },
615
+ {
616
+ "epoch": 0.10234798314268513,
617
+ "grad_norm": 0.8769268989562988,
618
+ "learning_rate": 1.7952793834296726e-05,
619
+ "loss": 1.2832,
620
+ "step": 425
621
+ },
622
+ {
623
+ "epoch": 0.10355207706201083,
624
+ "grad_norm": 0.7283303141593933,
625
+ "learning_rate": 1.7928709055876688e-05,
626
+ "loss": 1.3163,
627
+ "step": 430
628
+ },
629
+ {
630
+ "epoch": 0.10475617098133655,
631
+ "grad_norm": 0.6984531283378601,
632
+ "learning_rate": 1.7904624277456647e-05,
633
+ "loss": 1.1873,
634
+ "step": 435
635
+ },
636
+ {
637
+ "epoch": 0.10596026490066225,
638
+ "grad_norm": 0.7543785572052002,
639
+ "learning_rate": 1.7880539499036613e-05,
640
+ "loss": 1.2617,
641
+ "step": 440
642
+ },
643
+ {
644
+ "epoch": 0.10716435881998795,
645
+ "grad_norm": 0.7481861710548401,
646
+ "learning_rate": 1.785645472061657e-05,
647
+ "loss": 1.1792,
648
+ "step": 445
649
+ },
650
+ {
651
+ "epoch": 0.10836845273931367,
652
+ "grad_norm": 0.7607905268669128,
653
+ "learning_rate": 1.7832369942196534e-05,
654
+ "loss": 1.1987,
655
+ "step": 450
656
+ },
657
+ {
658
+ "epoch": 0.10957254665863937,
659
+ "grad_norm": 0.7427639365196228,
660
+ "learning_rate": 1.7808285163776496e-05,
661
+ "loss": 1.2142,
662
+ "step": 455
663
+ },
664
+ {
665
+ "epoch": 0.11077664057796509,
666
+ "grad_norm": 0.7348979711532593,
667
+ "learning_rate": 1.7784200385356455e-05,
668
+ "loss": 1.2727,
669
+ "step": 460
670
+ },
671
+ {
672
+ "epoch": 0.11198073449729079,
673
+ "grad_norm": 0.7192914485931396,
674
+ "learning_rate": 1.7760115606936417e-05,
675
+ "loss": 1.2593,
676
+ "step": 465
677
+ },
678
+ {
679
+ "epoch": 0.11318482841661649,
680
+ "grad_norm": 0.6672971248626709,
681
+ "learning_rate": 1.773603082851638e-05,
682
+ "loss": 1.1905,
683
+ "step": 470
684
+ },
685
+ {
686
+ "epoch": 0.11438892233594221,
687
+ "grad_norm": 0.7585046291351318,
688
+ "learning_rate": 1.771194605009634e-05,
689
+ "loss": 1.3045,
690
+ "step": 475
691
+ },
692
+ {
693
+ "epoch": 0.11559301625526791,
694
+ "grad_norm": 0.7200559973716736,
695
+ "learning_rate": 1.76878612716763e-05,
696
+ "loss": 1.1416,
697
+ "step": 480
698
+ },
699
+ {
700
+ "epoch": 0.11679711017459361,
701
+ "grad_norm": 0.7220486998558044,
702
+ "learning_rate": 1.7663776493256263e-05,
703
+ "loss": 1.2749,
704
+ "step": 485
705
+ },
706
+ {
707
+ "epoch": 0.11800120409391933,
708
+ "grad_norm": 0.750998318195343,
709
+ "learning_rate": 1.7639691714836226e-05,
710
+ "loss": 1.2823,
711
+ "step": 490
712
+ },
713
+ {
714
+ "epoch": 0.11920529801324503,
715
+ "grad_norm": 0.705752432346344,
716
+ "learning_rate": 1.7615606936416188e-05,
717
+ "loss": 1.1662,
718
+ "step": 495
719
+ },
720
+ {
721
+ "epoch": 0.12040939193257075,
722
+ "grad_norm": 0.7431082129478455,
723
+ "learning_rate": 1.7591522157996147e-05,
724
+ "loss": 1.203,
725
+ "step": 500
726
+ },
727
+ {
728
+ "epoch": 0.12161348585189645,
729
+ "grad_norm": 0.7106168866157532,
730
+ "learning_rate": 1.756743737957611e-05,
731
+ "loss": 1.162,
732
+ "step": 505
733
+ },
734
+ {
735
+ "epoch": 0.12281757977122215,
736
+ "grad_norm": 0.7027086615562439,
737
+ "learning_rate": 1.754335260115607e-05,
738
+ "loss": 1.2636,
739
+ "step": 510
740
+ },
741
+ {
742
+ "epoch": 0.12402167369054787,
743
+ "grad_norm": 0.774591863155365,
744
+ "learning_rate": 1.7519267822736034e-05,
745
+ "loss": 1.2517,
746
+ "step": 515
747
+ },
748
+ {
749
+ "epoch": 0.12522576760987358,
750
+ "grad_norm": 0.6927476525306702,
751
+ "learning_rate": 1.7495183044315993e-05,
752
+ "loss": 1.1396,
753
+ "step": 520
754
+ },
755
+ {
756
+ "epoch": 0.12642986152919927,
757
+ "grad_norm": 0.6912556886672974,
758
+ "learning_rate": 1.7471098265895955e-05,
759
+ "loss": 1.1658,
760
+ "step": 525
761
+ },
762
+ {
763
+ "epoch": 0.127633955448525,
764
+ "grad_norm": 0.6708123683929443,
765
+ "learning_rate": 1.7447013487475918e-05,
766
+ "loss": 1.1288,
767
+ "step": 530
768
+ },
769
+ {
770
+ "epoch": 0.1288380493678507,
771
+ "grad_norm": 0.6942213177680969,
772
+ "learning_rate": 1.7422928709055877e-05,
773
+ "loss": 1.1174,
774
+ "step": 535
775
+ },
776
+ {
777
+ "epoch": 0.1300421432871764,
778
+ "grad_norm": 0.7080583572387695,
779
+ "learning_rate": 1.739884393063584e-05,
780
+ "loss": 1.2279,
781
+ "step": 540
782
+ },
783
+ {
784
+ "epoch": 0.1312462372065021,
785
+ "grad_norm": 0.8131110668182373,
786
+ "learning_rate": 1.73747591522158e-05,
787
+ "loss": 1.2525,
788
+ "step": 545
789
+ },
790
+ {
791
+ "epoch": 0.13245033112582782,
792
+ "grad_norm": 0.7450209259986877,
793
+ "learning_rate": 1.735067437379576e-05,
794
+ "loss": 1.1506,
795
+ "step": 550
796
+ },
797
+ {
798
+ "epoch": 0.1336544250451535,
799
+ "grad_norm": 0.7066523432731628,
800
+ "learning_rate": 1.7326589595375726e-05,
801
+ "loss": 1.2383,
802
+ "step": 555
803
+ },
804
+ {
805
+ "epoch": 0.13485851896447923,
806
+ "grad_norm": 0.7041836977005005,
807
+ "learning_rate": 1.7302504816955685e-05,
808
+ "loss": 1.1911,
809
+ "step": 560
810
+ },
811
+ {
812
+ "epoch": 0.13606261288380495,
813
+ "grad_norm": 0.672095000743866,
814
+ "learning_rate": 1.7278420038535647e-05,
815
+ "loss": 1.1839,
816
+ "step": 565
817
+ },
818
+ {
819
+ "epoch": 0.13726670680313063,
820
+ "grad_norm": 0.7928294539451599,
821
+ "learning_rate": 1.725433526011561e-05,
822
+ "loss": 1.3347,
823
+ "step": 570
824
+ },
825
+ {
826
+ "epoch": 0.13847080072245635,
827
+ "grad_norm": 0.7465523481369019,
828
+ "learning_rate": 1.723025048169557e-05,
829
+ "loss": 1.134,
830
+ "step": 575
831
+ },
832
+ {
833
+ "epoch": 0.13967489464178207,
834
+ "grad_norm": 0.7117074131965637,
835
+ "learning_rate": 1.720616570327553e-05,
836
+ "loss": 1.3538,
837
+ "step": 580
838
+ },
839
+ {
840
+ "epoch": 0.14087898856110775,
841
+ "grad_norm": 0.7291346192359924,
842
+ "learning_rate": 1.7182080924855493e-05,
843
+ "loss": 1.1339,
844
+ "step": 585
845
+ },
846
+ {
847
+ "epoch": 0.14208308248043347,
848
+ "grad_norm": 0.7633514404296875,
849
+ "learning_rate": 1.7157996146435455e-05,
850
+ "loss": 1.2007,
851
+ "step": 590
852
+ },
853
+ {
854
+ "epoch": 0.1432871763997592,
855
+ "grad_norm": 0.7896651029586792,
856
+ "learning_rate": 1.7133911368015418e-05,
857
+ "loss": 1.2656,
858
+ "step": 595
859
+ },
860
+ {
861
+ "epoch": 0.1444912703190849,
862
+ "grad_norm": 0.7436801791191101,
863
+ "learning_rate": 1.7109826589595377e-05,
864
+ "loss": 1.28,
865
+ "step": 600
866
+ },
867
+ {
868
+ "epoch": 0.1444912703190849,
869
+ "eval_loss": 1.23267662525177,
870
+ "eval_runtime": 10.4568,
871
+ "eval_samples_per_second": 9.563,
872
+ "eval_steps_per_second": 1.243,
873
+ "step": 600
874
+ },
875
+ {
876
+ "epoch": 0.1456953642384106,
877
+ "grad_norm": 0.7123108506202698,
878
+ "learning_rate": 1.708574181117534e-05,
879
+ "loss": 1.1687,
880
+ "step": 605
881
+ },
882
+ {
883
+ "epoch": 0.1468994581577363,
884
+ "grad_norm": 0.7362635731697083,
885
+ "learning_rate": 1.70616570327553e-05,
886
+ "loss": 1.2471,
887
+ "step": 610
888
+ },
889
+ {
890
+ "epoch": 0.14810355207706202,
891
+ "grad_norm": 0.6770139336585999,
892
+ "learning_rate": 1.703757225433526e-05,
893
+ "loss": 1.149,
894
+ "step": 615
895
+ },
896
+ {
897
+ "epoch": 0.1493076459963877,
898
+ "grad_norm": 0.7292876243591309,
899
+ "learning_rate": 1.7013487475915223e-05,
900
+ "loss": 1.2957,
901
+ "step": 620
902
+ },
903
+ {
904
+ "epoch": 0.15051173991571343,
905
+ "grad_norm": 0.7105087637901306,
906
+ "learning_rate": 1.6989402697495185e-05,
907
+ "loss": 1.1462,
908
+ "step": 625
909
+ },
910
+ {
911
+ "epoch": 0.15171583383503914,
912
+ "grad_norm": 0.7218934297561646,
913
+ "learning_rate": 1.6965317919075147e-05,
914
+ "loss": 1.2181,
915
+ "step": 630
916
+ },
917
+ {
918
+ "epoch": 0.15291992775436483,
919
+ "grad_norm": 0.7794796824455261,
920
+ "learning_rate": 1.6941233140655106e-05,
921
+ "loss": 1.174,
922
+ "step": 635
923
+ },
924
+ {
925
+ "epoch": 0.15412402167369055,
926
+ "grad_norm": 0.6424360871315002,
927
+ "learning_rate": 1.691714836223507e-05,
928
+ "loss": 1.2149,
929
+ "step": 640
930
+ },
931
+ {
932
+ "epoch": 0.15532811559301626,
933
+ "grad_norm": 0.7393337488174438,
934
+ "learning_rate": 1.689306358381503e-05,
935
+ "loss": 1.2259,
936
+ "step": 645
937
+ },
938
+ {
939
+ "epoch": 0.15653220951234195,
940
+ "grad_norm": 0.7363811731338501,
941
+ "learning_rate": 1.686897880539499e-05,
942
+ "loss": 1.2087,
943
+ "step": 650
944
+ },
945
+ {
946
+ "epoch": 0.15773630343166767,
947
+ "grad_norm": 0.7336999773979187,
948
+ "learning_rate": 1.6844894026974952e-05,
949
+ "loss": 1.2526,
950
+ "step": 655
951
+ },
952
+ {
953
+ "epoch": 0.15894039735099338,
954
+ "grad_norm": 0.7329117655754089,
955
+ "learning_rate": 1.6820809248554915e-05,
956
+ "loss": 1.233,
957
+ "step": 660
958
+ },
959
+ {
960
+ "epoch": 0.16014449127031907,
961
+ "grad_norm": 0.7724847197532654,
962
+ "learning_rate": 1.6796724470134877e-05,
963
+ "loss": 1.2506,
964
+ "step": 665
965
+ },
966
+ {
967
+ "epoch": 0.1613485851896448,
968
+ "grad_norm": 0.7213071584701538,
969
+ "learning_rate": 1.677263969171484e-05,
970
+ "loss": 1.2021,
971
+ "step": 670
972
+ },
973
+ {
974
+ "epoch": 0.1625526791089705,
975
+ "grad_norm": 0.6397632360458374,
976
+ "learning_rate": 1.6748554913294798e-05,
977
+ "loss": 1.1909,
978
+ "step": 675
979
+ },
980
+ {
981
+ "epoch": 0.16375677302829622,
982
+ "grad_norm": 0.6840227842330933,
983
+ "learning_rate": 1.672447013487476e-05,
984
+ "loss": 1.1838,
985
+ "step": 680
986
+ },
987
+ {
988
+ "epoch": 0.1649608669476219,
989
+ "grad_norm": 0.7317889332771301,
990
+ "learning_rate": 1.6700385356454723e-05,
991
+ "loss": 1.169,
992
+ "step": 685
993
+ },
994
+ {
995
+ "epoch": 0.16616496086694763,
996
+ "grad_norm": 0.7134217619895935,
997
+ "learning_rate": 1.6676300578034682e-05,
998
+ "loss": 1.2711,
999
+ "step": 690
1000
+ },
1001
+ {
1002
+ "epoch": 0.16736905478627334,
1003
+ "grad_norm": 0.7525413036346436,
1004
+ "learning_rate": 1.6652215799614644e-05,
1005
+ "loss": 1.2183,
1006
+ "step": 695
1007
+ },
1008
+ {
1009
+ "epoch": 0.16857314870559903,
1010
+ "grad_norm": 0.7243443131446838,
1011
+ "learning_rate": 1.6628131021194607e-05,
1012
+ "loss": 1.2585,
1013
+ "step": 700
1014
+ },
1015
+ {
1016
+ "epoch": 0.16977724262492475,
1017
+ "grad_norm": 0.7554174661636353,
1018
+ "learning_rate": 1.660404624277457e-05,
1019
+ "loss": 1.1662,
1020
+ "step": 705
1021
+ },
1022
+ {
1023
+ "epoch": 0.17098133654425046,
1024
+ "grad_norm": 0.6472408175468445,
1025
+ "learning_rate": 1.657996146435453e-05,
1026
+ "loss": 1.1471,
1027
+ "step": 710
1028
+ },
1029
+ {
1030
+ "epoch": 0.17218543046357615,
1031
+ "grad_norm": 0.7514950037002563,
1032
+ "learning_rate": 1.655587668593449e-05,
1033
+ "loss": 1.157,
1034
+ "step": 715
1035
+ },
1036
+ {
1037
+ "epoch": 0.17338952438290187,
1038
+ "grad_norm": 0.6952060461044312,
1039
+ "learning_rate": 1.6531791907514452e-05,
1040
+ "loss": 1.2981,
1041
+ "step": 720
1042
+ },
1043
+ {
1044
+ "epoch": 0.17459361830222758,
1045
+ "grad_norm": 0.832417905330658,
1046
+ "learning_rate": 1.6507707129094415e-05,
1047
+ "loss": 1.2457,
1048
+ "step": 725
1049
+ },
1050
+ {
1051
+ "epoch": 0.17579771222155327,
1052
+ "grad_norm": 0.6825481653213501,
1053
+ "learning_rate": 1.6483622350674374e-05,
1054
+ "loss": 1.2632,
1055
+ "step": 730
1056
+ },
1057
+ {
1058
+ "epoch": 0.177001806140879,
1059
+ "grad_norm": 0.8267092704772949,
1060
+ "learning_rate": 1.6459537572254336e-05,
1061
+ "loss": 1.0368,
1062
+ "step": 735
1063
+ },
1064
+ {
1065
+ "epoch": 0.1782059000602047,
1066
+ "grad_norm": 0.7782288789749146,
1067
+ "learning_rate": 1.64354527938343e-05,
1068
+ "loss": 1.201,
1069
+ "step": 740
1070
+ },
1071
+ {
1072
+ "epoch": 0.1794099939795304,
1073
+ "grad_norm": 0.7548500299453735,
1074
+ "learning_rate": 1.641136801541426e-05,
1075
+ "loss": 1.1831,
1076
+ "step": 745
1077
+ },
1078
+ {
1079
+ "epoch": 0.1806140878988561,
1080
+ "grad_norm": 0.6344689726829529,
1081
+ "learning_rate": 1.638728323699422e-05,
1082
+ "loss": 1.1985,
1083
+ "step": 750
1084
+ },
1085
+ {
1086
+ "epoch": 0.18181818181818182,
1087
+ "grad_norm": 0.7326412796974182,
1088
+ "learning_rate": 1.6363198458574182e-05,
1089
+ "loss": 1.3051,
1090
+ "step": 755
1091
+ },
1092
+ {
1093
+ "epoch": 0.18302227573750754,
1094
+ "grad_norm": 0.6460621953010559,
1095
+ "learning_rate": 1.6339113680154144e-05,
1096
+ "loss": 1.1524,
1097
+ "step": 760
1098
+ },
1099
+ {
1100
+ "epoch": 0.18422636965683323,
1101
+ "grad_norm": 0.6921172142028809,
1102
+ "learning_rate": 1.6315028901734103e-05,
1103
+ "loss": 1.29,
1104
+ "step": 765
1105
+ },
1106
+ {
1107
+ "epoch": 0.18543046357615894,
1108
+ "grad_norm": 0.7872111201286316,
1109
+ "learning_rate": 1.629094412331407e-05,
1110
+ "loss": 1.1187,
1111
+ "step": 770
1112
+ },
1113
+ {
1114
+ "epoch": 0.18663455749548466,
1115
+ "grad_norm": 0.7098067402839661,
1116
+ "learning_rate": 1.6266859344894028e-05,
1117
+ "loss": 1.2006,
1118
+ "step": 775
1119
+ },
1120
+ {
1121
+ "epoch": 0.18783865141481035,
1122
+ "grad_norm": 0.7701388001441956,
1123
+ "learning_rate": 1.624277456647399e-05,
1124
+ "loss": 1.1609,
1125
+ "step": 780
1126
+ },
1127
+ {
1128
+ "epoch": 0.18904274533413606,
1129
+ "grad_norm": 0.7027246952056885,
1130
+ "learning_rate": 1.6218689788053953e-05,
1131
+ "loss": 1.2083,
1132
+ "step": 785
1133
+ },
1134
+ {
1135
+ "epoch": 0.19024683925346178,
1136
+ "grad_norm": 0.7161267995834351,
1137
+ "learning_rate": 1.619460500963391e-05,
1138
+ "loss": 1.131,
1139
+ "step": 790
1140
+ },
1141
+ {
1142
+ "epoch": 0.19145093317278747,
1143
+ "grad_norm": 0.8034841418266296,
1144
+ "learning_rate": 1.6170520231213874e-05,
1145
+ "loss": 1.2107,
1146
+ "step": 795
1147
+ },
1148
+ {
1149
+ "epoch": 0.19265502709211318,
1150
+ "grad_norm": 0.7797010540962219,
1151
+ "learning_rate": 1.6146435452793836e-05,
1152
+ "loss": 1.2931,
1153
+ "step": 800
1154
+ },
1155
+ {
1156
+ "epoch": 0.19265502709211318,
1157
+ "eval_loss": 1.2178008556365967,
1158
+ "eval_runtime": 10.4492,
1159
+ "eval_samples_per_second": 9.57,
1160
+ "eval_steps_per_second": 1.244,
1161
+ "step": 800
1162
+ },
1163
+ {
1164
+ "epoch": 0.1938591210114389,
1165
+ "grad_norm": 0.6586723923683167,
1166
+ "learning_rate": 1.6122350674373795e-05,
1167
+ "loss": 1.2007,
1168
+ "step": 805
1169
+ },
1170
+ {
1171
+ "epoch": 0.1950632149307646,
1172
+ "grad_norm": 0.7271411418914795,
1173
+ "learning_rate": 1.6098265895953758e-05,
1174
+ "loss": 1.1049,
1175
+ "step": 810
1176
+ },
1177
+ {
1178
+ "epoch": 0.1962673088500903,
1179
+ "grad_norm": 0.743730902671814,
1180
+ "learning_rate": 1.607418111753372e-05,
1181
+ "loss": 1.1815,
1182
+ "step": 815
1183
+ },
1184
+ {
1185
+ "epoch": 0.19747140276941602,
1186
+ "grad_norm": 0.7163010835647583,
1187
+ "learning_rate": 1.6050096339113682e-05,
1188
+ "loss": 1.1635,
1189
+ "step": 820
1190
+ },
1191
+ {
1192
+ "epoch": 0.1986754966887417,
1193
+ "grad_norm": 0.7846854329109192,
1194
+ "learning_rate": 1.6026011560693645e-05,
1195
+ "loss": 1.1576,
1196
+ "step": 825
1197
+ },
1198
+ {
1199
+ "epoch": 0.19987959060806743,
1200
+ "grad_norm": 0.755066990852356,
1201
+ "learning_rate": 1.6001926782273604e-05,
1202
+ "loss": 1.2626,
1203
+ "step": 830
1204
+ },
1205
+ {
1206
+ "epoch": 0.20108368452739314,
1207
+ "grad_norm": 0.6981719732284546,
1208
+ "learning_rate": 1.5977842003853566e-05,
1209
+ "loss": 1.1883,
1210
+ "step": 835
1211
+ },
1212
+ {
1213
+ "epoch": 0.20228777844671886,
1214
+ "grad_norm": 0.712697446346283,
1215
+ "learning_rate": 1.5953757225433528e-05,
1216
+ "loss": 1.1539,
1217
+ "step": 840
1218
+ },
1219
+ {
1220
+ "epoch": 0.20349187236604455,
1221
+ "grad_norm": 0.8146225810050964,
1222
+ "learning_rate": 1.592967244701349e-05,
1223
+ "loss": 1.161,
1224
+ "step": 845
1225
+ },
1226
+ {
1227
+ "epoch": 0.20469596628537026,
1228
+ "grad_norm": 0.7796756625175476,
1229
+ "learning_rate": 1.590558766859345e-05,
1230
+ "loss": 1.1536,
1231
+ "step": 850
1232
+ },
1233
+ {
1234
+ "epoch": 0.20590006020469598,
1235
+ "grad_norm": 0.7026313543319702,
1236
+ "learning_rate": 1.5881502890173412e-05,
1237
+ "loss": 1.2206,
1238
+ "step": 855
1239
+ },
1240
+ {
1241
+ "epoch": 0.20710415412402167,
1242
+ "grad_norm": 0.6918179988861084,
1243
+ "learning_rate": 1.5857418111753374e-05,
1244
+ "loss": 1.0569,
1245
+ "step": 860
1246
+ },
1247
+ {
1248
+ "epoch": 0.20830824804334738,
1249
+ "grad_norm": 0.7711547613143921,
1250
+ "learning_rate": 1.5833333333333333e-05,
1251
+ "loss": 1.1826,
1252
+ "step": 865
1253
+ },
1254
+ {
1255
+ "epoch": 0.2095123419626731,
1256
+ "grad_norm": 0.6604825854301453,
1257
+ "learning_rate": 1.5809248554913295e-05,
1258
+ "loss": 1.1743,
1259
+ "step": 870
1260
+ },
1261
+ {
1262
+ "epoch": 0.2107164358819988,
1263
+ "grad_norm": 0.7115917205810547,
1264
+ "learning_rate": 1.5785163776493258e-05,
1265
+ "loss": 1.2815,
1266
+ "step": 875
1267
+ },
1268
+ {
1269
+ "epoch": 0.2119205298013245,
1270
+ "grad_norm": 0.69574373960495,
1271
+ "learning_rate": 1.5761078998073217e-05,
1272
+ "loss": 1.2131,
1273
+ "step": 880
1274
+ },
1275
+ {
1276
+ "epoch": 0.21312462372065022,
1277
+ "grad_norm": 0.6801597476005554,
1278
+ "learning_rate": 1.5736994219653182e-05,
1279
+ "loss": 1.3022,
1280
+ "step": 885
1281
+ },
1282
+ {
1283
+ "epoch": 0.2143287176399759,
1284
+ "grad_norm": 0.7042235136032104,
1285
+ "learning_rate": 1.571290944123314e-05,
1286
+ "loss": 1.1936,
1287
+ "step": 890
1288
+ },
1289
+ {
1290
+ "epoch": 0.21553281155930162,
1291
+ "grad_norm": 0.6610032916069031,
1292
+ "learning_rate": 1.5688824662813104e-05,
1293
+ "loss": 1.1972,
1294
+ "step": 895
1295
+ },
1296
+ {
1297
+ "epoch": 0.21673690547862734,
1298
+ "grad_norm": 0.8218846321105957,
1299
+ "learning_rate": 1.5664739884393066e-05,
1300
+ "loss": 1.2951,
1301
+ "step": 900
1302
+ },
1303
+ {
1304
+ "epoch": 0.21794099939795303,
1305
+ "grad_norm": 0.7200918197631836,
1306
+ "learning_rate": 1.5640655105973025e-05,
1307
+ "loss": 1.1256,
1308
+ "step": 905
1309
+ },
1310
+ {
1311
+ "epoch": 0.21914509331727874,
1312
+ "grad_norm": 0.6200802326202393,
1313
+ "learning_rate": 1.5616570327552987e-05,
1314
+ "loss": 1.2071,
1315
+ "step": 910
1316
+ },
1317
+ {
1318
+ "epoch": 0.22034918723660446,
1319
+ "grad_norm": 0.7431606650352478,
1320
+ "learning_rate": 1.559248554913295e-05,
1321
+ "loss": 1.2569,
1322
+ "step": 915
1323
+ },
1324
+ {
1325
+ "epoch": 0.22155328115593018,
1326
+ "grad_norm": 0.802537202835083,
1327
+ "learning_rate": 1.5568400770712912e-05,
1328
+ "loss": 1.1342,
1329
+ "step": 920
1330
+ },
1331
+ {
1332
+ "epoch": 0.22275737507525586,
1333
+ "grad_norm": 0.7078563570976257,
1334
+ "learning_rate": 1.5544315992292874e-05,
1335
+ "loss": 1.1928,
1336
+ "step": 925
1337
+ },
1338
+ {
1339
+ "epoch": 0.22396146899458158,
1340
+ "grad_norm": 0.6626926064491272,
1341
+ "learning_rate": 1.5520231213872833e-05,
1342
+ "loss": 1.1295,
1343
+ "step": 930
1344
+ },
1345
+ {
1346
+ "epoch": 0.2251655629139073,
1347
+ "grad_norm": 0.664715588092804,
1348
+ "learning_rate": 1.5496146435452796e-05,
1349
+ "loss": 1.1616,
1350
+ "step": 935
1351
+ },
1352
+ {
1353
+ "epoch": 0.22636965683323299,
1354
+ "grad_norm": 0.7077795267105103,
1355
+ "learning_rate": 1.5472061657032758e-05,
1356
+ "loss": 1.24,
1357
+ "step": 940
1358
+ },
1359
+ {
1360
+ "epoch": 0.2275737507525587,
1361
+ "grad_norm": 0.6813915371894836,
1362
+ "learning_rate": 1.5447976878612717e-05,
1363
+ "loss": 1.1807,
1364
+ "step": 945
1365
+ },
1366
+ {
1367
+ "epoch": 0.22877784467188442,
1368
+ "grad_norm": 0.7245080471038818,
1369
+ "learning_rate": 1.542389210019268e-05,
1370
+ "loss": 1.1737,
1371
+ "step": 950
1372
+ },
1373
+ {
1374
+ "epoch": 0.2299819385912101,
1375
+ "grad_norm": 0.7851933836936951,
1376
+ "learning_rate": 1.539980732177264e-05,
1377
+ "loss": 1.0702,
1378
+ "step": 955
1379
+ },
1380
+ {
1381
+ "epoch": 0.23118603251053582,
1382
+ "grad_norm": 0.7153802514076233,
1383
+ "learning_rate": 1.5375722543352604e-05,
1384
+ "loss": 1.1302,
1385
+ "step": 960
1386
+ },
1387
+ {
1388
+ "epoch": 0.23239012642986154,
1389
+ "grad_norm": 0.8112071752548218,
1390
+ "learning_rate": 1.5351637764932563e-05,
1391
+ "loss": 1.1591,
1392
+ "step": 965
1393
+ },
1394
+ {
1395
+ "epoch": 0.23359422034918723,
1396
+ "grad_norm": 0.7842305898666382,
1397
+ "learning_rate": 1.5327552986512525e-05,
1398
+ "loss": 1.1695,
1399
+ "step": 970
1400
+ },
1401
+ {
1402
+ "epoch": 0.23479831426851294,
1403
+ "grad_norm": 0.6949911713600159,
1404
+ "learning_rate": 1.5303468208092487e-05,
1405
+ "loss": 1.1726,
1406
+ "step": 975
1407
+ },
1408
+ {
1409
+ "epoch": 0.23600240818783866,
1410
+ "grad_norm": 0.7541187405586243,
1411
+ "learning_rate": 1.5279383429672446e-05,
1412
+ "loss": 1.2209,
1413
+ "step": 980
1414
+ },
1415
+ {
1416
+ "epoch": 0.23720650210716435,
1417
+ "grad_norm": 0.7082781791687012,
1418
+ "learning_rate": 1.525529865125241e-05,
1419
+ "loss": 1.2685,
1420
+ "step": 985
1421
+ },
1422
+ {
1423
+ "epoch": 0.23841059602649006,
1424
+ "grad_norm": 0.6862682104110718,
1425
+ "learning_rate": 1.5231213872832371e-05,
1426
+ "loss": 1.1569,
1427
+ "step": 990
1428
+ },
1429
+ {
1430
+ "epoch": 0.23961468994581578,
1431
+ "grad_norm": 0.836452066898346,
1432
+ "learning_rate": 1.5207129094412332e-05,
1433
+ "loss": 1.2158,
1434
+ "step": 995
1435
+ },
1436
+ {
1437
+ "epoch": 0.2408187838651415,
1438
+ "grad_norm": 0.8637534379959106,
1439
+ "learning_rate": 1.5183044315992294e-05,
1440
+ "loss": 1.1891,
1441
+ "step": 1000
1442
+ },
1443
+ {
1444
+ "epoch": 0.2408187838651415,
1445
+ "eval_loss": 1.2066378593444824,
1446
+ "eval_runtime": 10.4478,
1447
+ "eval_samples_per_second": 9.571,
1448
+ "eval_steps_per_second": 1.244,
1449
+ "step": 1000
1450
+ }
1451
+ ],
1452
+ "logging_steps": 5,
1453
+ "max_steps": 4152,
1454
+ "num_input_tokens_seen": 0,
1455
+ "num_train_epochs": 1,
1456
+ "save_steps": 500,
1457
+ "stateful_callbacks": {
1458
+ "TrainerControl": {
1459
+ "args": {
1460
+ "should_epoch_stop": false,
1461
+ "should_evaluate": false,
1462
+ "should_log": false,
1463
+ "should_save": true,
1464
+ "should_training_stop": false
1465
+ },
1466
+ "attributes": {}
1467
+ }
1468
+ },
1469
+ "total_flos": 4883747019944832.0,
1470
+ "train_batch_size": 1,
1471
+ "trial_name": null,
1472
+ "trial_params": null
1473
+ }
checkpoint-1000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86602ecfa5f4389c74519a3cbca0206cb3e9a395b2bd494b341ef1183c4e9454
3
+ size 5816
checkpoint-1000/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-1500/config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "ewre324/ewre324-Thinker-SmolLM2-135M-Instruct-Reasoning",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "head_dim": 64,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 576,
13
+ "initializer_range": 0.041666666666666664,
14
+ "intermediate_size": 1536,
15
+ "is_llama_config": true,
16
+ "max_position_embeddings": 8192,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "num_attention_heads": 9,
20
+ "num_hidden_layers": 30,
21
+ "num_key_value_heads": 3,
22
+ "pad_token_id": 2,
23
+ "pretraining_tp": 1,
24
+ "rms_norm_eps": 1e-05,
25
+ "rope_interleaved": false,
26
+ "rope_scaling": null,
27
+ "rope_theta": 100000,
28
+ "tie_word_embeddings": true,
29
+ "torch_dtype": "float32",
30
+ "transformers.js_config": {
31
+ "kv_cache_dtype": {
32
+ "fp16": "float16",
33
+ "q4f16": "float16"
34
+ }
35
+ },
36
+ "transformers_version": "4.49.0.dev0",
37
+ "use_cache": false,
38
+ "vocab_size": 49152
39
+ }
checkpoint-1500/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 2,
6
+ "transformers_version": "4.49.0.dev0"
7
+ }
checkpoint-1500/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-1500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:314fd6cfdf450b337374590f29ef732d89e5a2f7c123c7c0db6274bcad1ad04c
3
+ size 538090408
checkpoint-1500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1efd974b5d20d8a02ff45412c5819d7d1ddd214a51d5a3c42a512d72fea842b5
3
+ size 1076349050
checkpoint-1500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9899ccda7f0d8d9511991180b93aab508ce6e8489de708c88ad1188e7e1d90d6
3
+ size 14244
checkpoint-1500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f7acaaba5bf80cb5d5e1a88c4be3ebca794cd4bacb62bf046245137790e9a740
3
+ size 1064
checkpoint-1500/special_tokens_map.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>"
5
+ ],
6
+ "bos_token": {
7
+ "content": "<|im_start|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false
12
+ },
13
+ "eos_token": {
14
+ "content": "<|im_end|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ },
20
+ "pad_token": "<|im_end|>",
21
+ "unk_token": {
22
+ "content": "<|endoftext|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false
27
+ }
28
+ }
checkpoint-1500/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-1500/tokenizer_config.json ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "0": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "1": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "2": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "3": {
29
+ "content": "<repo_name>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "4": {
37
+ "content": "<reponame>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "5": {
45
+ "content": "<file_sep>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "6": {
53
+ "content": "<filename>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "7": {
61
+ "content": "<gh_stars>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "8": {
69
+ "content": "<issue_start>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "9": {
77
+ "content": "<issue_comment>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "10": {
85
+ "content": "<issue_closed>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "11": {
93
+ "content": "<jupyter_start>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "12": {
101
+ "content": "<jupyter_text>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "13": {
109
+ "content": "<jupyter_code>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "14": {
117
+ "content": "<jupyter_output>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": true
123
+ },
124
+ "15": {
125
+ "content": "<jupyter_script>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": true
131
+ },
132
+ "16": {
133
+ "content": "<empty_output>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": true
139
+ }
140
+ },
141
+ "additional_special_tokens": [
142
+ "<|im_start|>",
143
+ "<|im_end|>"
144
+ ],
145
+ "bos_token": "<|im_start|>",
146
+ "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
147
+ "clean_up_tokenization_spaces": false,
148
+ "eos_token": "<|im_end|>",
149
+ "extra_special_tokens": {},
150
+ "max_length": 2048,
151
+ "model_max_length": 2048,
152
+ "pad_to_multiple_of": null,
153
+ "pad_token": "<|im_end|>",
154
+ "pad_token_type_id": 0,
155
+ "padding_side": "right",
156
+ "stride": 0,
157
+ "tokenizer_class": "GPT2Tokenizer",
158
+ "truncation_side": "right",
159
+ "truncation_strategy": "longest_first",
160
+ "unk_token": "<|endoftext|>",
161
+ "vocab_size": 49152
162
+ }
checkpoint-1500/trainer_state.json ADDED
@@ -0,0 +1,2189 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.3612281757977122,
5
+ "eval_steps": 200,
6
+ "global_step": 1500,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0012040939193257074,
13
+ "grad_norm": 1.2087849378585815,
14
+ "learning_rate": 1.9975915221579964e-05,
15
+ "loss": 1.7501,
16
+ "step": 5
17
+ },
18
+ {
19
+ "epoch": 0.002408187838651415,
20
+ "grad_norm": 1.0137544870376587,
21
+ "learning_rate": 1.9951830443159926e-05,
22
+ "loss": 1.3294,
23
+ "step": 10
24
+ },
25
+ {
26
+ "epoch": 0.003612281757977122,
27
+ "grad_norm": 0.8906710147857666,
28
+ "learning_rate": 1.9927745664739885e-05,
29
+ "loss": 1.4793,
30
+ "step": 15
31
+ },
32
+ {
33
+ "epoch": 0.00481637567730283,
34
+ "grad_norm": 0.8175792098045349,
35
+ "learning_rate": 1.9903660886319848e-05,
36
+ "loss": 1.4537,
37
+ "step": 20
38
+ },
39
+ {
40
+ "epoch": 0.006020469596628537,
41
+ "grad_norm": 0.8326422572135925,
42
+ "learning_rate": 1.987957610789981e-05,
43
+ "loss": 1.5579,
44
+ "step": 25
45
+ },
46
+ {
47
+ "epoch": 0.007224563515954244,
48
+ "grad_norm": 0.8402993083000183,
49
+ "learning_rate": 1.985549132947977e-05,
50
+ "loss": 1.4146,
51
+ "step": 30
52
+ },
53
+ {
54
+ "epoch": 0.008428657435279952,
55
+ "grad_norm": 0.9211377501487732,
56
+ "learning_rate": 1.983140655105973e-05,
57
+ "loss": 1.4986,
58
+ "step": 35
59
+ },
60
+ {
61
+ "epoch": 0.00963275135460566,
62
+ "grad_norm": 0.7729477286338806,
63
+ "learning_rate": 1.9807321772639694e-05,
64
+ "loss": 1.3949,
65
+ "step": 40
66
+ },
67
+ {
68
+ "epoch": 0.010836845273931367,
69
+ "grad_norm": 0.8256559371948242,
70
+ "learning_rate": 1.9783236994219656e-05,
71
+ "loss": 1.2517,
72
+ "step": 45
73
+ },
74
+ {
75
+ "epoch": 0.012040939193257074,
76
+ "grad_norm": 0.7404232025146484,
77
+ "learning_rate": 1.9759152215799615e-05,
78
+ "loss": 1.4761,
79
+ "step": 50
80
+ },
81
+ {
82
+ "epoch": 0.013245033112582781,
83
+ "grad_norm": 0.773309588432312,
84
+ "learning_rate": 1.9735067437379577e-05,
85
+ "loss": 1.227,
86
+ "step": 55
87
+ },
88
+ {
89
+ "epoch": 0.014449127031908489,
90
+ "grad_norm": 0.7535562515258789,
91
+ "learning_rate": 1.971098265895954e-05,
92
+ "loss": 1.3443,
93
+ "step": 60
94
+ },
95
+ {
96
+ "epoch": 0.015653220951234198,
97
+ "grad_norm": 0.8561655282974243,
98
+ "learning_rate": 1.9686897880539502e-05,
99
+ "loss": 1.3229,
100
+ "step": 65
101
+ },
102
+ {
103
+ "epoch": 0.016857314870559904,
104
+ "grad_norm": 0.7888396978378296,
105
+ "learning_rate": 1.966281310211946e-05,
106
+ "loss": 1.4274,
107
+ "step": 70
108
+ },
109
+ {
110
+ "epoch": 0.018061408789885613,
111
+ "grad_norm": 0.7886033654212952,
112
+ "learning_rate": 1.9638728323699423e-05,
113
+ "loss": 1.4469,
114
+ "step": 75
115
+ },
116
+ {
117
+ "epoch": 0.01926550270921132,
118
+ "grad_norm": 0.8100034594535828,
119
+ "learning_rate": 1.9614643545279385e-05,
120
+ "loss": 1.314,
121
+ "step": 80
122
+ },
123
+ {
124
+ "epoch": 0.020469596628537028,
125
+ "grad_norm": 0.7786366939544678,
126
+ "learning_rate": 1.9590558766859348e-05,
127
+ "loss": 1.2631,
128
+ "step": 85
129
+ },
130
+ {
131
+ "epoch": 0.021673690547862733,
132
+ "grad_norm": 0.8259329795837402,
133
+ "learning_rate": 1.9566473988439307e-05,
134
+ "loss": 1.4006,
135
+ "step": 90
136
+ },
137
+ {
138
+ "epoch": 0.022877784467188442,
139
+ "grad_norm": 0.7402159571647644,
140
+ "learning_rate": 1.954238921001927e-05,
141
+ "loss": 1.416,
142
+ "step": 95
143
+ },
144
+ {
145
+ "epoch": 0.024081878386514148,
146
+ "grad_norm": 0.8619882464408875,
147
+ "learning_rate": 1.951830443159923e-05,
148
+ "loss": 1.2583,
149
+ "step": 100
150
+ },
151
+ {
152
+ "epoch": 0.025285972305839857,
153
+ "grad_norm": 0.7436647415161133,
154
+ "learning_rate": 1.949421965317919e-05,
155
+ "loss": 1.3527,
156
+ "step": 105
157
+ },
158
+ {
159
+ "epoch": 0.026490066225165563,
160
+ "grad_norm": 0.7261799573898315,
161
+ "learning_rate": 1.9470134874759156e-05,
162
+ "loss": 1.4234,
163
+ "step": 110
164
+ },
165
+ {
166
+ "epoch": 0.027694160144491272,
167
+ "grad_norm": 0.8648309707641602,
168
+ "learning_rate": 1.9446050096339115e-05,
169
+ "loss": 1.2541,
170
+ "step": 115
171
+ },
172
+ {
173
+ "epoch": 0.028898254063816978,
174
+ "grad_norm": 0.7091889381408691,
175
+ "learning_rate": 1.9421965317919077e-05,
176
+ "loss": 1.2473,
177
+ "step": 120
178
+ },
179
+ {
180
+ "epoch": 0.030102347983142687,
181
+ "grad_norm": 0.7450091242790222,
182
+ "learning_rate": 1.939788053949904e-05,
183
+ "loss": 1.2555,
184
+ "step": 125
185
+ },
186
+ {
187
+ "epoch": 0.031306441902468396,
188
+ "grad_norm": 0.8190895318984985,
189
+ "learning_rate": 1.9373795761079e-05,
190
+ "loss": 1.2874,
191
+ "step": 130
192
+ },
193
+ {
194
+ "epoch": 0.0325105358217941,
195
+ "grad_norm": 0.7458487749099731,
196
+ "learning_rate": 1.934971098265896e-05,
197
+ "loss": 1.4139,
198
+ "step": 135
199
+ },
200
+ {
201
+ "epoch": 0.03371462974111981,
202
+ "grad_norm": 0.806698203086853,
203
+ "learning_rate": 1.9325626204238923e-05,
204
+ "loss": 1.4061,
205
+ "step": 140
206
+ },
207
+ {
208
+ "epoch": 0.034918723660445516,
209
+ "grad_norm": 0.7941407561302185,
210
+ "learning_rate": 1.9301541425818882e-05,
211
+ "loss": 1.2819,
212
+ "step": 145
213
+ },
214
+ {
215
+ "epoch": 0.036122817579771226,
216
+ "grad_norm": 0.6824349164962769,
217
+ "learning_rate": 1.9277456647398845e-05,
218
+ "loss": 1.3025,
219
+ "step": 150
220
+ },
221
+ {
222
+ "epoch": 0.03732691149909693,
223
+ "grad_norm": 0.8056986927986145,
224
+ "learning_rate": 1.9253371868978807e-05,
225
+ "loss": 1.1617,
226
+ "step": 155
227
+ },
228
+ {
229
+ "epoch": 0.03853100541842264,
230
+ "grad_norm": 0.7486716508865356,
231
+ "learning_rate": 1.922928709055877e-05,
232
+ "loss": 1.3523,
233
+ "step": 160
234
+ },
235
+ {
236
+ "epoch": 0.039735099337748346,
237
+ "grad_norm": 0.7773532867431641,
238
+ "learning_rate": 1.920520231213873e-05,
239
+ "loss": 1.3013,
240
+ "step": 165
241
+ },
242
+ {
243
+ "epoch": 0.040939193257074055,
244
+ "grad_norm": 0.7498481869697571,
245
+ "learning_rate": 1.918111753371869e-05,
246
+ "loss": 1.2563,
247
+ "step": 170
248
+ },
249
+ {
250
+ "epoch": 0.04214328717639976,
251
+ "grad_norm": 0.7634468078613281,
252
+ "learning_rate": 1.9157032755298653e-05,
253
+ "loss": 1.2365,
254
+ "step": 175
255
+ },
256
+ {
257
+ "epoch": 0.04334738109572547,
258
+ "grad_norm": 0.6561328172683716,
259
+ "learning_rate": 1.9132947976878615e-05,
260
+ "loss": 1.3058,
261
+ "step": 180
262
+ },
263
+ {
264
+ "epoch": 0.044551475015051176,
265
+ "grad_norm": 0.7443628907203674,
266
+ "learning_rate": 1.9108863198458578e-05,
267
+ "loss": 1.3264,
268
+ "step": 185
269
+ },
270
+ {
271
+ "epoch": 0.045755568934376885,
272
+ "grad_norm": 0.6697310209274292,
273
+ "learning_rate": 1.9084778420038536e-05,
274
+ "loss": 1.1029,
275
+ "step": 190
276
+ },
277
+ {
278
+ "epoch": 0.04695966285370259,
279
+ "grad_norm": 0.7119889259338379,
280
+ "learning_rate": 1.90606936416185e-05,
281
+ "loss": 1.2503,
282
+ "step": 195
283
+ },
284
+ {
285
+ "epoch": 0.048163756773028296,
286
+ "grad_norm": 0.8120700120925903,
287
+ "learning_rate": 1.903660886319846e-05,
288
+ "loss": 1.246,
289
+ "step": 200
290
+ },
291
+ {
292
+ "epoch": 0.048163756773028296,
293
+ "eval_loss": 1.2944953441619873,
294
+ "eval_runtime": 10.5151,
295
+ "eval_samples_per_second": 9.51,
296
+ "eval_steps_per_second": 1.236,
297
+ "step": 200
298
+ },
299
+ {
300
+ "epoch": 0.049367850692354005,
301
+ "grad_norm": 0.7876719832420349,
302
+ "learning_rate": 1.901252408477842e-05,
303
+ "loss": 1.2097,
304
+ "step": 205
305
+ },
306
+ {
307
+ "epoch": 0.050571944611679714,
308
+ "grad_norm": 0.7367354035377502,
309
+ "learning_rate": 1.8988439306358382e-05,
310
+ "loss": 1.3332,
311
+ "step": 210
312
+ },
313
+ {
314
+ "epoch": 0.05177603853100542,
315
+ "grad_norm": 0.7384558320045471,
316
+ "learning_rate": 1.8964354527938345e-05,
317
+ "loss": 1.3897,
318
+ "step": 215
319
+ },
320
+ {
321
+ "epoch": 0.052980132450331126,
322
+ "grad_norm": 0.8261380195617676,
323
+ "learning_rate": 1.8940269749518304e-05,
324
+ "loss": 1.2677,
325
+ "step": 220
326
+ },
327
+ {
328
+ "epoch": 0.054184226369656835,
329
+ "grad_norm": 0.7237825989723206,
330
+ "learning_rate": 1.891618497109827e-05,
331
+ "loss": 1.2919,
332
+ "step": 225
333
+ },
334
+ {
335
+ "epoch": 0.055388320288982544,
336
+ "grad_norm": 0.7296762466430664,
337
+ "learning_rate": 1.889210019267823e-05,
338
+ "loss": 1.2977,
339
+ "step": 230
340
+ },
341
+ {
342
+ "epoch": 0.056592414208308246,
343
+ "grad_norm": 0.8224279880523682,
344
+ "learning_rate": 1.886801541425819e-05,
345
+ "loss": 1.3751,
346
+ "step": 235
347
+ },
348
+ {
349
+ "epoch": 0.057796508127633955,
350
+ "grad_norm": 0.7233418822288513,
351
+ "learning_rate": 1.8843930635838153e-05,
352
+ "loss": 1.293,
353
+ "step": 240
354
+ },
355
+ {
356
+ "epoch": 0.059000602046959665,
357
+ "grad_norm": 0.7015657424926758,
358
+ "learning_rate": 1.8819845857418112e-05,
359
+ "loss": 1.3093,
360
+ "step": 245
361
+ },
362
+ {
363
+ "epoch": 0.060204695966285374,
364
+ "grad_norm": 0.8122280240058899,
365
+ "learning_rate": 1.8795761078998074e-05,
366
+ "loss": 1.3979,
367
+ "step": 250
368
+ },
369
+ {
370
+ "epoch": 0.061408789885611076,
371
+ "grad_norm": 0.7582533955574036,
372
+ "learning_rate": 1.8771676300578037e-05,
373
+ "loss": 1.3147,
374
+ "step": 255
375
+ },
376
+ {
377
+ "epoch": 0.06261288380493679,
378
+ "grad_norm": 0.746967077255249,
379
+ "learning_rate": 1.8747591522158e-05,
380
+ "loss": 1.1785,
381
+ "step": 260
382
+ },
383
+ {
384
+ "epoch": 0.0638169777242625,
385
+ "grad_norm": 0.7588161826133728,
386
+ "learning_rate": 1.8723506743737958e-05,
387
+ "loss": 1.3648,
388
+ "step": 265
389
+ },
390
+ {
391
+ "epoch": 0.0650210716435882,
392
+ "grad_norm": 0.696869432926178,
393
+ "learning_rate": 1.869942196531792e-05,
394
+ "loss": 1.2854,
395
+ "step": 270
396
+ },
397
+ {
398
+ "epoch": 0.06622516556291391,
399
+ "grad_norm": 0.8297455906867981,
400
+ "learning_rate": 1.8675337186897883e-05,
401
+ "loss": 1.2011,
402
+ "step": 275
403
+ },
404
+ {
405
+ "epoch": 0.06742925948223961,
406
+ "grad_norm": 0.7183229327201843,
407
+ "learning_rate": 1.8651252408477845e-05,
408
+ "loss": 1.1944,
409
+ "step": 280
410
+ },
411
+ {
412
+ "epoch": 0.06863335340156532,
413
+ "grad_norm": 0.7092033624649048,
414
+ "learning_rate": 1.8627167630057804e-05,
415
+ "loss": 1.3296,
416
+ "step": 285
417
+ },
418
+ {
419
+ "epoch": 0.06983744732089103,
420
+ "grad_norm": 0.7024344801902771,
421
+ "learning_rate": 1.8603082851637766e-05,
422
+ "loss": 1.2349,
423
+ "step": 290
424
+ },
425
+ {
426
+ "epoch": 0.07104154124021674,
427
+ "grad_norm": 0.7504032254219055,
428
+ "learning_rate": 1.857899807321773e-05,
429
+ "loss": 1.1986,
430
+ "step": 295
431
+ },
432
+ {
433
+ "epoch": 0.07224563515954245,
434
+ "grad_norm": 0.7584034204483032,
435
+ "learning_rate": 1.855491329479769e-05,
436
+ "loss": 1.2998,
437
+ "step": 300
438
+ },
439
+ {
440
+ "epoch": 0.07344972907886815,
441
+ "grad_norm": 0.7563251852989197,
442
+ "learning_rate": 1.853082851637765e-05,
443
+ "loss": 1.3703,
444
+ "step": 305
445
+ },
446
+ {
447
+ "epoch": 0.07465382299819386,
448
+ "grad_norm": 0.728660523891449,
449
+ "learning_rate": 1.8506743737957612e-05,
450
+ "loss": 1.2916,
451
+ "step": 310
452
+ },
453
+ {
454
+ "epoch": 0.07585791691751957,
455
+ "grad_norm": 0.6596492528915405,
456
+ "learning_rate": 1.8482658959537575e-05,
457
+ "loss": 1.1821,
458
+ "step": 315
459
+ },
460
+ {
461
+ "epoch": 0.07706201083684527,
462
+ "grad_norm": 0.7137461304664612,
463
+ "learning_rate": 1.8458574181117533e-05,
464
+ "loss": 1.1982,
465
+ "step": 320
466
+ },
467
+ {
468
+ "epoch": 0.07826610475617098,
469
+ "grad_norm": 0.7411230206489563,
470
+ "learning_rate": 1.8434489402697496e-05,
471
+ "loss": 1.2711,
472
+ "step": 325
473
+ },
474
+ {
475
+ "epoch": 0.07947019867549669,
476
+ "grad_norm": 0.7566368579864502,
477
+ "learning_rate": 1.8410404624277458e-05,
478
+ "loss": 1.1603,
479
+ "step": 330
480
+ },
481
+ {
482
+ "epoch": 0.0806742925948224,
483
+ "grad_norm": 0.7235690355300903,
484
+ "learning_rate": 1.838631984585742e-05,
485
+ "loss": 1.1993,
486
+ "step": 335
487
+ },
488
+ {
489
+ "epoch": 0.08187838651414811,
490
+ "grad_norm": 0.7648451328277588,
491
+ "learning_rate": 1.8362235067437383e-05,
492
+ "loss": 1.134,
493
+ "step": 340
494
+ },
495
+ {
496
+ "epoch": 0.08308248043347381,
497
+ "grad_norm": 0.740710973739624,
498
+ "learning_rate": 1.8338150289017342e-05,
499
+ "loss": 1.1533,
500
+ "step": 345
501
+ },
502
+ {
503
+ "epoch": 0.08428657435279951,
504
+ "grad_norm": 0.7610028982162476,
505
+ "learning_rate": 1.8314065510597304e-05,
506
+ "loss": 1.3126,
507
+ "step": 350
508
+ },
509
+ {
510
+ "epoch": 0.08549066827212523,
511
+ "grad_norm": 0.6956422924995422,
512
+ "learning_rate": 1.8289980732177266e-05,
513
+ "loss": 1.2272,
514
+ "step": 355
515
+ },
516
+ {
517
+ "epoch": 0.08669476219145093,
518
+ "grad_norm": 0.6914700269699097,
519
+ "learning_rate": 1.8265895953757225e-05,
520
+ "loss": 1.3049,
521
+ "step": 360
522
+ },
523
+ {
524
+ "epoch": 0.08789885611077664,
525
+ "grad_norm": 0.7015173435211182,
526
+ "learning_rate": 1.8241811175337188e-05,
527
+ "loss": 1.3569,
528
+ "step": 365
529
+ },
530
+ {
531
+ "epoch": 0.08910295003010235,
532
+ "grad_norm": 0.6985088586807251,
533
+ "learning_rate": 1.821772639691715e-05,
534
+ "loss": 1.2307,
535
+ "step": 370
536
+ },
537
+ {
538
+ "epoch": 0.09030704394942805,
539
+ "grad_norm": 0.7313199043273926,
540
+ "learning_rate": 1.8193641618497112e-05,
541
+ "loss": 1.1783,
542
+ "step": 375
543
+ },
544
+ {
545
+ "epoch": 0.09151113786875377,
546
+ "grad_norm": 0.7657830715179443,
547
+ "learning_rate": 1.8169556840077075e-05,
548
+ "loss": 1.2228,
549
+ "step": 380
550
+ },
551
+ {
552
+ "epoch": 0.09271523178807947,
553
+ "grad_norm": 0.8051377534866333,
554
+ "learning_rate": 1.8145472061657034e-05,
555
+ "loss": 1.25,
556
+ "step": 385
557
+ },
558
+ {
559
+ "epoch": 0.09391932570740517,
560
+ "grad_norm": 0.8114392161369324,
561
+ "learning_rate": 1.8121387283236996e-05,
562
+ "loss": 1.2347,
563
+ "step": 390
564
+ },
565
+ {
566
+ "epoch": 0.09512341962673089,
567
+ "grad_norm": 0.7304120659828186,
568
+ "learning_rate": 1.809730250481696e-05,
569
+ "loss": 1.2437,
570
+ "step": 395
571
+ },
572
+ {
573
+ "epoch": 0.09632751354605659,
574
+ "grad_norm": 0.8418581485748291,
575
+ "learning_rate": 1.8073217726396917e-05,
576
+ "loss": 1.3255,
577
+ "step": 400
578
+ },
579
+ {
580
+ "epoch": 0.09632751354605659,
581
+ "eval_loss": 1.255526065826416,
582
+ "eval_runtime": 10.4675,
583
+ "eval_samples_per_second": 9.553,
584
+ "eval_steps_per_second": 1.242,
585
+ "step": 400
586
+ },
587
+ {
588
+ "epoch": 0.0975316074653823,
589
+ "grad_norm": 0.7186573147773743,
590
+ "learning_rate": 1.804913294797688e-05,
591
+ "loss": 1.2859,
592
+ "step": 405
593
+ },
594
+ {
595
+ "epoch": 0.09873570138470801,
596
+ "grad_norm": 0.7544155716896057,
597
+ "learning_rate": 1.8025048169556842e-05,
598
+ "loss": 1.2771,
599
+ "step": 410
600
+ },
601
+ {
602
+ "epoch": 0.09993979530403371,
603
+ "grad_norm": 0.700529158115387,
604
+ "learning_rate": 1.8000963391136804e-05,
605
+ "loss": 1.2696,
606
+ "step": 415
607
+ },
608
+ {
609
+ "epoch": 0.10114388922335943,
610
+ "grad_norm": 0.6894243955612183,
611
+ "learning_rate": 1.7976878612716763e-05,
612
+ "loss": 1.2556,
613
+ "step": 420
614
+ },
615
+ {
616
+ "epoch": 0.10234798314268513,
617
+ "grad_norm": 0.8769268989562988,
618
+ "learning_rate": 1.7952793834296726e-05,
619
+ "loss": 1.2832,
620
+ "step": 425
621
+ },
622
+ {
623
+ "epoch": 0.10355207706201083,
624
+ "grad_norm": 0.7283303141593933,
625
+ "learning_rate": 1.7928709055876688e-05,
626
+ "loss": 1.3163,
627
+ "step": 430
628
+ },
629
+ {
630
+ "epoch": 0.10475617098133655,
631
+ "grad_norm": 0.6984531283378601,
632
+ "learning_rate": 1.7904624277456647e-05,
633
+ "loss": 1.1873,
634
+ "step": 435
635
+ },
636
+ {
637
+ "epoch": 0.10596026490066225,
638
+ "grad_norm": 0.7543785572052002,
639
+ "learning_rate": 1.7880539499036613e-05,
640
+ "loss": 1.2617,
641
+ "step": 440
642
+ },
643
+ {
644
+ "epoch": 0.10716435881998795,
645
+ "grad_norm": 0.7481861710548401,
646
+ "learning_rate": 1.785645472061657e-05,
647
+ "loss": 1.1792,
648
+ "step": 445
649
+ },
650
+ {
651
+ "epoch": 0.10836845273931367,
652
+ "grad_norm": 0.7607905268669128,
653
+ "learning_rate": 1.7832369942196534e-05,
654
+ "loss": 1.1987,
655
+ "step": 450
656
+ },
657
+ {
658
+ "epoch": 0.10957254665863937,
659
+ "grad_norm": 0.7427639365196228,
660
+ "learning_rate": 1.7808285163776496e-05,
661
+ "loss": 1.2142,
662
+ "step": 455
663
+ },
664
+ {
665
+ "epoch": 0.11077664057796509,
666
+ "grad_norm": 0.7348979711532593,
667
+ "learning_rate": 1.7784200385356455e-05,
668
+ "loss": 1.2727,
669
+ "step": 460
670
+ },
671
+ {
672
+ "epoch": 0.11198073449729079,
673
+ "grad_norm": 0.7192914485931396,
674
+ "learning_rate": 1.7760115606936417e-05,
675
+ "loss": 1.2593,
676
+ "step": 465
677
+ },
678
+ {
679
+ "epoch": 0.11318482841661649,
680
+ "grad_norm": 0.6672971248626709,
681
+ "learning_rate": 1.773603082851638e-05,
682
+ "loss": 1.1905,
683
+ "step": 470
684
+ },
685
+ {
686
+ "epoch": 0.11438892233594221,
687
+ "grad_norm": 0.7585046291351318,
688
+ "learning_rate": 1.771194605009634e-05,
689
+ "loss": 1.3045,
690
+ "step": 475
691
+ },
692
+ {
693
+ "epoch": 0.11559301625526791,
694
+ "grad_norm": 0.7200559973716736,
695
+ "learning_rate": 1.76878612716763e-05,
696
+ "loss": 1.1416,
697
+ "step": 480
698
+ },
699
+ {
700
+ "epoch": 0.11679711017459361,
701
+ "grad_norm": 0.7220486998558044,
702
+ "learning_rate": 1.7663776493256263e-05,
703
+ "loss": 1.2749,
704
+ "step": 485
705
+ },
706
+ {
707
+ "epoch": 0.11800120409391933,
708
+ "grad_norm": 0.750998318195343,
709
+ "learning_rate": 1.7639691714836226e-05,
710
+ "loss": 1.2823,
711
+ "step": 490
712
+ },
713
+ {
714
+ "epoch": 0.11920529801324503,
715
+ "grad_norm": 0.705752432346344,
716
+ "learning_rate": 1.7615606936416188e-05,
717
+ "loss": 1.1662,
718
+ "step": 495
719
+ },
720
+ {
721
+ "epoch": 0.12040939193257075,
722
+ "grad_norm": 0.7431082129478455,
723
+ "learning_rate": 1.7591522157996147e-05,
724
+ "loss": 1.203,
725
+ "step": 500
726
+ },
727
+ {
728
+ "epoch": 0.12161348585189645,
729
+ "grad_norm": 0.7106168866157532,
730
+ "learning_rate": 1.756743737957611e-05,
731
+ "loss": 1.162,
732
+ "step": 505
733
+ },
734
+ {
735
+ "epoch": 0.12281757977122215,
736
+ "grad_norm": 0.7027086615562439,
737
+ "learning_rate": 1.754335260115607e-05,
738
+ "loss": 1.2636,
739
+ "step": 510
740
+ },
741
+ {
742
+ "epoch": 0.12402167369054787,
743
+ "grad_norm": 0.774591863155365,
744
+ "learning_rate": 1.7519267822736034e-05,
745
+ "loss": 1.2517,
746
+ "step": 515
747
+ },
748
+ {
749
+ "epoch": 0.12522576760987358,
750
+ "grad_norm": 0.6927476525306702,
751
+ "learning_rate": 1.7495183044315993e-05,
752
+ "loss": 1.1396,
753
+ "step": 520
754
+ },
755
+ {
756
+ "epoch": 0.12642986152919927,
757
+ "grad_norm": 0.6912556886672974,
758
+ "learning_rate": 1.7471098265895955e-05,
759
+ "loss": 1.1658,
760
+ "step": 525
761
+ },
762
+ {
763
+ "epoch": 0.127633955448525,
764
+ "grad_norm": 0.6708123683929443,
765
+ "learning_rate": 1.7447013487475918e-05,
766
+ "loss": 1.1288,
767
+ "step": 530
768
+ },
769
+ {
770
+ "epoch": 0.1288380493678507,
771
+ "grad_norm": 0.6942213177680969,
772
+ "learning_rate": 1.7422928709055877e-05,
773
+ "loss": 1.1174,
774
+ "step": 535
775
+ },
776
+ {
777
+ "epoch": 0.1300421432871764,
778
+ "grad_norm": 0.7080583572387695,
779
+ "learning_rate": 1.739884393063584e-05,
780
+ "loss": 1.2279,
781
+ "step": 540
782
+ },
783
+ {
784
+ "epoch": 0.1312462372065021,
785
+ "grad_norm": 0.8131110668182373,
786
+ "learning_rate": 1.73747591522158e-05,
787
+ "loss": 1.2525,
788
+ "step": 545
789
+ },
790
+ {
791
+ "epoch": 0.13245033112582782,
792
+ "grad_norm": 0.7450209259986877,
793
+ "learning_rate": 1.735067437379576e-05,
794
+ "loss": 1.1506,
795
+ "step": 550
796
+ },
797
+ {
798
+ "epoch": 0.1336544250451535,
799
+ "grad_norm": 0.7066523432731628,
800
+ "learning_rate": 1.7326589595375726e-05,
801
+ "loss": 1.2383,
802
+ "step": 555
803
+ },
804
+ {
805
+ "epoch": 0.13485851896447923,
806
+ "grad_norm": 0.7041836977005005,
807
+ "learning_rate": 1.7302504816955685e-05,
808
+ "loss": 1.1911,
809
+ "step": 560
810
+ },
811
+ {
812
+ "epoch": 0.13606261288380495,
813
+ "grad_norm": 0.672095000743866,
814
+ "learning_rate": 1.7278420038535647e-05,
815
+ "loss": 1.1839,
816
+ "step": 565
817
+ },
818
+ {
819
+ "epoch": 0.13726670680313063,
820
+ "grad_norm": 0.7928294539451599,
821
+ "learning_rate": 1.725433526011561e-05,
822
+ "loss": 1.3347,
823
+ "step": 570
824
+ },
825
+ {
826
+ "epoch": 0.13847080072245635,
827
+ "grad_norm": 0.7465523481369019,
828
+ "learning_rate": 1.723025048169557e-05,
829
+ "loss": 1.134,
830
+ "step": 575
831
+ },
832
+ {
833
+ "epoch": 0.13967489464178207,
834
+ "grad_norm": 0.7117074131965637,
835
+ "learning_rate": 1.720616570327553e-05,
836
+ "loss": 1.3538,
837
+ "step": 580
838
+ },
839
+ {
840
+ "epoch": 0.14087898856110775,
841
+ "grad_norm": 0.7291346192359924,
842
+ "learning_rate": 1.7182080924855493e-05,
843
+ "loss": 1.1339,
844
+ "step": 585
845
+ },
846
+ {
847
+ "epoch": 0.14208308248043347,
848
+ "grad_norm": 0.7633514404296875,
849
+ "learning_rate": 1.7157996146435455e-05,
850
+ "loss": 1.2007,
851
+ "step": 590
852
+ },
853
+ {
854
+ "epoch": 0.1432871763997592,
855
+ "grad_norm": 0.7896651029586792,
856
+ "learning_rate": 1.7133911368015418e-05,
857
+ "loss": 1.2656,
858
+ "step": 595
859
+ },
860
+ {
861
+ "epoch": 0.1444912703190849,
862
+ "grad_norm": 0.7436801791191101,
863
+ "learning_rate": 1.7109826589595377e-05,
864
+ "loss": 1.28,
865
+ "step": 600
866
+ },
867
+ {
868
+ "epoch": 0.1444912703190849,
869
+ "eval_loss": 1.23267662525177,
870
+ "eval_runtime": 10.4568,
871
+ "eval_samples_per_second": 9.563,
872
+ "eval_steps_per_second": 1.243,
873
+ "step": 600
874
+ },
875
+ {
876
+ "epoch": 0.1456953642384106,
877
+ "grad_norm": 0.7123108506202698,
878
+ "learning_rate": 1.708574181117534e-05,
879
+ "loss": 1.1687,
880
+ "step": 605
881
+ },
882
+ {
883
+ "epoch": 0.1468994581577363,
884
+ "grad_norm": 0.7362635731697083,
885
+ "learning_rate": 1.70616570327553e-05,
886
+ "loss": 1.2471,
887
+ "step": 610
888
+ },
889
+ {
890
+ "epoch": 0.14810355207706202,
891
+ "grad_norm": 0.6770139336585999,
892
+ "learning_rate": 1.703757225433526e-05,
893
+ "loss": 1.149,
894
+ "step": 615
895
+ },
896
+ {
897
+ "epoch": 0.1493076459963877,
898
+ "grad_norm": 0.7292876243591309,
899
+ "learning_rate": 1.7013487475915223e-05,
900
+ "loss": 1.2957,
901
+ "step": 620
902
+ },
903
+ {
904
+ "epoch": 0.15051173991571343,
905
+ "grad_norm": 0.7105087637901306,
906
+ "learning_rate": 1.6989402697495185e-05,
907
+ "loss": 1.1462,
908
+ "step": 625
909
+ },
910
+ {
911
+ "epoch": 0.15171583383503914,
912
+ "grad_norm": 0.7218934297561646,
913
+ "learning_rate": 1.6965317919075147e-05,
914
+ "loss": 1.2181,
915
+ "step": 630
916
+ },
917
+ {
918
+ "epoch": 0.15291992775436483,
919
+ "grad_norm": 0.7794796824455261,
920
+ "learning_rate": 1.6941233140655106e-05,
921
+ "loss": 1.174,
922
+ "step": 635
923
+ },
924
+ {
925
+ "epoch": 0.15412402167369055,
926
+ "grad_norm": 0.6424360871315002,
927
+ "learning_rate": 1.691714836223507e-05,
928
+ "loss": 1.2149,
929
+ "step": 640
930
+ },
931
+ {
932
+ "epoch": 0.15532811559301626,
933
+ "grad_norm": 0.7393337488174438,
934
+ "learning_rate": 1.689306358381503e-05,
935
+ "loss": 1.2259,
936
+ "step": 645
937
+ },
938
+ {
939
+ "epoch": 0.15653220951234195,
940
+ "grad_norm": 0.7363811731338501,
941
+ "learning_rate": 1.686897880539499e-05,
942
+ "loss": 1.2087,
943
+ "step": 650
944
+ },
945
+ {
946
+ "epoch": 0.15773630343166767,
947
+ "grad_norm": 0.7336999773979187,
948
+ "learning_rate": 1.6844894026974952e-05,
949
+ "loss": 1.2526,
950
+ "step": 655
951
+ },
952
+ {
953
+ "epoch": 0.15894039735099338,
954
+ "grad_norm": 0.7329117655754089,
955
+ "learning_rate": 1.6820809248554915e-05,
956
+ "loss": 1.233,
957
+ "step": 660
958
+ },
959
+ {
960
+ "epoch": 0.16014449127031907,
961
+ "grad_norm": 0.7724847197532654,
962
+ "learning_rate": 1.6796724470134877e-05,
963
+ "loss": 1.2506,
964
+ "step": 665
965
+ },
966
+ {
967
+ "epoch": 0.1613485851896448,
968
+ "grad_norm": 0.7213071584701538,
969
+ "learning_rate": 1.677263969171484e-05,
970
+ "loss": 1.2021,
971
+ "step": 670
972
+ },
973
+ {
974
+ "epoch": 0.1625526791089705,
975
+ "grad_norm": 0.6397632360458374,
976
+ "learning_rate": 1.6748554913294798e-05,
977
+ "loss": 1.1909,
978
+ "step": 675
979
+ },
980
+ {
981
+ "epoch": 0.16375677302829622,
982
+ "grad_norm": 0.6840227842330933,
983
+ "learning_rate": 1.672447013487476e-05,
984
+ "loss": 1.1838,
985
+ "step": 680
986
+ },
987
+ {
988
+ "epoch": 0.1649608669476219,
989
+ "grad_norm": 0.7317889332771301,
990
+ "learning_rate": 1.6700385356454723e-05,
991
+ "loss": 1.169,
992
+ "step": 685
993
+ },
994
+ {
995
+ "epoch": 0.16616496086694763,
996
+ "grad_norm": 0.7134217619895935,
997
+ "learning_rate": 1.6676300578034682e-05,
998
+ "loss": 1.2711,
999
+ "step": 690
1000
+ },
1001
+ {
1002
+ "epoch": 0.16736905478627334,
1003
+ "grad_norm": 0.7525413036346436,
1004
+ "learning_rate": 1.6652215799614644e-05,
1005
+ "loss": 1.2183,
1006
+ "step": 695
1007
+ },
1008
+ {
1009
+ "epoch": 0.16857314870559903,
1010
+ "grad_norm": 0.7243443131446838,
1011
+ "learning_rate": 1.6628131021194607e-05,
1012
+ "loss": 1.2585,
1013
+ "step": 700
1014
+ },
1015
+ {
1016
+ "epoch": 0.16977724262492475,
1017
+ "grad_norm": 0.7554174661636353,
1018
+ "learning_rate": 1.660404624277457e-05,
1019
+ "loss": 1.1662,
1020
+ "step": 705
1021
+ },
1022
+ {
1023
+ "epoch": 0.17098133654425046,
1024
+ "grad_norm": 0.6472408175468445,
1025
+ "learning_rate": 1.657996146435453e-05,
1026
+ "loss": 1.1471,
1027
+ "step": 710
1028
+ },
1029
+ {
1030
+ "epoch": 0.17218543046357615,
1031
+ "grad_norm": 0.7514950037002563,
1032
+ "learning_rate": 1.655587668593449e-05,
1033
+ "loss": 1.157,
1034
+ "step": 715
1035
+ },
1036
+ {
1037
+ "epoch": 0.17338952438290187,
1038
+ "grad_norm": 0.6952060461044312,
1039
+ "learning_rate": 1.6531791907514452e-05,
1040
+ "loss": 1.2981,
1041
+ "step": 720
1042
+ },
1043
+ {
1044
+ "epoch": 0.17459361830222758,
1045
+ "grad_norm": 0.832417905330658,
1046
+ "learning_rate": 1.6507707129094415e-05,
1047
+ "loss": 1.2457,
1048
+ "step": 725
1049
+ },
1050
+ {
1051
+ "epoch": 0.17579771222155327,
1052
+ "grad_norm": 0.6825481653213501,
1053
+ "learning_rate": 1.6483622350674374e-05,
1054
+ "loss": 1.2632,
1055
+ "step": 730
1056
+ },
1057
+ {
1058
+ "epoch": 0.177001806140879,
1059
+ "grad_norm": 0.8267092704772949,
1060
+ "learning_rate": 1.6459537572254336e-05,
1061
+ "loss": 1.0368,
1062
+ "step": 735
1063
+ },
1064
+ {
1065
+ "epoch": 0.1782059000602047,
1066
+ "grad_norm": 0.7782288789749146,
1067
+ "learning_rate": 1.64354527938343e-05,
1068
+ "loss": 1.201,
1069
+ "step": 740
1070
+ },
1071
+ {
1072
+ "epoch": 0.1794099939795304,
1073
+ "grad_norm": 0.7548500299453735,
1074
+ "learning_rate": 1.641136801541426e-05,
1075
+ "loss": 1.1831,
1076
+ "step": 745
1077
+ },
1078
+ {
1079
+ "epoch": 0.1806140878988561,
1080
+ "grad_norm": 0.6344689726829529,
1081
+ "learning_rate": 1.638728323699422e-05,
1082
+ "loss": 1.1985,
1083
+ "step": 750
1084
+ },
1085
+ {
1086
+ "epoch": 0.18181818181818182,
1087
+ "grad_norm": 0.7326412796974182,
1088
+ "learning_rate": 1.6363198458574182e-05,
1089
+ "loss": 1.3051,
1090
+ "step": 755
1091
+ },
1092
+ {
1093
+ "epoch": 0.18302227573750754,
1094
+ "grad_norm": 0.6460621953010559,
1095
+ "learning_rate": 1.6339113680154144e-05,
1096
+ "loss": 1.1524,
1097
+ "step": 760
1098
+ },
1099
+ {
1100
+ "epoch": 0.18422636965683323,
1101
+ "grad_norm": 0.6921172142028809,
1102
+ "learning_rate": 1.6315028901734103e-05,
1103
+ "loss": 1.29,
1104
+ "step": 765
1105
+ },
1106
+ {
1107
+ "epoch": 0.18543046357615894,
1108
+ "grad_norm": 0.7872111201286316,
1109
+ "learning_rate": 1.629094412331407e-05,
1110
+ "loss": 1.1187,
1111
+ "step": 770
1112
+ },
1113
+ {
1114
+ "epoch": 0.18663455749548466,
1115
+ "grad_norm": 0.7098067402839661,
1116
+ "learning_rate": 1.6266859344894028e-05,
1117
+ "loss": 1.2006,
1118
+ "step": 775
1119
+ },
1120
+ {
1121
+ "epoch": 0.18783865141481035,
1122
+ "grad_norm": 0.7701388001441956,
1123
+ "learning_rate": 1.624277456647399e-05,
1124
+ "loss": 1.1609,
1125
+ "step": 780
1126
+ },
1127
+ {
1128
+ "epoch": 0.18904274533413606,
1129
+ "grad_norm": 0.7027246952056885,
1130
+ "learning_rate": 1.6218689788053953e-05,
1131
+ "loss": 1.2083,
1132
+ "step": 785
1133
+ },
1134
+ {
1135
+ "epoch": 0.19024683925346178,
1136
+ "grad_norm": 0.7161267995834351,
1137
+ "learning_rate": 1.619460500963391e-05,
1138
+ "loss": 1.131,
1139
+ "step": 790
1140
+ },
1141
+ {
1142
+ "epoch": 0.19145093317278747,
1143
+ "grad_norm": 0.8034841418266296,
1144
+ "learning_rate": 1.6170520231213874e-05,
1145
+ "loss": 1.2107,
1146
+ "step": 795
1147
+ },
1148
+ {
1149
+ "epoch": 0.19265502709211318,
1150
+ "grad_norm": 0.7797010540962219,
1151
+ "learning_rate": 1.6146435452793836e-05,
1152
+ "loss": 1.2931,
1153
+ "step": 800
1154
+ },
1155
+ {
1156
+ "epoch": 0.19265502709211318,
1157
+ "eval_loss": 1.2178008556365967,
1158
+ "eval_runtime": 10.4492,
1159
+ "eval_samples_per_second": 9.57,
1160
+ "eval_steps_per_second": 1.244,
1161
+ "step": 800
1162
+ },
1163
+ {
1164
+ "epoch": 0.1938591210114389,
1165
+ "grad_norm": 0.6586723923683167,
1166
+ "learning_rate": 1.6122350674373795e-05,
1167
+ "loss": 1.2007,
1168
+ "step": 805
1169
+ },
1170
+ {
1171
+ "epoch": 0.1950632149307646,
1172
+ "grad_norm": 0.7271411418914795,
1173
+ "learning_rate": 1.6098265895953758e-05,
1174
+ "loss": 1.1049,
1175
+ "step": 810
1176
+ },
1177
+ {
1178
+ "epoch": 0.1962673088500903,
1179
+ "grad_norm": 0.743730902671814,
1180
+ "learning_rate": 1.607418111753372e-05,
1181
+ "loss": 1.1815,
1182
+ "step": 815
1183
+ },
1184
+ {
1185
+ "epoch": 0.19747140276941602,
1186
+ "grad_norm": 0.7163010835647583,
1187
+ "learning_rate": 1.6050096339113682e-05,
1188
+ "loss": 1.1635,
1189
+ "step": 820
1190
+ },
1191
+ {
1192
+ "epoch": 0.1986754966887417,
1193
+ "grad_norm": 0.7846854329109192,
1194
+ "learning_rate": 1.6026011560693645e-05,
1195
+ "loss": 1.1576,
1196
+ "step": 825
1197
+ },
1198
+ {
1199
+ "epoch": 0.19987959060806743,
1200
+ "grad_norm": 0.755066990852356,
1201
+ "learning_rate": 1.6001926782273604e-05,
1202
+ "loss": 1.2626,
1203
+ "step": 830
1204
+ },
1205
+ {
1206
+ "epoch": 0.20108368452739314,
1207
+ "grad_norm": 0.6981719732284546,
1208
+ "learning_rate": 1.5977842003853566e-05,
1209
+ "loss": 1.1883,
1210
+ "step": 835
1211
+ },
1212
+ {
1213
+ "epoch": 0.20228777844671886,
1214
+ "grad_norm": 0.712697446346283,
1215
+ "learning_rate": 1.5953757225433528e-05,
1216
+ "loss": 1.1539,
1217
+ "step": 840
1218
+ },
1219
+ {
1220
+ "epoch": 0.20349187236604455,
1221
+ "grad_norm": 0.8146225810050964,
1222
+ "learning_rate": 1.592967244701349e-05,
1223
+ "loss": 1.161,
1224
+ "step": 845
1225
+ },
1226
+ {
1227
+ "epoch": 0.20469596628537026,
1228
+ "grad_norm": 0.7796756625175476,
1229
+ "learning_rate": 1.590558766859345e-05,
1230
+ "loss": 1.1536,
1231
+ "step": 850
1232
+ },
1233
+ {
1234
+ "epoch": 0.20590006020469598,
1235
+ "grad_norm": 0.7026313543319702,
1236
+ "learning_rate": 1.5881502890173412e-05,
1237
+ "loss": 1.2206,
1238
+ "step": 855
1239
+ },
1240
+ {
1241
+ "epoch": 0.20710415412402167,
1242
+ "grad_norm": 0.6918179988861084,
1243
+ "learning_rate": 1.5857418111753374e-05,
1244
+ "loss": 1.0569,
1245
+ "step": 860
1246
+ },
1247
+ {
1248
+ "epoch": 0.20830824804334738,
1249
+ "grad_norm": 0.7711547613143921,
1250
+ "learning_rate": 1.5833333333333333e-05,
1251
+ "loss": 1.1826,
1252
+ "step": 865
1253
+ },
1254
+ {
1255
+ "epoch": 0.2095123419626731,
1256
+ "grad_norm": 0.6604825854301453,
1257
+ "learning_rate": 1.5809248554913295e-05,
1258
+ "loss": 1.1743,
1259
+ "step": 870
1260
+ },
1261
+ {
1262
+ "epoch": 0.2107164358819988,
1263
+ "grad_norm": 0.7115917205810547,
1264
+ "learning_rate": 1.5785163776493258e-05,
1265
+ "loss": 1.2815,
1266
+ "step": 875
1267
+ },
1268
+ {
1269
+ "epoch": 0.2119205298013245,
1270
+ "grad_norm": 0.69574373960495,
1271
+ "learning_rate": 1.5761078998073217e-05,
1272
+ "loss": 1.2131,
1273
+ "step": 880
1274
+ },
1275
+ {
1276
+ "epoch": 0.21312462372065022,
1277
+ "grad_norm": 0.6801597476005554,
1278
+ "learning_rate": 1.5736994219653182e-05,
1279
+ "loss": 1.3022,
1280
+ "step": 885
1281
+ },
1282
+ {
1283
+ "epoch": 0.2143287176399759,
1284
+ "grad_norm": 0.7042235136032104,
1285
+ "learning_rate": 1.571290944123314e-05,
1286
+ "loss": 1.1936,
1287
+ "step": 890
1288
+ },
1289
+ {
1290
+ "epoch": 0.21553281155930162,
1291
+ "grad_norm": 0.6610032916069031,
1292
+ "learning_rate": 1.5688824662813104e-05,
1293
+ "loss": 1.1972,
1294
+ "step": 895
1295
+ },
1296
+ {
1297
+ "epoch": 0.21673690547862734,
1298
+ "grad_norm": 0.8218846321105957,
1299
+ "learning_rate": 1.5664739884393066e-05,
1300
+ "loss": 1.2951,
1301
+ "step": 900
1302
+ },
1303
+ {
1304
+ "epoch": 0.21794099939795303,
1305
+ "grad_norm": 0.7200918197631836,
1306
+ "learning_rate": 1.5640655105973025e-05,
1307
+ "loss": 1.1256,
1308
+ "step": 905
1309
+ },
1310
+ {
1311
+ "epoch": 0.21914509331727874,
1312
+ "grad_norm": 0.6200802326202393,
1313
+ "learning_rate": 1.5616570327552987e-05,
1314
+ "loss": 1.2071,
1315
+ "step": 910
1316
+ },
1317
+ {
1318
+ "epoch": 0.22034918723660446,
1319
+ "grad_norm": 0.7431606650352478,
1320
+ "learning_rate": 1.559248554913295e-05,
1321
+ "loss": 1.2569,
1322
+ "step": 915
1323
+ },
1324
+ {
1325
+ "epoch": 0.22155328115593018,
1326
+ "grad_norm": 0.802537202835083,
1327
+ "learning_rate": 1.5568400770712912e-05,
1328
+ "loss": 1.1342,
1329
+ "step": 920
1330
+ },
1331
+ {
1332
+ "epoch": 0.22275737507525586,
1333
+ "grad_norm": 0.7078563570976257,
1334
+ "learning_rate": 1.5544315992292874e-05,
1335
+ "loss": 1.1928,
1336
+ "step": 925
1337
+ },
1338
+ {
1339
+ "epoch": 0.22396146899458158,
1340
+ "grad_norm": 0.6626926064491272,
1341
+ "learning_rate": 1.5520231213872833e-05,
1342
+ "loss": 1.1295,
1343
+ "step": 930
1344
+ },
1345
+ {
1346
+ "epoch": 0.2251655629139073,
1347
+ "grad_norm": 0.664715588092804,
1348
+ "learning_rate": 1.5496146435452796e-05,
1349
+ "loss": 1.1616,
1350
+ "step": 935
1351
+ },
1352
+ {
1353
+ "epoch": 0.22636965683323299,
1354
+ "grad_norm": 0.7077795267105103,
1355
+ "learning_rate": 1.5472061657032758e-05,
1356
+ "loss": 1.24,
1357
+ "step": 940
1358
+ },
1359
+ {
1360
+ "epoch": 0.2275737507525587,
1361
+ "grad_norm": 0.6813915371894836,
1362
+ "learning_rate": 1.5447976878612717e-05,
1363
+ "loss": 1.1807,
1364
+ "step": 945
1365
+ },
1366
+ {
1367
+ "epoch": 0.22877784467188442,
1368
+ "grad_norm": 0.7245080471038818,
1369
+ "learning_rate": 1.542389210019268e-05,
1370
+ "loss": 1.1737,
1371
+ "step": 950
1372
+ },
1373
+ {
1374
+ "epoch": 0.2299819385912101,
1375
+ "grad_norm": 0.7851933836936951,
1376
+ "learning_rate": 1.539980732177264e-05,
1377
+ "loss": 1.0702,
1378
+ "step": 955
1379
+ },
1380
+ {
1381
+ "epoch": 0.23118603251053582,
1382
+ "grad_norm": 0.7153802514076233,
1383
+ "learning_rate": 1.5375722543352604e-05,
1384
+ "loss": 1.1302,
1385
+ "step": 960
1386
+ },
1387
+ {
1388
+ "epoch": 0.23239012642986154,
1389
+ "grad_norm": 0.8112071752548218,
1390
+ "learning_rate": 1.5351637764932563e-05,
1391
+ "loss": 1.1591,
1392
+ "step": 965
1393
+ },
1394
+ {
1395
+ "epoch": 0.23359422034918723,
1396
+ "grad_norm": 0.7842305898666382,
1397
+ "learning_rate": 1.5327552986512525e-05,
1398
+ "loss": 1.1695,
1399
+ "step": 970
1400
+ },
1401
+ {
1402
+ "epoch": 0.23479831426851294,
1403
+ "grad_norm": 0.6949911713600159,
1404
+ "learning_rate": 1.5303468208092487e-05,
1405
+ "loss": 1.1726,
1406
+ "step": 975
1407
+ },
1408
+ {
1409
+ "epoch": 0.23600240818783866,
1410
+ "grad_norm": 0.7541187405586243,
1411
+ "learning_rate": 1.5279383429672446e-05,
1412
+ "loss": 1.2209,
1413
+ "step": 980
1414
+ },
1415
+ {
1416
+ "epoch": 0.23720650210716435,
1417
+ "grad_norm": 0.7082781791687012,
1418
+ "learning_rate": 1.525529865125241e-05,
1419
+ "loss": 1.2685,
1420
+ "step": 985
1421
+ },
1422
+ {
1423
+ "epoch": 0.23841059602649006,
1424
+ "grad_norm": 0.6862682104110718,
1425
+ "learning_rate": 1.5231213872832371e-05,
1426
+ "loss": 1.1569,
1427
+ "step": 990
1428
+ },
1429
+ {
1430
+ "epoch": 0.23961468994581578,
1431
+ "grad_norm": 0.836452066898346,
1432
+ "learning_rate": 1.5207129094412332e-05,
1433
+ "loss": 1.2158,
1434
+ "step": 995
1435
+ },
1436
+ {
1437
+ "epoch": 0.2408187838651415,
1438
+ "grad_norm": 0.8637534379959106,
1439
+ "learning_rate": 1.5183044315992294e-05,
1440
+ "loss": 1.1891,
1441
+ "step": 1000
1442
+ },
1443
+ {
1444
+ "epoch": 0.2408187838651415,
1445
+ "eval_loss": 1.2066378593444824,
1446
+ "eval_runtime": 10.4478,
1447
+ "eval_samples_per_second": 9.571,
1448
+ "eval_steps_per_second": 1.244,
1449
+ "step": 1000
1450
+ },
1451
+ {
1452
+ "epoch": 0.24202287778446718,
1453
+ "grad_norm": 0.8397049903869629,
1454
+ "learning_rate": 1.5158959537572255e-05,
1455
+ "loss": 1.2199,
1456
+ "step": 1005
1457
+ },
1458
+ {
1459
+ "epoch": 0.2432269717037929,
1460
+ "grad_norm": 0.7865377068519592,
1461
+ "learning_rate": 1.5134874759152215e-05,
1462
+ "loss": 1.2102,
1463
+ "step": 1010
1464
+ },
1465
+ {
1466
+ "epoch": 0.24443106562311862,
1467
+ "grad_norm": 0.6602733731269836,
1468
+ "learning_rate": 1.511078998073218e-05,
1469
+ "loss": 1.1512,
1470
+ "step": 1015
1471
+ },
1472
+ {
1473
+ "epoch": 0.2456351595424443,
1474
+ "grad_norm": 0.69524747133255,
1475
+ "learning_rate": 1.508670520231214e-05,
1476
+ "loss": 1.2387,
1477
+ "step": 1020
1478
+ },
1479
+ {
1480
+ "epoch": 0.24683925346177002,
1481
+ "grad_norm": 0.6567270755767822,
1482
+ "learning_rate": 1.50626204238921e-05,
1483
+ "loss": 1.159,
1484
+ "step": 1025
1485
+ },
1486
+ {
1487
+ "epoch": 0.24804334738109574,
1488
+ "grad_norm": 0.6918655633926392,
1489
+ "learning_rate": 1.5038535645472063e-05,
1490
+ "loss": 1.236,
1491
+ "step": 1030
1492
+ },
1493
+ {
1494
+ "epoch": 0.24924744130042142,
1495
+ "grad_norm": 0.748435378074646,
1496
+ "learning_rate": 1.5014450867052024e-05,
1497
+ "loss": 1.1422,
1498
+ "step": 1035
1499
+ },
1500
+ {
1501
+ "epoch": 0.25045153521974717,
1502
+ "grad_norm": 0.8211308717727661,
1503
+ "learning_rate": 1.4990366088631986e-05,
1504
+ "loss": 1.1775,
1505
+ "step": 1040
1506
+ },
1507
+ {
1508
+ "epoch": 0.25165562913907286,
1509
+ "grad_norm": 0.6989510655403137,
1510
+ "learning_rate": 1.4966281310211948e-05,
1511
+ "loss": 1.2473,
1512
+ "step": 1045
1513
+ },
1514
+ {
1515
+ "epoch": 0.25285972305839854,
1516
+ "grad_norm": 0.6981227993965149,
1517
+ "learning_rate": 1.4942196531791909e-05,
1518
+ "loss": 1.1887,
1519
+ "step": 1050
1520
+ },
1521
+ {
1522
+ "epoch": 0.2540638169777243,
1523
+ "grad_norm": 0.6409928202629089,
1524
+ "learning_rate": 1.4918111753371871e-05,
1525
+ "loss": 1.1323,
1526
+ "step": 1055
1527
+ },
1528
+ {
1529
+ "epoch": 0.25526791089705,
1530
+ "grad_norm": 0.8026551604270935,
1531
+ "learning_rate": 1.4894026974951832e-05,
1532
+ "loss": 1.2205,
1533
+ "step": 1060
1534
+ },
1535
+ {
1536
+ "epoch": 0.25647200481637566,
1537
+ "grad_norm": 0.7427927255630493,
1538
+ "learning_rate": 1.4869942196531793e-05,
1539
+ "loss": 1.3051,
1540
+ "step": 1065
1541
+ },
1542
+ {
1543
+ "epoch": 0.2576760987357014,
1544
+ "grad_norm": 0.7286926507949829,
1545
+ "learning_rate": 1.4845857418111755e-05,
1546
+ "loss": 1.1952,
1547
+ "step": 1070
1548
+ },
1549
+ {
1550
+ "epoch": 0.2588801926550271,
1551
+ "grad_norm": 0.717553436756134,
1552
+ "learning_rate": 1.4821772639691716e-05,
1553
+ "loss": 1.1395,
1554
+ "step": 1075
1555
+ },
1556
+ {
1557
+ "epoch": 0.2600842865743528,
1558
+ "grad_norm": 0.6549174189567566,
1559
+ "learning_rate": 1.4797687861271676e-05,
1560
+ "loss": 1.1677,
1561
+ "step": 1080
1562
+ },
1563
+ {
1564
+ "epoch": 0.26128838049367853,
1565
+ "grad_norm": 0.7400743365287781,
1566
+ "learning_rate": 1.477360308285164e-05,
1567
+ "loss": 1.239,
1568
+ "step": 1085
1569
+ },
1570
+ {
1571
+ "epoch": 0.2624924744130042,
1572
+ "grad_norm": 0.8405967950820923,
1573
+ "learning_rate": 1.4749518304431601e-05,
1574
+ "loss": 1.1379,
1575
+ "step": 1090
1576
+ },
1577
+ {
1578
+ "epoch": 0.2636965683323299,
1579
+ "grad_norm": 0.7308910489082336,
1580
+ "learning_rate": 1.4725433526011562e-05,
1581
+ "loss": 1.1709,
1582
+ "step": 1095
1583
+ },
1584
+ {
1585
+ "epoch": 0.26490066225165565,
1586
+ "grad_norm": 0.7904026508331299,
1587
+ "learning_rate": 1.4701348747591524e-05,
1588
+ "loss": 1.2348,
1589
+ "step": 1100
1590
+ },
1591
+ {
1592
+ "epoch": 0.26610475617098134,
1593
+ "grad_norm": 0.7531107664108276,
1594
+ "learning_rate": 1.4677263969171484e-05,
1595
+ "loss": 1.1702,
1596
+ "step": 1105
1597
+ },
1598
+ {
1599
+ "epoch": 0.267308850090307,
1600
+ "grad_norm": 0.7167130708694458,
1601
+ "learning_rate": 1.4653179190751445e-05,
1602
+ "loss": 1.1417,
1603
+ "step": 1110
1604
+ },
1605
+ {
1606
+ "epoch": 0.26851294400963277,
1607
+ "grad_norm": 0.7529981732368469,
1608
+ "learning_rate": 1.4629094412331407e-05,
1609
+ "loss": 1.3126,
1610
+ "step": 1115
1611
+ },
1612
+ {
1613
+ "epoch": 0.26971703792895846,
1614
+ "grad_norm": 0.7192036509513855,
1615
+ "learning_rate": 1.4605009633911368e-05,
1616
+ "loss": 1.2149,
1617
+ "step": 1120
1618
+ },
1619
+ {
1620
+ "epoch": 0.27092113184828415,
1621
+ "grad_norm": 0.7200658321380615,
1622
+ "learning_rate": 1.458092485549133e-05,
1623
+ "loss": 1.1958,
1624
+ "step": 1125
1625
+ },
1626
+ {
1627
+ "epoch": 0.2721252257676099,
1628
+ "grad_norm": 0.7063027024269104,
1629
+ "learning_rate": 1.4556840077071293e-05,
1630
+ "loss": 1.1266,
1631
+ "step": 1130
1632
+ },
1633
+ {
1634
+ "epoch": 0.2733293196869356,
1635
+ "grad_norm": 0.7414741516113281,
1636
+ "learning_rate": 1.4532755298651253e-05,
1637
+ "loss": 1.2046,
1638
+ "step": 1135
1639
+ },
1640
+ {
1641
+ "epoch": 0.27453341360626127,
1642
+ "grad_norm": 0.7233819365501404,
1643
+ "learning_rate": 1.4508670520231216e-05,
1644
+ "loss": 1.3376,
1645
+ "step": 1140
1646
+ },
1647
+ {
1648
+ "epoch": 0.275737507525587,
1649
+ "grad_norm": 0.737486720085144,
1650
+ "learning_rate": 1.4484585741811176e-05,
1651
+ "loss": 1.2529,
1652
+ "step": 1145
1653
+ },
1654
+ {
1655
+ "epoch": 0.2769416014449127,
1656
+ "grad_norm": 0.6850925087928772,
1657
+ "learning_rate": 1.4460500963391137e-05,
1658
+ "loss": 1.1784,
1659
+ "step": 1150
1660
+ },
1661
+ {
1662
+ "epoch": 0.2781456953642384,
1663
+ "grad_norm": 0.6742652058601379,
1664
+ "learning_rate": 1.4436416184971101e-05,
1665
+ "loss": 1.0412,
1666
+ "step": 1155
1667
+ },
1668
+ {
1669
+ "epoch": 0.27934978928356413,
1670
+ "grad_norm": 0.7630188465118408,
1671
+ "learning_rate": 1.4412331406551062e-05,
1672
+ "loss": 1.2421,
1673
+ "step": 1160
1674
+ },
1675
+ {
1676
+ "epoch": 0.2805538832028898,
1677
+ "grad_norm": 0.7039279341697693,
1678
+ "learning_rate": 1.4388246628131022e-05,
1679
+ "loss": 1.2076,
1680
+ "step": 1165
1681
+ },
1682
+ {
1683
+ "epoch": 0.2817579771222155,
1684
+ "grad_norm": 0.6968697905540466,
1685
+ "learning_rate": 1.4364161849710985e-05,
1686
+ "loss": 1.2709,
1687
+ "step": 1170
1688
+ },
1689
+ {
1690
+ "epoch": 0.28296207104154125,
1691
+ "grad_norm": 0.757686972618103,
1692
+ "learning_rate": 1.4340077071290945e-05,
1693
+ "loss": 1.2237,
1694
+ "step": 1175
1695
+ },
1696
+ {
1697
+ "epoch": 0.28416616496086694,
1698
+ "grad_norm": 0.70340895652771,
1699
+ "learning_rate": 1.4315992292870906e-05,
1700
+ "loss": 1.1979,
1701
+ "step": 1180
1702
+ },
1703
+ {
1704
+ "epoch": 0.28537025888019263,
1705
+ "grad_norm": 0.7078365683555603,
1706
+ "learning_rate": 1.4291907514450868e-05,
1707
+ "loss": 1.1801,
1708
+ "step": 1185
1709
+ },
1710
+ {
1711
+ "epoch": 0.2865743527995184,
1712
+ "grad_norm": 0.671120822429657,
1713
+ "learning_rate": 1.4267822736030829e-05,
1714
+ "loss": 1.0744,
1715
+ "step": 1190
1716
+ },
1717
+ {
1718
+ "epoch": 0.28777844671884406,
1719
+ "grad_norm": 0.6995368599891663,
1720
+ "learning_rate": 1.424373795761079e-05,
1721
+ "loss": 1.1057,
1722
+ "step": 1195
1723
+ },
1724
+ {
1725
+ "epoch": 0.2889825406381698,
1726
+ "grad_norm": 0.8009471297264099,
1727
+ "learning_rate": 1.4219653179190754e-05,
1728
+ "loss": 1.2852,
1729
+ "step": 1200
1730
+ },
1731
+ {
1732
+ "epoch": 0.2889825406381698,
1733
+ "eval_loss": 1.1969135999679565,
1734
+ "eval_runtime": 10.4824,
1735
+ "eval_samples_per_second": 9.54,
1736
+ "eval_steps_per_second": 1.24,
1737
+ "step": 1200
1738
+ },
1739
+ {
1740
+ "epoch": 0.2901866345574955,
1741
+ "grad_norm": 0.8135602474212646,
1742
+ "learning_rate": 1.4195568400770714e-05,
1743
+ "loss": 1.1475,
1744
+ "step": 1205
1745
+ },
1746
+ {
1747
+ "epoch": 0.2913907284768212,
1748
+ "grad_norm": 0.6879997849464417,
1749
+ "learning_rate": 1.4171483622350675e-05,
1750
+ "loss": 1.2054,
1751
+ "step": 1210
1752
+ },
1753
+ {
1754
+ "epoch": 0.2925948223961469,
1755
+ "grad_norm": 0.774534285068512,
1756
+ "learning_rate": 1.4147398843930637e-05,
1757
+ "loss": 1.1133,
1758
+ "step": 1215
1759
+ },
1760
+ {
1761
+ "epoch": 0.2937989163154726,
1762
+ "grad_norm": 0.7712222933769226,
1763
+ "learning_rate": 1.4123314065510598e-05,
1764
+ "loss": 1.2367,
1765
+ "step": 1220
1766
+ },
1767
+ {
1768
+ "epoch": 0.2950030102347983,
1769
+ "grad_norm": 0.7632959485054016,
1770
+ "learning_rate": 1.4099229287090558e-05,
1771
+ "loss": 1.2142,
1772
+ "step": 1225
1773
+ },
1774
+ {
1775
+ "epoch": 0.29620710415412405,
1776
+ "grad_norm": 0.6896396279335022,
1777
+ "learning_rate": 1.4075144508670523e-05,
1778
+ "loss": 1.2692,
1779
+ "step": 1230
1780
+ },
1781
+ {
1782
+ "epoch": 0.29741119807344973,
1783
+ "grad_norm": 0.7135001420974731,
1784
+ "learning_rate": 1.4051059730250483e-05,
1785
+ "loss": 1.1922,
1786
+ "step": 1235
1787
+ },
1788
+ {
1789
+ "epoch": 0.2986152919927754,
1790
+ "grad_norm": 0.7016868591308594,
1791
+ "learning_rate": 1.4026974951830444e-05,
1792
+ "loss": 1.2775,
1793
+ "step": 1240
1794
+ },
1795
+ {
1796
+ "epoch": 0.29981938591210117,
1797
+ "grad_norm": 0.7863327860832214,
1798
+ "learning_rate": 1.4002890173410406e-05,
1799
+ "loss": 1.193,
1800
+ "step": 1245
1801
+ },
1802
+ {
1803
+ "epoch": 0.30102347983142685,
1804
+ "grad_norm": 0.7253124713897705,
1805
+ "learning_rate": 1.3978805394990367e-05,
1806
+ "loss": 1.1557,
1807
+ "step": 1250
1808
+ },
1809
+ {
1810
+ "epoch": 0.30222757375075254,
1811
+ "grad_norm": 0.7214757204055786,
1812
+ "learning_rate": 1.3954720616570329e-05,
1813
+ "loss": 1.2212,
1814
+ "step": 1255
1815
+ },
1816
+ {
1817
+ "epoch": 0.3034316676700783,
1818
+ "grad_norm": 0.7036615014076233,
1819
+ "learning_rate": 1.393063583815029e-05,
1820
+ "loss": 1.3103,
1821
+ "step": 1260
1822
+ },
1823
+ {
1824
+ "epoch": 0.304635761589404,
1825
+ "grad_norm": 0.7152957320213318,
1826
+ "learning_rate": 1.390655105973025e-05,
1827
+ "loss": 1.3316,
1828
+ "step": 1265
1829
+ },
1830
+ {
1831
+ "epoch": 0.30583985550872966,
1832
+ "grad_norm": 0.7779828906059265,
1833
+ "learning_rate": 1.3882466281310214e-05,
1834
+ "loss": 1.2773,
1835
+ "step": 1270
1836
+ },
1837
+ {
1838
+ "epoch": 0.3070439494280554,
1839
+ "grad_norm": 0.7995044589042664,
1840
+ "learning_rate": 1.3858381502890175e-05,
1841
+ "loss": 1.297,
1842
+ "step": 1275
1843
+ },
1844
+ {
1845
+ "epoch": 0.3082480433473811,
1846
+ "grad_norm": 0.7020440697669983,
1847
+ "learning_rate": 1.3834296724470136e-05,
1848
+ "loss": 1.1593,
1849
+ "step": 1280
1850
+ },
1851
+ {
1852
+ "epoch": 0.3094521372667068,
1853
+ "grad_norm": 0.7052685022354126,
1854
+ "learning_rate": 1.3810211946050098e-05,
1855
+ "loss": 1.1415,
1856
+ "step": 1285
1857
+ },
1858
+ {
1859
+ "epoch": 0.3106562311860325,
1860
+ "grad_norm": 0.7097790837287903,
1861
+ "learning_rate": 1.3786127167630059e-05,
1862
+ "loss": 1.0992,
1863
+ "step": 1290
1864
+ },
1865
+ {
1866
+ "epoch": 0.3118603251053582,
1867
+ "grad_norm": 0.7684621214866638,
1868
+ "learning_rate": 1.376204238921002e-05,
1869
+ "loss": 1.2256,
1870
+ "step": 1295
1871
+ },
1872
+ {
1873
+ "epoch": 0.3130644190246839,
1874
+ "grad_norm": 0.746864914894104,
1875
+ "learning_rate": 1.3737957610789983e-05,
1876
+ "loss": 1.2155,
1877
+ "step": 1300
1878
+ },
1879
+ {
1880
+ "epoch": 0.31426851294400965,
1881
+ "grad_norm": 0.733629584312439,
1882
+ "learning_rate": 1.3713872832369944e-05,
1883
+ "loss": 1.1287,
1884
+ "step": 1305
1885
+ },
1886
+ {
1887
+ "epoch": 0.31547260686333534,
1888
+ "grad_norm": 0.6500353217124939,
1889
+ "learning_rate": 1.3689788053949905e-05,
1890
+ "loss": 1.267,
1891
+ "step": 1310
1892
+ },
1893
+ {
1894
+ "epoch": 0.316676700782661,
1895
+ "grad_norm": 0.7301894426345825,
1896
+ "learning_rate": 1.3665703275529867e-05,
1897
+ "loss": 1.2167,
1898
+ "step": 1315
1899
+ },
1900
+ {
1901
+ "epoch": 0.31788079470198677,
1902
+ "grad_norm": 0.677039623260498,
1903
+ "learning_rate": 1.3641618497109828e-05,
1904
+ "loss": 1.1037,
1905
+ "step": 1320
1906
+ },
1907
+ {
1908
+ "epoch": 0.31908488862131246,
1909
+ "grad_norm": 0.7031332850456238,
1910
+ "learning_rate": 1.3617533718689788e-05,
1911
+ "loss": 1.1252,
1912
+ "step": 1325
1913
+ },
1914
+ {
1915
+ "epoch": 0.32028898254063815,
1916
+ "grad_norm": 0.8084255456924438,
1917
+ "learning_rate": 1.359344894026975e-05,
1918
+ "loss": 1.1728,
1919
+ "step": 1330
1920
+ },
1921
+ {
1922
+ "epoch": 0.3214930764599639,
1923
+ "grad_norm": 0.6915790438652039,
1924
+ "learning_rate": 1.3569364161849711e-05,
1925
+ "loss": 1.1552,
1926
+ "step": 1335
1927
+ },
1928
+ {
1929
+ "epoch": 0.3226971703792896,
1930
+ "grad_norm": 0.742514431476593,
1931
+ "learning_rate": 1.3545279383429672e-05,
1932
+ "loss": 1.1406,
1933
+ "step": 1340
1934
+ },
1935
+ {
1936
+ "epoch": 0.32390126429861527,
1937
+ "grad_norm": 0.6682133674621582,
1938
+ "learning_rate": 1.3521194605009636e-05,
1939
+ "loss": 1.1642,
1940
+ "step": 1345
1941
+ },
1942
+ {
1943
+ "epoch": 0.325105358217941,
1944
+ "grad_norm": 0.7231415510177612,
1945
+ "learning_rate": 1.3497109826589597e-05,
1946
+ "loss": 1.253,
1947
+ "step": 1350
1948
+ },
1949
+ {
1950
+ "epoch": 0.3263094521372667,
1951
+ "grad_norm": 0.7634385824203491,
1952
+ "learning_rate": 1.3473025048169559e-05,
1953
+ "loss": 1.2276,
1954
+ "step": 1355
1955
+ },
1956
+ {
1957
+ "epoch": 0.32751354605659244,
1958
+ "grad_norm": 0.6695632934570312,
1959
+ "learning_rate": 1.344894026974952e-05,
1960
+ "loss": 1.1864,
1961
+ "step": 1360
1962
+ },
1963
+ {
1964
+ "epoch": 0.32871763997591813,
1965
+ "grad_norm": 0.7134613394737244,
1966
+ "learning_rate": 1.342485549132948e-05,
1967
+ "loss": 1.1097,
1968
+ "step": 1365
1969
+ },
1970
+ {
1971
+ "epoch": 0.3299217338952438,
1972
+ "grad_norm": 0.6643304228782654,
1973
+ "learning_rate": 1.3400770712909442e-05,
1974
+ "loss": 1.2438,
1975
+ "step": 1370
1976
+ },
1977
+ {
1978
+ "epoch": 0.33112582781456956,
1979
+ "grad_norm": 0.6821620464324951,
1980
+ "learning_rate": 1.3376685934489405e-05,
1981
+ "loss": 1.2137,
1982
+ "step": 1375
1983
+ },
1984
+ {
1985
+ "epoch": 0.33232992173389525,
1986
+ "grad_norm": 0.7436670660972595,
1987
+ "learning_rate": 1.3352601156069365e-05,
1988
+ "loss": 1.0745,
1989
+ "step": 1380
1990
+ },
1991
+ {
1992
+ "epoch": 0.33353401565322094,
1993
+ "grad_norm": 0.7283557653427124,
1994
+ "learning_rate": 1.3328516377649328e-05,
1995
+ "loss": 1.04,
1996
+ "step": 1385
1997
+ },
1998
+ {
1999
+ "epoch": 0.3347381095725467,
2000
+ "grad_norm": 0.6368799805641174,
2001
+ "learning_rate": 1.3304431599229288e-05,
2002
+ "loss": 1.1829,
2003
+ "step": 1390
2004
+ },
2005
+ {
2006
+ "epoch": 0.33594220349187237,
2007
+ "grad_norm": 0.8481898307800293,
2008
+ "learning_rate": 1.3280346820809249e-05,
2009
+ "loss": 1.0901,
2010
+ "step": 1395
2011
+ },
2012
+ {
2013
+ "epoch": 0.33714629741119806,
2014
+ "grad_norm": 0.7311575412750244,
2015
+ "learning_rate": 1.3256262042389211e-05,
2016
+ "loss": 1.1135,
2017
+ "step": 1400
2018
+ },
2019
+ {
2020
+ "epoch": 0.33714629741119806,
2021
+ "eval_loss": 1.1897846460342407,
2022
+ "eval_runtime": 10.4886,
2023
+ "eval_samples_per_second": 9.534,
2024
+ "eval_steps_per_second": 1.239,
2025
+ "step": 1400
2026
+ },
2027
+ {
2028
+ "epoch": 0.3383503913305238,
2029
+ "grad_norm": 0.7002428770065308,
2030
+ "learning_rate": 1.3232177263969172e-05,
2031
+ "loss": 1.1384,
2032
+ "step": 1405
2033
+ },
2034
+ {
2035
+ "epoch": 0.3395544852498495,
2036
+ "grad_norm": 0.7509122490882874,
2037
+ "learning_rate": 1.3208092485549133e-05,
2038
+ "loss": 1.1667,
2039
+ "step": 1410
2040
+ },
2041
+ {
2042
+ "epoch": 0.3407585791691752,
2043
+ "grad_norm": 0.7156943082809448,
2044
+ "learning_rate": 1.3184007707129097e-05,
2045
+ "loss": 1.0665,
2046
+ "step": 1415
2047
+ },
2048
+ {
2049
+ "epoch": 0.3419626730885009,
2050
+ "grad_norm": 0.8622729182243347,
2051
+ "learning_rate": 1.3159922928709057e-05,
2052
+ "loss": 1.1503,
2053
+ "step": 1420
2054
+ },
2055
+ {
2056
+ "epoch": 0.3431667670078266,
2057
+ "grad_norm": 0.7578678131103516,
2058
+ "learning_rate": 1.3135838150289018e-05,
2059
+ "loss": 1.0761,
2060
+ "step": 1425
2061
+ },
2062
+ {
2063
+ "epoch": 0.3443708609271523,
2064
+ "grad_norm": 0.7222859263420105,
2065
+ "learning_rate": 1.311175337186898e-05,
2066
+ "loss": 1.2214,
2067
+ "step": 1430
2068
+ },
2069
+ {
2070
+ "epoch": 0.34557495484647804,
2071
+ "grad_norm": 0.7335066795349121,
2072
+ "learning_rate": 1.3087668593448941e-05,
2073
+ "loss": 1.1857,
2074
+ "step": 1435
2075
+ },
2076
+ {
2077
+ "epoch": 0.34677904876580373,
2078
+ "grad_norm": 0.6828716397285461,
2079
+ "learning_rate": 1.3063583815028902e-05,
2080
+ "loss": 1.2078,
2081
+ "step": 1440
2082
+ },
2083
+ {
2084
+ "epoch": 0.3479831426851294,
2085
+ "grad_norm": 0.8298027515411377,
2086
+ "learning_rate": 1.3039499036608864e-05,
2087
+ "loss": 1.2375,
2088
+ "step": 1445
2089
+ },
2090
+ {
2091
+ "epoch": 0.34918723660445516,
2092
+ "grad_norm": 0.7282932996749878,
2093
+ "learning_rate": 1.3015414258188825e-05,
2094
+ "loss": 1.1631,
2095
+ "step": 1450
2096
+ },
2097
+ {
2098
+ "epoch": 0.35039133052378085,
2099
+ "grad_norm": 0.6626558303833008,
2100
+ "learning_rate": 1.2991329479768787e-05,
2101
+ "loss": 1.1374,
2102
+ "step": 1455
2103
+ },
2104
+ {
2105
+ "epoch": 0.35159542444310654,
2106
+ "grad_norm": 0.8166629672050476,
2107
+ "learning_rate": 1.296724470134875e-05,
2108
+ "loss": 1.2521,
2109
+ "step": 1460
2110
+ },
2111
+ {
2112
+ "epoch": 0.3527995183624323,
2113
+ "grad_norm": 0.688026487827301,
2114
+ "learning_rate": 1.294315992292871e-05,
2115
+ "loss": 1.1008,
2116
+ "step": 1465
2117
+ },
2118
+ {
2119
+ "epoch": 0.354003612281758,
2120
+ "grad_norm": 0.7461696267127991,
2121
+ "learning_rate": 1.2919075144508672e-05,
2122
+ "loss": 1.0932,
2123
+ "step": 1470
2124
+ },
2125
+ {
2126
+ "epoch": 0.35520770620108366,
2127
+ "grad_norm": 0.7331961989402771,
2128
+ "learning_rate": 1.2894990366088633e-05,
2129
+ "loss": 1.0613,
2130
+ "step": 1475
2131
+ },
2132
+ {
2133
+ "epoch": 0.3564118001204094,
2134
+ "grad_norm": 0.7345579862594604,
2135
+ "learning_rate": 1.2870905587668594e-05,
2136
+ "loss": 1.3048,
2137
+ "step": 1480
2138
+ },
2139
+ {
2140
+ "epoch": 0.3576158940397351,
2141
+ "grad_norm": 0.8133093118667603,
2142
+ "learning_rate": 1.2846820809248558e-05,
2143
+ "loss": 1.2435,
2144
+ "step": 1485
2145
+ },
2146
+ {
2147
+ "epoch": 0.3588199879590608,
2148
+ "grad_norm": 0.7404915690422058,
2149
+ "learning_rate": 1.2822736030828518e-05,
2150
+ "loss": 1.1767,
2151
+ "step": 1490
2152
+ },
2153
+ {
2154
+ "epoch": 0.3600240818783865,
2155
+ "grad_norm": 0.8186049461364746,
2156
+ "learning_rate": 1.2798651252408479e-05,
2157
+ "loss": 1.0789,
2158
+ "step": 1495
2159
+ },
2160
+ {
2161
+ "epoch": 0.3612281757977122,
2162
+ "grad_norm": 0.8047255277633667,
2163
+ "learning_rate": 1.2774566473988441e-05,
2164
+ "loss": 1.1343,
2165
+ "step": 1500
2166
+ }
2167
+ ],
2168
+ "logging_steps": 5,
2169
+ "max_steps": 4152,
2170
+ "num_input_tokens_seen": 0,
2171
+ "num_train_epochs": 1,
2172
+ "save_steps": 500,
2173
+ "stateful_callbacks": {
2174
+ "TrainerControl": {
2175
+ "args": {
2176
+ "should_epoch_stop": false,
2177
+ "should_evaluate": false,
2178
+ "should_log": false,
2179
+ "should_save": true,
2180
+ "should_training_stop": false
2181
+ },
2182
+ "attributes": {}
2183
+ }
2184
+ },
2185
+ "total_flos": 7331181236669952.0,
2186
+ "train_batch_size": 1,
2187
+ "trial_name": null,
2188
+ "trial_params": null
2189
+ }
checkpoint-1500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86602ecfa5f4389c74519a3cbca0206cb3e9a395b2bd494b341ef1183c4e9454
3
+ size 5816
checkpoint-1500/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-2000/config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "ewre324/ewre324-Thinker-SmolLM2-135M-Instruct-Reasoning",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "head_dim": 64,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 576,
13
+ "initializer_range": 0.041666666666666664,
14
+ "intermediate_size": 1536,
15
+ "is_llama_config": true,
16
+ "max_position_embeddings": 8192,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "num_attention_heads": 9,
20
+ "num_hidden_layers": 30,
21
+ "num_key_value_heads": 3,
22
+ "pad_token_id": 2,
23
+ "pretraining_tp": 1,
24
+ "rms_norm_eps": 1e-05,
25
+ "rope_interleaved": false,
26
+ "rope_scaling": null,
27
+ "rope_theta": 100000,
28
+ "tie_word_embeddings": true,
29
+ "torch_dtype": "float32",
30
+ "transformers.js_config": {
31
+ "kv_cache_dtype": {
32
+ "fp16": "float16",
33
+ "q4f16": "float16"
34
+ }
35
+ },
36
+ "transformers_version": "4.49.0.dev0",
37
+ "use_cache": false,
38
+ "vocab_size": 49152
39
+ }
checkpoint-2000/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 2,
6
+ "transformers_version": "4.49.0.dev0"
7
+ }
checkpoint-2000/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-2000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f76c3be189d0fbc24247515de8ddfcacdf3152495192477e2f06ce3d583904a5
3
+ size 538090408
checkpoint-2000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:554127328c47f0b87512ab17ded0cf0957e5f3c358936ef9ab2b29777902af58
3
+ size 1076349050
checkpoint-2000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5edb34d031c0c2b447f3eaadb401a4c1e7e7e6d8c096e28b7092e01a8bd48c92
3
+ size 14244
checkpoint-2000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a6a91dbc2ee0881caca567a1c7d4bda579188bc179e9e0441bb8880b35c4bd4
3
+ size 1064
checkpoint-2000/special_tokens_map.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>"
5
+ ],
6
+ "bos_token": {
7
+ "content": "<|im_start|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false
12
+ },
13
+ "eos_token": {
14
+ "content": "<|im_end|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ },
20
+ "pad_token": "<|im_end|>",
21
+ "unk_token": {
22
+ "content": "<|endoftext|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false
27
+ }
28
+ }
checkpoint-2000/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-2000/tokenizer_config.json ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "0": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "1": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "2": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "3": {
29
+ "content": "<repo_name>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "4": {
37
+ "content": "<reponame>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "5": {
45
+ "content": "<file_sep>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "6": {
53
+ "content": "<filename>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "7": {
61
+ "content": "<gh_stars>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "8": {
69
+ "content": "<issue_start>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "9": {
77
+ "content": "<issue_comment>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "10": {
85
+ "content": "<issue_closed>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "11": {
93
+ "content": "<jupyter_start>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "12": {
101
+ "content": "<jupyter_text>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "13": {
109
+ "content": "<jupyter_code>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "14": {
117
+ "content": "<jupyter_output>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": true
123
+ },
124
+ "15": {
125
+ "content": "<jupyter_script>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": true
131
+ },
132
+ "16": {
133
+ "content": "<empty_output>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": true
139
+ }
140
+ },
141
+ "additional_special_tokens": [
142
+ "<|im_start|>",
143
+ "<|im_end|>"
144
+ ],
145
+ "bos_token": "<|im_start|>",
146
+ "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
147
+ "clean_up_tokenization_spaces": false,
148
+ "eos_token": "<|im_end|>",
149
+ "extra_special_tokens": {},
150
+ "max_length": 2048,
151
+ "model_max_length": 2048,
152
+ "pad_to_multiple_of": null,
153
+ "pad_token": "<|im_end|>",
154
+ "pad_token_type_id": 0,
155
+ "padding_side": "right",
156
+ "stride": 0,
157
+ "tokenizer_class": "GPT2Tokenizer",
158
+ "truncation_side": "right",
159
+ "truncation_strategy": "longest_first",
160
+ "unk_token": "<|endoftext|>",
161
+ "vocab_size": 49152
162
+ }
checkpoint-2000/trainer_state.json ADDED
@@ -0,0 +1,2913 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.481637567730283,
5
+ "eval_steps": 200,
6
+ "global_step": 2000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0012040939193257074,
13
+ "grad_norm": 1.2087849378585815,
14
+ "learning_rate": 1.9975915221579964e-05,
15
+ "loss": 1.7501,
16
+ "step": 5
17
+ },
18
+ {
19
+ "epoch": 0.002408187838651415,
20
+ "grad_norm": 1.0137544870376587,
21
+ "learning_rate": 1.9951830443159926e-05,
22
+ "loss": 1.3294,
23
+ "step": 10
24
+ },
25
+ {
26
+ "epoch": 0.003612281757977122,
27
+ "grad_norm": 0.8906710147857666,
28
+ "learning_rate": 1.9927745664739885e-05,
29
+ "loss": 1.4793,
30
+ "step": 15
31
+ },
32
+ {
33
+ "epoch": 0.00481637567730283,
34
+ "grad_norm": 0.8175792098045349,
35
+ "learning_rate": 1.9903660886319848e-05,
36
+ "loss": 1.4537,
37
+ "step": 20
38
+ },
39
+ {
40
+ "epoch": 0.006020469596628537,
41
+ "grad_norm": 0.8326422572135925,
42
+ "learning_rate": 1.987957610789981e-05,
43
+ "loss": 1.5579,
44
+ "step": 25
45
+ },
46
+ {
47
+ "epoch": 0.007224563515954244,
48
+ "grad_norm": 0.8402993083000183,
49
+ "learning_rate": 1.985549132947977e-05,
50
+ "loss": 1.4146,
51
+ "step": 30
52
+ },
53
+ {
54
+ "epoch": 0.008428657435279952,
55
+ "grad_norm": 0.9211377501487732,
56
+ "learning_rate": 1.983140655105973e-05,
57
+ "loss": 1.4986,
58
+ "step": 35
59
+ },
60
+ {
61
+ "epoch": 0.00963275135460566,
62
+ "grad_norm": 0.7729477286338806,
63
+ "learning_rate": 1.9807321772639694e-05,
64
+ "loss": 1.3949,
65
+ "step": 40
66
+ },
67
+ {
68
+ "epoch": 0.010836845273931367,
69
+ "grad_norm": 0.8256559371948242,
70
+ "learning_rate": 1.9783236994219656e-05,
71
+ "loss": 1.2517,
72
+ "step": 45
73
+ },
74
+ {
75
+ "epoch": 0.012040939193257074,
76
+ "grad_norm": 0.7404232025146484,
77
+ "learning_rate": 1.9759152215799615e-05,
78
+ "loss": 1.4761,
79
+ "step": 50
80
+ },
81
+ {
82
+ "epoch": 0.013245033112582781,
83
+ "grad_norm": 0.773309588432312,
84
+ "learning_rate": 1.9735067437379577e-05,
85
+ "loss": 1.227,
86
+ "step": 55
87
+ },
88
+ {
89
+ "epoch": 0.014449127031908489,
90
+ "grad_norm": 0.7535562515258789,
91
+ "learning_rate": 1.971098265895954e-05,
92
+ "loss": 1.3443,
93
+ "step": 60
94
+ },
95
+ {
96
+ "epoch": 0.015653220951234198,
97
+ "grad_norm": 0.8561655282974243,
98
+ "learning_rate": 1.9686897880539502e-05,
99
+ "loss": 1.3229,
100
+ "step": 65
101
+ },
102
+ {
103
+ "epoch": 0.016857314870559904,
104
+ "grad_norm": 0.7888396978378296,
105
+ "learning_rate": 1.966281310211946e-05,
106
+ "loss": 1.4274,
107
+ "step": 70
108
+ },
109
+ {
110
+ "epoch": 0.018061408789885613,
111
+ "grad_norm": 0.7886033654212952,
112
+ "learning_rate": 1.9638728323699423e-05,
113
+ "loss": 1.4469,
114
+ "step": 75
115
+ },
116
+ {
117
+ "epoch": 0.01926550270921132,
118
+ "grad_norm": 0.8100034594535828,
119
+ "learning_rate": 1.9614643545279385e-05,
120
+ "loss": 1.314,
121
+ "step": 80
122
+ },
123
+ {
124
+ "epoch": 0.020469596628537028,
125
+ "grad_norm": 0.7786366939544678,
126
+ "learning_rate": 1.9590558766859348e-05,
127
+ "loss": 1.2631,
128
+ "step": 85
129
+ },
130
+ {
131
+ "epoch": 0.021673690547862733,
132
+ "grad_norm": 0.8259329795837402,
133
+ "learning_rate": 1.9566473988439307e-05,
134
+ "loss": 1.4006,
135
+ "step": 90
136
+ },
137
+ {
138
+ "epoch": 0.022877784467188442,
139
+ "grad_norm": 0.7402159571647644,
140
+ "learning_rate": 1.954238921001927e-05,
141
+ "loss": 1.416,
142
+ "step": 95
143
+ },
144
+ {
145
+ "epoch": 0.024081878386514148,
146
+ "grad_norm": 0.8619882464408875,
147
+ "learning_rate": 1.951830443159923e-05,
148
+ "loss": 1.2583,
149
+ "step": 100
150
+ },
151
+ {
152
+ "epoch": 0.025285972305839857,
153
+ "grad_norm": 0.7436647415161133,
154
+ "learning_rate": 1.949421965317919e-05,
155
+ "loss": 1.3527,
156
+ "step": 105
157
+ },
158
+ {
159
+ "epoch": 0.026490066225165563,
160
+ "grad_norm": 0.7261799573898315,
161
+ "learning_rate": 1.9470134874759156e-05,
162
+ "loss": 1.4234,
163
+ "step": 110
164
+ },
165
+ {
166
+ "epoch": 0.027694160144491272,
167
+ "grad_norm": 0.8648309707641602,
168
+ "learning_rate": 1.9446050096339115e-05,
169
+ "loss": 1.2541,
170
+ "step": 115
171
+ },
172
+ {
173
+ "epoch": 0.028898254063816978,
174
+ "grad_norm": 0.7091889381408691,
175
+ "learning_rate": 1.9421965317919077e-05,
176
+ "loss": 1.2473,
177
+ "step": 120
178
+ },
179
+ {
180
+ "epoch": 0.030102347983142687,
181
+ "grad_norm": 0.7450091242790222,
182
+ "learning_rate": 1.939788053949904e-05,
183
+ "loss": 1.2555,
184
+ "step": 125
185
+ },
186
+ {
187
+ "epoch": 0.031306441902468396,
188
+ "grad_norm": 0.8190895318984985,
189
+ "learning_rate": 1.9373795761079e-05,
190
+ "loss": 1.2874,
191
+ "step": 130
192
+ },
193
+ {
194
+ "epoch": 0.0325105358217941,
195
+ "grad_norm": 0.7458487749099731,
196
+ "learning_rate": 1.934971098265896e-05,
197
+ "loss": 1.4139,
198
+ "step": 135
199
+ },
200
+ {
201
+ "epoch": 0.03371462974111981,
202
+ "grad_norm": 0.806698203086853,
203
+ "learning_rate": 1.9325626204238923e-05,
204
+ "loss": 1.4061,
205
+ "step": 140
206
+ },
207
+ {
208
+ "epoch": 0.034918723660445516,
209
+ "grad_norm": 0.7941407561302185,
210
+ "learning_rate": 1.9301541425818882e-05,
211
+ "loss": 1.2819,
212
+ "step": 145
213
+ },
214
+ {
215
+ "epoch": 0.036122817579771226,
216
+ "grad_norm": 0.6824349164962769,
217
+ "learning_rate": 1.9277456647398845e-05,
218
+ "loss": 1.3025,
219
+ "step": 150
220
+ },
221
+ {
222
+ "epoch": 0.03732691149909693,
223
+ "grad_norm": 0.8056986927986145,
224
+ "learning_rate": 1.9253371868978807e-05,
225
+ "loss": 1.1617,
226
+ "step": 155
227
+ },
228
+ {
229
+ "epoch": 0.03853100541842264,
230
+ "grad_norm": 0.7486716508865356,
231
+ "learning_rate": 1.922928709055877e-05,
232
+ "loss": 1.3523,
233
+ "step": 160
234
+ },
235
+ {
236
+ "epoch": 0.039735099337748346,
237
+ "grad_norm": 0.7773532867431641,
238
+ "learning_rate": 1.920520231213873e-05,
239
+ "loss": 1.3013,
240
+ "step": 165
241
+ },
242
+ {
243
+ "epoch": 0.040939193257074055,
244
+ "grad_norm": 0.7498481869697571,
245
+ "learning_rate": 1.918111753371869e-05,
246
+ "loss": 1.2563,
247
+ "step": 170
248
+ },
249
+ {
250
+ "epoch": 0.04214328717639976,
251
+ "grad_norm": 0.7634468078613281,
252
+ "learning_rate": 1.9157032755298653e-05,
253
+ "loss": 1.2365,
254
+ "step": 175
255
+ },
256
+ {
257
+ "epoch": 0.04334738109572547,
258
+ "grad_norm": 0.6561328172683716,
259
+ "learning_rate": 1.9132947976878615e-05,
260
+ "loss": 1.3058,
261
+ "step": 180
262
+ },
263
+ {
264
+ "epoch": 0.044551475015051176,
265
+ "grad_norm": 0.7443628907203674,
266
+ "learning_rate": 1.9108863198458578e-05,
267
+ "loss": 1.3264,
268
+ "step": 185
269
+ },
270
+ {
271
+ "epoch": 0.045755568934376885,
272
+ "grad_norm": 0.6697310209274292,
273
+ "learning_rate": 1.9084778420038536e-05,
274
+ "loss": 1.1029,
275
+ "step": 190
276
+ },
277
+ {
278
+ "epoch": 0.04695966285370259,
279
+ "grad_norm": 0.7119889259338379,
280
+ "learning_rate": 1.90606936416185e-05,
281
+ "loss": 1.2503,
282
+ "step": 195
283
+ },
284
+ {
285
+ "epoch": 0.048163756773028296,
286
+ "grad_norm": 0.8120700120925903,
287
+ "learning_rate": 1.903660886319846e-05,
288
+ "loss": 1.246,
289
+ "step": 200
290
+ },
291
+ {
292
+ "epoch": 0.048163756773028296,
293
+ "eval_loss": 1.2944953441619873,
294
+ "eval_runtime": 10.5151,
295
+ "eval_samples_per_second": 9.51,
296
+ "eval_steps_per_second": 1.236,
297
+ "step": 200
298
+ },
299
+ {
300
+ "epoch": 0.049367850692354005,
301
+ "grad_norm": 0.7876719832420349,
302
+ "learning_rate": 1.901252408477842e-05,
303
+ "loss": 1.2097,
304
+ "step": 205
305
+ },
306
+ {
307
+ "epoch": 0.050571944611679714,
308
+ "grad_norm": 0.7367354035377502,
309
+ "learning_rate": 1.8988439306358382e-05,
310
+ "loss": 1.3332,
311
+ "step": 210
312
+ },
313
+ {
314
+ "epoch": 0.05177603853100542,
315
+ "grad_norm": 0.7384558320045471,
316
+ "learning_rate": 1.8964354527938345e-05,
317
+ "loss": 1.3897,
318
+ "step": 215
319
+ },
320
+ {
321
+ "epoch": 0.052980132450331126,
322
+ "grad_norm": 0.8261380195617676,
323
+ "learning_rate": 1.8940269749518304e-05,
324
+ "loss": 1.2677,
325
+ "step": 220
326
+ },
327
+ {
328
+ "epoch": 0.054184226369656835,
329
+ "grad_norm": 0.7237825989723206,
330
+ "learning_rate": 1.891618497109827e-05,
331
+ "loss": 1.2919,
332
+ "step": 225
333
+ },
334
+ {
335
+ "epoch": 0.055388320288982544,
336
+ "grad_norm": 0.7296762466430664,
337
+ "learning_rate": 1.889210019267823e-05,
338
+ "loss": 1.2977,
339
+ "step": 230
340
+ },
341
+ {
342
+ "epoch": 0.056592414208308246,
343
+ "grad_norm": 0.8224279880523682,
344
+ "learning_rate": 1.886801541425819e-05,
345
+ "loss": 1.3751,
346
+ "step": 235
347
+ },
348
+ {
349
+ "epoch": 0.057796508127633955,
350
+ "grad_norm": 0.7233418822288513,
351
+ "learning_rate": 1.8843930635838153e-05,
352
+ "loss": 1.293,
353
+ "step": 240
354
+ },
355
+ {
356
+ "epoch": 0.059000602046959665,
357
+ "grad_norm": 0.7015657424926758,
358
+ "learning_rate": 1.8819845857418112e-05,
359
+ "loss": 1.3093,
360
+ "step": 245
361
+ },
362
+ {
363
+ "epoch": 0.060204695966285374,
364
+ "grad_norm": 0.8122280240058899,
365
+ "learning_rate": 1.8795761078998074e-05,
366
+ "loss": 1.3979,
367
+ "step": 250
368
+ },
369
+ {
370
+ "epoch": 0.061408789885611076,
371
+ "grad_norm": 0.7582533955574036,
372
+ "learning_rate": 1.8771676300578037e-05,
373
+ "loss": 1.3147,
374
+ "step": 255
375
+ },
376
+ {
377
+ "epoch": 0.06261288380493679,
378
+ "grad_norm": 0.746967077255249,
379
+ "learning_rate": 1.8747591522158e-05,
380
+ "loss": 1.1785,
381
+ "step": 260
382
+ },
383
+ {
384
+ "epoch": 0.0638169777242625,
385
+ "grad_norm": 0.7588161826133728,
386
+ "learning_rate": 1.8723506743737958e-05,
387
+ "loss": 1.3648,
388
+ "step": 265
389
+ },
390
+ {
391
+ "epoch": 0.0650210716435882,
392
+ "grad_norm": 0.696869432926178,
393
+ "learning_rate": 1.869942196531792e-05,
394
+ "loss": 1.2854,
395
+ "step": 270
396
+ },
397
+ {
398
+ "epoch": 0.06622516556291391,
399
+ "grad_norm": 0.8297455906867981,
400
+ "learning_rate": 1.8675337186897883e-05,
401
+ "loss": 1.2011,
402
+ "step": 275
403
+ },
404
+ {
405
+ "epoch": 0.06742925948223961,
406
+ "grad_norm": 0.7183229327201843,
407
+ "learning_rate": 1.8651252408477845e-05,
408
+ "loss": 1.1944,
409
+ "step": 280
410
+ },
411
+ {
412
+ "epoch": 0.06863335340156532,
413
+ "grad_norm": 0.7092033624649048,
414
+ "learning_rate": 1.8627167630057804e-05,
415
+ "loss": 1.3296,
416
+ "step": 285
417
+ },
418
+ {
419
+ "epoch": 0.06983744732089103,
420
+ "grad_norm": 0.7024344801902771,
421
+ "learning_rate": 1.8603082851637766e-05,
422
+ "loss": 1.2349,
423
+ "step": 290
424
+ },
425
+ {
426
+ "epoch": 0.07104154124021674,
427
+ "grad_norm": 0.7504032254219055,
428
+ "learning_rate": 1.857899807321773e-05,
429
+ "loss": 1.1986,
430
+ "step": 295
431
+ },
432
+ {
433
+ "epoch": 0.07224563515954245,
434
+ "grad_norm": 0.7584034204483032,
435
+ "learning_rate": 1.855491329479769e-05,
436
+ "loss": 1.2998,
437
+ "step": 300
438
+ },
439
+ {
440
+ "epoch": 0.07344972907886815,
441
+ "grad_norm": 0.7563251852989197,
442
+ "learning_rate": 1.853082851637765e-05,
443
+ "loss": 1.3703,
444
+ "step": 305
445
+ },
446
+ {
447
+ "epoch": 0.07465382299819386,
448
+ "grad_norm": 0.728660523891449,
449
+ "learning_rate": 1.8506743737957612e-05,
450
+ "loss": 1.2916,
451
+ "step": 310
452
+ },
453
+ {
454
+ "epoch": 0.07585791691751957,
455
+ "grad_norm": 0.6596492528915405,
456
+ "learning_rate": 1.8482658959537575e-05,
457
+ "loss": 1.1821,
458
+ "step": 315
459
+ },
460
+ {
461
+ "epoch": 0.07706201083684527,
462
+ "grad_norm": 0.7137461304664612,
463
+ "learning_rate": 1.8458574181117533e-05,
464
+ "loss": 1.1982,
465
+ "step": 320
466
+ },
467
+ {
468
+ "epoch": 0.07826610475617098,
469
+ "grad_norm": 0.7411230206489563,
470
+ "learning_rate": 1.8434489402697496e-05,
471
+ "loss": 1.2711,
472
+ "step": 325
473
+ },
474
+ {
475
+ "epoch": 0.07947019867549669,
476
+ "grad_norm": 0.7566368579864502,
477
+ "learning_rate": 1.8410404624277458e-05,
478
+ "loss": 1.1603,
479
+ "step": 330
480
+ },
481
+ {
482
+ "epoch": 0.0806742925948224,
483
+ "grad_norm": 0.7235690355300903,
484
+ "learning_rate": 1.838631984585742e-05,
485
+ "loss": 1.1993,
486
+ "step": 335
487
+ },
488
+ {
489
+ "epoch": 0.08187838651414811,
490
+ "grad_norm": 0.7648451328277588,
491
+ "learning_rate": 1.8362235067437383e-05,
492
+ "loss": 1.134,
493
+ "step": 340
494
+ },
495
+ {
496
+ "epoch": 0.08308248043347381,
497
+ "grad_norm": 0.740710973739624,
498
+ "learning_rate": 1.8338150289017342e-05,
499
+ "loss": 1.1533,
500
+ "step": 345
501
+ },
502
+ {
503
+ "epoch": 0.08428657435279951,
504
+ "grad_norm": 0.7610028982162476,
505
+ "learning_rate": 1.8314065510597304e-05,
506
+ "loss": 1.3126,
507
+ "step": 350
508
+ },
509
+ {
510
+ "epoch": 0.08549066827212523,
511
+ "grad_norm": 0.6956422924995422,
512
+ "learning_rate": 1.8289980732177266e-05,
513
+ "loss": 1.2272,
514
+ "step": 355
515
+ },
516
+ {
517
+ "epoch": 0.08669476219145093,
518
+ "grad_norm": 0.6914700269699097,
519
+ "learning_rate": 1.8265895953757225e-05,
520
+ "loss": 1.3049,
521
+ "step": 360
522
+ },
523
+ {
524
+ "epoch": 0.08789885611077664,
525
+ "grad_norm": 0.7015173435211182,
526
+ "learning_rate": 1.8241811175337188e-05,
527
+ "loss": 1.3569,
528
+ "step": 365
529
+ },
530
+ {
531
+ "epoch": 0.08910295003010235,
532
+ "grad_norm": 0.6985088586807251,
533
+ "learning_rate": 1.821772639691715e-05,
534
+ "loss": 1.2307,
535
+ "step": 370
536
+ },
537
+ {
538
+ "epoch": 0.09030704394942805,
539
+ "grad_norm": 0.7313199043273926,
540
+ "learning_rate": 1.8193641618497112e-05,
541
+ "loss": 1.1783,
542
+ "step": 375
543
+ },
544
+ {
545
+ "epoch": 0.09151113786875377,
546
+ "grad_norm": 0.7657830715179443,
547
+ "learning_rate": 1.8169556840077075e-05,
548
+ "loss": 1.2228,
549
+ "step": 380
550
+ },
551
+ {
552
+ "epoch": 0.09271523178807947,
553
+ "grad_norm": 0.8051377534866333,
554
+ "learning_rate": 1.8145472061657034e-05,
555
+ "loss": 1.25,
556
+ "step": 385
557
+ },
558
+ {
559
+ "epoch": 0.09391932570740517,
560
+ "grad_norm": 0.8114392161369324,
561
+ "learning_rate": 1.8121387283236996e-05,
562
+ "loss": 1.2347,
563
+ "step": 390
564
+ },
565
+ {
566
+ "epoch": 0.09512341962673089,
567
+ "grad_norm": 0.7304120659828186,
568
+ "learning_rate": 1.809730250481696e-05,
569
+ "loss": 1.2437,
570
+ "step": 395
571
+ },
572
+ {
573
+ "epoch": 0.09632751354605659,
574
+ "grad_norm": 0.8418581485748291,
575
+ "learning_rate": 1.8073217726396917e-05,
576
+ "loss": 1.3255,
577
+ "step": 400
578
+ },
579
+ {
580
+ "epoch": 0.09632751354605659,
581
+ "eval_loss": 1.255526065826416,
582
+ "eval_runtime": 10.4675,
583
+ "eval_samples_per_second": 9.553,
584
+ "eval_steps_per_second": 1.242,
585
+ "step": 400
586
+ },
587
+ {
588
+ "epoch": 0.0975316074653823,
589
+ "grad_norm": 0.7186573147773743,
590
+ "learning_rate": 1.804913294797688e-05,
591
+ "loss": 1.2859,
592
+ "step": 405
593
+ },
594
+ {
595
+ "epoch": 0.09873570138470801,
596
+ "grad_norm": 0.7544155716896057,
597
+ "learning_rate": 1.8025048169556842e-05,
598
+ "loss": 1.2771,
599
+ "step": 410
600
+ },
601
+ {
602
+ "epoch": 0.09993979530403371,
603
+ "grad_norm": 0.700529158115387,
604
+ "learning_rate": 1.8000963391136804e-05,
605
+ "loss": 1.2696,
606
+ "step": 415
607
+ },
608
+ {
609
+ "epoch": 0.10114388922335943,
610
+ "grad_norm": 0.6894243955612183,
611
+ "learning_rate": 1.7976878612716763e-05,
612
+ "loss": 1.2556,
613
+ "step": 420
614
+ },
615
+ {
616
+ "epoch": 0.10234798314268513,
617
+ "grad_norm": 0.8769268989562988,
618
+ "learning_rate": 1.7952793834296726e-05,
619
+ "loss": 1.2832,
620
+ "step": 425
621
+ },
622
+ {
623
+ "epoch": 0.10355207706201083,
624
+ "grad_norm": 0.7283303141593933,
625
+ "learning_rate": 1.7928709055876688e-05,
626
+ "loss": 1.3163,
627
+ "step": 430
628
+ },
629
+ {
630
+ "epoch": 0.10475617098133655,
631
+ "grad_norm": 0.6984531283378601,
632
+ "learning_rate": 1.7904624277456647e-05,
633
+ "loss": 1.1873,
634
+ "step": 435
635
+ },
636
+ {
637
+ "epoch": 0.10596026490066225,
638
+ "grad_norm": 0.7543785572052002,
639
+ "learning_rate": 1.7880539499036613e-05,
640
+ "loss": 1.2617,
641
+ "step": 440
642
+ },
643
+ {
644
+ "epoch": 0.10716435881998795,
645
+ "grad_norm": 0.7481861710548401,
646
+ "learning_rate": 1.785645472061657e-05,
647
+ "loss": 1.1792,
648
+ "step": 445
649
+ },
650
+ {
651
+ "epoch": 0.10836845273931367,
652
+ "grad_norm": 0.7607905268669128,
653
+ "learning_rate": 1.7832369942196534e-05,
654
+ "loss": 1.1987,
655
+ "step": 450
656
+ },
657
+ {
658
+ "epoch": 0.10957254665863937,
659
+ "grad_norm": 0.7427639365196228,
660
+ "learning_rate": 1.7808285163776496e-05,
661
+ "loss": 1.2142,
662
+ "step": 455
663
+ },
664
+ {
665
+ "epoch": 0.11077664057796509,
666
+ "grad_norm": 0.7348979711532593,
667
+ "learning_rate": 1.7784200385356455e-05,
668
+ "loss": 1.2727,
669
+ "step": 460
670
+ },
671
+ {
672
+ "epoch": 0.11198073449729079,
673
+ "grad_norm": 0.7192914485931396,
674
+ "learning_rate": 1.7760115606936417e-05,
675
+ "loss": 1.2593,
676
+ "step": 465
677
+ },
678
+ {
679
+ "epoch": 0.11318482841661649,
680
+ "grad_norm": 0.6672971248626709,
681
+ "learning_rate": 1.773603082851638e-05,
682
+ "loss": 1.1905,
683
+ "step": 470
684
+ },
685
+ {
686
+ "epoch": 0.11438892233594221,
687
+ "grad_norm": 0.7585046291351318,
688
+ "learning_rate": 1.771194605009634e-05,
689
+ "loss": 1.3045,
690
+ "step": 475
691
+ },
692
+ {
693
+ "epoch": 0.11559301625526791,
694
+ "grad_norm": 0.7200559973716736,
695
+ "learning_rate": 1.76878612716763e-05,
696
+ "loss": 1.1416,
697
+ "step": 480
698
+ },
699
+ {
700
+ "epoch": 0.11679711017459361,
701
+ "grad_norm": 0.7220486998558044,
702
+ "learning_rate": 1.7663776493256263e-05,
703
+ "loss": 1.2749,
704
+ "step": 485
705
+ },
706
+ {
707
+ "epoch": 0.11800120409391933,
708
+ "grad_norm": 0.750998318195343,
709
+ "learning_rate": 1.7639691714836226e-05,
710
+ "loss": 1.2823,
711
+ "step": 490
712
+ },
713
+ {
714
+ "epoch": 0.11920529801324503,
715
+ "grad_norm": 0.705752432346344,
716
+ "learning_rate": 1.7615606936416188e-05,
717
+ "loss": 1.1662,
718
+ "step": 495
719
+ },
720
+ {
721
+ "epoch": 0.12040939193257075,
722
+ "grad_norm": 0.7431082129478455,
723
+ "learning_rate": 1.7591522157996147e-05,
724
+ "loss": 1.203,
725
+ "step": 500
726
+ },
727
+ {
728
+ "epoch": 0.12161348585189645,
729
+ "grad_norm": 0.7106168866157532,
730
+ "learning_rate": 1.756743737957611e-05,
731
+ "loss": 1.162,
732
+ "step": 505
733
+ },
734
+ {
735
+ "epoch": 0.12281757977122215,
736
+ "grad_norm": 0.7027086615562439,
737
+ "learning_rate": 1.754335260115607e-05,
738
+ "loss": 1.2636,
739
+ "step": 510
740
+ },
741
+ {
742
+ "epoch": 0.12402167369054787,
743
+ "grad_norm": 0.774591863155365,
744
+ "learning_rate": 1.7519267822736034e-05,
745
+ "loss": 1.2517,
746
+ "step": 515
747
+ },
748
+ {
749
+ "epoch": 0.12522576760987358,
750
+ "grad_norm": 0.6927476525306702,
751
+ "learning_rate": 1.7495183044315993e-05,
752
+ "loss": 1.1396,
753
+ "step": 520
754
+ },
755
+ {
756
+ "epoch": 0.12642986152919927,
757
+ "grad_norm": 0.6912556886672974,
758
+ "learning_rate": 1.7471098265895955e-05,
759
+ "loss": 1.1658,
760
+ "step": 525
761
+ },
762
+ {
763
+ "epoch": 0.127633955448525,
764
+ "grad_norm": 0.6708123683929443,
765
+ "learning_rate": 1.7447013487475918e-05,
766
+ "loss": 1.1288,
767
+ "step": 530
768
+ },
769
+ {
770
+ "epoch": 0.1288380493678507,
771
+ "grad_norm": 0.6942213177680969,
772
+ "learning_rate": 1.7422928709055877e-05,
773
+ "loss": 1.1174,
774
+ "step": 535
775
+ },
776
+ {
777
+ "epoch": 0.1300421432871764,
778
+ "grad_norm": 0.7080583572387695,
779
+ "learning_rate": 1.739884393063584e-05,
780
+ "loss": 1.2279,
781
+ "step": 540
782
+ },
783
+ {
784
+ "epoch": 0.1312462372065021,
785
+ "grad_norm": 0.8131110668182373,
786
+ "learning_rate": 1.73747591522158e-05,
787
+ "loss": 1.2525,
788
+ "step": 545
789
+ },
790
+ {
791
+ "epoch": 0.13245033112582782,
792
+ "grad_norm": 0.7450209259986877,
793
+ "learning_rate": 1.735067437379576e-05,
794
+ "loss": 1.1506,
795
+ "step": 550
796
+ },
797
+ {
798
+ "epoch": 0.1336544250451535,
799
+ "grad_norm": 0.7066523432731628,
800
+ "learning_rate": 1.7326589595375726e-05,
801
+ "loss": 1.2383,
802
+ "step": 555
803
+ },
804
+ {
805
+ "epoch": 0.13485851896447923,
806
+ "grad_norm": 0.7041836977005005,
807
+ "learning_rate": 1.7302504816955685e-05,
808
+ "loss": 1.1911,
809
+ "step": 560
810
+ },
811
+ {
812
+ "epoch": 0.13606261288380495,
813
+ "grad_norm": 0.672095000743866,
814
+ "learning_rate": 1.7278420038535647e-05,
815
+ "loss": 1.1839,
816
+ "step": 565
817
+ },
818
+ {
819
+ "epoch": 0.13726670680313063,
820
+ "grad_norm": 0.7928294539451599,
821
+ "learning_rate": 1.725433526011561e-05,
822
+ "loss": 1.3347,
823
+ "step": 570
824
+ },
825
+ {
826
+ "epoch": 0.13847080072245635,
827
+ "grad_norm": 0.7465523481369019,
828
+ "learning_rate": 1.723025048169557e-05,
829
+ "loss": 1.134,
830
+ "step": 575
831
+ },
832
+ {
833
+ "epoch": 0.13967489464178207,
834
+ "grad_norm": 0.7117074131965637,
835
+ "learning_rate": 1.720616570327553e-05,
836
+ "loss": 1.3538,
837
+ "step": 580
838
+ },
839
+ {
840
+ "epoch": 0.14087898856110775,
841
+ "grad_norm": 0.7291346192359924,
842
+ "learning_rate": 1.7182080924855493e-05,
843
+ "loss": 1.1339,
844
+ "step": 585
845
+ },
846
+ {
847
+ "epoch": 0.14208308248043347,
848
+ "grad_norm": 0.7633514404296875,
849
+ "learning_rate": 1.7157996146435455e-05,
850
+ "loss": 1.2007,
851
+ "step": 590
852
+ },
853
+ {
854
+ "epoch": 0.1432871763997592,
855
+ "grad_norm": 0.7896651029586792,
856
+ "learning_rate": 1.7133911368015418e-05,
857
+ "loss": 1.2656,
858
+ "step": 595
859
+ },
860
+ {
861
+ "epoch": 0.1444912703190849,
862
+ "grad_norm": 0.7436801791191101,
863
+ "learning_rate": 1.7109826589595377e-05,
864
+ "loss": 1.28,
865
+ "step": 600
866
+ },
867
+ {
868
+ "epoch": 0.1444912703190849,
869
+ "eval_loss": 1.23267662525177,
870
+ "eval_runtime": 10.4568,
871
+ "eval_samples_per_second": 9.563,
872
+ "eval_steps_per_second": 1.243,
873
+ "step": 600
874
+ },
875
+ {
876
+ "epoch": 0.1456953642384106,
877
+ "grad_norm": 0.7123108506202698,
878
+ "learning_rate": 1.708574181117534e-05,
879
+ "loss": 1.1687,
880
+ "step": 605
881
+ },
882
+ {
883
+ "epoch": 0.1468994581577363,
884
+ "grad_norm": 0.7362635731697083,
885
+ "learning_rate": 1.70616570327553e-05,
886
+ "loss": 1.2471,
887
+ "step": 610
888
+ },
889
+ {
890
+ "epoch": 0.14810355207706202,
891
+ "grad_norm": 0.6770139336585999,
892
+ "learning_rate": 1.703757225433526e-05,
893
+ "loss": 1.149,
894
+ "step": 615
895
+ },
896
+ {
897
+ "epoch": 0.1493076459963877,
898
+ "grad_norm": 0.7292876243591309,
899
+ "learning_rate": 1.7013487475915223e-05,
900
+ "loss": 1.2957,
901
+ "step": 620
902
+ },
903
+ {
904
+ "epoch": 0.15051173991571343,
905
+ "grad_norm": 0.7105087637901306,
906
+ "learning_rate": 1.6989402697495185e-05,
907
+ "loss": 1.1462,
908
+ "step": 625
909
+ },
910
+ {
911
+ "epoch": 0.15171583383503914,
912
+ "grad_norm": 0.7218934297561646,
913
+ "learning_rate": 1.6965317919075147e-05,
914
+ "loss": 1.2181,
915
+ "step": 630
916
+ },
917
+ {
918
+ "epoch": 0.15291992775436483,
919
+ "grad_norm": 0.7794796824455261,
920
+ "learning_rate": 1.6941233140655106e-05,
921
+ "loss": 1.174,
922
+ "step": 635
923
+ },
924
+ {
925
+ "epoch": 0.15412402167369055,
926
+ "grad_norm": 0.6424360871315002,
927
+ "learning_rate": 1.691714836223507e-05,
928
+ "loss": 1.2149,
929
+ "step": 640
930
+ },
931
+ {
932
+ "epoch": 0.15532811559301626,
933
+ "grad_norm": 0.7393337488174438,
934
+ "learning_rate": 1.689306358381503e-05,
935
+ "loss": 1.2259,
936
+ "step": 645
937
+ },
938
+ {
939
+ "epoch": 0.15653220951234195,
940
+ "grad_norm": 0.7363811731338501,
941
+ "learning_rate": 1.686897880539499e-05,
942
+ "loss": 1.2087,
943
+ "step": 650
944
+ },
945
+ {
946
+ "epoch": 0.15773630343166767,
947
+ "grad_norm": 0.7336999773979187,
948
+ "learning_rate": 1.6844894026974952e-05,
949
+ "loss": 1.2526,
950
+ "step": 655
951
+ },
952
+ {
953
+ "epoch": 0.15894039735099338,
954
+ "grad_norm": 0.7329117655754089,
955
+ "learning_rate": 1.6820809248554915e-05,
956
+ "loss": 1.233,
957
+ "step": 660
958
+ },
959
+ {
960
+ "epoch": 0.16014449127031907,
961
+ "grad_norm": 0.7724847197532654,
962
+ "learning_rate": 1.6796724470134877e-05,
963
+ "loss": 1.2506,
964
+ "step": 665
965
+ },
966
+ {
967
+ "epoch": 0.1613485851896448,
968
+ "grad_norm": 0.7213071584701538,
969
+ "learning_rate": 1.677263969171484e-05,
970
+ "loss": 1.2021,
971
+ "step": 670
972
+ },
973
+ {
974
+ "epoch": 0.1625526791089705,
975
+ "grad_norm": 0.6397632360458374,
976
+ "learning_rate": 1.6748554913294798e-05,
977
+ "loss": 1.1909,
978
+ "step": 675
979
+ },
980
+ {
981
+ "epoch": 0.16375677302829622,
982
+ "grad_norm": 0.6840227842330933,
983
+ "learning_rate": 1.672447013487476e-05,
984
+ "loss": 1.1838,
985
+ "step": 680
986
+ },
987
+ {
988
+ "epoch": 0.1649608669476219,
989
+ "grad_norm": 0.7317889332771301,
990
+ "learning_rate": 1.6700385356454723e-05,
991
+ "loss": 1.169,
992
+ "step": 685
993
+ },
994
+ {
995
+ "epoch": 0.16616496086694763,
996
+ "grad_norm": 0.7134217619895935,
997
+ "learning_rate": 1.6676300578034682e-05,
998
+ "loss": 1.2711,
999
+ "step": 690
1000
+ },
1001
+ {
1002
+ "epoch": 0.16736905478627334,
1003
+ "grad_norm": 0.7525413036346436,
1004
+ "learning_rate": 1.6652215799614644e-05,
1005
+ "loss": 1.2183,
1006
+ "step": 695
1007
+ },
1008
+ {
1009
+ "epoch": 0.16857314870559903,
1010
+ "grad_norm": 0.7243443131446838,
1011
+ "learning_rate": 1.6628131021194607e-05,
1012
+ "loss": 1.2585,
1013
+ "step": 700
1014
+ },
1015
+ {
1016
+ "epoch": 0.16977724262492475,
1017
+ "grad_norm": 0.7554174661636353,
1018
+ "learning_rate": 1.660404624277457e-05,
1019
+ "loss": 1.1662,
1020
+ "step": 705
1021
+ },
1022
+ {
1023
+ "epoch": 0.17098133654425046,
1024
+ "grad_norm": 0.6472408175468445,
1025
+ "learning_rate": 1.657996146435453e-05,
1026
+ "loss": 1.1471,
1027
+ "step": 710
1028
+ },
1029
+ {
1030
+ "epoch": 0.17218543046357615,
1031
+ "grad_norm": 0.7514950037002563,
1032
+ "learning_rate": 1.655587668593449e-05,
1033
+ "loss": 1.157,
1034
+ "step": 715
1035
+ },
1036
+ {
1037
+ "epoch": 0.17338952438290187,
1038
+ "grad_norm": 0.6952060461044312,
1039
+ "learning_rate": 1.6531791907514452e-05,
1040
+ "loss": 1.2981,
1041
+ "step": 720
1042
+ },
1043
+ {
1044
+ "epoch": 0.17459361830222758,
1045
+ "grad_norm": 0.832417905330658,
1046
+ "learning_rate": 1.6507707129094415e-05,
1047
+ "loss": 1.2457,
1048
+ "step": 725
1049
+ },
1050
+ {
1051
+ "epoch": 0.17579771222155327,
1052
+ "grad_norm": 0.6825481653213501,
1053
+ "learning_rate": 1.6483622350674374e-05,
1054
+ "loss": 1.2632,
1055
+ "step": 730
1056
+ },
1057
+ {
1058
+ "epoch": 0.177001806140879,
1059
+ "grad_norm": 0.8267092704772949,
1060
+ "learning_rate": 1.6459537572254336e-05,
1061
+ "loss": 1.0368,
1062
+ "step": 735
1063
+ },
1064
+ {
1065
+ "epoch": 0.1782059000602047,
1066
+ "grad_norm": 0.7782288789749146,
1067
+ "learning_rate": 1.64354527938343e-05,
1068
+ "loss": 1.201,
1069
+ "step": 740
1070
+ },
1071
+ {
1072
+ "epoch": 0.1794099939795304,
1073
+ "grad_norm": 0.7548500299453735,
1074
+ "learning_rate": 1.641136801541426e-05,
1075
+ "loss": 1.1831,
1076
+ "step": 745
1077
+ },
1078
+ {
1079
+ "epoch": 0.1806140878988561,
1080
+ "grad_norm": 0.6344689726829529,
1081
+ "learning_rate": 1.638728323699422e-05,
1082
+ "loss": 1.1985,
1083
+ "step": 750
1084
+ },
1085
+ {
1086
+ "epoch": 0.18181818181818182,
1087
+ "grad_norm": 0.7326412796974182,
1088
+ "learning_rate": 1.6363198458574182e-05,
1089
+ "loss": 1.3051,
1090
+ "step": 755
1091
+ },
1092
+ {
1093
+ "epoch": 0.18302227573750754,
1094
+ "grad_norm": 0.6460621953010559,
1095
+ "learning_rate": 1.6339113680154144e-05,
1096
+ "loss": 1.1524,
1097
+ "step": 760
1098
+ },
1099
+ {
1100
+ "epoch": 0.18422636965683323,
1101
+ "grad_norm": 0.6921172142028809,
1102
+ "learning_rate": 1.6315028901734103e-05,
1103
+ "loss": 1.29,
1104
+ "step": 765
1105
+ },
1106
+ {
1107
+ "epoch": 0.18543046357615894,
1108
+ "grad_norm": 0.7872111201286316,
1109
+ "learning_rate": 1.629094412331407e-05,
1110
+ "loss": 1.1187,
1111
+ "step": 770
1112
+ },
1113
+ {
1114
+ "epoch": 0.18663455749548466,
1115
+ "grad_norm": 0.7098067402839661,
1116
+ "learning_rate": 1.6266859344894028e-05,
1117
+ "loss": 1.2006,
1118
+ "step": 775
1119
+ },
1120
+ {
1121
+ "epoch": 0.18783865141481035,
1122
+ "grad_norm": 0.7701388001441956,
1123
+ "learning_rate": 1.624277456647399e-05,
1124
+ "loss": 1.1609,
1125
+ "step": 780
1126
+ },
1127
+ {
1128
+ "epoch": 0.18904274533413606,
1129
+ "grad_norm": 0.7027246952056885,
1130
+ "learning_rate": 1.6218689788053953e-05,
1131
+ "loss": 1.2083,
1132
+ "step": 785
1133
+ },
1134
+ {
1135
+ "epoch": 0.19024683925346178,
1136
+ "grad_norm": 0.7161267995834351,
1137
+ "learning_rate": 1.619460500963391e-05,
1138
+ "loss": 1.131,
1139
+ "step": 790
1140
+ },
1141
+ {
1142
+ "epoch": 0.19145093317278747,
1143
+ "grad_norm": 0.8034841418266296,
1144
+ "learning_rate": 1.6170520231213874e-05,
1145
+ "loss": 1.2107,
1146
+ "step": 795
1147
+ },
1148
+ {
1149
+ "epoch": 0.19265502709211318,
1150
+ "grad_norm": 0.7797010540962219,
1151
+ "learning_rate": 1.6146435452793836e-05,
1152
+ "loss": 1.2931,
1153
+ "step": 800
1154
+ },
1155
+ {
1156
+ "epoch": 0.19265502709211318,
1157
+ "eval_loss": 1.2178008556365967,
1158
+ "eval_runtime": 10.4492,
1159
+ "eval_samples_per_second": 9.57,
1160
+ "eval_steps_per_second": 1.244,
1161
+ "step": 800
1162
+ },
1163
+ {
1164
+ "epoch": 0.1938591210114389,
1165
+ "grad_norm": 0.6586723923683167,
1166
+ "learning_rate": 1.6122350674373795e-05,
1167
+ "loss": 1.2007,
1168
+ "step": 805
1169
+ },
1170
+ {
1171
+ "epoch": 0.1950632149307646,
1172
+ "grad_norm": 0.7271411418914795,
1173
+ "learning_rate": 1.6098265895953758e-05,
1174
+ "loss": 1.1049,
1175
+ "step": 810
1176
+ },
1177
+ {
1178
+ "epoch": 0.1962673088500903,
1179
+ "grad_norm": 0.743730902671814,
1180
+ "learning_rate": 1.607418111753372e-05,
1181
+ "loss": 1.1815,
1182
+ "step": 815
1183
+ },
1184
+ {
1185
+ "epoch": 0.19747140276941602,
1186
+ "grad_norm": 0.7163010835647583,
1187
+ "learning_rate": 1.6050096339113682e-05,
1188
+ "loss": 1.1635,
1189
+ "step": 820
1190
+ },
1191
+ {
1192
+ "epoch": 0.1986754966887417,
1193
+ "grad_norm": 0.7846854329109192,
1194
+ "learning_rate": 1.6026011560693645e-05,
1195
+ "loss": 1.1576,
1196
+ "step": 825
1197
+ },
1198
+ {
1199
+ "epoch": 0.19987959060806743,
1200
+ "grad_norm": 0.755066990852356,
1201
+ "learning_rate": 1.6001926782273604e-05,
1202
+ "loss": 1.2626,
1203
+ "step": 830
1204
+ },
1205
+ {
1206
+ "epoch": 0.20108368452739314,
1207
+ "grad_norm": 0.6981719732284546,
1208
+ "learning_rate": 1.5977842003853566e-05,
1209
+ "loss": 1.1883,
1210
+ "step": 835
1211
+ },
1212
+ {
1213
+ "epoch": 0.20228777844671886,
1214
+ "grad_norm": 0.712697446346283,
1215
+ "learning_rate": 1.5953757225433528e-05,
1216
+ "loss": 1.1539,
1217
+ "step": 840
1218
+ },
1219
+ {
1220
+ "epoch": 0.20349187236604455,
1221
+ "grad_norm": 0.8146225810050964,
1222
+ "learning_rate": 1.592967244701349e-05,
1223
+ "loss": 1.161,
1224
+ "step": 845
1225
+ },
1226
+ {
1227
+ "epoch": 0.20469596628537026,
1228
+ "grad_norm": 0.7796756625175476,
1229
+ "learning_rate": 1.590558766859345e-05,
1230
+ "loss": 1.1536,
1231
+ "step": 850
1232
+ },
1233
+ {
1234
+ "epoch": 0.20590006020469598,
1235
+ "grad_norm": 0.7026313543319702,
1236
+ "learning_rate": 1.5881502890173412e-05,
1237
+ "loss": 1.2206,
1238
+ "step": 855
1239
+ },
1240
+ {
1241
+ "epoch": 0.20710415412402167,
1242
+ "grad_norm": 0.6918179988861084,
1243
+ "learning_rate": 1.5857418111753374e-05,
1244
+ "loss": 1.0569,
1245
+ "step": 860
1246
+ },
1247
+ {
1248
+ "epoch": 0.20830824804334738,
1249
+ "grad_norm": 0.7711547613143921,
1250
+ "learning_rate": 1.5833333333333333e-05,
1251
+ "loss": 1.1826,
1252
+ "step": 865
1253
+ },
1254
+ {
1255
+ "epoch": 0.2095123419626731,
1256
+ "grad_norm": 0.6604825854301453,
1257
+ "learning_rate": 1.5809248554913295e-05,
1258
+ "loss": 1.1743,
1259
+ "step": 870
1260
+ },
1261
+ {
1262
+ "epoch": 0.2107164358819988,
1263
+ "grad_norm": 0.7115917205810547,
1264
+ "learning_rate": 1.5785163776493258e-05,
1265
+ "loss": 1.2815,
1266
+ "step": 875
1267
+ },
1268
+ {
1269
+ "epoch": 0.2119205298013245,
1270
+ "grad_norm": 0.69574373960495,
1271
+ "learning_rate": 1.5761078998073217e-05,
1272
+ "loss": 1.2131,
1273
+ "step": 880
1274
+ },
1275
+ {
1276
+ "epoch": 0.21312462372065022,
1277
+ "grad_norm": 0.6801597476005554,
1278
+ "learning_rate": 1.5736994219653182e-05,
1279
+ "loss": 1.3022,
1280
+ "step": 885
1281
+ },
1282
+ {
1283
+ "epoch": 0.2143287176399759,
1284
+ "grad_norm": 0.7042235136032104,
1285
+ "learning_rate": 1.571290944123314e-05,
1286
+ "loss": 1.1936,
1287
+ "step": 890
1288
+ },
1289
+ {
1290
+ "epoch": 0.21553281155930162,
1291
+ "grad_norm": 0.6610032916069031,
1292
+ "learning_rate": 1.5688824662813104e-05,
1293
+ "loss": 1.1972,
1294
+ "step": 895
1295
+ },
1296
+ {
1297
+ "epoch": 0.21673690547862734,
1298
+ "grad_norm": 0.8218846321105957,
1299
+ "learning_rate": 1.5664739884393066e-05,
1300
+ "loss": 1.2951,
1301
+ "step": 900
1302
+ },
1303
+ {
1304
+ "epoch": 0.21794099939795303,
1305
+ "grad_norm": 0.7200918197631836,
1306
+ "learning_rate": 1.5640655105973025e-05,
1307
+ "loss": 1.1256,
1308
+ "step": 905
1309
+ },
1310
+ {
1311
+ "epoch": 0.21914509331727874,
1312
+ "grad_norm": 0.6200802326202393,
1313
+ "learning_rate": 1.5616570327552987e-05,
1314
+ "loss": 1.2071,
1315
+ "step": 910
1316
+ },
1317
+ {
1318
+ "epoch": 0.22034918723660446,
1319
+ "grad_norm": 0.7431606650352478,
1320
+ "learning_rate": 1.559248554913295e-05,
1321
+ "loss": 1.2569,
1322
+ "step": 915
1323
+ },
1324
+ {
1325
+ "epoch": 0.22155328115593018,
1326
+ "grad_norm": 0.802537202835083,
1327
+ "learning_rate": 1.5568400770712912e-05,
1328
+ "loss": 1.1342,
1329
+ "step": 920
1330
+ },
1331
+ {
1332
+ "epoch": 0.22275737507525586,
1333
+ "grad_norm": 0.7078563570976257,
1334
+ "learning_rate": 1.5544315992292874e-05,
1335
+ "loss": 1.1928,
1336
+ "step": 925
1337
+ },
1338
+ {
1339
+ "epoch": 0.22396146899458158,
1340
+ "grad_norm": 0.6626926064491272,
1341
+ "learning_rate": 1.5520231213872833e-05,
1342
+ "loss": 1.1295,
1343
+ "step": 930
1344
+ },
1345
+ {
1346
+ "epoch": 0.2251655629139073,
1347
+ "grad_norm": 0.664715588092804,
1348
+ "learning_rate": 1.5496146435452796e-05,
1349
+ "loss": 1.1616,
1350
+ "step": 935
1351
+ },
1352
+ {
1353
+ "epoch": 0.22636965683323299,
1354
+ "grad_norm": 0.7077795267105103,
1355
+ "learning_rate": 1.5472061657032758e-05,
1356
+ "loss": 1.24,
1357
+ "step": 940
1358
+ },
1359
+ {
1360
+ "epoch": 0.2275737507525587,
1361
+ "grad_norm": 0.6813915371894836,
1362
+ "learning_rate": 1.5447976878612717e-05,
1363
+ "loss": 1.1807,
1364
+ "step": 945
1365
+ },
1366
+ {
1367
+ "epoch": 0.22877784467188442,
1368
+ "grad_norm": 0.7245080471038818,
1369
+ "learning_rate": 1.542389210019268e-05,
1370
+ "loss": 1.1737,
1371
+ "step": 950
1372
+ },
1373
+ {
1374
+ "epoch": 0.2299819385912101,
1375
+ "grad_norm": 0.7851933836936951,
1376
+ "learning_rate": 1.539980732177264e-05,
1377
+ "loss": 1.0702,
1378
+ "step": 955
1379
+ },
1380
+ {
1381
+ "epoch": 0.23118603251053582,
1382
+ "grad_norm": 0.7153802514076233,
1383
+ "learning_rate": 1.5375722543352604e-05,
1384
+ "loss": 1.1302,
1385
+ "step": 960
1386
+ },
1387
+ {
1388
+ "epoch": 0.23239012642986154,
1389
+ "grad_norm": 0.8112071752548218,
1390
+ "learning_rate": 1.5351637764932563e-05,
1391
+ "loss": 1.1591,
1392
+ "step": 965
1393
+ },
1394
+ {
1395
+ "epoch": 0.23359422034918723,
1396
+ "grad_norm": 0.7842305898666382,
1397
+ "learning_rate": 1.5327552986512525e-05,
1398
+ "loss": 1.1695,
1399
+ "step": 970
1400
+ },
1401
+ {
1402
+ "epoch": 0.23479831426851294,
1403
+ "grad_norm": 0.6949911713600159,
1404
+ "learning_rate": 1.5303468208092487e-05,
1405
+ "loss": 1.1726,
1406
+ "step": 975
1407
+ },
1408
+ {
1409
+ "epoch": 0.23600240818783866,
1410
+ "grad_norm": 0.7541187405586243,
1411
+ "learning_rate": 1.5279383429672446e-05,
1412
+ "loss": 1.2209,
1413
+ "step": 980
1414
+ },
1415
+ {
1416
+ "epoch": 0.23720650210716435,
1417
+ "grad_norm": 0.7082781791687012,
1418
+ "learning_rate": 1.525529865125241e-05,
1419
+ "loss": 1.2685,
1420
+ "step": 985
1421
+ },
1422
+ {
1423
+ "epoch": 0.23841059602649006,
1424
+ "grad_norm": 0.6862682104110718,
1425
+ "learning_rate": 1.5231213872832371e-05,
1426
+ "loss": 1.1569,
1427
+ "step": 990
1428
+ },
1429
+ {
1430
+ "epoch": 0.23961468994581578,
1431
+ "grad_norm": 0.836452066898346,
1432
+ "learning_rate": 1.5207129094412332e-05,
1433
+ "loss": 1.2158,
1434
+ "step": 995
1435
+ },
1436
+ {
1437
+ "epoch": 0.2408187838651415,
1438
+ "grad_norm": 0.8637534379959106,
1439
+ "learning_rate": 1.5183044315992294e-05,
1440
+ "loss": 1.1891,
1441
+ "step": 1000
1442
+ },
1443
+ {
1444
+ "epoch": 0.2408187838651415,
1445
+ "eval_loss": 1.2066378593444824,
1446
+ "eval_runtime": 10.4478,
1447
+ "eval_samples_per_second": 9.571,
1448
+ "eval_steps_per_second": 1.244,
1449
+ "step": 1000
1450
+ },
1451
+ {
1452
+ "epoch": 0.24202287778446718,
1453
+ "grad_norm": 0.8397049903869629,
1454
+ "learning_rate": 1.5158959537572255e-05,
1455
+ "loss": 1.2199,
1456
+ "step": 1005
1457
+ },
1458
+ {
1459
+ "epoch": 0.2432269717037929,
1460
+ "grad_norm": 0.7865377068519592,
1461
+ "learning_rate": 1.5134874759152215e-05,
1462
+ "loss": 1.2102,
1463
+ "step": 1010
1464
+ },
1465
+ {
1466
+ "epoch": 0.24443106562311862,
1467
+ "grad_norm": 0.6602733731269836,
1468
+ "learning_rate": 1.511078998073218e-05,
1469
+ "loss": 1.1512,
1470
+ "step": 1015
1471
+ },
1472
+ {
1473
+ "epoch": 0.2456351595424443,
1474
+ "grad_norm": 0.69524747133255,
1475
+ "learning_rate": 1.508670520231214e-05,
1476
+ "loss": 1.2387,
1477
+ "step": 1020
1478
+ },
1479
+ {
1480
+ "epoch": 0.24683925346177002,
1481
+ "grad_norm": 0.6567270755767822,
1482
+ "learning_rate": 1.50626204238921e-05,
1483
+ "loss": 1.159,
1484
+ "step": 1025
1485
+ },
1486
+ {
1487
+ "epoch": 0.24804334738109574,
1488
+ "grad_norm": 0.6918655633926392,
1489
+ "learning_rate": 1.5038535645472063e-05,
1490
+ "loss": 1.236,
1491
+ "step": 1030
1492
+ },
1493
+ {
1494
+ "epoch": 0.24924744130042142,
1495
+ "grad_norm": 0.748435378074646,
1496
+ "learning_rate": 1.5014450867052024e-05,
1497
+ "loss": 1.1422,
1498
+ "step": 1035
1499
+ },
1500
+ {
1501
+ "epoch": 0.25045153521974717,
1502
+ "grad_norm": 0.8211308717727661,
1503
+ "learning_rate": 1.4990366088631986e-05,
1504
+ "loss": 1.1775,
1505
+ "step": 1040
1506
+ },
1507
+ {
1508
+ "epoch": 0.25165562913907286,
1509
+ "grad_norm": 0.6989510655403137,
1510
+ "learning_rate": 1.4966281310211948e-05,
1511
+ "loss": 1.2473,
1512
+ "step": 1045
1513
+ },
1514
+ {
1515
+ "epoch": 0.25285972305839854,
1516
+ "grad_norm": 0.6981227993965149,
1517
+ "learning_rate": 1.4942196531791909e-05,
1518
+ "loss": 1.1887,
1519
+ "step": 1050
1520
+ },
1521
+ {
1522
+ "epoch": 0.2540638169777243,
1523
+ "grad_norm": 0.6409928202629089,
1524
+ "learning_rate": 1.4918111753371871e-05,
1525
+ "loss": 1.1323,
1526
+ "step": 1055
1527
+ },
1528
+ {
1529
+ "epoch": 0.25526791089705,
1530
+ "grad_norm": 0.8026551604270935,
1531
+ "learning_rate": 1.4894026974951832e-05,
1532
+ "loss": 1.2205,
1533
+ "step": 1060
1534
+ },
1535
+ {
1536
+ "epoch": 0.25647200481637566,
1537
+ "grad_norm": 0.7427927255630493,
1538
+ "learning_rate": 1.4869942196531793e-05,
1539
+ "loss": 1.3051,
1540
+ "step": 1065
1541
+ },
1542
+ {
1543
+ "epoch": 0.2576760987357014,
1544
+ "grad_norm": 0.7286926507949829,
1545
+ "learning_rate": 1.4845857418111755e-05,
1546
+ "loss": 1.1952,
1547
+ "step": 1070
1548
+ },
1549
+ {
1550
+ "epoch": 0.2588801926550271,
1551
+ "grad_norm": 0.717553436756134,
1552
+ "learning_rate": 1.4821772639691716e-05,
1553
+ "loss": 1.1395,
1554
+ "step": 1075
1555
+ },
1556
+ {
1557
+ "epoch": 0.2600842865743528,
1558
+ "grad_norm": 0.6549174189567566,
1559
+ "learning_rate": 1.4797687861271676e-05,
1560
+ "loss": 1.1677,
1561
+ "step": 1080
1562
+ },
1563
+ {
1564
+ "epoch": 0.26128838049367853,
1565
+ "grad_norm": 0.7400743365287781,
1566
+ "learning_rate": 1.477360308285164e-05,
1567
+ "loss": 1.239,
1568
+ "step": 1085
1569
+ },
1570
+ {
1571
+ "epoch": 0.2624924744130042,
1572
+ "grad_norm": 0.8405967950820923,
1573
+ "learning_rate": 1.4749518304431601e-05,
1574
+ "loss": 1.1379,
1575
+ "step": 1090
1576
+ },
1577
+ {
1578
+ "epoch": 0.2636965683323299,
1579
+ "grad_norm": 0.7308910489082336,
1580
+ "learning_rate": 1.4725433526011562e-05,
1581
+ "loss": 1.1709,
1582
+ "step": 1095
1583
+ },
1584
+ {
1585
+ "epoch": 0.26490066225165565,
1586
+ "grad_norm": 0.7904026508331299,
1587
+ "learning_rate": 1.4701348747591524e-05,
1588
+ "loss": 1.2348,
1589
+ "step": 1100
1590
+ },
1591
+ {
1592
+ "epoch": 0.26610475617098134,
1593
+ "grad_norm": 0.7531107664108276,
1594
+ "learning_rate": 1.4677263969171484e-05,
1595
+ "loss": 1.1702,
1596
+ "step": 1105
1597
+ },
1598
+ {
1599
+ "epoch": 0.267308850090307,
1600
+ "grad_norm": 0.7167130708694458,
1601
+ "learning_rate": 1.4653179190751445e-05,
1602
+ "loss": 1.1417,
1603
+ "step": 1110
1604
+ },
1605
+ {
1606
+ "epoch": 0.26851294400963277,
1607
+ "grad_norm": 0.7529981732368469,
1608
+ "learning_rate": 1.4629094412331407e-05,
1609
+ "loss": 1.3126,
1610
+ "step": 1115
1611
+ },
1612
+ {
1613
+ "epoch": 0.26971703792895846,
1614
+ "grad_norm": 0.7192036509513855,
1615
+ "learning_rate": 1.4605009633911368e-05,
1616
+ "loss": 1.2149,
1617
+ "step": 1120
1618
+ },
1619
+ {
1620
+ "epoch": 0.27092113184828415,
1621
+ "grad_norm": 0.7200658321380615,
1622
+ "learning_rate": 1.458092485549133e-05,
1623
+ "loss": 1.1958,
1624
+ "step": 1125
1625
+ },
1626
+ {
1627
+ "epoch": 0.2721252257676099,
1628
+ "grad_norm": 0.7063027024269104,
1629
+ "learning_rate": 1.4556840077071293e-05,
1630
+ "loss": 1.1266,
1631
+ "step": 1130
1632
+ },
1633
+ {
1634
+ "epoch": 0.2733293196869356,
1635
+ "grad_norm": 0.7414741516113281,
1636
+ "learning_rate": 1.4532755298651253e-05,
1637
+ "loss": 1.2046,
1638
+ "step": 1135
1639
+ },
1640
+ {
1641
+ "epoch": 0.27453341360626127,
1642
+ "grad_norm": 0.7233819365501404,
1643
+ "learning_rate": 1.4508670520231216e-05,
1644
+ "loss": 1.3376,
1645
+ "step": 1140
1646
+ },
1647
+ {
1648
+ "epoch": 0.275737507525587,
1649
+ "grad_norm": 0.737486720085144,
1650
+ "learning_rate": 1.4484585741811176e-05,
1651
+ "loss": 1.2529,
1652
+ "step": 1145
1653
+ },
1654
+ {
1655
+ "epoch": 0.2769416014449127,
1656
+ "grad_norm": 0.6850925087928772,
1657
+ "learning_rate": 1.4460500963391137e-05,
1658
+ "loss": 1.1784,
1659
+ "step": 1150
1660
+ },
1661
+ {
1662
+ "epoch": 0.2781456953642384,
1663
+ "grad_norm": 0.6742652058601379,
1664
+ "learning_rate": 1.4436416184971101e-05,
1665
+ "loss": 1.0412,
1666
+ "step": 1155
1667
+ },
1668
+ {
1669
+ "epoch": 0.27934978928356413,
1670
+ "grad_norm": 0.7630188465118408,
1671
+ "learning_rate": 1.4412331406551062e-05,
1672
+ "loss": 1.2421,
1673
+ "step": 1160
1674
+ },
1675
+ {
1676
+ "epoch": 0.2805538832028898,
1677
+ "grad_norm": 0.7039279341697693,
1678
+ "learning_rate": 1.4388246628131022e-05,
1679
+ "loss": 1.2076,
1680
+ "step": 1165
1681
+ },
1682
+ {
1683
+ "epoch": 0.2817579771222155,
1684
+ "grad_norm": 0.6968697905540466,
1685
+ "learning_rate": 1.4364161849710985e-05,
1686
+ "loss": 1.2709,
1687
+ "step": 1170
1688
+ },
1689
+ {
1690
+ "epoch": 0.28296207104154125,
1691
+ "grad_norm": 0.757686972618103,
1692
+ "learning_rate": 1.4340077071290945e-05,
1693
+ "loss": 1.2237,
1694
+ "step": 1175
1695
+ },
1696
+ {
1697
+ "epoch": 0.28416616496086694,
1698
+ "grad_norm": 0.70340895652771,
1699
+ "learning_rate": 1.4315992292870906e-05,
1700
+ "loss": 1.1979,
1701
+ "step": 1180
1702
+ },
1703
+ {
1704
+ "epoch": 0.28537025888019263,
1705
+ "grad_norm": 0.7078365683555603,
1706
+ "learning_rate": 1.4291907514450868e-05,
1707
+ "loss": 1.1801,
1708
+ "step": 1185
1709
+ },
1710
+ {
1711
+ "epoch": 0.2865743527995184,
1712
+ "grad_norm": 0.671120822429657,
1713
+ "learning_rate": 1.4267822736030829e-05,
1714
+ "loss": 1.0744,
1715
+ "step": 1190
1716
+ },
1717
+ {
1718
+ "epoch": 0.28777844671884406,
1719
+ "grad_norm": 0.6995368599891663,
1720
+ "learning_rate": 1.424373795761079e-05,
1721
+ "loss": 1.1057,
1722
+ "step": 1195
1723
+ },
1724
+ {
1725
+ "epoch": 0.2889825406381698,
1726
+ "grad_norm": 0.8009471297264099,
1727
+ "learning_rate": 1.4219653179190754e-05,
1728
+ "loss": 1.2852,
1729
+ "step": 1200
1730
+ },
1731
+ {
1732
+ "epoch": 0.2889825406381698,
1733
+ "eval_loss": 1.1969135999679565,
1734
+ "eval_runtime": 10.4824,
1735
+ "eval_samples_per_second": 9.54,
1736
+ "eval_steps_per_second": 1.24,
1737
+ "step": 1200
1738
+ },
1739
+ {
1740
+ "epoch": 0.2901866345574955,
1741
+ "grad_norm": 0.8135602474212646,
1742
+ "learning_rate": 1.4195568400770714e-05,
1743
+ "loss": 1.1475,
1744
+ "step": 1205
1745
+ },
1746
+ {
1747
+ "epoch": 0.2913907284768212,
1748
+ "grad_norm": 0.6879997849464417,
1749
+ "learning_rate": 1.4171483622350675e-05,
1750
+ "loss": 1.2054,
1751
+ "step": 1210
1752
+ },
1753
+ {
1754
+ "epoch": 0.2925948223961469,
1755
+ "grad_norm": 0.774534285068512,
1756
+ "learning_rate": 1.4147398843930637e-05,
1757
+ "loss": 1.1133,
1758
+ "step": 1215
1759
+ },
1760
+ {
1761
+ "epoch": 0.2937989163154726,
1762
+ "grad_norm": 0.7712222933769226,
1763
+ "learning_rate": 1.4123314065510598e-05,
1764
+ "loss": 1.2367,
1765
+ "step": 1220
1766
+ },
1767
+ {
1768
+ "epoch": 0.2950030102347983,
1769
+ "grad_norm": 0.7632959485054016,
1770
+ "learning_rate": 1.4099229287090558e-05,
1771
+ "loss": 1.2142,
1772
+ "step": 1225
1773
+ },
1774
+ {
1775
+ "epoch": 0.29620710415412405,
1776
+ "grad_norm": 0.6896396279335022,
1777
+ "learning_rate": 1.4075144508670523e-05,
1778
+ "loss": 1.2692,
1779
+ "step": 1230
1780
+ },
1781
+ {
1782
+ "epoch": 0.29741119807344973,
1783
+ "grad_norm": 0.7135001420974731,
1784
+ "learning_rate": 1.4051059730250483e-05,
1785
+ "loss": 1.1922,
1786
+ "step": 1235
1787
+ },
1788
+ {
1789
+ "epoch": 0.2986152919927754,
1790
+ "grad_norm": 0.7016868591308594,
1791
+ "learning_rate": 1.4026974951830444e-05,
1792
+ "loss": 1.2775,
1793
+ "step": 1240
1794
+ },
1795
+ {
1796
+ "epoch": 0.29981938591210117,
1797
+ "grad_norm": 0.7863327860832214,
1798
+ "learning_rate": 1.4002890173410406e-05,
1799
+ "loss": 1.193,
1800
+ "step": 1245
1801
+ },
1802
+ {
1803
+ "epoch": 0.30102347983142685,
1804
+ "grad_norm": 0.7253124713897705,
1805
+ "learning_rate": 1.3978805394990367e-05,
1806
+ "loss": 1.1557,
1807
+ "step": 1250
1808
+ },
1809
+ {
1810
+ "epoch": 0.30222757375075254,
1811
+ "grad_norm": 0.7214757204055786,
1812
+ "learning_rate": 1.3954720616570329e-05,
1813
+ "loss": 1.2212,
1814
+ "step": 1255
1815
+ },
1816
+ {
1817
+ "epoch": 0.3034316676700783,
1818
+ "grad_norm": 0.7036615014076233,
1819
+ "learning_rate": 1.393063583815029e-05,
1820
+ "loss": 1.3103,
1821
+ "step": 1260
1822
+ },
1823
+ {
1824
+ "epoch": 0.304635761589404,
1825
+ "grad_norm": 0.7152957320213318,
1826
+ "learning_rate": 1.390655105973025e-05,
1827
+ "loss": 1.3316,
1828
+ "step": 1265
1829
+ },
1830
+ {
1831
+ "epoch": 0.30583985550872966,
1832
+ "grad_norm": 0.7779828906059265,
1833
+ "learning_rate": 1.3882466281310214e-05,
1834
+ "loss": 1.2773,
1835
+ "step": 1270
1836
+ },
1837
+ {
1838
+ "epoch": 0.3070439494280554,
1839
+ "grad_norm": 0.7995044589042664,
1840
+ "learning_rate": 1.3858381502890175e-05,
1841
+ "loss": 1.297,
1842
+ "step": 1275
1843
+ },
1844
+ {
1845
+ "epoch": 0.3082480433473811,
1846
+ "grad_norm": 0.7020440697669983,
1847
+ "learning_rate": 1.3834296724470136e-05,
1848
+ "loss": 1.1593,
1849
+ "step": 1280
1850
+ },
1851
+ {
1852
+ "epoch": 0.3094521372667068,
1853
+ "grad_norm": 0.7052685022354126,
1854
+ "learning_rate": 1.3810211946050098e-05,
1855
+ "loss": 1.1415,
1856
+ "step": 1285
1857
+ },
1858
+ {
1859
+ "epoch": 0.3106562311860325,
1860
+ "grad_norm": 0.7097790837287903,
1861
+ "learning_rate": 1.3786127167630059e-05,
1862
+ "loss": 1.0992,
1863
+ "step": 1290
1864
+ },
1865
+ {
1866
+ "epoch": 0.3118603251053582,
1867
+ "grad_norm": 0.7684621214866638,
1868
+ "learning_rate": 1.376204238921002e-05,
1869
+ "loss": 1.2256,
1870
+ "step": 1295
1871
+ },
1872
+ {
1873
+ "epoch": 0.3130644190246839,
1874
+ "grad_norm": 0.746864914894104,
1875
+ "learning_rate": 1.3737957610789983e-05,
1876
+ "loss": 1.2155,
1877
+ "step": 1300
1878
+ },
1879
+ {
1880
+ "epoch": 0.31426851294400965,
1881
+ "grad_norm": 0.733629584312439,
1882
+ "learning_rate": 1.3713872832369944e-05,
1883
+ "loss": 1.1287,
1884
+ "step": 1305
1885
+ },
1886
+ {
1887
+ "epoch": 0.31547260686333534,
1888
+ "grad_norm": 0.6500353217124939,
1889
+ "learning_rate": 1.3689788053949905e-05,
1890
+ "loss": 1.267,
1891
+ "step": 1310
1892
+ },
1893
+ {
1894
+ "epoch": 0.316676700782661,
1895
+ "grad_norm": 0.7301894426345825,
1896
+ "learning_rate": 1.3665703275529867e-05,
1897
+ "loss": 1.2167,
1898
+ "step": 1315
1899
+ },
1900
+ {
1901
+ "epoch": 0.31788079470198677,
1902
+ "grad_norm": 0.677039623260498,
1903
+ "learning_rate": 1.3641618497109828e-05,
1904
+ "loss": 1.1037,
1905
+ "step": 1320
1906
+ },
1907
+ {
1908
+ "epoch": 0.31908488862131246,
1909
+ "grad_norm": 0.7031332850456238,
1910
+ "learning_rate": 1.3617533718689788e-05,
1911
+ "loss": 1.1252,
1912
+ "step": 1325
1913
+ },
1914
+ {
1915
+ "epoch": 0.32028898254063815,
1916
+ "grad_norm": 0.8084255456924438,
1917
+ "learning_rate": 1.359344894026975e-05,
1918
+ "loss": 1.1728,
1919
+ "step": 1330
1920
+ },
1921
+ {
1922
+ "epoch": 0.3214930764599639,
1923
+ "grad_norm": 0.6915790438652039,
1924
+ "learning_rate": 1.3569364161849711e-05,
1925
+ "loss": 1.1552,
1926
+ "step": 1335
1927
+ },
1928
+ {
1929
+ "epoch": 0.3226971703792896,
1930
+ "grad_norm": 0.742514431476593,
1931
+ "learning_rate": 1.3545279383429672e-05,
1932
+ "loss": 1.1406,
1933
+ "step": 1340
1934
+ },
1935
+ {
1936
+ "epoch": 0.32390126429861527,
1937
+ "grad_norm": 0.6682133674621582,
1938
+ "learning_rate": 1.3521194605009636e-05,
1939
+ "loss": 1.1642,
1940
+ "step": 1345
1941
+ },
1942
+ {
1943
+ "epoch": 0.325105358217941,
1944
+ "grad_norm": 0.7231415510177612,
1945
+ "learning_rate": 1.3497109826589597e-05,
1946
+ "loss": 1.253,
1947
+ "step": 1350
1948
+ },
1949
+ {
1950
+ "epoch": 0.3263094521372667,
1951
+ "grad_norm": 0.7634385824203491,
1952
+ "learning_rate": 1.3473025048169559e-05,
1953
+ "loss": 1.2276,
1954
+ "step": 1355
1955
+ },
1956
+ {
1957
+ "epoch": 0.32751354605659244,
1958
+ "grad_norm": 0.6695632934570312,
1959
+ "learning_rate": 1.344894026974952e-05,
1960
+ "loss": 1.1864,
1961
+ "step": 1360
1962
+ },
1963
+ {
1964
+ "epoch": 0.32871763997591813,
1965
+ "grad_norm": 0.7134613394737244,
1966
+ "learning_rate": 1.342485549132948e-05,
1967
+ "loss": 1.1097,
1968
+ "step": 1365
1969
+ },
1970
+ {
1971
+ "epoch": 0.3299217338952438,
1972
+ "grad_norm": 0.6643304228782654,
1973
+ "learning_rate": 1.3400770712909442e-05,
1974
+ "loss": 1.2438,
1975
+ "step": 1370
1976
+ },
1977
+ {
1978
+ "epoch": 0.33112582781456956,
1979
+ "grad_norm": 0.6821620464324951,
1980
+ "learning_rate": 1.3376685934489405e-05,
1981
+ "loss": 1.2137,
1982
+ "step": 1375
1983
+ },
1984
+ {
1985
+ "epoch": 0.33232992173389525,
1986
+ "grad_norm": 0.7436670660972595,
1987
+ "learning_rate": 1.3352601156069365e-05,
1988
+ "loss": 1.0745,
1989
+ "step": 1380
1990
+ },
1991
+ {
1992
+ "epoch": 0.33353401565322094,
1993
+ "grad_norm": 0.7283557653427124,
1994
+ "learning_rate": 1.3328516377649328e-05,
1995
+ "loss": 1.04,
1996
+ "step": 1385
1997
+ },
1998
+ {
1999
+ "epoch": 0.3347381095725467,
2000
+ "grad_norm": 0.6368799805641174,
2001
+ "learning_rate": 1.3304431599229288e-05,
2002
+ "loss": 1.1829,
2003
+ "step": 1390
2004
+ },
2005
+ {
2006
+ "epoch": 0.33594220349187237,
2007
+ "grad_norm": 0.8481898307800293,
2008
+ "learning_rate": 1.3280346820809249e-05,
2009
+ "loss": 1.0901,
2010
+ "step": 1395
2011
+ },
2012
+ {
2013
+ "epoch": 0.33714629741119806,
2014
+ "grad_norm": 0.7311575412750244,
2015
+ "learning_rate": 1.3256262042389211e-05,
2016
+ "loss": 1.1135,
2017
+ "step": 1400
2018
+ },
2019
+ {
2020
+ "epoch": 0.33714629741119806,
2021
+ "eval_loss": 1.1897846460342407,
2022
+ "eval_runtime": 10.4886,
2023
+ "eval_samples_per_second": 9.534,
2024
+ "eval_steps_per_second": 1.239,
2025
+ "step": 1400
2026
+ },
2027
+ {
2028
+ "epoch": 0.3383503913305238,
2029
+ "grad_norm": 0.7002428770065308,
2030
+ "learning_rate": 1.3232177263969172e-05,
2031
+ "loss": 1.1384,
2032
+ "step": 1405
2033
+ },
2034
+ {
2035
+ "epoch": 0.3395544852498495,
2036
+ "grad_norm": 0.7509122490882874,
2037
+ "learning_rate": 1.3208092485549133e-05,
2038
+ "loss": 1.1667,
2039
+ "step": 1410
2040
+ },
2041
+ {
2042
+ "epoch": 0.3407585791691752,
2043
+ "grad_norm": 0.7156943082809448,
2044
+ "learning_rate": 1.3184007707129097e-05,
2045
+ "loss": 1.0665,
2046
+ "step": 1415
2047
+ },
2048
+ {
2049
+ "epoch": 0.3419626730885009,
2050
+ "grad_norm": 0.8622729182243347,
2051
+ "learning_rate": 1.3159922928709057e-05,
2052
+ "loss": 1.1503,
2053
+ "step": 1420
2054
+ },
2055
+ {
2056
+ "epoch": 0.3431667670078266,
2057
+ "grad_norm": 0.7578678131103516,
2058
+ "learning_rate": 1.3135838150289018e-05,
2059
+ "loss": 1.0761,
2060
+ "step": 1425
2061
+ },
2062
+ {
2063
+ "epoch": 0.3443708609271523,
2064
+ "grad_norm": 0.7222859263420105,
2065
+ "learning_rate": 1.311175337186898e-05,
2066
+ "loss": 1.2214,
2067
+ "step": 1430
2068
+ },
2069
+ {
2070
+ "epoch": 0.34557495484647804,
2071
+ "grad_norm": 0.7335066795349121,
2072
+ "learning_rate": 1.3087668593448941e-05,
2073
+ "loss": 1.1857,
2074
+ "step": 1435
2075
+ },
2076
+ {
2077
+ "epoch": 0.34677904876580373,
2078
+ "grad_norm": 0.6828716397285461,
2079
+ "learning_rate": 1.3063583815028902e-05,
2080
+ "loss": 1.2078,
2081
+ "step": 1440
2082
+ },
2083
+ {
2084
+ "epoch": 0.3479831426851294,
2085
+ "grad_norm": 0.8298027515411377,
2086
+ "learning_rate": 1.3039499036608864e-05,
2087
+ "loss": 1.2375,
2088
+ "step": 1445
2089
+ },
2090
+ {
2091
+ "epoch": 0.34918723660445516,
2092
+ "grad_norm": 0.7282932996749878,
2093
+ "learning_rate": 1.3015414258188825e-05,
2094
+ "loss": 1.1631,
2095
+ "step": 1450
2096
+ },
2097
+ {
2098
+ "epoch": 0.35039133052378085,
2099
+ "grad_norm": 0.6626558303833008,
2100
+ "learning_rate": 1.2991329479768787e-05,
2101
+ "loss": 1.1374,
2102
+ "step": 1455
2103
+ },
2104
+ {
2105
+ "epoch": 0.35159542444310654,
2106
+ "grad_norm": 0.8166629672050476,
2107
+ "learning_rate": 1.296724470134875e-05,
2108
+ "loss": 1.2521,
2109
+ "step": 1460
2110
+ },
2111
+ {
2112
+ "epoch": 0.3527995183624323,
2113
+ "grad_norm": 0.688026487827301,
2114
+ "learning_rate": 1.294315992292871e-05,
2115
+ "loss": 1.1008,
2116
+ "step": 1465
2117
+ },
2118
+ {
2119
+ "epoch": 0.354003612281758,
2120
+ "grad_norm": 0.7461696267127991,
2121
+ "learning_rate": 1.2919075144508672e-05,
2122
+ "loss": 1.0932,
2123
+ "step": 1470
2124
+ },
2125
+ {
2126
+ "epoch": 0.35520770620108366,
2127
+ "grad_norm": 0.7331961989402771,
2128
+ "learning_rate": 1.2894990366088633e-05,
2129
+ "loss": 1.0613,
2130
+ "step": 1475
2131
+ },
2132
+ {
2133
+ "epoch": 0.3564118001204094,
2134
+ "grad_norm": 0.7345579862594604,
2135
+ "learning_rate": 1.2870905587668594e-05,
2136
+ "loss": 1.3048,
2137
+ "step": 1480
2138
+ },
2139
+ {
2140
+ "epoch": 0.3576158940397351,
2141
+ "grad_norm": 0.8133093118667603,
2142
+ "learning_rate": 1.2846820809248558e-05,
2143
+ "loss": 1.2435,
2144
+ "step": 1485
2145
+ },
2146
+ {
2147
+ "epoch": 0.3588199879590608,
2148
+ "grad_norm": 0.7404915690422058,
2149
+ "learning_rate": 1.2822736030828518e-05,
2150
+ "loss": 1.1767,
2151
+ "step": 1490
2152
+ },
2153
+ {
2154
+ "epoch": 0.3600240818783865,
2155
+ "grad_norm": 0.8186049461364746,
2156
+ "learning_rate": 1.2798651252408479e-05,
2157
+ "loss": 1.0789,
2158
+ "step": 1495
2159
+ },
2160
+ {
2161
+ "epoch": 0.3612281757977122,
2162
+ "grad_norm": 0.8047255277633667,
2163
+ "learning_rate": 1.2774566473988441e-05,
2164
+ "loss": 1.1343,
2165
+ "step": 1500
2166
+ },
2167
+ {
2168
+ "epoch": 0.3624322697170379,
2169
+ "grad_norm": 0.6876709461212158,
2170
+ "learning_rate": 1.2750481695568402e-05,
2171
+ "loss": 1.0948,
2172
+ "step": 1505
2173
+ },
2174
+ {
2175
+ "epoch": 0.36363636363636365,
2176
+ "grad_norm": 0.7112250328063965,
2177
+ "learning_rate": 1.2726396917148362e-05,
2178
+ "loss": 1.1495,
2179
+ "step": 1510
2180
+ },
2181
+ {
2182
+ "epoch": 0.36484045755568933,
2183
+ "grad_norm": 0.6787546873092651,
2184
+ "learning_rate": 1.2702312138728325e-05,
2185
+ "loss": 1.2422,
2186
+ "step": 1515
2187
+ },
2188
+ {
2189
+ "epoch": 0.3660445514750151,
2190
+ "grad_norm": 0.677456796169281,
2191
+ "learning_rate": 1.2678227360308285e-05,
2192
+ "loss": 1.2153,
2193
+ "step": 1520
2194
+ },
2195
+ {
2196
+ "epoch": 0.36724864539434077,
2197
+ "grad_norm": 0.6819108724594116,
2198
+ "learning_rate": 1.2654142581888246e-05,
2199
+ "loss": 1.0196,
2200
+ "step": 1525
2201
+ },
2202
+ {
2203
+ "epoch": 0.36845273931366646,
2204
+ "grad_norm": 0.7156027555465698,
2205
+ "learning_rate": 1.263005780346821e-05,
2206
+ "loss": 1.2116,
2207
+ "step": 1530
2208
+ },
2209
+ {
2210
+ "epoch": 0.3696568332329922,
2211
+ "grad_norm": 0.7217147350311279,
2212
+ "learning_rate": 1.260597302504817e-05,
2213
+ "loss": 1.1021,
2214
+ "step": 1535
2215
+ },
2216
+ {
2217
+ "epoch": 0.3708609271523179,
2218
+ "grad_norm": 0.6850091814994812,
2219
+ "learning_rate": 1.2581888246628131e-05,
2220
+ "loss": 1.1054,
2221
+ "step": 1540
2222
+ },
2223
+ {
2224
+ "epoch": 0.3720650210716436,
2225
+ "grad_norm": 0.7076934576034546,
2226
+ "learning_rate": 1.2557803468208094e-05,
2227
+ "loss": 1.1533,
2228
+ "step": 1545
2229
+ },
2230
+ {
2231
+ "epoch": 0.3732691149909693,
2232
+ "grad_norm": 0.7233012318611145,
2233
+ "learning_rate": 1.2533718689788054e-05,
2234
+ "loss": 1.2163,
2235
+ "step": 1550
2236
+ },
2237
+ {
2238
+ "epoch": 0.374473208910295,
2239
+ "grad_norm": 0.8073462843894958,
2240
+ "learning_rate": 1.2509633911368015e-05,
2241
+ "loss": 1.2245,
2242
+ "step": 1555
2243
+ },
2244
+ {
2245
+ "epoch": 0.3756773028296207,
2246
+ "grad_norm": 0.72356116771698,
2247
+ "learning_rate": 1.2485549132947979e-05,
2248
+ "loss": 1.1593,
2249
+ "step": 1560
2250
+ },
2251
+ {
2252
+ "epoch": 0.37688139674894644,
2253
+ "grad_norm": 0.645554780960083,
2254
+ "learning_rate": 1.246146435452794e-05,
2255
+ "loss": 1.1967,
2256
+ "step": 1565
2257
+ },
2258
+ {
2259
+ "epoch": 0.37808549066827213,
2260
+ "grad_norm": 0.6589043140411377,
2261
+ "learning_rate": 1.2437379576107902e-05,
2262
+ "loss": 1.2093,
2263
+ "step": 1570
2264
+ },
2265
+ {
2266
+ "epoch": 0.3792895845875978,
2267
+ "grad_norm": 0.6996996402740479,
2268
+ "learning_rate": 1.2413294797687863e-05,
2269
+ "loss": 1.1498,
2270
+ "step": 1575
2271
+ },
2272
+ {
2273
+ "epoch": 0.38049367850692356,
2274
+ "grad_norm": 0.6371117234230042,
2275
+ "learning_rate": 1.2389210019267823e-05,
2276
+ "loss": 1.1499,
2277
+ "step": 1580
2278
+ },
2279
+ {
2280
+ "epoch": 0.38169777242624925,
2281
+ "grad_norm": 0.6780497431755066,
2282
+ "learning_rate": 1.2365125240847786e-05,
2283
+ "loss": 1.1191,
2284
+ "step": 1585
2285
+ },
2286
+ {
2287
+ "epoch": 0.38290186634557494,
2288
+ "grad_norm": 0.7249863743782043,
2289
+ "learning_rate": 1.2341040462427746e-05,
2290
+ "loss": 1.2254,
2291
+ "step": 1590
2292
+ },
2293
+ {
2294
+ "epoch": 0.3841059602649007,
2295
+ "grad_norm": 0.6600310206413269,
2296
+ "learning_rate": 1.2316955684007707e-05,
2297
+ "loss": 1.0585,
2298
+ "step": 1595
2299
+ },
2300
+ {
2301
+ "epoch": 0.38531005418422637,
2302
+ "grad_norm": 0.6883255839347839,
2303
+ "learning_rate": 1.2292870905587671e-05,
2304
+ "loss": 1.1318,
2305
+ "step": 1600
2306
+ },
2307
+ {
2308
+ "epoch": 0.38531005418422637,
2309
+ "eval_loss": 1.1825724840164185,
2310
+ "eval_runtime": 10.5009,
2311
+ "eval_samples_per_second": 9.523,
2312
+ "eval_steps_per_second": 1.238,
2313
+ "step": 1600
2314
+ },
2315
+ {
2316
+ "epoch": 0.38651414810355206,
2317
+ "grad_norm": 0.7173585295677185,
2318
+ "learning_rate": 1.2268786127167632e-05,
2319
+ "loss": 1.284,
2320
+ "step": 1605
2321
+ },
2322
+ {
2323
+ "epoch": 0.3877182420228778,
2324
+ "grad_norm": 0.7643466591835022,
2325
+ "learning_rate": 1.2244701348747592e-05,
2326
+ "loss": 1.3699,
2327
+ "step": 1610
2328
+ },
2329
+ {
2330
+ "epoch": 0.3889223359422035,
2331
+ "grad_norm": 0.690148115158081,
2332
+ "learning_rate": 1.2220616570327555e-05,
2333
+ "loss": 1.1335,
2334
+ "step": 1615
2335
+ },
2336
+ {
2337
+ "epoch": 0.3901264298615292,
2338
+ "grad_norm": 0.6676989197731018,
2339
+ "learning_rate": 1.2196531791907515e-05,
2340
+ "loss": 1.1887,
2341
+ "step": 1620
2342
+ },
2343
+ {
2344
+ "epoch": 0.3913305237808549,
2345
+ "grad_norm": 1.0724517107009888,
2346
+ "learning_rate": 1.2172447013487476e-05,
2347
+ "loss": 1.2212,
2348
+ "step": 1625
2349
+ },
2350
+ {
2351
+ "epoch": 0.3925346177001806,
2352
+ "grad_norm": 0.6685481071472168,
2353
+ "learning_rate": 1.214836223506744e-05,
2354
+ "loss": 1.0551,
2355
+ "step": 1630
2356
+ },
2357
+ {
2358
+ "epoch": 0.3937387116195063,
2359
+ "grad_norm": 0.7440733313560486,
2360
+ "learning_rate": 1.21242774566474e-05,
2361
+ "loss": 1.1565,
2362
+ "step": 1635
2363
+ },
2364
+ {
2365
+ "epoch": 0.39494280553883204,
2366
+ "grad_norm": 0.7820375561714172,
2367
+ "learning_rate": 1.2100192678227361e-05,
2368
+ "loss": 1.0733,
2369
+ "step": 1640
2370
+ },
2371
+ {
2372
+ "epoch": 0.39614689945815773,
2373
+ "grad_norm": 0.7016002535820007,
2374
+ "learning_rate": 1.2076107899807323e-05,
2375
+ "loss": 1.1914,
2376
+ "step": 1645
2377
+ },
2378
+ {
2379
+ "epoch": 0.3973509933774834,
2380
+ "grad_norm": 0.8267973065376282,
2381
+ "learning_rate": 1.2052023121387284e-05,
2382
+ "loss": 1.1687,
2383
+ "step": 1650
2384
+ },
2385
+ {
2386
+ "epoch": 0.39855508729680916,
2387
+ "grad_norm": 0.7492110729217529,
2388
+ "learning_rate": 1.2027938342967245e-05,
2389
+ "loss": 1.2462,
2390
+ "step": 1655
2391
+ },
2392
+ {
2393
+ "epoch": 0.39975918121613485,
2394
+ "grad_norm": 0.6964198350906372,
2395
+ "learning_rate": 1.2003853564547207e-05,
2396
+ "loss": 1.1537,
2397
+ "step": 1660
2398
+ },
2399
+ {
2400
+ "epoch": 0.40096327513546054,
2401
+ "grad_norm": 0.7141043543815613,
2402
+ "learning_rate": 1.1979768786127168e-05,
2403
+ "loss": 1.1106,
2404
+ "step": 1665
2405
+ },
2406
+ {
2407
+ "epoch": 0.4021673690547863,
2408
+ "grad_norm": 0.7486653327941895,
2409
+ "learning_rate": 1.1955684007707128e-05,
2410
+ "loss": 1.1255,
2411
+ "step": 1670
2412
+ },
2413
+ {
2414
+ "epoch": 0.40337146297411197,
2415
+ "grad_norm": 0.68989098072052,
2416
+ "learning_rate": 1.1931599229287092e-05,
2417
+ "loss": 1.1491,
2418
+ "step": 1675
2419
+ },
2420
+ {
2421
+ "epoch": 0.4045755568934377,
2422
+ "grad_norm": 0.7117523550987244,
2423
+ "learning_rate": 1.1907514450867053e-05,
2424
+ "loss": 1.2564,
2425
+ "step": 1680
2426
+ },
2427
+ {
2428
+ "epoch": 0.4057796508127634,
2429
+ "grad_norm": 0.7256306409835815,
2430
+ "learning_rate": 1.1883429672447015e-05,
2431
+ "loss": 1.2544,
2432
+ "step": 1685
2433
+ },
2434
+ {
2435
+ "epoch": 0.4069837447320891,
2436
+ "grad_norm": 0.6354299187660217,
2437
+ "learning_rate": 1.1859344894026976e-05,
2438
+ "loss": 1.1011,
2439
+ "step": 1690
2440
+ },
2441
+ {
2442
+ "epoch": 0.40818783865141484,
2443
+ "grad_norm": 0.7653043270111084,
2444
+ "learning_rate": 1.1835260115606937e-05,
2445
+ "loss": 1.1881,
2446
+ "step": 1695
2447
+ },
2448
+ {
2449
+ "epoch": 0.4093919325707405,
2450
+ "grad_norm": 0.6969402432441711,
2451
+ "learning_rate": 1.1811175337186899e-05,
2452
+ "loss": 1.2509,
2453
+ "step": 1700
2454
+ },
2455
+ {
2456
+ "epoch": 0.4105960264900662,
2457
+ "grad_norm": 0.7759403586387634,
2458
+ "learning_rate": 1.1787090558766861e-05,
2459
+ "loss": 1.1777,
2460
+ "step": 1705
2461
+ },
2462
+ {
2463
+ "epoch": 0.41180012040939196,
2464
+ "grad_norm": 0.8177191019058228,
2465
+ "learning_rate": 1.1763005780346822e-05,
2466
+ "loss": 1.2677,
2467
+ "step": 1710
2468
+ },
2469
+ {
2470
+ "epoch": 0.41300421432871764,
2471
+ "grad_norm": 0.6352664828300476,
2472
+ "learning_rate": 1.1738921001926784e-05,
2473
+ "loss": 1.107,
2474
+ "step": 1715
2475
+ },
2476
+ {
2477
+ "epoch": 0.41420830824804333,
2478
+ "grad_norm": 0.7194890975952148,
2479
+ "learning_rate": 1.1714836223506745e-05,
2480
+ "loss": 1.1646,
2481
+ "step": 1720
2482
+ },
2483
+ {
2484
+ "epoch": 0.4154124021673691,
2485
+ "grad_norm": 0.6629904508590698,
2486
+ "learning_rate": 1.1690751445086706e-05,
2487
+ "loss": 1.1938,
2488
+ "step": 1725
2489
+ },
2490
+ {
2491
+ "epoch": 0.41661649608669477,
2492
+ "grad_norm": 0.6937288641929626,
2493
+ "learning_rate": 1.1666666666666668e-05,
2494
+ "loss": 1.1254,
2495
+ "step": 1730
2496
+ },
2497
+ {
2498
+ "epoch": 0.41782059000602045,
2499
+ "grad_norm": 0.6822580099105835,
2500
+ "learning_rate": 1.1642581888246629e-05,
2501
+ "loss": 1.1878,
2502
+ "step": 1735
2503
+ },
2504
+ {
2505
+ "epoch": 0.4190246839253462,
2506
+ "grad_norm": 0.8951644897460938,
2507
+ "learning_rate": 1.161849710982659e-05,
2508
+ "loss": 1.0153,
2509
+ "step": 1740
2510
+ },
2511
+ {
2512
+ "epoch": 0.4202287778446719,
2513
+ "grad_norm": 0.7985068559646606,
2514
+ "learning_rate": 1.1594412331406553e-05,
2515
+ "loss": 1.2309,
2516
+ "step": 1745
2517
+ },
2518
+ {
2519
+ "epoch": 0.4214328717639976,
2520
+ "grad_norm": 0.7742875814437866,
2521
+ "learning_rate": 1.1570327552986514e-05,
2522
+ "loss": 1.2868,
2523
+ "step": 1750
2524
+ },
2525
+ {
2526
+ "epoch": 0.4226369656833233,
2527
+ "grad_norm": 0.7290802001953125,
2528
+ "learning_rate": 1.1546242774566474e-05,
2529
+ "loss": 1.1529,
2530
+ "step": 1755
2531
+ },
2532
+ {
2533
+ "epoch": 0.423841059602649,
2534
+ "grad_norm": 0.7852470278739929,
2535
+ "learning_rate": 1.1522157996146437e-05,
2536
+ "loss": 1.1758,
2537
+ "step": 1760
2538
+ },
2539
+ {
2540
+ "epoch": 0.4250451535219747,
2541
+ "grad_norm": 0.6864736676216125,
2542
+ "learning_rate": 1.1498073217726397e-05,
2543
+ "loss": 1.173,
2544
+ "step": 1765
2545
+ },
2546
+ {
2547
+ "epoch": 0.42624924744130044,
2548
+ "grad_norm": 0.6752791404724121,
2549
+ "learning_rate": 1.1473988439306358e-05,
2550
+ "loss": 1.101,
2551
+ "step": 1770
2552
+ },
2553
+ {
2554
+ "epoch": 0.4274533413606261,
2555
+ "grad_norm": 0.6282265186309814,
2556
+ "learning_rate": 1.144990366088632e-05,
2557
+ "loss": 1.0218,
2558
+ "step": 1775
2559
+ },
2560
+ {
2561
+ "epoch": 0.4286574352799518,
2562
+ "grad_norm": 0.7045950889587402,
2563
+ "learning_rate": 1.1425818882466281e-05,
2564
+ "loss": 1.2667,
2565
+ "step": 1780
2566
+ },
2567
+ {
2568
+ "epoch": 0.42986152919927756,
2569
+ "grad_norm": 0.7136130928993225,
2570
+ "learning_rate": 1.1401734104046245e-05,
2571
+ "loss": 1.2445,
2572
+ "step": 1785
2573
+ },
2574
+ {
2575
+ "epoch": 0.43106562311860325,
2576
+ "grad_norm": 0.7001535296440125,
2577
+ "learning_rate": 1.1377649325626206e-05,
2578
+ "loss": 1.2076,
2579
+ "step": 1790
2580
+ },
2581
+ {
2582
+ "epoch": 0.43226971703792894,
2583
+ "grad_norm": 0.7172293066978455,
2584
+ "learning_rate": 1.1353564547206166e-05,
2585
+ "loss": 1.0828,
2586
+ "step": 1795
2587
+ },
2588
+ {
2589
+ "epoch": 0.4334738109572547,
2590
+ "grad_norm": 0.7160560488700867,
2591
+ "learning_rate": 1.1329479768786129e-05,
2592
+ "loss": 1.1676,
2593
+ "step": 1800
2594
+ },
2595
+ {
2596
+ "epoch": 0.4334738109572547,
2597
+ "eval_loss": 1.1778274774551392,
2598
+ "eval_runtime": 10.4751,
2599
+ "eval_samples_per_second": 9.546,
2600
+ "eval_steps_per_second": 1.241,
2601
+ "step": 1800
2602
+ },
2603
+ {
2604
+ "epoch": 0.43467790487658037,
2605
+ "grad_norm": 0.7364829778671265,
2606
+ "learning_rate": 1.130539499036609e-05,
2607
+ "loss": 1.1337,
2608
+ "step": 1805
2609
+ },
2610
+ {
2611
+ "epoch": 0.43588199879590606,
2612
+ "grad_norm": 0.7415775656700134,
2613
+ "learning_rate": 1.128131021194605e-05,
2614
+ "loss": 1.1882,
2615
+ "step": 1810
2616
+ },
2617
+ {
2618
+ "epoch": 0.4370860927152318,
2619
+ "grad_norm": 0.7825621366500854,
2620
+ "learning_rate": 1.1257225433526014e-05,
2621
+ "loss": 1.1627,
2622
+ "step": 1815
2623
+ },
2624
+ {
2625
+ "epoch": 0.4382901866345575,
2626
+ "grad_norm": 0.7862880229949951,
2627
+ "learning_rate": 1.1233140655105975e-05,
2628
+ "loss": 1.2287,
2629
+ "step": 1820
2630
+ },
2631
+ {
2632
+ "epoch": 0.4394942805538832,
2633
+ "grad_norm": 0.7740750908851624,
2634
+ "learning_rate": 1.1209055876685935e-05,
2635
+ "loss": 1.1488,
2636
+ "step": 1825
2637
+ },
2638
+ {
2639
+ "epoch": 0.4406983744732089,
2640
+ "grad_norm": 0.8551940321922302,
2641
+ "learning_rate": 1.1184971098265898e-05,
2642
+ "loss": 1.1894,
2643
+ "step": 1830
2644
+ },
2645
+ {
2646
+ "epoch": 0.4419024683925346,
2647
+ "grad_norm": 0.7718561291694641,
2648
+ "learning_rate": 1.1160886319845858e-05,
2649
+ "loss": 1.1825,
2650
+ "step": 1835
2651
+ },
2652
+ {
2653
+ "epoch": 0.44310656231186035,
2654
+ "grad_norm": 0.7665084600448608,
2655
+ "learning_rate": 1.1136801541425819e-05,
2656
+ "loss": 1.0931,
2657
+ "step": 1840
2658
+ },
2659
+ {
2660
+ "epoch": 0.44431065623118604,
2661
+ "grad_norm": 0.7155455350875854,
2662
+ "learning_rate": 1.1112716763005781e-05,
2663
+ "loss": 1.2208,
2664
+ "step": 1845
2665
+ },
2666
+ {
2667
+ "epoch": 0.44551475015051173,
2668
+ "grad_norm": 0.6576226353645325,
2669
+ "learning_rate": 1.1088631984585742e-05,
2670
+ "loss": 1.1294,
2671
+ "step": 1850
2672
+ },
2673
+ {
2674
+ "epoch": 0.4467188440698375,
2675
+ "grad_norm": 0.6880900859832764,
2676
+ "learning_rate": 1.1064547206165703e-05,
2677
+ "loss": 1.0646,
2678
+ "step": 1855
2679
+ },
2680
+ {
2681
+ "epoch": 0.44792293798916316,
2682
+ "grad_norm": 0.633608341217041,
2683
+ "learning_rate": 1.1040462427745667e-05,
2684
+ "loss": 1.0871,
2685
+ "step": 1860
2686
+ },
2687
+ {
2688
+ "epoch": 0.44912703190848885,
2689
+ "grad_norm": 0.6906594634056091,
2690
+ "learning_rate": 1.1016377649325627e-05,
2691
+ "loss": 1.1449,
2692
+ "step": 1865
2693
+ },
2694
+ {
2695
+ "epoch": 0.4503311258278146,
2696
+ "grad_norm": 0.7329970598220825,
2697
+ "learning_rate": 1.0992292870905588e-05,
2698
+ "loss": 1.1158,
2699
+ "step": 1870
2700
+ },
2701
+ {
2702
+ "epoch": 0.4515352197471403,
2703
+ "grad_norm": 0.7347983121871948,
2704
+ "learning_rate": 1.096820809248555e-05,
2705
+ "loss": 1.1252,
2706
+ "step": 1875
2707
+ },
2708
+ {
2709
+ "epoch": 0.45273931366646597,
2710
+ "grad_norm": 0.7085864543914795,
2711
+ "learning_rate": 1.094412331406551e-05,
2712
+ "loss": 1.1535,
2713
+ "step": 1880
2714
+ },
2715
+ {
2716
+ "epoch": 0.4539434075857917,
2717
+ "grad_norm": 0.7692695260047913,
2718
+ "learning_rate": 1.0920038535645471e-05,
2719
+ "loss": 1.1632,
2720
+ "step": 1885
2721
+ },
2722
+ {
2723
+ "epoch": 0.4551475015051174,
2724
+ "grad_norm": 0.7081477046012878,
2725
+ "learning_rate": 1.0895953757225436e-05,
2726
+ "loss": 1.1269,
2727
+ "step": 1890
2728
+ },
2729
+ {
2730
+ "epoch": 0.4563515954244431,
2731
+ "grad_norm": 0.663478434085846,
2732
+ "learning_rate": 1.0871868978805396e-05,
2733
+ "loss": 1.1256,
2734
+ "step": 1895
2735
+ },
2736
+ {
2737
+ "epoch": 0.45755568934376883,
2738
+ "grad_norm": 0.7210192084312439,
2739
+ "learning_rate": 1.0847784200385358e-05,
2740
+ "loss": 1.1887,
2741
+ "step": 1900
2742
+ },
2743
+ {
2744
+ "epoch": 0.4587597832630945,
2745
+ "grad_norm": 0.7571423649787903,
2746
+ "learning_rate": 1.0823699421965319e-05,
2747
+ "loss": 1.1451,
2748
+ "step": 1905
2749
+ },
2750
+ {
2751
+ "epoch": 0.4599638771824202,
2752
+ "grad_norm": 0.79608154296875,
2753
+ "learning_rate": 1.079961464354528e-05,
2754
+ "loss": 1.1874,
2755
+ "step": 1910
2756
+ },
2757
+ {
2758
+ "epoch": 0.46116797110174595,
2759
+ "grad_norm": 0.857712984085083,
2760
+ "learning_rate": 1.0775529865125242e-05,
2761
+ "loss": 1.1669,
2762
+ "step": 1915
2763
+ },
2764
+ {
2765
+ "epoch": 0.46237206502107164,
2766
+ "grad_norm": 0.644782304763794,
2767
+ "learning_rate": 1.0751445086705203e-05,
2768
+ "loss": 1.1627,
2769
+ "step": 1920
2770
+ },
2771
+ {
2772
+ "epoch": 0.46357615894039733,
2773
+ "grad_norm": 0.6408756971359253,
2774
+ "learning_rate": 1.0727360308285163e-05,
2775
+ "loss": 1.069,
2776
+ "step": 1925
2777
+ },
2778
+ {
2779
+ "epoch": 0.4647802528597231,
2780
+ "grad_norm": 0.6929676532745361,
2781
+ "learning_rate": 1.0703275529865127e-05,
2782
+ "loss": 1.1435,
2783
+ "step": 1930
2784
+ },
2785
+ {
2786
+ "epoch": 0.46598434677904876,
2787
+ "grad_norm": 0.7538990378379822,
2788
+ "learning_rate": 1.0679190751445088e-05,
2789
+ "loss": 1.2604,
2790
+ "step": 1935
2791
+ },
2792
+ {
2793
+ "epoch": 0.46718844069837445,
2794
+ "grad_norm": 0.6711621880531311,
2795
+ "learning_rate": 1.0655105973025049e-05,
2796
+ "loss": 1.1923,
2797
+ "step": 1940
2798
+ },
2799
+ {
2800
+ "epoch": 0.4683925346177002,
2801
+ "grad_norm": 0.6384220123291016,
2802
+ "learning_rate": 1.0631021194605011e-05,
2803
+ "loss": 1.1269,
2804
+ "step": 1945
2805
+ },
2806
+ {
2807
+ "epoch": 0.4695966285370259,
2808
+ "grad_norm": 0.682374119758606,
2809
+ "learning_rate": 1.0606936416184972e-05,
2810
+ "loss": 1.1268,
2811
+ "step": 1950
2812
+ },
2813
+ {
2814
+ "epoch": 0.4708007224563516,
2815
+ "grad_norm": 0.7387133240699768,
2816
+ "learning_rate": 1.0582851637764932e-05,
2817
+ "loss": 1.1043,
2818
+ "step": 1955
2819
+ },
2820
+ {
2821
+ "epoch": 0.4720048163756773,
2822
+ "grad_norm": 0.6632955074310303,
2823
+ "learning_rate": 1.0558766859344896e-05,
2824
+ "loss": 1.1148,
2825
+ "step": 1960
2826
+ },
2827
+ {
2828
+ "epoch": 0.473208910295003,
2829
+ "grad_norm": 0.6747167706489563,
2830
+ "learning_rate": 1.0534682080924857e-05,
2831
+ "loss": 1.1788,
2832
+ "step": 1965
2833
+ },
2834
+ {
2835
+ "epoch": 0.4744130042143287,
2836
+ "grad_norm": 0.6498105525970459,
2837
+ "learning_rate": 1.0510597302504818e-05,
2838
+ "loss": 1.2199,
2839
+ "step": 1970
2840
+ },
2841
+ {
2842
+ "epoch": 0.47561709813365444,
2843
+ "grad_norm": 0.72532057762146,
2844
+ "learning_rate": 1.048651252408478e-05,
2845
+ "loss": 0.9953,
2846
+ "step": 1975
2847
+ },
2848
+ {
2849
+ "epoch": 0.4768211920529801,
2850
+ "grad_norm": 0.7126530408859253,
2851
+ "learning_rate": 1.046242774566474e-05,
2852
+ "loss": 1.1014,
2853
+ "step": 1980
2854
+ },
2855
+ {
2856
+ "epoch": 0.4780252859723058,
2857
+ "grad_norm": 0.8094469904899597,
2858
+ "learning_rate": 1.0438342967244701e-05,
2859
+ "loss": 1.0567,
2860
+ "step": 1985
2861
+ },
2862
+ {
2863
+ "epoch": 0.47922937989163156,
2864
+ "grad_norm": 0.7181186676025391,
2865
+ "learning_rate": 1.0414258188824664e-05,
2866
+ "loss": 1.1077,
2867
+ "step": 1990
2868
+ },
2869
+ {
2870
+ "epoch": 0.48043347381095725,
2871
+ "grad_norm": 0.6791627407073975,
2872
+ "learning_rate": 1.0390173410404624e-05,
2873
+ "loss": 1.1272,
2874
+ "step": 1995
2875
+ },
2876
+ {
2877
+ "epoch": 0.481637567730283,
2878
+ "grad_norm": 0.7202663421630859,
2879
+ "learning_rate": 1.0366088631984585e-05,
2880
+ "loss": 1.139,
2881
+ "step": 2000
2882
+ },
2883
+ {
2884
+ "epoch": 0.481637567730283,
2885
+ "eval_loss": 1.1731120347976685,
2886
+ "eval_runtime": 10.4779,
2887
+ "eval_samples_per_second": 9.544,
2888
+ "eval_steps_per_second": 1.241,
2889
+ "step": 2000
2890
+ }
2891
+ ],
2892
+ "logging_steps": 5,
2893
+ "max_steps": 4152,
2894
+ "num_input_tokens_seen": 0,
2895
+ "num_train_epochs": 1,
2896
+ "save_steps": 500,
2897
+ "stateful_callbacks": {
2898
+ "TrainerControl": {
2899
+ "args": {
2900
+ "should_epoch_stop": false,
2901
+ "should_evaluate": false,
2902
+ "should_log": false,
2903
+ "should_save": true,
2904
+ "should_training_stop": false
2905
+ },
2906
+ "attributes": {}
2907
+ }
2908
+ },
2909
+ "total_flos": 9755901083039616.0,
2910
+ "train_batch_size": 1,
2911
+ "trial_name": null,
2912
+ "trial_params": null
2913
+ }
checkpoint-2000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:86602ecfa5f4389c74519a3cbca0206cb3e9a395b2bd494b341ef1183c4e9454
3
+ size 5816
checkpoint-2000/vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-2500/config.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "ewre324/ewre324-Thinker-SmolLM2-135M-Instruct-Reasoning",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "head_dim": 64,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 576,
13
+ "initializer_range": 0.041666666666666664,
14
+ "intermediate_size": 1536,
15
+ "is_llama_config": true,
16
+ "max_position_embeddings": 8192,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "num_attention_heads": 9,
20
+ "num_hidden_layers": 30,
21
+ "num_key_value_heads": 3,
22
+ "pad_token_id": 2,
23
+ "pretraining_tp": 1,
24
+ "rms_norm_eps": 1e-05,
25
+ "rope_interleaved": false,
26
+ "rope_scaling": null,
27
+ "rope_theta": 100000,
28
+ "tie_word_embeddings": true,
29
+ "torch_dtype": "float32",
30
+ "transformers.js_config": {
31
+ "kv_cache_dtype": {
32
+ "fp16": "float16",
33
+ "q4f16": "float16"
34
+ }
35
+ },
36
+ "transformers_version": "4.49.0.dev0",
37
+ "use_cache": false,
38
+ "vocab_size": 49152
39
+ }
checkpoint-2500/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 2,
6
+ "transformers_version": "4.49.0.dev0"
7
+ }
checkpoint-2500/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-2500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:571dc96d919eb9409f1ba2c6b035386239d5662f6d38758d3a46deae5e37b5ec
3
+ size 538090408
checkpoint-2500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:469dd6597ed712043bd385bbe1f480fb6e96128932c9c68a82b641b2d0e7f39f
3
+ size 1076349050
checkpoint-2500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7682299c566684ea51cf26f0c86b6ffaa3c0bc63cbdf84674b29a2c62ac72143
3
+ size 14244
checkpoint-2500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed5b1b6578240bf12045c4122be0a86777e9482729c1838710b7ba020d56f195
3
+ size 1064
checkpoint-2500/special_tokens_map.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>"
5
+ ],
6
+ "bos_token": {
7
+ "content": "<|im_start|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false
12
+ },
13
+ "eos_token": {
14
+ "content": "<|im_end|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ },
20
+ "pad_token": "<|im_end|>",
21
+ "unk_token": {
22
+ "content": "<|endoftext|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false
27
+ }
28
+ }
checkpoint-2500/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-2500/tokenizer_config.json ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "0": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "1": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "2": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ },
28
+ "3": {
29
+ "content": "<repo_name>",
30
+ "lstrip": false,
31
+ "normalized": false,
32
+ "rstrip": false,
33
+ "single_word": false,
34
+ "special": true
35
+ },
36
+ "4": {
37
+ "content": "<reponame>",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false,
42
+ "special": true
43
+ },
44
+ "5": {
45
+ "content": "<file_sep>",
46
+ "lstrip": false,
47
+ "normalized": false,
48
+ "rstrip": false,
49
+ "single_word": false,
50
+ "special": true
51
+ },
52
+ "6": {
53
+ "content": "<filename>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false,
58
+ "special": true
59
+ },
60
+ "7": {
61
+ "content": "<gh_stars>",
62
+ "lstrip": false,
63
+ "normalized": false,
64
+ "rstrip": false,
65
+ "single_word": false,
66
+ "special": true
67
+ },
68
+ "8": {
69
+ "content": "<issue_start>",
70
+ "lstrip": false,
71
+ "normalized": false,
72
+ "rstrip": false,
73
+ "single_word": false,
74
+ "special": true
75
+ },
76
+ "9": {
77
+ "content": "<issue_comment>",
78
+ "lstrip": false,
79
+ "normalized": false,
80
+ "rstrip": false,
81
+ "single_word": false,
82
+ "special": true
83
+ },
84
+ "10": {
85
+ "content": "<issue_closed>",
86
+ "lstrip": false,
87
+ "normalized": false,
88
+ "rstrip": false,
89
+ "single_word": false,
90
+ "special": true
91
+ },
92
+ "11": {
93
+ "content": "<jupyter_start>",
94
+ "lstrip": false,
95
+ "normalized": false,
96
+ "rstrip": false,
97
+ "single_word": false,
98
+ "special": true
99
+ },
100
+ "12": {
101
+ "content": "<jupyter_text>",
102
+ "lstrip": false,
103
+ "normalized": false,
104
+ "rstrip": false,
105
+ "single_word": false,
106
+ "special": true
107
+ },
108
+ "13": {
109
+ "content": "<jupyter_code>",
110
+ "lstrip": false,
111
+ "normalized": false,
112
+ "rstrip": false,
113
+ "single_word": false,
114
+ "special": true
115
+ },
116
+ "14": {
117
+ "content": "<jupyter_output>",
118
+ "lstrip": false,
119
+ "normalized": false,
120
+ "rstrip": false,
121
+ "single_word": false,
122
+ "special": true
123
+ },
124
+ "15": {
125
+ "content": "<jupyter_script>",
126
+ "lstrip": false,
127
+ "normalized": false,
128
+ "rstrip": false,
129
+ "single_word": false,
130
+ "special": true
131
+ },
132
+ "16": {
133
+ "content": "<empty_output>",
134
+ "lstrip": false,
135
+ "normalized": false,
136
+ "rstrip": false,
137
+ "single_word": false,
138
+ "special": true
139
+ }
140
+ },
141
+ "additional_special_tokens": [
142
+ "<|im_start|>",
143
+ "<|im_end|>"
144
+ ],
145
+ "bos_token": "<|im_start|>",
146
+ "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
147
+ "clean_up_tokenization_spaces": false,
148
+ "eos_token": "<|im_end|>",
149
+ "extra_special_tokens": {},
150
+ "max_length": 2048,
151
+ "model_max_length": 2048,
152
+ "pad_to_multiple_of": null,
153
+ "pad_token": "<|im_end|>",
154
+ "pad_token_type_id": 0,
155
+ "padding_side": "right",
156
+ "stride": 0,
157
+ "tokenizer_class": "GPT2Tokenizer",
158
+ "truncation_side": "right",
159
+ "truncation_strategy": "longest_first",
160
+ "unk_token": "<|endoftext|>",
161
+ "vocab_size": 49152
162
+ }
checkpoint-2500/trainer_state.json ADDED
@@ -0,0 +1,3629 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.6020469596628537,
5
+ "eval_steps": 200,
6
+ "global_step": 2500,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0012040939193257074,
13
+ "grad_norm": 1.2087849378585815,
14
+ "learning_rate": 1.9975915221579964e-05,
15
+ "loss": 1.7501,
16
+ "step": 5
17
+ },
18
+ {
19
+ "epoch": 0.002408187838651415,
20
+ "grad_norm": 1.0137544870376587,
21
+ "learning_rate": 1.9951830443159926e-05,
22
+ "loss": 1.3294,
23
+ "step": 10
24
+ },
25
+ {
26
+ "epoch": 0.003612281757977122,
27
+ "grad_norm": 0.8906710147857666,
28
+ "learning_rate": 1.9927745664739885e-05,
29
+ "loss": 1.4793,
30
+ "step": 15
31
+ },
32
+ {
33
+ "epoch": 0.00481637567730283,
34
+ "grad_norm": 0.8175792098045349,
35
+ "learning_rate": 1.9903660886319848e-05,
36
+ "loss": 1.4537,
37
+ "step": 20
38
+ },
39
+ {
40
+ "epoch": 0.006020469596628537,
41
+ "grad_norm": 0.8326422572135925,
42
+ "learning_rate": 1.987957610789981e-05,
43
+ "loss": 1.5579,
44
+ "step": 25
45
+ },
46
+ {
47
+ "epoch": 0.007224563515954244,
48
+ "grad_norm": 0.8402993083000183,
49
+ "learning_rate": 1.985549132947977e-05,
50
+ "loss": 1.4146,
51
+ "step": 30
52
+ },
53
+ {
54
+ "epoch": 0.008428657435279952,
55
+ "grad_norm": 0.9211377501487732,
56
+ "learning_rate": 1.983140655105973e-05,
57
+ "loss": 1.4986,
58
+ "step": 35
59
+ },
60
+ {
61
+ "epoch": 0.00963275135460566,
62
+ "grad_norm": 0.7729477286338806,
63
+ "learning_rate": 1.9807321772639694e-05,
64
+ "loss": 1.3949,
65
+ "step": 40
66
+ },
67
+ {
68
+ "epoch": 0.010836845273931367,
69
+ "grad_norm": 0.8256559371948242,
70
+ "learning_rate": 1.9783236994219656e-05,
71
+ "loss": 1.2517,
72
+ "step": 45
73
+ },
74
+ {
75
+ "epoch": 0.012040939193257074,
76
+ "grad_norm": 0.7404232025146484,
77
+ "learning_rate": 1.9759152215799615e-05,
78
+ "loss": 1.4761,
79
+ "step": 50
80
+ },
81
+ {
82
+ "epoch": 0.013245033112582781,
83
+ "grad_norm": 0.773309588432312,
84
+ "learning_rate": 1.9735067437379577e-05,
85
+ "loss": 1.227,
86
+ "step": 55
87
+ },
88
+ {
89
+ "epoch": 0.014449127031908489,
90
+ "grad_norm": 0.7535562515258789,
91
+ "learning_rate": 1.971098265895954e-05,
92
+ "loss": 1.3443,
93
+ "step": 60
94
+ },
95
+ {
96
+ "epoch": 0.015653220951234198,
97
+ "grad_norm": 0.8561655282974243,
98
+ "learning_rate": 1.9686897880539502e-05,
99
+ "loss": 1.3229,
100
+ "step": 65
101
+ },
102
+ {
103
+ "epoch": 0.016857314870559904,
104
+ "grad_norm": 0.7888396978378296,
105
+ "learning_rate": 1.966281310211946e-05,
106
+ "loss": 1.4274,
107
+ "step": 70
108
+ },
109
+ {
110
+ "epoch": 0.018061408789885613,
111
+ "grad_norm": 0.7886033654212952,
112
+ "learning_rate": 1.9638728323699423e-05,
113
+ "loss": 1.4469,
114
+ "step": 75
115
+ },
116
+ {
117
+ "epoch": 0.01926550270921132,
118
+ "grad_norm": 0.8100034594535828,
119
+ "learning_rate": 1.9614643545279385e-05,
120
+ "loss": 1.314,
121
+ "step": 80
122
+ },
123
+ {
124
+ "epoch": 0.020469596628537028,
125
+ "grad_norm": 0.7786366939544678,
126
+ "learning_rate": 1.9590558766859348e-05,
127
+ "loss": 1.2631,
128
+ "step": 85
129
+ },
130
+ {
131
+ "epoch": 0.021673690547862733,
132
+ "grad_norm": 0.8259329795837402,
133
+ "learning_rate": 1.9566473988439307e-05,
134
+ "loss": 1.4006,
135
+ "step": 90
136
+ },
137
+ {
138
+ "epoch": 0.022877784467188442,
139
+ "grad_norm": 0.7402159571647644,
140
+ "learning_rate": 1.954238921001927e-05,
141
+ "loss": 1.416,
142
+ "step": 95
143
+ },
144
+ {
145
+ "epoch": 0.024081878386514148,
146
+ "grad_norm": 0.8619882464408875,
147
+ "learning_rate": 1.951830443159923e-05,
148
+ "loss": 1.2583,
149
+ "step": 100
150
+ },
151
+ {
152
+ "epoch": 0.025285972305839857,
153
+ "grad_norm": 0.7436647415161133,
154
+ "learning_rate": 1.949421965317919e-05,
155
+ "loss": 1.3527,
156
+ "step": 105
157
+ },
158
+ {
159
+ "epoch": 0.026490066225165563,
160
+ "grad_norm": 0.7261799573898315,
161
+ "learning_rate": 1.9470134874759156e-05,
162
+ "loss": 1.4234,
163
+ "step": 110
164
+ },
165
+ {
166
+ "epoch": 0.027694160144491272,
167
+ "grad_norm": 0.8648309707641602,
168
+ "learning_rate": 1.9446050096339115e-05,
169
+ "loss": 1.2541,
170
+ "step": 115
171
+ },
172
+ {
173
+ "epoch": 0.028898254063816978,
174
+ "grad_norm": 0.7091889381408691,
175
+ "learning_rate": 1.9421965317919077e-05,
176
+ "loss": 1.2473,
177
+ "step": 120
178
+ },
179
+ {
180
+ "epoch": 0.030102347983142687,
181
+ "grad_norm": 0.7450091242790222,
182
+ "learning_rate": 1.939788053949904e-05,
183
+ "loss": 1.2555,
184
+ "step": 125
185
+ },
186
+ {
187
+ "epoch": 0.031306441902468396,
188
+ "grad_norm": 0.8190895318984985,
189
+ "learning_rate": 1.9373795761079e-05,
190
+ "loss": 1.2874,
191
+ "step": 130
192
+ },
193
+ {
194
+ "epoch": 0.0325105358217941,
195
+ "grad_norm": 0.7458487749099731,
196
+ "learning_rate": 1.934971098265896e-05,
197
+ "loss": 1.4139,
198
+ "step": 135
199
+ },
200
+ {
201
+ "epoch": 0.03371462974111981,
202
+ "grad_norm": 0.806698203086853,
203
+ "learning_rate": 1.9325626204238923e-05,
204
+ "loss": 1.4061,
205
+ "step": 140
206
+ },
207
+ {
208
+ "epoch": 0.034918723660445516,
209
+ "grad_norm": 0.7941407561302185,
210
+ "learning_rate": 1.9301541425818882e-05,
211
+ "loss": 1.2819,
212
+ "step": 145
213
+ },
214
+ {
215
+ "epoch": 0.036122817579771226,
216
+ "grad_norm": 0.6824349164962769,
217
+ "learning_rate": 1.9277456647398845e-05,
218
+ "loss": 1.3025,
219
+ "step": 150
220
+ },
221
+ {
222
+ "epoch": 0.03732691149909693,
223
+ "grad_norm": 0.8056986927986145,
224
+ "learning_rate": 1.9253371868978807e-05,
225
+ "loss": 1.1617,
226
+ "step": 155
227
+ },
228
+ {
229
+ "epoch": 0.03853100541842264,
230
+ "grad_norm": 0.7486716508865356,
231
+ "learning_rate": 1.922928709055877e-05,
232
+ "loss": 1.3523,
233
+ "step": 160
234
+ },
235
+ {
236
+ "epoch": 0.039735099337748346,
237
+ "grad_norm": 0.7773532867431641,
238
+ "learning_rate": 1.920520231213873e-05,
239
+ "loss": 1.3013,
240
+ "step": 165
241
+ },
242
+ {
243
+ "epoch": 0.040939193257074055,
244
+ "grad_norm": 0.7498481869697571,
245
+ "learning_rate": 1.918111753371869e-05,
246
+ "loss": 1.2563,
247
+ "step": 170
248
+ },
249
+ {
250
+ "epoch": 0.04214328717639976,
251
+ "grad_norm": 0.7634468078613281,
252
+ "learning_rate": 1.9157032755298653e-05,
253
+ "loss": 1.2365,
254
+ "step": 175
255
+ },
256
+ {
257
+ "epoch": 0.04334738109572547,
258
+ "grad_norm": 0.6561328172683716,
259
+ "learning_rate": 1.9132947976878615e-05,
260
+ "loss": 1.3058,
261
+ "step": 180
262
+ },
263
+ {
264
+ "epoch": 0.044551475015051176,
265
+ "grad_norm": 0.7443628907203674,
266
+ "learning_rate": 1.9108863198458578e-05,
267
+ "loss": 1.3264,
268
+ "step": 185
269
+ },
270
+ {
271
+ "epoch": 0.045755568934376885,
272
+ "grad_norm": 0.6697310209274292,
273
+ "learning_rate": 1.9084778420038536e-05,
274
+ "loss": 1.1029,
275
+ "step": 190
276
+ },
277
+ {
278
+ "epoch": 0.04695966285370259,
279
+ "grad_norm": 0.7119889259338379,
280
+ "learning_rate": 1.90606936416185e-05,
281
+ "loss": 1.2503,
282
+ "step": 195
283
+ },
284
+ {
285
+ "epoch": 0.048163756773028296,
286
+ "grad_norm": 0.8120700120925903,
287
+ "learning_rate": 1.903660886319846e-05,
288
+ "loss": 1.246,
289
+ "step": 200
290
+ },
291
+ {
292
+ "epoch": 0.048163756773028296,
293
+ "eval_loss": 1.2944953441619873,
294
+ "eval_runtime": 10.5151,
295
+ "eval_samples_per_second": 9.51,
296
+ "eval_steps_per_second": 1.236,
297
+ "step": 200
298
+ },
299
+ {
300
+ "epoch": 0.049367850692354005,
301
+ "grad_norm": 0.7876719832420349,
302
+ "learning_rate": 1.901252408477842e-05,
303
+ "loss": 1.2097,
304
+ "step": 205
305
+ },
306
+ {
307
+ "epoch": 0.050571944611679714,
308
+ "grad_norm": 0.7367354035377502,
309
+ "learning_rate": 1.8988439306358382e-05,
310
+ "loss": 1.3332,
311
+ "step": 210
312
+ },
313
+ {
314
+ "epoch": 0.05177603853100542,
315
+ "grad_norm": 0.7384558320045471,
316
+ "learning_rate": 1.8964354527938345e-05,
317
+ "loss": 1.3897,
318
+ "step": 215
319
+ },
320
+ {
321
+ "epoch": 0.052980132450331126,
322
+ "grad_norm": 0.8261380195617676,
323
+ "learning_rate": 1.8940269749518304e-05,
324
+ "loss": 1.2677,
325
+ "step": 220
326
+ },
327
+ {
328
+ "epoch": 0.054184226369656835,
329
+ "grad_norm": 0.7237825989723206,
330
+ "learning_rate": 1.891618497109827e-05,
331
+ "loss": 1.2919,
332
+ "step": 225
333
+ },
334
+ {
335
+ "epoch": 0.055388320288982544,
336
+ "grad_norm": 0.7296762466430664,
337
+ "learning_rate": 1.889210019267823e-05,
338
+ "loss": 1.2977,
339
+ "step": 230
340
+ },
341
+ {
342
+ "epoch": 0.056592414208308246,
343
+ "grad_norm": 0.8224279880523682,
344
+ "learning_rate": 1.886801541425819e-05,
345
+ "loss": 1.3751,
346
+ "step": 235
347
+ },
348
+ {
349
+ "epoch": 0.057796508127633955,
350
+ "grad_norm": 0.7233418822288513,
351
+ "learning_rate": 1.8843930635838153e-05,
352
+ "loss": 1.293,
353
+ "step": 240
354
+ },
355
+ {
356
+ "epoch": 0.059000602046959665,
357
+ "grad_norm": 0.7015657424926758,
358
+ "learning_rate": 1.8819845857418112e-05,
359
+ "loss": 1.3093,
360
+ "step": 245
361
+ },
362
+ {
363
+ "epoch": 0.060204695966285374,
364
+ "grad_norm": 0.8122280240058899,
365
+ "learning_rate": 1.8795761078998074e-05,
366
+ "loss": 1.3979,
367
+ "step": 250
368
+ },
369
+ {
370
+ "epoch": 0.061408789885611076,
371
+ "grad_norm": 0.7582533955574036,
372
+ "learning_rate": 1.8771676300578037e-05,
373
+ "loss": 1.3147,
374
+ "step": 255
375
+ },
376
+ {
377
+ "epoch": 0.06261288380493679,
378
+ "grad_norm": 0.746967077255249,
379
+ "learning_rate": 1.8747591522158e-05,
380
+ "loss": 1.1785,
381
+ "step": 260
382
+ },
383
+ {
384
+ "epoch": 0.0638169777242625,
385
+ "grad_norm": 0.7588161826133728,
386
+ "learning_rate": 1.8723506743737958e-05,
387
+ "loss": 1.3648,
388
+ "step": 265
389
+ },
390
+ {
391
+ "epoch": 0.0650210716435882,
392
+ "grad_norm": 0.696869432926178,
393
+ "learning_rate": 1.869942196531792e-05,
394
+ "loss": 1.2854,
395
+ "step": 270
396
+ },
397
+ {
398
+ "epoch": 0.06622516556291391,
399
+ "grad_norm": 0.8297455906867981,
400
+ "learning_rate": 1.8675337186897883e-05,
401
+ "loss": 1.2011,
402
+ "step": 275
403
+ },
404
+ {
405
+ "epoch": 0.06742925948223961,
406
+ "grad_norm": 0.7183229327201843,
407
+ "learning_rate": 1.8651252408477845e-05,
408
+ "loss": 1.1944,
409
+ "step": 280
410
+ },
411
+ {
412
+ "epoch": 0.06863335340156532,
413
+ "grad_norm": 0.7092033624649048,
414
+ "learning_rate": 1.8627167630057804e-05,
415
+ "loss": 1.3296,
416
+ "step": 285
417
+ },
418
+ {
419
+ "epoch": 0.06983744732089103,
420
+ "grad_norm": 0.7024344801902771,
421
+ "learning_rate": 1.8603082851637766e-05,
422
+ "loss": 1.2349,
423
+ "step": 290
424
+ },
425
+ {
426
+ "epoch": 0.07104154124021674,
427
+ "grad_norm": 0.7504032254219055,
428
+ "learning_rate": 1.857899807321773e-05,
429
+ "loss": 1.1986,
430
+ "step": 295
431
+ },
432
+ {
433
+ "epoch": 0.07224563515954245,
434
+ "grad_norm": 0.7584034204483032,
435
+ "learning_rate": 1.855491329479769e-05,
436
+ "loss": 1.2998,
437
+ "step": 300
438
+ },
439
+ {
440
+ "epoch": 0.07344972907886815,
441
+ "grad_norm": 0.7563251852989197,
442
+ "learning_rate": 1.853082851637765e-05,
443
+ "loss": 1.3703,
444
+ "step": 305
445
+ },
446
+ {
447
+ "epoch": 0.07465382299819386,
448
+ "grad_norm": 0.728660523891449,
449
+ "learning_rate": 1.8506743737957612e-05,
450
+ "loss": 1.2916,
451
+ "step": 310
452
+ },
453
+ {
454
+ "epoch": 0.07585791691751957,
455
+ "grad_norm": 0.6596492528915405,
456
+ "learning_rate": 1.8482658959537575e-05,
457
+ "loss": 1.1821,
458
+ "step": 315
459
+ },
460
+ {
461
+ "epoch": 0.07706201083684527,
462
+ "grad_norm": 0.7137461304664612,
463
+ "learning_rate": 1.8458574181117533e-05,
464
+ "loss": 1.1982,
465
+ "step": 320
466
+ },
467
+ {
468
+ "epoch": 0.07826610475617098,
469
+ "grad_norm": 0.7411230206489563,
470
+ "learning_rate": 1.8434489402697496e-05,
471
+ "loss": 1.2711,
472
+ "step": 325
473
+ },
474
+ {
475
+ "epoch": 0.07947019867549669,
476
+ "grad_norm": 0.7566368579864502,
477
+ "learning_rate": 1.8410404624277458e-05,
478
+ "loss": 1.1603,
479
+ "step": 330
480
+ },
481
+ {
482
+ "epoch": 0.0806742925948224,
483
+ "grad_norm": 0.7235690355300903,
484
+ "learning_rate": 1.838631984585742e-05,
485
+ "loss": 1.1993,
486
+ "step": 335
487
+ },
488
+ {
489
+ "epoch": 0.08187838651414811,
490
+ "grad_norm": 0.7648451328277588,
491
+ "learning_rate": 1.8362235067437383e-05,
492
+ "loss": 1.134,
493
+ "step": 340
494
+ },
495
+ {
496
+ "epoch": 0.08308248043347381,
497
+ "grad_norm": 0.740710973739624,
498
+ "learning_rate": 1.8338150289017342e-05,
499
+ "loss": 1.1533,
500
+ "step": 345
501
+ },
502
+ {
503
+ "epoch": 0.08428657435279951,
504
+ "grad_norm": 0.7610028982162476,
505
+ "learning_rate": 1.8314065510597304e-05,
506
+ "loss": 1.3126,
507
+ "step": 350
508
+ },
509
+ {
510
+ "epoch": 0.08549066827212523,
511
+ "grad_norm": 0.6956422924995422,
512
+ "learning_rate": 1.8289980732177266e-05,
513
+ "loss": 1.2272,
514
+ "step": 355
515
+ },
516
+ {
517
+ "epoch": 0.08669476219145093,
518
+ "grad_norm": 0.6914700269699097,
519
+ "learning_rate": 1.8265895953757225e-05,
520
+ "loss": 1.3049,
521
+ "step": 360
522
+ },
523
+ {
524
+ "epoch": 0.08789885611077664,
525
+ "grad_norm": 0.7015173435211182,
526
+ "learning_rate": 1.8241811175337188e-05,
527
+ "loss": 1.3569,
528
+ "step": 365
529
+ },
530
+ {
531
+ "epoch": 0.08910295003010235,
532
+ "grad_norm": 0.6985088586807251,
533
+ "learning_rate": 1.821772639691715e-05,
534
+ "loss": 1.2307,
535
+ "step": 370
536
+ },
537
+ {
538
+ "epoch": 0.09030704394942805,
539
+ "grad_norm": 0.7313199043273926,
540
+ "learning_rate": 1.8193641618497112e-05,
541
+ "loss": 1.1783,
542
+ "step": 375
543
+ },
544
+ {
545
+ "epoch": 0.09151113786875377,
546
+ "grad_norm": 0.7657830715179443,
547
+ "learning_rate": 1.8169556840077075e-05,
548
+ "loss": 1.2228,
549
+ "step": 380
550
+ },
551
+ {
552
+ "epoch": 0.09271523178807947,
553
+ "grad_norm": 0.8051377534866333,
554
+ "learning_rate": 1.8145472061657034e-05,
555
+ "loss": 1.25,
556
+ "step": 385
557
+ },
558
+ {
559
+ "epoch": 0.09391932570740517,
560
+ "grad_norm": 0.8114392161369324,
561
+ "learning_rate": 1.8121387283236996e-05,
562
+ "loss": 1.2347,
563
+ "step": 390
564
+ },
565
+ {
566
+ "epoch": 0.09512341962673089,
567
+ "grad_norm": 0.7304120659828186,
568
+ "learning_rate": 1.809730250481696e-05,
569
+ "loss": 1.2437,
570
+ "step": 395
571
+ },
572
+ {
573
+ "epoch": 0.09632751354605659,
574
+ "grad_norm": 0.8418581485748291,
575
+ "learning_rate": 1.8073217726396917e-05,
576
+ "loss": 1.3255,
577
+ "step": 400
578
+ },
579
+ {
580
+ "epoch": 0.09632751354605659,
581
+ "eval_loss": 1.255526065826416,
582
+ "eval_runtime": 10.4675,
583
+ "eval_samples_per_second": 9.553,
584
+ "eval_steps_per_second": 1.242,
585
+ "step": 400
586
+ },
587
+ {
588
+ "epoch": 0.0975316074653823,
589
+ "grad_norm": 0.7186573147773743,
590
+ "learning_rate": 1.804913294797688e-05,
591
+ "loss": 1.2859,
592
+ "step": 405
593
+ },
594
+ {
595
+ "epoch": 0.09873570138470801,
596
+ "grad_norm": 0.7544155716896057,
597
+ "learning_rate": 1.8025048169556842e-05,
598
+ "loss": 1.2771,
599
+ "step": 410
600
+ },
601
+ {
602
+ "epoch": 0.09993979530403371,
603
+ "grad_norm": 0.700529158115387,
604
+ "learning_rate": 1.8000963391136804e-05,
605
+ "loss": 1.2696,
606
+ "step": 415
607
+ },
608
+ {
609
+ "epoch": 0.10114388922335943,
610
+ "grad_norm": 0.6894243955612183,
611
+ "learning_rate": 1.7976878612716763e-05,
612
+ "loss": 1.2556,
613
+ "step": 420
614
+ },
615
+ {
616
+ "epoch": 0.10234798314268513,
617
+ "grad_norm": 0.8769268989562988,
618
+ "learning_rate": 1.7952793834296726e-05,
619
+ "loss": 1.2832,
620
+ "step": 425
621
+ },
622
+ {
623
+ "epoch": 0.10355207706201083,
624
+ "grad_norm": 0.7283303141593933,
625
+ "learning_rate": 1.7928709055876688e-05,
626
+ "loss": 1.3163,
627
+ "step": 430
628
+ },
629
+ {
630
+ "epoch": 0.10475617098133655,
631
+ "grad_norm": 0.6984531283378601,
632
+ "learning_rate": 1.7904624277456647e-05,
633
+ "loss": 1.1873,
634
+ "step": 435
635
+ },
636
+ {
637
+ "epoch": 0.10596026490066225,
638
+ "grad_norm": 0.7543785572052002,
639
+ "learning_rate": 1.7880539499036613e-05,
640
+ "loss": 1.2617,
641
+ "step": 440
642
+ },
643
+ {
644
+ "epoch": 0.10716435881998795,
645
+ "grad_norm": 0.7481861710548401,
646
+ "learning_rate": 1.785645472061657e-05,
647
+ "loss": 1.1792,
648
+ "step": 445
649
+ },
650
+ {
651
+ "epoch": 0.10836845273931367,
652
+ "grad_norm": 0.7607905268669128,
653
+ "learning_rate": 1.7832369942196534e-05,
654
+ "loss": 1.1987,
655
+ "step": 450
656
+ },
657
+ {
658
+ "epoch": 0.10957254665863937,
659
+ "grad_norm": 0.7427639365196228,
660
+ "learning_rate": 1.7808285163776496e-05,
661
+ "loss": 1.2142,
662
+ "step": 455
663
+ },
664
+ {
665
+ "epoch": 0.11077664057796509,
666
+ "grad_norm": 0.7348979711532593,
667
+ "learning_rate": 1.7784200385356455e-05,
668
+ "loss": 1.2727,
669
+ "step": 460
670
+ },
671
+ {
672
+ "epoch": 0.11198073449729079,
673
+ "grad_norm": 0.7192914485931396,
674
+ "learning_rate": 1.7760115606936417e-05,
675
+ "loss": 1.2593,
676
+ "step": 465
677
+ },
678
+ {
679
+ "epoch": 0.11318482841661649,
680
+ "grad_norm": 0.6672971248626709,
681
+ "learning_rate": 1.773603082851638e-05,
682
+ "loss": 1.1905,
683
+ "step": 470
684
+ },
685
+ {
686
+ "epoch": 0.11438892233594221,
687
+ "grad_norm": 0.7585046291351318,
688
+ "learning_rate": 1.771194605009634e-05,
689
+ "loss": 1.3045,
690
+ "step": 475
691
+ },
692
+ {
693
+ "epoch": 0.11559301625526791,
694
+ "grad_norm": 0.7200559973716736,
695
+ "learning_rate": 1.76878612716763e-05,
696
+ "loss": 1.1416,
697
+ "step": 480
698
+ },
699
+ {
700
+ "epoch": 0.11679711017459361,
701
+ "grad_norm": 0.7220486998558044,
702
+ "learning_rate": 1.7663776493256263e-05,
703
+ "loss": 1.2749,
704
+ "step": 485
705
+ },
706
+ {
707
+ "epoch": 0.11800120409391933,
708
+ "grad_norm": 0.750998318195343,
709
+ "learning_rate": 1.7639691714836226e-05,
710
+ "loss": 1.2823,
711
+ "step": 490
712
+ },
713
+ {
714
+ "epoch": 0.11920529801324503,
715
+ "grad_norm": 0.705752432346344,
716
+ "learning_rate": 1.7615606936416188e-05,
717
+ "loss": 1.1662,
718
+ "step": 495
719
+ },
720
+ {
721
+ "epoch": 0.12040939193257075,
722
+ "grad_norm": 0.7431082129478455,
723
+ "learning_rate": 1.7591522157996147e-05,
724
+ "loss": 1.203,
725
+ "step": 500
726
+ },
727
+ {
728
+ "epoch": 0.12161348585189645,
729
+ "grad_norm": 0.7106168866157532,
730
+ "learning_rate": 1.756743737957611e-05,
731
+ "loss": 1.162,
732
+ "step": 505
733
+ },
734
+ {
735
+ "epoch": 0.12281757977122215,
736
+ "grad_norm": 0.7027086615562439,
737
+ "learning_rate": 1.754335260115607e-05,
738
+ "loss": 1.2636,
739
+ "step": 510
740
+ },
741
+ {
742
+ "epoch": 0.12402167369054787,
743
+ "grad_norm": 0.774591863155365,
744
+ "learning_rate": 1.7519267822736034e-05,
745
+ "loss": 1.2517,
746
+ "step": 515
747
+ },
748
+ {
749
+ "epoch": 0.12522576760987358,
750
+ "grad_norm": 0.6927476525306702,
751
+ "learning_rate": 1.7495183044315993e-05,
752
+ "loss": 1.1396,
753
+ "step": 520
754
+ },
755
+ {
756
+ "epoch": 0.12642986152919927,
757
+ "grad_norm": 0.6912556886672974,
758
+ "learning_rate": 1.7471098265895955e-05,
759
+ "loss": 1.1658,
760
+ "step": 525
761
+ },
762
+ {
763
+ "epoch": 0.127633955448525,
764
+ "grad_norm": 0.6708123683929443,
765
+ "learning_rate": 1.7447013487475918e-05,
766
+ "loss": 1.1288,
767
+ "step": 530
768
+ },
769
+ {
770
+ "epoch": 0.1288380493678507,
771
+ "grad_norm": 0.6942213177680969,
772
+ "learning_rate": 1.7422928709055877e-05,
773
+ "loss": 1.1174,
774
+ "step": 535
775
+ },
776
+ {
777
+ "epoch": 0.1300421432871764,
778
+ "grad_norm": 0.7080583572387695,
779
+ "learning_rate": 1.739884393063584e-05,
780
+ "loss": 1.2279,
781
+ "step": 540
782
+ },
783
+ {
784
+ "epoch": 0.1312462372065021,
785
+ "grad_norm": 0.8131110668182373,
786
+ "learning_rate": 1.73747591522158e-05,
787
+ "loss": 1.2525,
788
+ "step": 545
789
+ },
790
+ {
791
+ "epoch": 0.13245033112582782,
792
+ "grad_norm": 0.7450209259986877,
793
+ "learning_rate": 1.735067437379576e-05,
794
+ "loss": 1.1506,
795
+ "step": 550
796
+ },
797
+ {
798
+ "epoch": 0.1336544250451535,
799
+ "grad_norm": 0.7066523432731628,
800
+ "learning_rate": 1.7326589595375726e-05,
801
+ "loss": 1.2383,
802
+ "step": 555
803
+ },
804
+ {
805
+ "epoch": 0.13485851896447923,
806
+ "grad_norm": 0.7041836977005005,
807
+ "learning_rate": 1.7302504816955685e-05,
808
+ "loss": 1.1911,
809
+ "step": 560
810
+ },
811
+ {
812
+ "epoch": 0.13606261288380495,
813
+ "grad_norm": 0.672095000743866,
814
+ "learning_rate": 1.7278420038535647e-05,
815
+ "loss": 1.1839,
816
+ "step": 565
817
+ },
818
+ {
819
+ "epoch": 0.13726670680313063,
820
+ "grad_norm": 0.7928294539451599,
821
+ "learning_rate": 1.725433526011561e-05,
822
+ "loss": 1.3347,
823
+ "step": 570
824
+ },
825
+ {
826
+ "epoch": 0.13847080072245635,
827
+ "grad_norm": 0.7465523481369019,
828
+ "learning_rate": 1.723025048169557e-05,
829
+ "loss": 1.134,
830
+ "step": 575
831
+ },
832
+ {
833
+ "epoch": 0.13967489464178207,
834
+ "grad_norm": 0.7117074131965637,
835
+ "learning_rate": 1.720616570327553e-05,
836
+ "loss": 1.3538,
837
+ "step": 580
838
+ },
839
+ {
840
+ "epoch": 0.14087898856110775,
841
+ "grad_norm": 0.7291346192359924,
842
+ "learning_rate": 1.7182080924855493e-05,
843
+ "loss": 1.1339,
844
+ "step": 585
845
+ },
846
+ {
847
+ "epoch": 0.14208308248043347,
848
+ "grad_norm": 0.7633514404296875,
849
+ "learning_rate": 1.7157996146435455e-05,
850
+ "loss": 1.2007,
851
+ "step": 590
852
+ },
853
+ {
854
+ "epoch": 0.1432871763997592,
855
+ "grad_norm": 0.7896651029586792,
856
+ "learning_rate": 1.7133911368015418e-05,
857
+ "loss": 1.2656,
858
+ "step": 595
859
+ },
860
+ {
861
+ "epoch": 0.1444912703190849,
862
+ "grad_norm": 0.7436801791191101,
863
+ "learning_rate": 1.7109826589595377e-05,
864
+ "loss": 1.28,
865
+ "step": 600
866
+ },
867
+ {
868
+ "epoch": 0.1444912703190849,
869
+ "eval_loss": 1.23267662525177,
870
+ "eval_runtime": 10.4568,
871
+ "eval_samples_per_second": 9.563,
872
+ "eval_steps_per_second": 1.243,
873
+ "step": 600
874
+ },
875
+ {
876
+ "epoch": 0.1456953642384106,
877
+ "grad_norm": 0.7123108506202698,
878
+ "learning_rate": 1.708574181117534e-05,
879
+ "loss": 1.1687,
880
+ "step": 605
881
+ },
882
+ {
883
+ "epoch": 0.1468994581577363,
884
+ "grad_norm": 0.7362635731697083,
885
+ "learning_rate": 1.70616570327553e-05,
886
+ "loss": 1.2471,
887
+ "step": 610
888
+ },
889
+ {
890
+ "epoch": 0.14810355207706202,
891
+ "grad_norm": 0.6770139336585999,
892
+ "learning_rate": 1.703757225433526e-05,
893
+ "loss": 1.149,
894
+ "step": 615
895
+ },
896
+ {
897
+ "epoch": 0.1493076459963877,
898
+ "grad_norm": 0.7292876243591309,
899
+ "learning_rate": 1.7013487475915223e-05,
900
+ "loss": 1.2957,
901
+ "step": 620
902
+ },
903
+ {
904
+ "epoch": 0.15051173991571343,
905
+ "grad_norm": 0.7105087637901306,
906
+ "learning_rate": 1.6989402697495185e-05,
907
+ "loss": 1.1462,
908
+ "step": 625
909
+ },
910
+ {
911
+ "epoch": 0.15171583383503914,
912
+ "grad_norm": 0.7218934297561646,
913
+ "learning_rate": 1.6965317919075147e-05,
914
+ "loss": 1.2181,
915
+ "step": 630
916
+ },
917
+ {
918
+ "epoch": 0.15291992775436483,
919
+ "grad_norm": 0.7794796824455261,
920
+ "learning_rate": 1.6941233140655106e-05,
921
+ "loss": 1.174,
922
+ "step": 635
923
+ },
924
+ {
925
+ "epoch": 0.15412402167369055,
926
+ "grad_norm": 0.6424360871315002,
927
+ "learning_rate": 1.691714836223507e-05,
928
+ "loss": 1.2149,
929
+ "step": 640
930
+ },
931
+ {
932
+ "epoch": 0.15532811559301626,
933
+ "grad_norm": 0.7393337488174438,
934
+ "learning_rate": 1.689306358381503e-05,
935
+ "loss": 1.2259,
936
+ "step": 645
937
+ },
938
+ {
939
+ "epoch": 0.15653220951234195,
940
+ "grad_norm": 0.7363811731338501,
941
+ "learning_rate": 1.686897880539499e-05,
942
+ "loss": 1.2087,
943
+ "step": 650
944
+ },
945
+ {
946
+ "epoch": 0.15773630343166767,
947
+ "grad_norm": 0.7336999773979187,
948
+ "learning_rate": 1.6844894026974952e-05,
949
+ "loss": 1.2526,
950
+ "step": 655
951
+ },
952
+ {
953
+ "epoch": 0.15894039735099338,
954
+ "grad_norm": 0.7329117655754089,
955
+ "learning_rate": 1.6820809248554915e-05,
956
+ "loss": 1.233,
957
+ "step": 660
958
+ },
959
+ {
960
+ "epoch": 0.16014449127031907,
961
+ "grad_norm": 0.7724847197532654,
962
+ "learning_rate": 1.6796724470134877e-05,
963
+ "loss": 1.2506,
964
+ "step": 665
965
+ },
966
+ {
967
+ "epoch": 0.1613485851896448,
968
+ "grad_norm": 0.7213071584701538,
969
+ "learning_rate": 1.677263969171484e-05,
970
+ "loss": 1.2021,
971
+ "step": 670
972
+ },
973
+ {
974
+ "epoch": 0.1625526791089705,
975
+ "grad_norm": 0.6397632360458374,
976
+ "learning_rate": 1.6748554913294798e-05,
977
+ "loss": 1.1909,
978
+ "step": 675
979
+ },
980
+ {
981
+ "epoch": 0.16375677302829622,
982
+ "grad_norm": 0.6840227842330933,
983
+ "learning_rate": 1.672447013487476e-05,
984
+ "loss": 1.1838,
985
+ "step": 680
986
+ },
987
+ {
988
+ "epoch": 0.1649608669476219,
989
+ "grad_norm": 0.7317889332771301,
990
+ "learning_rate": 1.6700385356454723e-05,
991
+ "loss": 1.169,
992
+ "step": 685
993
+ },
994
+ {
995
+ "epoch": 0.16616496086694763,
996
+ "grad_norm": 0.7134217619895935,
997
+ "learning_rate": 1.6676300578034682e-05,
998
+ "loss": 1.2711,
999
+ "step": 690
1000
+ },
1001
+ {
1002
+ "epoch": 0.16736905478627334,
1003
+ "grad_norm": 0.7525413036346436,
1004
+ "learning_rate": 1.6652215799614644e-05,
1005
+ "loss": 1.2183,
1006
+ "step": 695
1007
+ },
1008
+ {
1009
+ "epoch": 0.16857314870559903,
1010
+ "grad_norm": 0.7243443131446838,
1011
+ "learning_rate": 1.6628131021194607e-05,
1012
+ "loss": 1.2585,
1013
+ "step": 700
1014
+ },
1015
+ {
1016
+ "epoch": 0.16977724262492475,
1017
+ "grad_norm": 0.7554174661636353,
1018
+ "learning_rate": 1.660404624277457e-05,
1019
+ "loss": 1.1662,
1020
+ "step": 705
1021
+ },
1022
+ {
1023
+ "epoch": 0.17098133654425046,
1024
+ "grad_norm": 0.6472408175468445,
1025
+ "learning_rate": 1.657996146435453e-05,
1026
+ "loss": 1.1471,
1027
+ "step": 710
1028
+ },
1029
+ {
1030
+ "epoch": 0.17218543046357615,
1031
+ "grad_norm": 0.7514950037002563,
1032
+ "learning_rate": 1.655587668593449e-05,
1033
+ "loss": 1.157,
1034
+ "step": 715
1035
+ },
1036
+ {
1037
+ "epoch": 0.17338952438290187,
1038
+ "grad_norm": 0.6952060461044312,
1039
+ "learning_rate": 1.6531791907514452e-05,
1040
+ "loss": 1.2981,
1041
+ "step": 720
1042
+ },
1043
+ {
1044
+ "epoch": 0.17459361830222758,
1045
+ "grad_norm": 0.832417905330658,
1046
+ "learning_rate": 1.6507707129094415e-05,
1047
+ "loss": 1.2457,
1048
+ "step": 725
1049
+ },
1050
+ {
1051
+ "epoch": 0.17579771222155327,
1052
+ "grad_norm": 0.6825481653213501,
1053
+ "learning_rate": 1.6483622350674374e-05,
1054
+ "loss": 1.2632,
1055
+ "step": 730
1056
+ },
1057
+ {
1058
+ "epoch": 0.177001806140879,
1059
+ "grad_norm": 0.8267092704772949,
1060
+ "learning_rate": 1.6459537572254336e-05,
1061
+ "loss": 1.0368,
1062
+ "step": 735
1063
+ },
1064
+ {
1065
+ "epoch": 0.1782059000602047,
1066
+ "grad_norm": 0.7782288789749146,
1067
+ "learning_rate": 1.64354527938343e-05,
1068
+ "loss": 1.201,
1069
+ "step": 740
1070
+ },
1071
+ {
1072
+ "epoch": 0.1794099939795304,
1073
+ "grad_norm": 0.7548500299453735,
1074
+ "learning_rate": 1.641136801541426e-05,
1075
+ "loss": 1.1831,
1076
+ "step": 745
1077
+ },
1078
+ {
1079
+ "epoch": 0.1806140878988561,
1080
+ "grad_norm": 0.6344689726829529,
1081
+ "learning_rate": 1.638728323699422e-05,
1082
+ "loss": 1.1985,
1083
+ "step": 750
1084
+ },
1085
+ {
1086
+ "epoch": 0.18181818181818182,
1087
+ "grad_norm": 0.7326412796974182,
1088
+ "learning_rate": 1.6363198458574182e-05,
1089
+ "loss": 1.3051,
1090
+ "step": 755
1091
+ },
1092
+ {
1093
+ "epoch": 0.18302227573750754,
1094
+ "grad_norm": 0.6460621953010559,
1095
+ "learning_rate": 1.6339113680154144e-05,
1096
+ "loss": 1.1524,
1097
+ "step": 760
1098
+ },
1099
+ {
1100
+ "epoch": 0.18422636965683323,
1101
+ "grad_norm": 0.6921172142028809,
1102
+ "learning_rate": 1.6315028901734103e-05,
1103
+ "loss": 1.29,
1104
+ "step": 765
1105
+ },
1106
+ {
1107
+ "epoch": 0.18543046357615894,
1108
+ "grad_norm": 0.7872111201286316,
1109
+ "learning_rate": 1.629094412331407e-05,
1110
+ "loss": 1.1187,
1111
+ "step": 770
1112
+ },
1113
+ {
1114
+ "epoch": 0.18663455749548466,
1115
+ "grad_norm": 0.7098067402839661,
1116
+ "learning_rate": 1.6266859344894028e-05,
1117
+ "loss": 1.2006,
1118
+ "step": 775
1119
+ },
1120
+ {
1121
+ "epoch": 0.18783865141481035,
1122
+ "grad_norm": 0.7701388001441956,
1123
+ "learning_rate": 1.624277456647399e-05,
1124
+ "loss": 1.1609,
1125
+ "step": 780
1126
+ },
1127
+ {
1128
+ "epoch": 0.18904274533413606,
1129
+ "grad_norm": 0.7027246952056885,
1130
+ "learning_rate": 1.6218689788053953e-05,
1131
+ "loss": 1.2083,
1132
+ "step": 785
1133
+ },
1134
+ {
1135
+ "epoch": 0.19024683925346178,
1136
+ "grad_norm": 0.7161267995834351,
1137
+ "learning_rate": 1.619460500963391e-05,
1138
+ "loss": 1.131,
1139
+ "step": 790
1140
+ },
1141
+ {
1142
+ "epoch": 0.19145093317278747,
1143
+ "grad_norm": 0.8034841418266296,
1144
+ "learning_rate": 1.6170520231213874e-05,
1145
+ "loss": 1.2107,
1146
+ "step": 795
1147
+ },
1148
+ {
1149
+ "epoch": 0.19265502709211318,
1150
+ "grad_norm": 0.7797010540962219,
1151
+ "learning_rate": 1.6146435452793836e-05,
1152
+ "loss": 1.2931,
1153
+ "step": 800
1154
+ },
1155
+ {
1156
+ "epoch": 0.19265502709211318,
1157
+ "eval_loss": 1.2178008556365967,
1158
+ "eval_runtime": 10.4492,
1159
+ "eval_samples_per_second": 9.57,
1160
+ "eval_steps_per_second": 1.244,
1161
+ "step": 800
1162
+ },
1163
+ {
1164
+ "epoch": 0.1938591210114389,
1165
+ "grad_norm": 0.6586723923683167,
1166
+ "learning_rate": 1.6122350674373795e-05,
1167
+ "loss": 1.2007,
1168
+ "step": 805
1169
+ },
1170
+ {
1171
+ "epoch": 0.1950632149307646,
1172
+ "grad_norm": 0.7271411418914795,
1173
+ "learning_rate": 1.6098265895953758e-05,
1174
+ "loss": 1.1049,
1175
+ "step": 810
1176
+ },
1177
+ {
1178
+ "epoch": 0.1962673088500903,
1179
+ "grad_norm": 0.743730902671814,
1180
+ "learning_rate": 1.607418111753372e-05,
1181
+ "loss": 1.1815,
1182
+ "step": 815
1183
+ },
1184
+ {
1185
+ "epoch": 0.19747140276941602,
1186
+ "grad_norm": 0.7163010835647583,
1187
+ "learning_rate": 1.6050096339113682e-05,
1188
+ "loss": 1.1635,
1189
+ "step": 820
1190
+ },
1191
+ {
1192
+ "epoch": 0.1986754966887417,
1193
+ "grad_norm": 0.7846854329109192,
1194
+ "learning_rate": 1.6026011560693645e-05,
1195
+ "loss": 1.1576,
1196
+ "step": 825
1197
+ },
1198
+ {
1199
+ "epoch": 0.19987959060806743,
1200
+ "grad_norm": 0.755066990852356,
1201
+ "learning_rate": 1.6001926782273604e-05,
1202
+ "loss": 1.2626,
1203
+ "step": 830
1204
+ },
1205
+ {
1206
+ "epoch": 0.20108368452739314,
1207
+ "grad_norm": 0.6981719732284546,
1208
+ "learning_rate": 1.5977842003853566e-05,
1209
+ "loss": 1.1883,
1210
+ "step": 835
1211
+ },
1212
+ {
1213
+ "epoch": 0.20228777844671886,
1214
+ "grad_norm": 0.712697446346283,
1215
+ "learning_rate": 1.5953757225433528e-05,
1216
+ "loss": 1.1539,
1217
+ "step": 840
1218
+ },
1219
+ {
1220
+ "epoch": 0.20349187236604455,
1221
+ "grad_norm": 0.8146225810050964,
1222
+ "learning_rate": 1.592967244701349e-05,
1223
+ "loss": 1.161,
1224
+ "step": 845
1225
+ },
1226
+ {
1227
+ "epoch": 0.20469596628537026,
1228
+ "grad_norm": 0.7796756625175476,
1229
+ "learning_rate": 1.590558766859345e-05,
1230
+ "loss": 1.1536,
1231
+ "step": 850
1232
+ },
1233
+ {
1234
+ "epoch": 0.20590006020469598,
1235
+ "grad_norm": 0.7026313543319702,
1236
+ "learning_rate": 1.5881502890173412e-05,
1237
+ "loss": 1.2206,
1238
+ "step": 855
1239
+ },
1240
+ {
1241
+ "epoch": 0.20710415412402167,
1242
+ "grad_norm": 0.6918179988861084,
1243
+ "learning_rate": 1.5857418111753374e-05,
1244
+ "loss": 1.0569,
1245
+ "step": 860
1246
+ },
1247
+ {
1248
+ "epoch": 0.20830824804334738,
1249
+ "grad_norm": 0.7711547613143921,
1250
+ "learning_rate": 1.5833333333333333e-05,
1251
+ "loss": 1.1826,
1252
+ "step": 865
1253
+ },
1254
+ {
1255
+ "epoch": 0.2095123419626731,
1256
+ "grad_norm": 0.6604825854301453,
1257
+ "learning_rate": 1.5809248554913295e-05,
1258
+ "loss": 1.1743,
1259
+ "step": 870
1260
+ },
1261
+ {
1262
+ "epoch": 0.2107164358819988,
1263
+ "grad_norm": 0.7115917205810547,
1264
+ "learning_rate": 1.5785163776493258e-05,
1265
+ "loss": 1.2815,
1266
+ "step": 875
1267
+ },
1268
+ {
1269
+ "epoch": 0.2119205298013245,
1270
+ "grad_norm": 0.69574373960495,
1271
+ "learning_rate": 1.5761078998073217e-05,
1272
+ "loss": 1.2131,
1273
+ "step": 880
1274
+ },
1275
+ {
1276
+ "epoch": 0.21312462372065022,
1277
+ "grad_norm": 0.6801597476005554,
1278
+ "learning_rate": 1.5736994219653182e-05,
1279
+ "loss": 1.3022,
1280
+ "step": 885
1281
+ },
1282
+ {
1283
+ "epoch": 0.2143287176399759,
1284
+ "grad_norm": 0.7042235136032104,
1285
+ "learning_rate": 1.571290944123314e-05,
1286
+ "loss": 1.1936,
1287
+ "step": 890
1288
+ },
1289
+ {
1290
+ "epoch": 0.21553281155930162,
1291
+ "grad_norm": 0.6610032916069031,
1292
+ "learning_rate": 1.5688824662813104e-05,
1293
+ "loss": 1.1972,
1294
+ "step": 895
1295
+ },
1296
+ {
1297
+ "epoch": 0.21673690547862734,
1298
+ "grad_norm": 0.8218846321105957,
1299
+ "learning_rate": 1.5664739884393066e-05,
1300
+ "loss": 1.2951,
1301
+ "step": 900
1302
+ },
1303
+ {
1304
+ "epoch": 0.21794099939795303,
1305
+ "grad_norm": 0.7200918197631836,
1306
+ "learning_rate": 1.5640655105973025e-05,
1307
+ "loss": 1.1256,
1308
+ "step": 905
1309
+ },
1310
+ {
1311
+ "epoch": 0.21914509331727874,
1312
+ "grad_norm": 0.6200802326202393,
1313
+ "learning_rate": 1.5616570327552987e-05,
1314
+ "loss": 1.2071,
1315
+ "step": 910
1316
+ },
1317
+ {
1318
+ "epoch": 0.22034918723660446,
1319
+ "grad_norm": 0.7431606650352478,
1320
+ "learning_rate": 1.559248554913295e-05,
1321
+ "loss": 1.2569,
1322
+ "step": 915
1323
+ },
1324
+ {
1325
+ "epoch": 0.22155328115593018,
1326
+ "grad_norm": 0.802537202835083,
1327
+ "learning_rate": 1.5568400770712912e-05,
1328
+ "loss": 1.1342,
1329
+ "step": 920
1330
+ },
1331
+ {
1332
+ "epoch": 0.22275737507525586,
1333
+ "grad_norm": 0.7078563570976257,
1334
+ "learning_rate": 1.5544315992292874e-05,
1335
+ "loss": 1.1928,
1336
+ "step": 925
1337
+ },
1338
+ {
1339
+ "epoch": 0.22396146899458158,
1340
+ "grad_norm": 0.6626926064491272,
1341
+ "learning_rate": 1.5520231213872833e-05,
1342
+ "loss": 1.1295,
1343
+ "step": 930
1344
+ },
1345
+ {
1346
+ "epoch": 0.2251655629139073,
1347
+ "grad_norm": 0.664715588092804,
1348
+ "learning_rate": 1.5496146435452796e-05,
1349
+ "loss": 1.1616,
1350
+ "step": 935
1351
+ },
1352
+ {
1353
+ "epoch": 0.22636965683323299,
1354
+ "grad_norm": 0.7077795267105103,
1355
+ "learning_rate": 1.5472061657032758e-05,
1356
+ "loss": 1.24,
1357
+ "step": 940
1358
+ },
1359
+ {
1360
+ "epoch": 0.2275737507525587,
1361
+ "grad_norm": 0.6813915371894836,
1362
+ "learning_rate": 1.5447976878612717e-05,
1363
+ "loss": 1.1807,
1364
+ "step": 945
1365
+ },
1366
+ {
1367
+ "epoch": 0.22877784467188442,
1368
+ "grad_norm": 0.7245080471038818,
1369
+ "learning_rate": 1.542389210019268e-05,
1370
+ "loss": 1.1737,
1371
+ "step": 950
1372
+ },
1373
+ {
1374
+ "epoch": 0.2299819385912101,
1375
+ "grad_norm": 0.7851933836936951,
1376
+ "learning_rate": 1.539980732177264e-05,
1377
+ "loss": 1.0702,
1378
+ "step": 955
1379
+ },
1380
+ {
1381
+ "epoch": 0.23118603251053582,
1382
+ "grad_norm": 0.7153802514076233,
1383
+ "learning_rate": 1.5375722543352604e-05,
1384
+ "loss": 1.1302,
1385
+ "step": 960
1386
+ },
1387
+ {
1388
+ "epoch": 0.23239012642986154,
1389
+ "grad_norm": 0.8112071752548218,
1390
+ "learning_rate": 1.5351637764932563e-05,
1391
+ "loss": 1.1591,
1392
+ "step": 965
1393
+ },
1394
+ {
1395
+ "epoch": 0.23359422034918723,
1396
+ "grad_norm": 0.7842305898666382,
1397
+ "learning_rate": 1.5327552986512525e-05,
1398
+ "loss": 1.1695,
1399
+ "step": 970
1400
+ },
1401
+ {
1402
+ "epoch": 0.23479831426851294,
1403
+ "grad_norm": 0.6949911713600159,
1404
+ "learning_rate": 1.5303468208092487e-05,
1405
+ "loss": 1.1726,
1406
+ "step": 975
1407
+ },
1408
+ {
1409
+ "epoch": 0.23600240818783866,
1410
+ "grad_norm": 0.7541187405586243,
1411
+ "learning_rate": 1.5279383429672446e-05,
1412
+ "loss": 1.2209,
1413
+ "step": 980
1414
+ },
1415
+ {
1416
+ "epoch": 0.23720650210716435,
1417
+ "grad_norm": 0.7082781791687012,
1418
+ "learning_rate": 1.525529865125241e-05,
1419
+ "loss": 1.2685,
1420
+ "step": 985
1421
+ },
1422
+ {
1423
+ "epoch": 0.23841059602649006,
1424
+ "grad_norm": 0.6862682104110718,
1425
+ "learning_rate": 1.5231213872832371e-05,
1426
+ "loss": 1.1569,
1427
+ "step": 990
1428
+ },
1429
+ {
1430
+ "epoch": 0.23961468994581578,
1431
+ "grad_norm": 0.836452066898346,
1432
+ "learning_rate": 1.5207129094412332e-05,
1433
+ "loss": 1.2158,
1434
+ "step": 995
1435
+ },
1436
+ {
1437
+ "epoch": 0.2408187838651415,
1438
+ "grad_norm": 0.8637534379959106,
1439
+ "learning_rate": 1.5183044315992294e-05,
1440
+ "loss": 1.1891,
1441
+ "step": 1000
1442
+ },
1443
+ {
1444
+ "epoch": 0.2408187838651415,
1445
+ "eval_loss": 1.2066378593444824,
1446
+ "eval_runtime": 10.4478,
1447
+ "eval_samples_per_second": 9.571,
1448
+ "eval_steps_per_second": 1.244,
1449
+ "step": 1000
1450
+ },
1451
+ {
1452
+ "epoch": 0.24202287778446718,
1453
+ "grad_norm": 0.8397049903869629,
1454
+ "learning_rate": 1.5158959537572255e-05,
1455
+ "loss": 1.2199,
1456
+ "step": 1005
1457
+ },
1458
+ {
1459
+ "epoch": 0.2432269717037929,
1460
+ "grad_norm": 0.7865377068519592,
1461
+ "learning_rate": 1.5134874759152215e-05,
1462
+ "loss": 1.2102,
1463
+ "step": 1010
1464
+ },
1465
+ {
1466
+ "epoch": 0.24443106562311862,
1467
+ "grad_norm": 0.6602733731269836,
1468
+ "learning_rate": 1.511078998073218e-05,
1469
+ "loss": 1.1512,
1470
+ "step": 1015
1471
+ },
1472
+ {
1473
+ "epoch": 0.2456351595424443,
1474
+ "grad_norm": 0.69524747133255,
1475
+ "learning_rate": 1.508670520231214e-05,
1476
+ "loss": 1.2387,
1477
+ "step": 1020
1478
+ },
1479
+ {
1480
+ "epoch": 0.24683925346177002,
1481
+ "grad_norm": 0.6567270755767822,
1482
+ "learning_rate": 1.50626204238921e-05,
1483
+ "loss": 1.159,
1484
+ "step": 1025
1485
+ },
1486
+ {
1487
+ "epoch": 0.24804334738109574,
1488
+ "grad_norm": 0.6918655633926392,
1489
+ "learning_rate": 1.5038535645472063e-05,
1490
+ "loss": 1.236,
1491
+ "step": 1030
1492
+ },
1493
+ {
1494
+ "epoch": 0.24924744130042142,
1495
+ "grad_norm": 0.748435378074646,
1496
+ "learning_rate": 1.5014450867052024e-05,
1497
+ "loss": 1.1422,
1498
+ "step": 1035
1499
+ },
1500
+ {
1501
+ "epoch": 0.25045153521974717,
1502
+ "grad_norm": 0.8211308717727661,
1503
+ "learning_rate": 1.4990366088631986e-05,
1504
+ "loss": 1.1775,
1505
+ "step": 1040
1506
+ },
1507
+ {
1508
+ "epoch": 0.25165562913907286,
1509
+ "grad_norm": 0.6989510655403137,
1510
+ "learning_rate": 1.4966281310211948e-05,
1511
+ "loss": 1.2473,
1512
+ "step": 1045
1513
+ },
1514
+ {
1515
+ "epoch": 0.25285972305839854,
1516
+ "grad_norm": 0.6981227993965149,
1517
+ "learning_rate": 1.4942196531791909e-05,
1518
+ "loss": 1.1887,
1519
+ "step": 1050
1520
+ },
1521
+ {
1522
+ "epoch": 0.2540638169777243,
1523
+ "grad_norm": 0.6409928202629089,
1524
+ "learning_rate": 1.4918111753371871e-05,
1525
+ "loss": 1.1323,
1526
+ "step": 1055
1527
+ },
1528
+ {
1529
+ "epoch": 0.25526791089705,
1530
+ "grad_norm": 0.8026551604270935,
1531
+ "learning_rate": 1.4894026974951832e-05,
1532
+ "loss": 1.2205,
1533
+ "step": 1060
1534
+ },
1535
+ {
1536
+ "epoch": 0.25647200481637566,
1537
+ "grad_norm": 0.7427927255630493,
1538
+ "learning_rate": 1.4869942196531793e-05,
1539
+ "loss": 1.3051,
1540
+ "step": 1065
1541
+ },
1542
+ {
1543
+ "epoch": 0.2576760987357014,
1544
+ "grad_norm": 0.7286926507949829,
1545
+ "learning_rate": 1.4845857418111755e-05,
1546
+ "loss": 1.1952,
1547
+ "step": 1070
1548
+ },
1549
+ {
1550
+ "epoch": 0.2588801926550271,
1551
+ "grad_norm": 0.717553436756134,
1552
+ "learning_rate": 1.4821772639691716e-05,
1553
+ "loss": 1.1395,
1554
+ "step": 1075
1555
+ },
1556
+ {
1557
+ "epoch": 0.2600842865743528,
1558
+ "grad_norm": 0.6549174189567566,
1559
+ "learning_rate": 1.4797687861271676e-05,
1560
+ "loss": 1.1677,
1561
+ "step": 1080
1562
+ },
1563
+ {
1564
+ "epoch": 0.26128838049367853,
1565
+ "grad_norm": 0.7400743365287781,
1566
+ "learning_rate": 1.477360308285164e-05,
1567
+ "loss": 1.239,
1568
+ "step": 1085
1569
+ },
1570
+ {
1571
+ "epoch": 0.2624924744130042,
1572
+ "grad_norm": 0.8405967950820923,
1573
+ "learning_rate": 1.4749518304431601e-05,
1574
+ "loss": 1.1379,
1575
+ "step": 1090
1576
+ },
1577
+ {
1578
+ "epoch": 0.2636965683323299,
1579
+ "grad_norm": 0.7308910489082336,
1580
+ "learning_rate": 1.4725433526011562e-05,
1581
+ "loss": 1.1709,
1582
+ "step": 1095
1583
+ },
1584
+ {
1585
+ "epoch": 0.26490066225165565,
1586
+ "grad_norm": 0.7904026508331299,
1587
+ "learning_rate": 1.4701348747591524e-05,
1588
+ "loss": 1.2348,
1589
+ "step": 1100
1590
+ },
1591
+ {
1592
+ "epoch": 0.26610475617098134,
1593
+ "grad_norm": 0.7531107664108276,
1594
+ "learning_rate": 1.4677263969171484e-05,
1595
+ "loss": 1.1702,
1596
+ "step": 1105
1597
+ },
1598
+ {
1599
+ "epoch": 0.267308850090307,
1600
+ "grad_norm": 0.7167130708694458,
1601
+ "learning_rate": 1.4653179190751445e-05,
1602
+ "loss": 1.1417,
1603
+ "step": 1110
1604
+ },
1605
+ {
1606
+ "epoch": 0.26851294400963277,
1607
+ "grad_norm": 0.7529981732368469,
1608
+ "learning_rate": 1.4629094412331407e-05,
1609
+ "loss": 1.3126,
1610
+ "step": 1115
1611
+ },
1612
+ {
1613
+ "epoch": 0.26971703792895846,
1614
+ "grad_norm": 0.7192036509513855,
1615
+ "learning_rate": 1.4605009633911368e-05,
1616
+ "loss": 1.2149,
1617
+ "step": 1120
1618
+ },
1619
+ {
1620
+ "epoch": 0.27092113184828415,
1621
+ "grad_norm": 0.7200658321380615,
1622
+ "learning_rate": 1.458092485549133e-05,
1623
+ "loss": 1.1958,
1624
+ "step": 1125
1625
+ },
1626
+ {
1627
+ "epoch": 0.2721252257676099,
1628
+ "grad_norm": 0.7063027024269104,
1629
+ "learning_rate": 1.4556840077071293e-05,
1630
+ "loss": 1.1266,
1631
+ "step": 1130
1632
+ },
1633
+ {
1634
+ "epoch": 0.2733293196869356,
1635
+ "grad_norm": 0.7414741516113281,
1636
+ "learning_rate": 1.4532755298651253e-05,
1637
+ "loss": 1.2046,
1638
+ "step": 1135
1639
+ },
1640
+ {
1641
+ "epoch": 0.27453341360626127,
1642
+ "grad_norm": 0.7233819365501404,
1643
+ "learning_rate": 1.4508670520231216e-05,
1644
+ "loss": 1.3376,
1645
+ "step": 1140
1646
+ },
1647
+ {
1648
+ "epoch": 0.275737507525587,
1649
+ "grad_norm": 0.737486720085144,
1650
+ "learning_rate": 1.4484585741811176e-05,
1651
+ "loss": 1.2529,
1652
+ "step": 1145
1653
+ },
1654
+ {
1655
+ "epoch": 0.2769416014449127,
1656
+ "grad_norm": 0.6850925087928772,
1657
+ "learning_rate": 1.4460500963391137e-05,
1658
+ "loss": 1.1784,
1659
+ "step": 1150
1660
+ },
1661
+ {
1662
+ "epoch": 0.2781456953642384,
1663
+ "grad_norm": 0.6742652058601379,
1664
+ "learning_rate": 1.4436416184971101e-05,
1665
+ "loss": 1.0412,
1666
+ "step": 1155
1667
+ },
1668
+ {
1669
+ "epoch": 0.27934978928356413,
1670
+ "grad_norm": 0.7630188465118408,
1671
+ "learning_rate": 1.4412331406551062e-05,
1672
+ "loss": 1.2421,
1673
+ "step": 1160
1674
+ },
1675
+ {
1676
+ "epoch": 0.2805538832028898,
1677
+ "grad_norm": 0.7039279341697693,
1678
+ "learning_rate": 1.4388246628131022e-05,
1679
+ "loss": 1.2076,
1680
+ "step": 1165
1681
+ },
1682
+ {
1683
+ "epoch": 0.2817579771222155,
1684
+ "grad_norm": 0.6968697905540466,
1685
+ "learning_rate": 1.4364161849710985e-05,
1686
+ "loss": 1.2709,
1687
+ "step": 1170
1688
+ },
1689
+ {
1690
+ "epoch": 0.28296207104154125,
1691
+ "grad_norm": 0.757686972618103,
1692
+ "learning_rate": 1.4340077071290945e-05,
1693
+ "loss": 1.2237,
1694
+ "step": 1175
1695
+ },
1696
+ {
1697
+ "epoch": 0.28416616496086694,
1698
+ "grad_norm": 0.70340895652771,
1699
+ "learning_rate": 1.4315992292870906e-05,
1700
+ "loss": 1.1979,
1701
+ "step": 1180
1702
+ },
1703
+ {
1704
+ "epoch": 0.28537025888019263,
1705
+ "grad_norm": 0.7078365683555603,
1706
+ "learning_rate": 1.4291907514450868e-05,
1707
+ "loss": 1.1801,
1708
+ "step": 1185
1709
+ },
1710
+ {
1711
+ "epoch": 0.2865743527995184,
1712
+ "grad_norm": 0.671120822429657,
1713
+ "learning_rate": 1.4267822736030829e-05,
1714
+ "loss": 1.0744,
1715
+ "step": 1190
1716
+ },
1717
+ {
1718
+ "epoch": 0.28777844671884406,
1719
+ "grad_norm": 0.6995368599891663,
1720
+ "learning_rate": 1.424373795761079e-05,
1721
+ "loss": 1.1057,
1722
+ "step": 1195
1723
+ },
1724
+ {
1725
+ "epoch": 0.2889825406381698,
1726
+ "grad_norm": 0.8009471297264099,
1727
+ "learning_rate": 1.4219653179190754e-05,
1728
+ "loss": 1.2852,
1729
+ "step": 1200
1730
+ },
1731
+ {
1732
+ "epoch": 0.2889825406381698,
1733
+ "eval_loss": 1.1969135999679565,
1734
+ "eval_runtime": 10.4824,
1735
+ "eval_samples_per_second": 9.54,
1736
+ "eval_steps_per_second": 1.24,
1737
+ "step": 1200
1738
+ },
1739
+ {
1740
+ "epoch": 0.2901866345574955,
1741
+ "grad_norm": 0.8135602474212646,
1742
+ "learning_rate": 1.4195568400770714e-05,
1743
+ "loss": 1.1475,
1744
+ "step": 1205
1745
+ },
1746
+ {
1747
+ "epoch": 0.2913907284768212,
1748
+ "grad_norm": 0.6879997849464417,
1749
+ "learning_rate": 1.4171483622350675e-05,
1750
+ "loss": 1.2054,
1751
+ "step": 1210
1752
+ },
1753
+ {
1754
+ "epoch": 0.2925948223961469,
1755
+ "grad_norm": 0.774534285068512,
1756
+ "learning_rate": 1.4147398843930637e-05,
1757
+ "loss": 1.1133,
1758
+ "step": 1215
1759
+ },
1760
+ {
1761
+ "epoch": 0.2937989163154726,
1762
+ "grad_norm": 0.7712222933769226,
1763
+ "learning_rate": 1.4123314065510598e-05,
1764
+ "loss": 1.2367,
1765
+ "step": 1220
1766
+ },
1767
+ {
1768
+ "epoch": 0.2950030102347983,
1769
+ "grad_norm": 0.7632959485054016,
1770
+ "learning_rate": 1.4099229287090558e-05,
1771
+ "loss": 1.2142,
1772
+ "step": 1225
1773
+ },
1774
+ {
1775
+ "epoch": 0.29620710415412405,
1776
+ "grad_norm": 0.6896396279335022,
1777
+ "learning_rate": 1.4075144508670523e-05,
1778
+ "loss": 1.2692,
1779
+ "step": 1230
1780
+ },
1781
+ {
1782
+ "epoch": 0.29741119807344973,
1783
+ "grad_norm": 0.7135001420974731,
1784
+ "learning_rate": 1.4051059730250483e-05,
1785
+ "loss": 1.1922,
1786
+ "step": 1235
1787
+ },
1788
+ {
1789
+ "epoch": 0.2986152919927754,
1790
+ "grad_norm": 0.7016868591308594,
1791
+ "learning_rate": 1.4026974951830444e-05,
1792
+ "loss": 1.2775,
1793
+ "step": 1240
1794
+ },
1795
+ {
1796
+ "epoch": 0.29981938591210117,
1797
+ "grad_norm": 0.7863327860832214,
1798
+ "learning_rate": 1.4002890173410406e-05,
1799
+ "loss": 1.193,
1800
+ "step": 1245
1801
+ },
1802
+ {
1803
+ "epoch": 0.30102347983142685,
1804
+ "grad_norm": 0.7253124713897705,
1805
+ "learning_rate": 1.3978805394990367e-05,
1806
+ "loss": 1.1557,
1807
+ "step": 1250
1808
+ },
1809
+ {
1810
+ "epoch": 0.30222757375075254,
1811
+ "grad_norm": 0.7214757204055786,
1812
+ "learning_rate": 1.3954720616570329e-05,
1813
+ "loss": 1.2212,
1814
+ "step": 1255
1815
+ },
1816
+ {
1817
+ "epoch": 0.3034316676700783,
1818
+ "grad_norm": 0.7036615014076233,
1819
+ "learning_rate": 1.393063583815029e-05,
1820
+ "loss": 1.3103,
1821
+ "step": 1260
1822
+ },
1823
+ {
1824
+ "epoch": 0.304635761589404,
1825
+ "grad_norm": 0.7152957320213318,
1826
+ "learning_rate": 1.390655105973025e-05,
1827
+ "loss": 1.3316,
1828
+ "step": 1265
1829
+ },
1830
+ {
1831
+ "epoch": 0.30583985550872966,
1832
+ "grad_norm": 0.7779828906059265,
1833
+ "learning_rate": 1.3882466281310214e-05,
1834
+ "loss": 1.2773,
1835
+ "step": 1270
1836
+ },
1837
+ {
1838
+ "epoch": 0.3070439494280554,
1839
+ "grad_norm": 0.7995044589042664,
1840
+ "learning_rate": 1.3858381502890175e-05,
1841
+ "loss": 1.297,
1842
+ "step": 1275
1843
+ },
1844
+ {
1845
+ "epoch": 0.3082480433473811,
1846
+ "grad_norm": 0.7020440697669983,
1847
+ "learning_rate": 1.3834296724470136e-05,
1848
+ "loss": 1.1593,
1849
+ "step": 1280
1850
+ },
1851
+ {
1852
+ "epoch": 0.3094521372667068,
1853
+ "grad_norm": 0.7052685022354126,
1854
+ "learning_rate": 1.3810211946050098e-05,
1855
+ "loss": 1.1415,
1856
+ "step": 1285
1857
+ },
1858
+ {
1859
+ "epoch": 0.3106562311860325,
1860
+ "grad_norm": 0.7097790837287903,
1861
+ "learning_rate": 1.3786127167630059e-05,
1862
+ "loss": 1.0992,
1863
+ "step": 1290
1864
+ },
1865
+ {
1866
+ "epoch": 0.3118603251053582,
1867
+ "grad_norm": 0.7684621214866638,
1868
+ "learning_rate": 1.376204238921002e-05,
1869
+ "loss": 1.2256,
1870
+ "step": 1295
1871
+ },
1872
+ {
1873
+ "epoch": 0.3130644190246839,
1874
+ "grad_norm": 0.746864914894104,
1875
+ "learning_rate": 1.3737957610789983e-05,
1876
+ "loss": 1.2155,
1877
+ "step": 1300
1878
+ },
1879
+ {
1880
+ "epoch": 0.31426851294400965,
1881
+ "grad_norm": 0.733629584312439,
1882
+ "learning_rate": 1.3713872832369944e-05,
1883
+ "loss": 1.1287,
1884
+ "step": 1305
1885
+ },
1886
+ {
1887
+ "epoch": 0.31547260686333534,
1888
+ "grad_norm": 0.6500353217124939,
1889
+ "learning_rate": 1.3689788053949905e-05,
1890
+ "loss": 1.267,
1891
+ "step": 1310
1892
+ },
1893
+ {
1894
+ "epoch": 0.316676700782661,
1895
+ "grad_norm": 0.7301894426345825,
1896
+ "learning_rate": 1.3665703275529867e-05,
1897
+ "loss": 1.2167,
1898
+ "step": 1315
1899
+ },
1900
+ {
1901
+ "epoch": 0.31788079470198677,
1902
+ "grad_norm": 0.677039623260498,
1903
+ "learning_rate": 1.3641618497109828e-05,
1904
+ "loss": 1.1037,
1905
+ "step": 1320
1906
+ },
1907
+ {
1908
+ "epoch": 0.31908488862131246,
1909
+ "grad_norm": 0.7031332850456238,
1910
+ "learning_rate": 1.3617533718689788e-05,
1911
+ "loss": 1.1252,
1912
+ "step": 1325
1913
+ },
1914
+ {
1915
+ "epoch": 0.32028898254063815,
1916
+ "grad_norm": 0.8084255456924438,
1917
+ "learning_rate": 1.359344894026975e-05,
1918
+ "loss": 1.1728,
1919
+ "step": 1330
1920
+ },
1921
+ {
1922
+ "epoch": 0.3214930764599639,
1923
+ "grad_norm": 0.6915790438652039,
1924
+ "learning_rate": 1.3569364161849711e-05,
1925
+ "loss": 1.1552,
1926
+ "step": 1335
1927
+ },
1928
+ {
1929
+ "epoch": 0.3226971703792896,
1930
+ "grad_norm": 0.742514431476593,
1931
+ "learning_rate": 1.3545279383429672e-05,
1932
+ "loss": 1.1406,
1933
+ "step": 1340
1934
+ },
1935
+ {
1936
+ "epoch": 0.32390126429861527,
1937
+ "grad_norm": 0.6682133674621582,
1938
+ "learning_rate": 1.3521194605009636e-05,
1939
+ "loss": 1.1642,
1940
+ "step": 1345
1941
+ },
1942
+ {
1943
+ "epoch": 0.325105358217941,
1944
+ "grad_norm": 0.7231415510177612,
1945
+ "learning_rate": 1.3497109826589597e-05,
1946
+ "loss": 1.253,
1947
+ "step": 1350
1948
+ },
1949
+ {
1950
+ "epoch": 0.3263094521372667,
1951
+ "grad_norm": 0.7634385824203491,
1952
+ "learning_rate": 1.3473025048169559e-05,
1953
+ "loss": 1.2276,
1954
+ "step": 1355
1955
+ },
1956
+ {
1957
+ "epoch": 0.32751354605659244,
1958
+ "grad_norm": 0.6695632934570312,
1959
+ "learning_rate": 1.344894026974952e-05,
1960
+ "loss": 1.1864,
1961
+ "step": 1360
1962
+ },
1963
+ {
1964
+ "epoch": 0.32871763997591813,
1965
+ "grad_norm": 0.7134613394737244,
1966
+ "learning_rate": 1.342485549132948e-05,
1967
+ "loss": 1.1097,
1968
+ "step": 1365
1969
+ },
1970
+ {
1971
+ "epoch": 0.3299217338952438,
1972
+ "grad_norm": 0.6643304228782654,
1973
+ "learning_rate": 1.3400770712909442e-05,
1974
+ "loss": 1.2438,
1975
+ "step": 1370
1976
+ },
1977
+ {
1978
+ "epoch": 0.33112582781456956,
1979
+ "grad_norm": 0.6821620464324951,
1980
+ "learning_rate": 1.3376685934489405e-05,
1981
+ "loss": 1.2137,
1982
+ "step": 1375
1983
+ },
1984
+ {
1985
+ "epoch": 0.33232992173389525,
1986
+ "grad_norm": 0.7436670660972595,
1987
+ "learning_rate": 1.3352601156069365e-05,
1988
+ "loss": 1.0745,
1989
+ "step": 1380
1990
+ },
1991
+ {
1992
+ "epoch": 0.33353401565322094,
1993
+ "grad_norm": 0.7283557653427124,
1994
+ "learning_rate": 1.3328516377649328e-05,
1995
+ "loss": 1.04,
1996
+ "step": 1385
1997
+ },
1998
+ {
1999
+ "epoch": 0.3347381095725467,
2000
+ "grad_norm": 0.6368799805641174,
2001
+ "learning_rate": 1.3304431599229288e-05,
2002
+ "loss": 1.1829,
2003
+ "step": 1390
2004
+ },
2005
+ {
2006
+ "epoch": 0.33594220349187237,
2007
+ "grad_norm": 0.8481898307800293,
2008
+ "learning_rate": 1.3280346820809249e-05,
2009
+ "loss": 1.0901,
2010
+ "step": 1395
2011
+ },
2012
+ {
2013
+ "epoch": 0.33714629741119806,
2014
+ "grad_norm": 0.7311575412750244,
2015
+ "learning_rate": 1.3256262042389211e-05,
2016
+ "loss": 1.1135,
2017
+ "step": 1400
2018
+ },
2019
+ {
2020
+ "epoch": 0.33714629741119806,
2021
+ "eval_loss": 1.1897846460342407,
2022
+ "eval_runtime": 10.4886,
2023
+ "eval_samples_per_second": 9.534,
2024
+ "eval_steps_per_second": 1.239,
2025
+ "step": 1400
2026
+ },
2027
+ {
2028
+ "epoch": 0.3383503913305238,
2029
+ "grad_norm": 0.7002428770065308,
2030
+ "learning_rate": 1.3232177263969172e-05,
2031
+ "loss": 1.1384,
2032
+ "step": 1405
2033
+ },
2034
+ {
2035
+ "epoch": 0.3395544852498495,
2036
+ "grad_norm": 0.7509122490882874,
2037
+ "learning_rate": 1.3208092485549133e-05,
2038
+ "loss": 1.1667,
2039
+ "step": 1410
2040
+ },
2041
+ {
2042
+ "epoch": 0.3407585791691752,
2043
+ "grad_norm": 0.7156943082809448,
2044
+ "learning_rate": 1.3184007707129097e-05,
2045
+ "loss": 1.0665,
2046
+ "step": 1415
2047
+ },
2048
+ {
2049
+ "epoch": 0.3419626730885009,
2050
+ "grad_norm": 0.8622729182243347,
2051
+ "learning_rate": 1.3159922928709057e-05,
2052
+ "loss": 1.1503,
2053
+ "step": 1420
2054
+ },
2055
+ {
2056
+ "epoch": 0.3431667670078266,
2057
+ "grad_norm": 0.7578678131103516,
2058
+ "learning_rate": 1.3135838150289018e-05,
2059
+ "loss": 1.0761,
2060
+ "step": 1425
2061
+ },
2062
+ {
2063
+ "epoch": 0.3443708609271523,
2064
+ "grad_norm": 0.7222859263420105,
2065
+ "learning_rate": 1.311175337186898e-05,
2066
+ "loss": 1.2214,
2067
+ "step": 1430
2068
+ },
2069
+ {
2070
+ "epoch": 0.34557495484647804,
2071
+ "grad_norm": 0.7335066795349121,
2072
+ "learning_rate": 1.3087668593448941e-05,
2073
+ "loss": 1.1857,
2074
+ "step": 1435
2075
+ },
2076
+ {
2077
+ "epoch": 0.34677904876580373,
2078
+ "grad_norm": 0.6828716397285461,
2079
+ "learning_rate": 1.3063583815028902e-05,
2080
+ "loss": 1.2078,
2081
+ "step": 1440
2082
+ },
2083
+ {
2084
+ "epoch": 0.3479831426851294,
2085
+ "grad_norm": 0.8298027515411377,
2086
+ "learning_rate": 1.3039499036608864e-05,
2087
+ "loss": 1.2375,
2088
+ "step": 1445
2089
+ },
2090
+ {
2091
+ "epoch": 0.34918723660445516,
2092
+ "grad_norm": 0.7282932996749878,
2093
+ "learning_rate": 1.3015414258188825e-05,
2094
+ "loss": 1.1631,
2095
+ "step": 1450
2096
+ },
2097
+ {
2098
+ "epoch": 0.35039133052378085,
2099
+ "grad_norm": 0.6626558303833008,
2100
+ "learning_rate": 1.2991329479768787e-05,
2101
+ "loss": 1.1374,
2102
+ "step": 1455
2103
+ },
2104
+ {
2105
+ "epoch": 0.35159542444310654,
2106
+ "grad_norm": 0.8166629672050476,
2107
+ "learning_rate": 1.296724470134875e-05,
2108
+ "loss": 1.2521,
2109
+ "step": 1460
2110
+ },
2111
+ {
2112
+ "epoch": 0.3527995183624323,
2113
+ "grad_norm": 0.688026487827301,
2114
+ "learning_rate": 1.294315992292871e-05,
2115
+ "loss": 1.1008,
2116
+ "step": 1465
2117
+ },
2118
+ {
2119
+ "epoch": 0.354003612281758,
2120
+ "grad_norm": 0.7461696267127991,
2121
+ "learning_rate": 1.2919075144508672e-05,
2122
+ "loss": 1.0932,
2123
+ "step": 1470
2124
+ },
2125
+ {
2126
+ "epoch": 0.35520770620108366,
2127
+ "grad_norm": 0.7331961989402771,
2128
+ "learning_rate": 1.2894990366088633e-05,
2129
+ "loss": 1.0613,
2130
+ "step": 1475
2131
+ },
2132
+ {
2133
+ "epoch": 0.3564118001204094,
2134
+ "grad_norm": 0.7345579862594604,
2135
+ "learning_rate": 1.2870905587668594e-05,
2136
+ "loss": 1.3048,
2137
+ "step": 1480
2138
+ },
2139
+ {
2140
+ "epoch": 0.3576158940397351,
2141
+ "grad_norm": 0.8133093118667603,
2142
+ "learning_rate": 1.2846820809248558e-05,
2143
+ "loss": 1.2435,
2144
+ "step": 1485
2145
+ },
2146
+ {
2147
+ "epoch": 0.3588199879590608,
2148
+ "grad_norm": 0.7404915690422058,
2149
+ "learning_rate": 1.2822736030828518e-05,
2150
+ "loss": 1.1767,
2151
+ "step": 1490
2152
+ },
2153
+ {
2154
+ "epoch": 0.3600240818783865,
2155
+ "grad_norm": 0.8186049461364746,
2156
+ "learning_rate": 1.2798651252408479e-05,
2157
+ "loss": 1.0789,
2158
+ "step": 1495
2159
+ },
2160
+ {
2161
+ "epoch": 0.3612281757977122,
2162
+ "grad_norm": 0.8047255277633667,
2163
+ "learning_rate": 1.2774566473988441e-05,
2164
+ "loss": 1.1343,
2165
+ "step": 1500
2166
+ },
2167
+ {
2168
+ "epoch": 0.3624322697170379,
2169
+ "grad_norm": 0.6876709461212158,
2170
+ "learning_rate": 1.2750481695568402e-05,
2171
+ "loss": 1.0948,
2172
+ "step": 1505
2173
+ },
2174
+ {
2175
+ "epoch": 0.36363636363636365,
2176
+ "grad_norm": 0.7112250328063965,
2177
+ "learning_rate": 1.2726396917148362e-05,
2178
+ "loss": 1.1495,
2179
+ "step": 1510
2180
+ },
2181
+ {
2182
+ "epoch": 0.36484045755568933,
2183
+ "grad_norm": 0.6787546873092651,
2184
+ "learning_rate": 1.2702312138728325e-05,
2185
+ "loss": 1.2422,
2186
+ "step": 1515
2187
+ },
2188
+ {
2189
+ "epoch": 0.3660445514750151,
2190
+ "grad_norm": 0.677456796169281,
2191
+ "learning_rate": 1.2678227360308285e-05,
2192
+ "loss": 1.2153,
2193
+ "step": 1520
2194
+ },
2195
+ {
2196
+ "epoch": 0.36724864539434077,
2197
+ "grad_norm": 0.6819108724594116,
2198
+ "learning_rate": 1.2654142581888246e-05,
2199
+ "loss": 1.0196,
2200
+ "step": 1525
2201
+ },
2202
+ {
2203
+ "epoch": 0.36845273931366646,
2204
+ "grad_norm": 0.7156027555465698,
2205
+ "learning_rate": 1.263005780346821e-05,
2206
+ "loss": 1.2116,
2207
+ "step": 1530
2208
+ },
2209
+ {
2210
+ "epoch": 0.3696568332329922,
2211
+ "grad_norm": 0.7217147350311279,
2212
+ "learning_rate": 1.260597302504817e-05,
2213
+ "loss": 1.1021,
2214
+ "step": 1535
2215
+ },
2216
+ {
2217
+ "epoch": 0.3708609271523179,
2218
+ "grad_norm": 0.6850091814994812,
2219
+ "learning_rate": 1.2581888246628131e-05,
2220
+ "loss": 1.1054,
2221
+ "step": 1540
2222
+ },
2223
+ {
2224
+ "epoch": 0.3720650210716436,
2225
+ "grad_norm": 0.7076934576034546,
2226
+ "learning_rate": 1.2557803468208094e-05,
2227
+ "loss": 1.1533,
2228
+ "step": 1545
2229
+ },
2230
+ {
2231
+ "epoch": 0.3732691149909693,
2232
+ "grad_norm": 0.7233012318611145,
2233
+ "learning_rate": 1.2533718689788054e-05,
2234
+ "loss": 1.2163,
2235
+ "step": 1550
2236
+ },
2237
+ {
2238
+ "epoch": 0.374473208910295,
2239
+ "grad_norm": 0.8073462843894958,
2240
+ "learning_rate": 1.2509633911368015e-05,
2241
+ "loss": 1.2245,
2242
+ "step": 1555
2243
+ },
2244
+ {
2245
+ "epoch": 0.3756773028296207,
2246
+ "grad_norm": 0.72356116771698,
2247
+ "learning_rate": 1.2485549132947979e-05,
2248
+ "loss": 1.1593,
2249
+ "step": 1560
2250
+ },
2251
+ {
2252
+ "epoch": 0.37688139674894644,
2253
+ "grad_norm": 0.645554780960083,
2254
+ "learning_rate": 1.246146435452794e-05,
2255
+ "loss": 1.1967,
2256
+ "step": 1565
2257
+ },
2258
+ {
2259
+ "epoch": 0.37808549066827213,
2260
+ "grad_norm": 0.6589043140411377,
2261
+ "learning_rate": 1.2437379576107902e-05,
2262
+ "loss": 1.2093,
2263
+ "step": 1570
2264
+ },
2265
+ {
2266
+ "epoch": 0.3792895845875978,
2267
+ "grad_norm": 0.6996996402740479,
2268
+ "learning_rate": 1.2413294797687863e-05,
2269
+ "loss": 1.1498,
2270
+ "step": 1575
2271
+ },
2272
+ {
2273
+ "epoch": 0.38049367850692356,
2274
+ "grad_norm": 0.6371117234230042,
2275
+ "learning_rate": 1.2389210019267823e-05,
2276
+ "loss": 1.1499,
2277
+ "step": 1580
2278
+ },
2279
+ {
2280
+ "epoch": 0.38169777242624925,
2281
+ "grad_norm": 0.6780497431755066,
2282
+ "learning_rate": 1.2365125240847786e-05,
2283
+ "loss": 1.1191,
2284
+ "step": 1585
2285
+ },
2286
+ {
2287
+ "epoch": 0.38290186634557494,
2288
+ "grad_norm": 0.7249863743782043,
2289
+ "learning_rate": 1.2341040462427746e-05,
2290
+ "loss": 1.2254,
2291
+ "step": 1590
2292
+ },
2293
+ {
2294
+ "epoch": 0.3841059602649007,
2295
+ "grad_norm": 0.6600310206413269,
2296
+ "learning_rate": 1.2316955684007707e-05,
2297
+ "loss": 1.0585,
2298
+ "step": 1595
2299
+ },
2300
+ {
2301
+ "epoch": 0.38531005418422637,
2302
+ "grad_norm": 0.6883255839347839,
2303
+ "learning_rate": 1.2292870905587671e-05,
2304
+ "loss": 1.1318,
2305
+ "step": 1600
2306
+ },
2307
+ {
2308
+ "epoch": 0.38531005418422637,
2309
+ "eval_loss": 1.1825724840164185,
2310
+ "eval_runtime": 10.5009,
2311
+ "eval_samples_per_second": 9.523,
2312
+ "eval_steps_per_second": 1.238,
2313
+ "step": 1600
2314
+ },
2315
+ {
2316
+ "epoch": 0.38651414810355206,
2317
+ "grad_norm": 0.7173585295677185,
2318
+ "learning_rate": 1.2268786127167632e-05,
2319
+ "loss": 1.284,
2320
+ "step": 1605
2321
+ },
2322
+ {
2323
+ "epoch": 0.3877182420228778,
2324
+ "grad_norm": 0.7643466591835022,
2325
+ "learning_rate": 1.2244701348747592e-05,
2326
+ "loss": 1.3699,
2327
+ "step": 1610
2328
+ },
2329
+ {
2330
+ "epoch": 0.3889223359422035,
2331
+ "grad_norm": 0.690148115158081,
2332
+ "learning_rate": 1.2220616570327555e-05,
2333
+ "loss": 1.1335,
2334
+ "step": 1615
2335
+ },
2336
+ {
2337
+ "epoch": 0.3901264298615292,
2338
+ "grad_norm": 0.6676989197731018,
2339
+ "learning_rate": 1.2196531791907515e-05,
2340
+ "loss": 1.1887,
2341
+ "step": 1620
2342
+ },
2343
+ {
2344
+ "epoch": 0.3913305237808549,
2345
+ "grad_norm": 1.0724517107009888,
2346
+ "learning_rate": 1.2172447013487476e-05,
2347
+ "loss": 1.2212,
2348
+ "step": 1625
2349
+ },
2350
+ {
2351
+ "epoch": 0.3925346177001806,
2352
+ "grad_norm": 0.6685481071472168,
2353
+ "learning_rate": 1.214836223506744e-05,
2354
+ "loss": 1.0551,
2355
+ "step": 1630
2356
+ },
2357
+ {
2358
+ "epoch": 0.3937387116195063,
2359
+ "grad_norm": 0.7440733313560486,
2360
+ "learning_rate": 1.21242774566474e-05,
2361
+ "loss": 1.1565,
2362
+ "step": 1635
2363
+ },
2364
+ {
2365
+ "epoch": 0.39494280553883204,
2366
+ "grad_norm": 0.7820375561714172,
2367
+ "learning_rate": 1.2100192678227361e-05,
2368
+ "loss": 1.0733,
2369
+ "step": 1640
2370
+ },
2371
+ {
2372
+ "epoch": 0.39614689945815773,
2373
+ "grad_norm": 0.7016002535820007,
2374
+ "learning_rate": 1.2076107899807323e-05,
2375
+ "loss": 1.1914,
2376
+ "step": 1645
2377
+ },
2378
+ {
2379
+ "epoch": 0.3973509933774834,
2380
+ "grad_norm": 0.8267973065376282,
2381
+ "learning_rate": 1.2052023121387284e-05,
2382
+ "loss": 1.1687,
2383
+ "step": 1650
2384
+ },
2385
+ {
2386
+ "epoch": 0.39855508729680916,
2387
+ "grad_norm": 0.7492110729217529,
2388
+ "learning_rate": 1.2027938342967245e-05,
2389
+ "loss": 1.2462,
2390
+ "step": 1655
2391
+ },
2392
+ {
2393
+ "epoch": 0.39975918121613485,
2394
+ "grad_norm": 0.6964198350906372,
2395
+ "learning_rate": 1.2003853564547207e-05,
2396
+ "loss": 1.1537,
2397
+ "step": 1660
2398
+ },
2399
+ {
2400
+ "epoch": 0.40096327513546054,
2401
+ "grad_norm": 0.7141043543815613,
2402
+ "learning_rate": 1.1979768786127168e-05,
2403
+ "loss": 1.1106,
2404
+ "step": 1665
2405
+ },
2406
+ {
2407
+ "epoch": 0.4021673690547863,
2408
+ "grad_norm": 0.7486653327941895,
2409
+ "learning_rate": 1.1955684007707128e-05,
2410
+ "loss": 1.1255,
2411
+ "step": 1670
2412
+ },
2413
+ {
2414
+ "epoch": 0.40337146297411197,
2415
+ "grad_norm": 0.68989098072052,
2416
+ "learning_rate": 1.1931599229287092e-05,
2417
+ "loss": 1.1491,
2418
+ "step": 1675
2419
+ },
2420
+ {
2421
+ "epoch": 0.4045755568934377,
2422
+ "grad_norm": 0.7117523550987244,
2423
+ "learning_rate": 1.1907514450867053e-05,
2424
+ "loss": 1.2564,
2425
+ "step": 1680
2426
+ },
2427
+ {
2428
+ "epoch": 0.4057796508127634,
2429
+ "grad_norm": 0.7256306409835815,
2430
+ "learning_rate": 1.1883429672447015e-05,
2431
+ "loss": 1.2544,
2432
+ "step": 1685
2433
+ },
2434
+ {
2435
+ "epoch": 0.4069837447320891,
2436
+ "grad_norm": 0.6354299187660217,
2437
+ "learning_rate": 1.1859344894026976e-05,
2438
+ "loss": 1.1011,
2439
+ "step": 1690
2440
+ },
2441
+ {
2442
+ "epoch": 0.40818783865141484,
2443
+ "grad_norm": 0.7653043270111084,
2444
+ "learning_rate": 1.1835260115606937e-05,
2445
+ "loss": 1.1881,
2446
+ "step": 1695
2447
+ },
2448
+ {
2449
+ "epoch": 0.4093919325707405,
2450
+ "grad_norm": 0.6969402432441711,
2451
+ "learning_rate": 1.1811175337186899e-05,
2452
+ "loss": 1.2509,
2453
+ "step": 1700
2454
+ },
2455
+ {
2456
+ "epoch": 0.4105960264900662,
2457
+ "grad_norm": 0.7759403586387634,
2458
+ "learning_rate": 1.1787090558766861e-05,
2459
+ "loss": 1.1777,
2460
+ "step": 1705
2461
+ },
2462
+ {
2463
+ "epoch": 0.41180012040939196,
2464
+ "grad_norm": 0.8177191019058228,
2465
+ "learning_rate": 1.1763005780346822e-05,
2466
+ "loss": 1.2677,
2467
+ "step": 1710
2468
+ },
2469
+ {
2470
+ "epoch": 0.41300421432871764,
2471
+ "grad_norm": 0.6352664828300476,
2472
+ "learning_rate": 1.1738921001926784e-05,
2473
+ "loss": 1.107,
2474
+ "step": 1715
2475
+ },
2476
+ {
2477
+ "epoch": 0.41420830824804333,
2478
+ "grad_norm": 0.7194890975952148,
2479
+ "learning_rate": 1.1714836223506745e-05,
2480
+ "loss": 1.1646,
2481
+ "step": 1720
2482
+ },
2483
+ {
2484
+ "epoch": 0.4154124021673691,
2485
+ "grad_norm": 0.6629904508590698,
2486
+ "learning_rate": 1.1690751445086706e-05,
2487
+ "loss": 1.1938,
2488
+ "step": 1725
2489
+ },
2490
+ {
2491
+ "epoch": 0.41661649608669477,
2492
+ "grad_norm": 0.6937288641929626,
2493
+ "learning_rate": 1.1666666666666668e-05,
2494
+ "loss": 1.1254,
2495
+ "step": 1730
2496
+ },
2497
+ {
2498
+ "epoch": 0.41782059000602045,
2499
+ "grad_norm": 0.6822580099105835,
2500
+ "learning_rate": 1.1642581888246629e-05,
2501
+ "loss": 1.1878,
2502
+ "step": 1735
2503
+ },
2504
+ {
2505
+ "epoch": 0.4190246839253462,
2506
+ "grad_norm": 0.8951644897460938,
2507
+ "learning_rate": 1.161849710982659e-05,
2508
+ "loss": 1.0153,
2509
+ "step": 1740
2510
+ },
2511
+ {
2512
+ "epoch": 0.4202287778446719,
2513
+ "grad_norm": 0.7985068559646606,
2514
+ "learning_rate": 1.1594412331406553e-05,
2515
+ "loss": 1.2309,
2516
+ "step": 1745
2517
+ },
2518
+ {
2519
+ "epoch": 0.4214328717639976,
2520
+ "grad_norm": 0.7742875814437866,
2521
+ "learning_rate": 1.1570327552986514e-05,
2522
+ "loss": 1.2868,
2523
+ "step": 1750
2524
+ },
2525
+ {
2526
+ "epoch": 0.4226369656833233,
2527
+ "grad_norm": 0.7290802001953125,
2528
+ "learning_rate": 1.1546242774566474e-05,
2529
+ "loss": 1.1529,
2530
+ "step": 1755
2531
+ },
2532
+ {
2533
+ "epoch": 0.423841059602649,
2534
+ "grad_norm": 0.7852470278739929,
2535
+ "learning_rate": 1.1522157996146437e-05,
2536
+ "loss": 1.1758,
2537
+ "step": 1760
2538
+ },
2539
+ {
2540
+ "epoch": 0.4250451535219747,
2541
+ "grad_norm": 0.6864736676216125,
2542
+ "learning_rate": 1.1498073217726397e-05,
2543
+ "loss": 1.173,
2544
+ "step": 1765
2545
+ },
2546
+ {
2547
+ "epoch": 0.42624924744130044,
2548
+ "grad_norm": 0.6752791404724121,
2549
+ "learning_rate": 1.1473988439306358e-05,
2550
+ "loss": 1.101,
2551
+ "step": 1770
2552
+ },
2553
+ {
2554
+ "epoch": 0.4274533413606261,
2555
+ "grad_norm": 0.6282265186309814,
2556
+ "learning_rate": 1.144990366088632e-05,
2557
+ "loss": 1.0218,
2558
+ "step": 1775
2559
+ },
2560
+ {
2561
+ "epoch": 0.4286574352799518,
2562
+ "grad_norm": 0.7045950889587402,
2563
+ "learning_rate": 1.1425818882466281e-05,
2564
+ "loss": 1.2667,
2565
+ "step": 1780
2566
+ },
2567
+ {
2568
+ "epoch": 0.42986152919927756,
2569
+ "grad_norm": 0.7136130928993225,
2570
+ "learning_rate": 1.1401734104046245e-05,
2571
+ "loss": 1.2445,
2572
+ "step": 1785
2573
+ },
2574
+ {
2575
+ "epoch": 0.43106562311860325,
2576
+ "grad_norm": 0.7001535296440125,
2577
+ "learning_rate": 1.1377649325626206e-05,
2578
+ "loss": 1.2076,
2579
+ "step": 1790
2580
+ },
2581
+ {
2582
+ "epoch": 0.43226971703792894,
2583
+ "grad_norm": 0.7172293066978455,
2584
+ "learning_rate": 1.1353564547206166e-05,
2585
+ "loss": 1.0828,
2586
+ "step": 1795
2587
+ },
2588
+ {
2589
+ "epoch": 0.4334738109572547,
2590
+ "grad_norm": 0.7160560488700867,
2591
+ "learning_rate": 1.1329479768786129e-05,
2592
+ "loss": 1.1676,
2593
+ "step": 1800
2594
+ },
2595
+ {
2596
+ "epoch": 0.4334738109572547,
2597
+ "eval_loss": 1.1778274774551392,
2598
+ "eval_runtime": 10.4751,
2599
+ "eval_samples_per_second": 9.546,
2600
+ "eval_steps_per_second": 1.241,
2601
+ "step": 1800
2602
+ },
2603
+ {
2604
+ "epoch": 0.43467790487658037,
2605
+ "grad_norm": 0.7364829778671265,
2606
+ "learning_rate": 1.130539499036609e-05,
2607
+ "loss": 1.1337,
2608
+ "step": 1805
2609
+ },
2610
+ {
2611
+ "epoch": 0.43588199879590606,
2612
+ "grad_norm": 0.7415775656700134,
2613
+ "learning_rate": 1.128131021194605e-05,
2614
+ "loss": 1.1882,
2615
+ "step": 1810
2616
+ },
2617
+ {
2618
+ "epoch": 0.4370860927152318,
2619
+ "grad_norm": 0.7825621366500854,
2620
+ "learning_rate": 1.1257225433526014e-05,
2621
+ "loss": 1.1627,
2622
+ "step": 1815
2623
+ },
2624
+ {
2625
+ "epoch": 0.4382901866345575,
2626
+ "grad_norm": 0.7862880229949951,
2627
+ "learning_rate": 1.1233140655105975e-05,
2628
+ "loss": 1.2287,
2629
+ "step": 1820
2630
+ },
2631
+ {
2632
+ "epoch": 0.4394942805538832,
2633
+ "grad_norm": 0.7740750908851624,
2634
+ "learning_rate": 1.1209055876685935e-05,
2635
+ "loss": 1.1488,
2636
+ "step": 1825
2637
+ },
2638
+ {
2639
+ "epoch": 0.4406983744732089,
2640
+ "grad_norm": 0.8551940321922302,
2641
+ "learning_rate": 1.1184971098265898e-05,
2642
+ "loss": 1.1894,
2643
+ "step": 1830
2644
+ },
2645
+ {
2646
+ "epoch": 0.4419024683925346,
2647
+ "grad_norm": 0.7718561291694641,
2648
+ "learning_rate": 1.1160886319845858e-05,
2649
+ "loss": 1.1825,
2650
+ "step": 1835
2651
+ },
2652
+ {
2653
+ "epoch": 0.44310656231186035,
2654
+ "grad_norm": 0.7665084600448608,
2655
+ "learning_rate": 1.1136801541425819e-05,
2656
+ "loss": 1.0931,
2657
+ "step": 1840
2658
+ },
2659
+ {
2660
+ "epoch": 0.44431065623118604,
2661
+ "grad_norm": 0.7155455350875854,
2662
+ "learning_rate": 1.1112716763005781e-05,
2663
+ "loss": 1.2208,
2664
+ "step": 1845
2665
+ },
2666
+ {
2667
+ "epoch": 0.44551475015051173,
2668
+ "grad_norm": 0.6576226353645325,
2669
+ "learning_rate": 1.1088631984585742e-05,
2670
+ "loss": 1.1294,
2671
+ "step": 1850
2672
+ },
2673
+ {
2674
+ "epoch": 0.4467188440698375,
2675
+ "grad_norm": 0.6880900859832764,
2676
+ "learning_rate": 1.1064547206165703e-05,
2677
+ "loss": 1.0646,
2678
+ "step": 1855
2679
+ },
2680
+ {
2681
+ "epoch": 0.44792293798916316,
2682
+ "grad_norm": 0.633608341217041,
2683
+ "learning_rate": 1.1040462427745667e-05,
2684
+ "loss": 1.0871,
2685
+ "step": 1860
2686
+ },
2687
+ {
2688
+ "epoch": 0.44912703190848885,
2689
+ "grad_norm": 0.6906594634056091,
2690
+ "learning_rate": 1.1016377649325627e-05,
2691
+ "loss": 1.1449,
2692
+ "step": 1865
2693
+ },
2694
+ {
2695
+ "epoch": 0.4503311258278146,
2696
+ "grad_norm": 0.7329970598220825,
2697
+ "learning_rate": 1.0992292870905588e-05,
2698
+ "loss": 1.1158,
2699
+ "step": 1870
2700
+ },
2701
+ {
2702
+ "epoch": 0.4515352197471403,
2703
+ "grad_norm": 0.7347983121871948,
2704
+ "learning_rate": 1.096820809248555e-05,
2705
+ "loss": 1.1252,
2706
+ "step": 1875
2707
+ },
2708
+ {
2709
+ "epoch": 0.45273931366646597,
2710
+ "grad_norm": 0.7085864543914795,
2711
+ "learning_rate": 1.094412331406551e-05,
2712
+ "loss": 1.1535,
2713
+ "step": 1880
2714
+ },
2715
+ {
2716
+ "epoch": 0.4539434075857917,
2717
+ "grad_norm": 0.7692695260047913,
2718
+ "learning_rate": 1.0920038535645471e-05,
2719
+ "loss": 1.1632,
2720
+ "step": 1885
2721
+ },
2722
+ {
2723
+ "epoch": 0.4551475015051174,
2724
+ "grad_norm": 0.7081477046012878,
2725
+ "learning_rate": 1.0895953757225436e-05,
2726
+ "loss": 1.1269,
2727
+ "step": 1890
2728
+ },
2729
+ {
2730
+ "epoch": 0.4563515954244431,
2731
+ "grad_norm": 0.663478434085846,
2732
+ "learning_rate": 1.0871868978805396e-05,
2733
+ "loss": 1.1256,
2734
+ "step": 1895
2735
+ },
2736
+ {
2737
+ "epoch": 0.45755568934376883,
2738
+ "grad_norm": 0.7210192084312439,
2739
+ "learning_rate": 1.0847784200385358e-05,
2740
+ "loss": 1.1887,
2741
+ "step": 1900
2742
+ },
2743
+ {
2744
+ "epoch": 0.4587597832630945,
2745
+ "grad_norm": 0.7571423649787903,
2746
+ "learning_rate": 1.0823699421965319e-05,
2747
+ "loss": 1.1451,
2748
+ "step": 1905
2749
+ },
2750
+ {
2751
+ "epoch": 0.4599638771824202,
2752
+ "grad_norm": 0.79608154296875,
2753
+ "learning_rate": 1.079961464354528e-05,
2754
+ "loss": 1.1874,
2755
+ "step": 1910
2756
+ },
2757
+ {
2758
+ "epoch": 0.46116797110174595,
2759
+ "grad_norm": 0.857712984085083,
2760
+ "learning_rate": 1.0775529865125242e-05,
2761
+ "loss": 1.1669,
2762
+ "step": 1915
2763
+ },
2764
+ {
2765
+ "epoch": 0.46237206502107164,
2766
+ "grad_norm": 0.644782304763794,
2767
+ "learning_rate": 1.0751445086705203e-05,
2768
+ "loss": 1.1627,
2769
+ "step": 1920
2770
+ },
2771
+ {
2772
+ "epoch": 0.46357615894039733,
2773
+ "grad_norm": 0.6408756971359253,
2774
+ "learning_rate": 1.0727360308285163e-05,
2775
+ "loss": 1.069,
2776
+ "step": 1925
2777
+ },
2778
+ {
2779
+ "epoch": 0.4647802528597231,
2780
+ "grad_norm": 0.6929676532745361,
2781
+ "learning_rate": 1.0703275529865127e-05,
2782
+ "loss": 1.1435,
2783
+ "step": 1930
2784
+ },
2785
+ {
2786
+ "epoch": 0.46598434677904876,
2787
+ "grad_norm": 0.7538990378379822,
2788
+ "learning_rate": 1.0679190751445088e-05,
2789
+ "loss": 1.2604,
2790
+ "step": 1935
2791
+ },
2792
+ {
2793
+ "epoch": 0.46718844069837445,
2794
+ "grad_norm": 0.6711621880531311,
2795
+ "learning_rate": 1.0655105973025049e-05,
2796
+ "loss": 1.1923,
2797
+ "step": 1940
2798
+ },
2799
+ {
2800
+ "epoch": 0.4683925346177002,
2801
+ "grad_norm": 0.6384220123291016,
2802
+ "learning_rate": 1.0631021194605011e-05,
2803
+ "loss": 1.1269,
2804
+ "step": 1945
2805
+ },
2806
+ {
2807
+ "epoch": 0.4695966285370259,
2808
+ "grad_norm": 0.682374119758606,
2809
+ "learning_rate": 1.0606936416184972e-05,
2810
+ "loss": 1.1268,
2811
+ "step": 1950
2812
+ },
2813
+ {
2814
+ "epoch": 0.4708007224563516,
2815
+ "grad_norm": 0.7387133240699768,
2816
+ "learning_rate": 1.0582851637764932e-05,
2817
+ "loss": 1.1043,
2818
+ "step": 1955
2819
+ },
2820
+ {
2821
+ "epoch": 0.4720048163756773,
2822
+ "grad_norm": 0.6632955074310303,
2823
+ "learning_rate": 1.0558766859344896e-05,
2824
+ "loss": 1.1148,
2825
+ "step": 1960
2826
+ },
2827
+ {
2828
+ "epoch": 0.473208910295003,
2829
+ "grad_norm": 0.6747167706489563,
2830
+ "learning_rate": 1.0534682080924857e-05,
2831
+ "loss": 1.1788,
2832
+ "step": 1965
2833
+ },
2834
+ {
2835
+ "epoch": 0.4744130042143287,
2836
+ "grad_norm": 0.6498105525970459,
2837
+ "learning_rate": 1.0510597302504818e-05,
2838
+ "loss": 1.2199,
2839
+ "step": 1970
2840
+ },
2841
+ {
2842
+ "epoch": 0.47561709813365444,
2843
+ "grad_norm": 0.72532057762146,
2844
+ "learning_rate": 1.048651252408478e-05,
2845
+ "loss": 0.9953,
2846
+ "step": 1975
2847
+ },
2848
+ {
2849
+ "epoch": 0.4768211920529801,
2850
+ "grad_norm": 0.7126530408859253,
2851
+ "learning_rate": 1.046242774566474e-05,
2852
+ "loss": 1.1014,
2853
+ "step": 1980
2854
+ },
2855
+ {
2856
+ "epoch": 0.4780252859723058,
2857
+ "grad_norm": 0.8094469904899597,
2858
+ "learning_rate": 1.0438342967244701e-05,
2859
+ "loss": 1.0567,
2860
+ "step": 1985
2861
+ },
2862
+ {
2863
+ "epoch": 0.47922937989163156,
2864
+ "grad_norm": 0.7181186676025391,
2865
+ "learning_rate": 1.0414258188824664e-05,
2866
+ "loss": 1.1077,
2867
+ "step": 1990
2868
+ },
2869
+ {
2870
+ "epoch": 0.48043347381095725,
2871
+ "grad_norm": 0.6791627407073975,
2872
+ "learning_rate": 1.0390173410404624e-05,
2873
+ "loss": 1.1272,
2874
+ "step": 1995
2875
+ },
2876
+ {
2877
+ "epoch": 0.481637567730283,
2878
+ "grad_norm": 0.7202663421630859,
2879
+ "learning_rate": 1.0366088631984585e-05,
2880
+ "loss": 1.139,
2881
+ "step": 2000
2882
+ },
2883
+ {
2884
+ "epoch": 0.481637567730283,
2885
+ "eval_loss": 1.1731120347976685,
2886
+ "eval_runtime": 10.4779,
2887
+ "eval_samples_per_second": 9.544,
2888
+ "eval_steps_per_second": 1.241,
2889
+ "step": 2000
2890
+ },
2891
+ {
2892
+ "epoch": 0.4828416616496087,
2893
+ "grad_norm": 0.7046011686325073,
2894
+ "learning_rate": 1.0342003853564549e-05,
2895
+ "loss": 1.3475,
2896
+ "step": 2005
2897
+ },
2898
+ {
2899
+ "epoch": 0.48404575556893437,
2900
+ "grad_norm": 0.7181674242019653,
2901
+ "learning_rate": 1.031791907514451e-05,
2902
+ "loss": 1.1532,
2903
+ "step": 2010
2904
+ },
2905
+ {
2906
+ "epoch": 0.4852498494882601,
2907
+ "grad_norm": 0.7380154132843018,
2908
+ "learning_rate": 1.0293834296724472e-05,
2909
+ "loss": 1.0837,
2910
+ "step": 2015
2911
+ },
2912
+ {
2913
+ "epoch": 0.4864539434075858,
2914
+ "grad_norm": 0.7488975524902344,
2915
+ "learning_rate": 1.0269749518304432e-05,
2916
+ "loss": 1.1603,
2917
+ "step": 2020
2918
+ },
2919
+ {
2920
+ "epoch": 0.4876580373269115,
2921
+ "grad_norm": 0.6608518958091736,
2922
+ "learning_rate": 1.0245664739884393e-05,
2923
+ "loss": 1.1098,
2924
+ "step": 2025
2925
+ },
2926
+ {
2927
+ "epoch": 0.48886213124623723,
2928
+ "grad_norm": 0.6861777305603027,
2929
+ "learning_rate": 1.0221579961464355e-05,
2930
+ "loss": 1.1446,
2931
+ "step": 2030
2932
+ },
2933
+ {
2934
+ "epoch": 0.4900662251655629,
2935
+ "grad_norm": 0.6870723962783813,
2936
+ "learning_rate": 1.0197495183044318e-05,
2937
+ "loss": 1.0822,
2938
+ "step": 2035
2939
+ },
2940
+ {
2941
+ "epoch": 0.4912703190848886,
2942
+ "grad_norm": 0.7478333711624146,
2943
+ "learning_rate": 1.0173410404624278e-05,
2944
+ "loss": 1.122,
2945
+ "step": 2040
2946
+ },
2947
+ {
2948
+ "epoch": 0.49247441300421435,
2949
+ "grad_norm": 0.7138153910636902,
2950
+ "learning_rate": 1.014932562620424e-05,
2951
+ "loss": 1.1535,
2952
+ "step": 2045
2953
+ },
2954
+ {
2955
+ "epoch": 0.49367850692354004,
2956
+ "grad_norm": 0.7198281288146973,
2957
+ "learning_rate": 1.0125240847784201e-05,
2958
+ "loss": 1.2071,
2959
+ "step": 2050
2960
+ },
2961
+ {
2962
+ "epoch": 0.4948826008428657,
2963
+ "grad_norm": 0.7242985367774963,
2964
+ "learning_rate": 1.0101156069364162e-05,
2965
+ "loss": 1.1206,
2966
+ "step": 2055
2967
+ },
2968
+ {
2969
+ "epoch": 0.49608669476219147,
2970
+ "grad_norm": 0.71178138256073,
2971
+ "learning_rate": 1.0077071290944124e-05,
2972
+ "loss": 1.1888,
2973
+ "step": 2060
2974
+ },
2975
+ {
2976
+ "epoch": 0.49729078868151716,
2977
+ "grad_norm": 0.7503355741500854,
2978
+ "learning_rate": 1.0052986512524085e-05,
2979
+ "loss": 1.152,
2980
+ "step": 2065
2981
+ },
2982
+ {
2983
+ "epoch": 0.49849488260084285,
2984
+ "grad_norm": 0.741552472114563,
2985
+ "learning_rate": 1.0028901734104046e-05,
2986
+ "loss": 1.1612,
2987
+ "step": 2070
2988
+ },
2989
+ {
2990
+ "epoch": 0.4996989765201686,
2991
+ "grad_norm": 0.7213262319564819,
2992
+ "learning_rate": 1.000481695568401e-05,
2993
+ "loss": 1.1795,
2994
+ "step": 2075
2995
+ },
2996
+ {
2997
+ "epoch": 0.5009030704394943,
2998
+ "grad_norm": 0.6946147680282593,
2999
+ "learning_rate": 9.98073217726397e-06,
3000
+ "loss": 1.1088,
3001
+ "step": 2080
3002
+ },
3003
+ {
3004
+ "epoch": 0.50210716435882,
3005
+ "grad_norm": 0.7090691924095154,
3006
+ "learning_rate": 9.956647398843931e-06,
3007
+ "loss": 1.1167,
3008
+ "step": 2085
3009
+ },
3010
+ {
3011
+ "epoch": 0.5033112582781457,
3012
+ "grad_norm": 0.7390459179878235,
3013
+ "learning_rate": 9.932562620423893e-06,
3014
+ "loss": 1.1359,
3015
+ "step": 2090
3016
+ },
3017
+ {
3018
+ "epoch": 0.5045153521974715,
3019
+ "grad_norm": 0.6658144593238831,
3020
+ "learning_rate": 9.908477842003854e-06,
3021
+ "loss": 1.0265,
3022
+ "step": 2095
3023
+ },
3024
+ {
3025
+ "epoch": 0.5057194461167971,
3026
+ "grad_norm": 0.7175756692886353,
3027
+ "learning_rate": 9.884393063583816e-06,
3028
+ "loss": 1.1079,
3029
+ "step": 2100
3030
+ },
3031
+ {
3032
+ "epoch": 0.5069235400361228,
3033
+ "grad_norm": 0.649573802947998,
3034
+ "learning_rate": 9.860308285163777e-06,
3035
+ "loss": 1.1319,
3036
+ "step": 2105
3037
+ },
3038
+ {
3039
+ "epoch": 0.5081276339554486,
3040
+ "grad_norm": 0.770150899887085,
3041
+ "learning_rate": 9.836223506743738e-06,
3042
+ "loss": 1.1809,
3043
+ "step": 2110
3044
+ },
3045
+ {
3046
+ "epoch": 0.5093317278747742,
3047
+ "grad_norm": 0.6186807155609131,
3048
+ "learning_rate": 9.8121387283237e-06,
3049
+ "loss": 1.1099,
3050
+ "step": 2115
3051
+ },
3052
+ {
3053
+ "epoch": 0.5105358217941,
3054
+ "grad_norm": 0.8233024477958679,
3055
+ "learning_rate": 9.788053949903662e-06,
3056
+ "loss": 1.0872,
3057
+ "step": 2120
3058
+ },
3059
+ {
3060
+ "epoch": 0.5117399157134257,
3061
+ "grad_norm": 0.7936651706695557,
3062
+ "learning_rate": 9.763969171483623e-06,
3063
+ "loss": 1.1372,
3064
+ "step": 2125
3065
+ },
3066
+ {
3067
+ "epoch": 0.5129440096327513,
3068
+ "grad_norm": 0.7144600749015808,
3069
+ "learning_rate": 9.739884393063585e-06,
3070
+ "loss": 1.207,
3071
+ "step": 2130
3072
+ },
3073
+ {
3074
+ "epoch": 0.5141481035520771,
3075
+ "grad_norm": 0.7522640228271484,
3076
+ "learning_rate": 9.715799614643546e-06,
3077
+ "loss": 1.0803,
3078
+ "step": 2135
3079
+ },
3080
+ {
3081
+ "epoch": 0.5153521974714028,
3082
+ "grad_norm": 0.7434991002082825,
3083
+ "learning_rate": 9.691714836223508e-06,
3084
+ "loss": 1.2786,
3085
+ "step": 2140
3086
+ },
3087
+ {
3088
+ "epoch": 0.5165562913907285,
3089
+ "grad_norm": 0.6706145405769348,
3090
+ "learning_rate": 9.667630057803469e-06,
3091
+ "loss": 1.1589,
3092
+ "step": 2145
3093
+ },
3094
+ {
3095
+ "epoch": 0.5177603853100542,
3096
+ "grad_norm": 0.7273653745651245,
3097
+ "learning_rate": 9.643545279383431e-06,
3098
+ "loss": 1.1075,
3099
+ "step": 2150
3100
+ },
3101
+ {
3102
+ "epoch": 0.5189644792293799,
3103
+ "grad_norm": 0.6501129269599915,
3104
+ "learning_rate": 9.619460500963392e-06,
3105
+ "loss": 1.1765,
3106
+ "step": 2155
3107
+ },
3108
+ {
3109
+ "epoch": 0.5201685731487056,
3110
+ "grad_norm": 0.6212615966796875,
3111
+ "learning_rate": 9.595375722543352e-06,
3112
+ "loss": 1.1055,
3113
+ "step": 2160
3114
+ },
3115
+ {
3116
+ "epoch": 0.5213726670680313,
3117
+ "grad_norm": 0.6511011123657227,
3118
+ "learning_rate": 9.571290944123315e-06,
3119
+ "loss": 1.2099,
3120
+ "step": 2165
3121
+ },
3122
+ {
3123
+ "epoch": 0.5225767609873571,
3124
+ "grad_norm": 0.6724874973297119,
3125
+ "learning_rate": 9.547206165703277e-06,
3126
+ "loss": 1.083,
3127
+ "step": 2170
3128
+ },
3129
+ {
3130
+ "epoch": 0.5237808549066827,
3131
+ "grad_norm": 0.8235754370689392,
3132
+ "learning_rate": 9.523121387283238e-06,
3133
+ "loss": 1.1804,
3134
+ "step": 2175
3135
+ },
3136
+ {
3137
+ "epoch": 0.5249849488260084,
3138
+ "grad_norm": 0.7417103052139282,
3139
+ "learning_rate": 9.499036608863198e-06,
3140
+ "loss": 1.1334,
3141
+ "step": 2180
3142
+ },
3143
+ {
3144
+ "epoch": 0.5261890427453342,
3145
+ "grad_norm": 0.6703997850418091,
3146
+ "learning_rate": 9.47495183044316e-06,
3147
+ "loss": 1.0683,
3148
+ "step": 2185
3149
+ },
3150
+ {
3151
+ "epoch": 0.5273931366646598,
3152
+ "grad_norm": 0.8259227275848389,
3153
+ "learning_rate": 9.450867052023123e-06,
3154
+ "loss": 1.2371,
3155
+ "step": 2190
3156
+ },
3157
+ {
3158
+ "epoch": 0.5285972305839856,
3159
+ "grad_norm": 0.7515948414802551,
3160
+ "learning_rate": 9.426782273603084e-06,
3161
+ "loss": 1.0862,
3162
+ "step": 2195
3163
+ },
3164
+ {
3165
+ "epoch": 0.5298013245033113,
3166
+ "grad_norm": 0.7639467716217041,
3167
+ "learning_rate": 9.402697495183044e-06,
3168
+ "loss": 1.1411,
3169
+ "step": 2200
3170
+ },
3171
+ {
3172
+ "epoch": 0.5298013245033113,
3173
+ "eval_loss": 1.1692317724227905,
3174
+ "eval_runtime": 10.416,
3175
+ "eval_samples_per_second": 9.601,
3176
+ "eval_steps_per_second": 1.248,
3177
+ "step": 2200
3178
+ },
3179
+ {
3180
+ "epoch": 0.5310054184226369,
3181
+ "grad_norm": 0.641248881816864,
3182
+ "learning_rate": 9.378612716763007e-06,
3183
+ "loss": 1.1405,
3184
+ "step": 2205
3185
+ },
3186
+ {
3187
+ "epoch": 0.5322095123419627,
3188
+ "grad_norm": 0.7671200037002563,
3189
+ "learning_rate": 9.354527938342967e-06,
3190
+ "loss": 1.2501,
3191
+ "step": 2210
3192
+ },
3193
+ {
3194
+ "epoch": 0.5334136062612884,
3195
+ "grad_norm": 0.7269463539123535,
3196
+ "learning_rate": 9.33044315992293e-06,
3197
+ "loss": 1.1345,
3198
+ "step": 2215
3199
+ },
3200
+ {
3201
+ "epoch": 0.534617700180614,
3202
+ "grad_norm": 0.7109377980232239,
3203
+ "learning_rate": 9.306358381502892e-06,
3204
+ "loss": 1.0514,
3205
+ "step": 2220
3206
+ },
3207
+ {
3208
+ "epoch": 0.5358217940999398,
3209
+ "grad_norm": 0.7129069566726685,
3210
+ "learning_rate": 9.282273603082853e-06,
3211
+ "loss": 1.1813,
3212
+ "step": 2225
3213
+ },
3214
+ {
3215
+ "epoch": 0.5370258880192655,
3216
+ "grad_norm": 0.6922154426574707,
3217
+ "learning_rate": 9.258188824662813e-06,
3218
+ "loss": 1.1343,
3219
+ "step": 2230
3220
+ },
3221
+ {
3222
+ "epoch": 0.5382299819385912,
3223
+ "grad_norm": 0.7257413864135742,
3224
+ "learning_rate": 9.234104046242776e-06,
3225
+ "loss": 1.0229,
3226
+ "step": 2235
3227
+ },
3228
+ {
3229
+ "epoch": 0.5394340758579169,
3230
+ "grad_norm": 0.7164874076843262,
3231
+ "learning_rate": 9.210019267822736e-06,
3232
+ "loss": 1.1988,
3233
+ "step": 2240
3234
+ },
3235
+ {
3236
+ "epoch": 0.5406381697772427,
3237
+ "grad_norm": 0.7674132585525513,
3238
+ "learning_rate": 9.185934489402699e-06,
3239
+ "loss": 1.2122,
3240
+ "step": 2245
3241
+ },
3242
+ {
3243
+ "epoch": 0.5418422636965683,
3244
+ "grad_norm": 0.6795153617858887,
3245
+ "learning_rate": 9.16184971098266e-06,
3246
+ "loss": 1.2543,
3247
+ "step": 2250
3248
+ },
3249
+ {
3250
+ "epoch": 0.543046357615894,
3251
+ "grad_norm": 0.7131598591804504,
3252
+ "learning_rate": 9.137764932562622e-06,
3253
+ "loss": 1.2006,
3254
+ "step": 2255
3255
+ },
3256
+ {
3257
+ "epoch": 0.5442504515352198,
3258
+ "grad_norm": 0.8124786615371704,
3259
+ "learning_rate": 9.113680154142582e-06,
3260
+ "loss": 1.1184,
3261
+ "step": 2260
3262
+ },
3263
+ {
3264
+ "epoch": 0.5454545454545454,
3265
+ "grad_norm": 0.7463451027870178,
3266
+ "learning_rate": 9.089595375722545e-06,
3267
+ "loss": 1.1329,
3268
+ "step": 2265
3269
+ },
3270
+ {
3271
+ "epoch": 0.5466586393738712,
3272
+ "grad_norm": 0.8033417463302612,
3273
+ "learning_rate": 9.065510597302505e-06,
3274
+ "loss": 1.0873,
3275
+ "step": 2270
3276
+ },
3277
+ {
3278
+ "epoch": 0.5478627332931969,
3279
+ "grad_norm": 0.7155200839042664,
3280
+ "learning_rate": 9.041425818882466e-06,
3281
+ "loss": 1.1056,
3282
+ "step": 2275
3283
+ },
3284
+ {
3285
+ "epoch": 0.5490668272125225,
3286
+ "grad_norm": 0.643032431602478,
3287
+ "learning_rate": 9.017341040462428e-06,
3288
+ "loss": 1.2173,
3289
+ "step": 2280
3290
+ },
3291
+ {
3292
+ "epoch": 0.5502709211318483,
3293
+ "grad_norm": 0.7842517495155334,
3294
+ "learning_rate": 8.99325626204239e-06,
3295
+ "loss": 1.2479,
3296
+ "step": 2285
3297
+ },
3298
+ {
3299
+ "epoch": 0.551475015051174,
3300
+ "grad_norm": 0.7327888011932373,
3301
+ "learning_rate": 8.969171483622351e-06,
3302
+ "loss": 1.2657,
3303
+ "step": 2290
3304
+ },
3305
+ {
3306
+ "epoch": 0.5526791089704997,
3307
+ "grad_norm": 0.8408945202827454,
3308
+ "learning_rate": 8.945086705202313e-06,
3309
+ "loss": 1.1306,
3310
+ "step": 2295
3311
+ },
3312
+ {
3313
+ "epoch": 0.5538832028898254,
3314
+ "grad_norm": 0.7568773031234741,
3315
+ "learning_rate": 8.921001926782274e-06,
3316
+ "loss": 1.1459,
3317
+ "step": 2300
3318
+ },
3319
+ {
3320
+ "epoch": 0.5550872968091511,
3321
+ "grad_norm": 0.7822588682174683,
3322
+ "learning_rate": 8.896917148362236e-06,
3323
+ "loss": 1.1774,
3324
+ "step": 2305
3325
+ },
3326
+ {
3327
+ "epoch": 0.5562913907284768,
3328
+ "grad_norm": 0.7185366749763489,
3329
+ "learning_rate": 8.872832369942197e-06,
3330
+ "loss": 1.1045,
3331
+ "step": 2310
3332
+ },
3333
+ {
3334
+ "epoch": 0.5574954846478025,
3335
+ "grad_norm": 0.70743328332901,
3336
+ "learning_rate": 8.84874759152216e-06,
3337
+ "loss": 1.1726,
3338
+ "step": 2315
3339
+ },
3340
+ {
3341
+ "epoch": 0.5586995785671283,
3342
+ "grad_norm": 0.8196027874946594,
3343
+ "learning_rate": 8.82466281310212e-06,
3344
+ "loss": 1.1076,
3345
+ "step": 2320
3346
+ },
3347
+ {
3348
+ "epoch": 0.5599036724864539,
3349
+ "grad_norm": 0.7110490798950195,
3350
+ "learning_rate": 8.80057803468208e-06,
3351
+ "loss": 1.1325,
3352
+ "step": 2325
3353
+ },
3354
+ {
3355
+ "epoch": 0.5611077664057796,
3356
+ "grad_norm": 0.709387481212616,
3357
+ "learning_rate": 8.776493256262043e-06,
3358
+ "loss": 1.1767,
3359
+ "step": 2330
3360
+ },
3361
+ {
3362
+ "epoch": 0.5623118603251054,
3363
+ "grad_norm": 0.6487988233566284,
3364
+ "learning_rate": 8.752408477842005e-06,
3365
+ "loss": 1.2378,
3366
+ "step": 2335
3367
+ },
3368
+ {
3369
+ "epoch": 0.563515954244431,
3370
+ "grad_norm": 0.8318617939949036,
3371
+ "learning_rate": 8.728323699421966e-06,
3372
+ "loss": 1.214,
3373
+ "step": 2340
3374
+ },
3375
+ {
3376
+ "epoch": 0.5647200481637568,
3377
+ "grad_norm": 0.7174832224845886,
3378
+ "learning_rate": 8.704238921001927e-06,
3379
+ "loss": 1.217,
3380
+ "step": 2345
3381
+ },
3382
+ {
3383
+ "epoch": 0.5659241420830825,
3384
+ "grad_norm": 0.7566967010498047,
3385
+ "learning_rate": 8.680154142581889e-06,
3386
+ "loss": 1.1078,
3387
+ "step": 2350
3388
+ },
3389
+ {
3390
+ "epoch": 0.5671282360024081,
3391
+ "grad_norm": 0.8358272314071655,
3392
+ "learning_rate": 8.656069364161851e-06,
3393
+ "loss": 1.1509,
3394
+ "step": 2355
3395
+ },
3396
+ {
3397
+ "epoch": 0.5683323299217339,
3398
+ "grad_norm": 0.6700050830841064,
3399
+ "learning_rate": 8.631984585741812e-06,
3400
+ "loss": 1.152,
3401
+ "step": 2360
3402
+ },
3403
+ {
3404
+ "epoch": 0.5695364238410596,
3405
+ "grad_norm": 0.7518517374992371,
3406
+ "learning_rate": 8.607899807321773e-06,
3407
+ "loss": 1.1596,
3408
+ "step": 2365
3409
+ },
3410
+ {
3411
+ "epoch": 0.5707405177603853,
3412
+ "grad_norm": 0.6849754452705383,
3413
+ "learning_rate": 8.583815028901735e-06,
3414
+ "loss": 1.3295,
3415
+ "step": 2370
3416
+ },
3417
+ {
3418
+ "epoch": 0.571944611679711,
3419
+ "grad_norm": 0.7027848362922668,
3420
+ "learning_rate": 8.559730250481696e-06,
3421
+ "loss": 1.1989,
3422
+ "step": 2375
3423
+ },
3424
+ {
3425
+ "epoch": 0.5731487055990367,
3426
+ "grad_norm": 0.6843424439430237,
3427
+ "learning_rate": 8.535645472061658e-06,
3428
+ "loss": 1.1571,
3429
+ "step": 2380
3430
+ },
3431
+ {
3432
+ "epoch": 0.5743527995183624,
3433
+ "grad_norm": 0.7068962454795837,
3434
+ "learning_rate": 8.51156069364162e-06,
3435
+ "loss": 1.1649,
3436
+ "step": 2385
3437
+ },
3438
+ {
3439
+ "epoch": 0.5755568934376881,
3440
+ "grad_norm": 0.7391172647476196,
3441
+ "learning_rate": 8.487475915221581e-06,
3442
+ "loss": 1.0886,
3443
+ "step": 2390
3444
+ },
3445
+ {
3446
+ "epoch": 0.5767609873570139,
3447
+ "grad_norm": 0.7446891069412231,
3448
+ "learning_rate": 8.463391136801542e-06,
3449
+ "loss": 1.0976,
3450
+ "step": 2395
3451
+ },
3452
+ {
3453
+ "epoch": 0.5779650812763396,
3454
+ "grad_norm": 0.6888987421989441,
3455
+ "learning_rate": 8.439306358381504e-06,
3456
+ "loss": 1.134,
3457
+ "step": 2400
3458
+ },
3459
+ {
3460
+ "epoch": 0.5779650812763396,
3461
+ "eval_loss": 1.1661710739135742,
3462
+ "eval_runtime": 10.4237,
3463
+ "eval_samples_per_second": 9.593,
3464
+ "eval_steps_per_second": 1.247,
3465
+ "step": 2400
3466
+ },
3467
+ {
3468
+ "epoch": 0.5791691751956652,
3469
+ "grad_norm": 0.7011299729347229,
3470
+ "learning_rate": 8.415221579961465e-06,
3471
+ "loss": 1.1068,
3472
+ "step": 2405
3473
+ },
3474
+ {
3475
+ "epoch": 0.580373269114991,
3476
+ "grad_norm": 0.8446539640426636,
3477
+ "learning_rate": 8.391136801541427e-06,
3478
+ "loss": 1.2224,
3479
+ "step": 2410
3480
+ },
3481
+ {
3482
+ "epoch": 0.5815773630343167,
3483
+ "grad_norm": 0.6970465183258057,
3484
+ "learning_rate": 8.367052023121387e-06,
3485
+ "loss": 1.1774,
3486
+ "step": 2415
3487
+ },
3488
+ {
3489
+ "epoch": 0.5827814569536424,
3490
+ "grad_norm": 0.7700527310371399,
3491
+ "learning_rate": 8.34296724470135e-06,
3492
+ "loss": 1.2252,
3493
+ "step": 2420
3494
+ },
3495
+ {
3496
+ "epoch": 0.5839855508729681,
3497
+ "grad_norm": 0.6577072143554688,
3498
+ "learning_rate": 8.31888246628131e-06,
3499
+ "loss": 1.1651,
3500
+ "step": 2425
3501
+ },
3502
+ {
3503
+ "epoch": 0.5851896447922939,
3504
+ "grad_norm": 0.7169828414916992,
3505
+ "learning_rate": 8.294797687861273e-06,
3506
+ "loss": 1.2504,
3507
+ "step": 2430
3508
+ },
3509
+ {
3510
+ "epoch": 0.5863937387116195,
3511
+ "grad_norm": 0.7923945188522339,
3512
+ "learning_rate": 8.270712909441233e-06,
3513
+ "loss": 1.0871,
3514
+ "step": 2435
3515
+ },
3516
+ {
3517
+ "epoch": 0.5875978326309452,
3518
+ "grad_norm": 0.716785728931427,
3519
+ "learning_rate": 8.246628131021194e-06,
3520
+ "loss": 1.1508,
3521
+ "step": 2440
3522
+ },
3523
+ {
3524
+ "epoch": 0.588801926550271,
3525
+ "grad_norm": 0.7191885709762573,
3526
+ "learning_rate": 8.222543352601156e-06,
3527
+ "loss": 1.1277,
3528
+ "step": 2445
3529
+ },
3530
+ {
3531
+ "epoch": 0.5900060204695966,
3532
+ "grad_norm": 0.7167068123817444,
3533
+ "learning_rate": 8.198458574181119e-06,
3534
+ "loss": 1.2251,
3535
+ "step": 2450
3536
+ },
3537
+ {
3538
+ "epoch": 0.5912101143889223,
3539
+ "grad_norm": 0.6932991743087769,
3540
+ "learning_rate": 8.17437379576108e-06,
3541
+ "loss": 1.0736,
3542
+ "step": 2455
3543
+ },
3544
+ {
3545
+ "epoch": 0.5924142083082481,
3546
+ "grad_norm": 0.6908733248710632,
3547
+ "learning_rate": 8.150289017341042e-06,
3548
+ "loss": 1.0718,
3549
+ "step": 2460
3550
+ },
3551
+ {
3552
+ "epoch": 0.5936183022275737,
3553
+ "grad_norm": 0.7101254463195801,
3554
+ "learning_rate": 8.126204238921002e-06,
3555
+ "loss": 1.1986,
3556
+ "step": 2465
3557
+ },
3558
+ {
3559
+ "epoch": 0.5948223961468995,
3560
+ "grad_norm": 0.7084072232246399,
3561
+ "learning_rate": 8.102119460500965e-06,
3562
+ "loss": 1.2554,
3563
+ "step": 2470
3564
+ },
3565
+ {
3566
+ "epoch": 0.5960264900662252,
3567
+ "grad_norm": 0.7392582297325134,
3568
+ "learning_rate": 8.078034682080925e-06,
3569
+ "loss": 1.1967,
3570
+ "step": 2475
3571
+ },
3572
+ {
3573
+ "epoch": 0.5972305839855508,
3574
+ "grad_norm": 0.7432191371917725,
3575
+ "learning_rate": 8.053949903660888e-06,
3576
+ "loss": 1.1868,
3577
+ "step": 2480
3578
+ },
3579
+ {
3580
+ "epoch": 0.5984346779048766,
3581
+ "grad_norm": 0.7011279463768005,
3582
+ "learning_rate": 8.029865125240848e-06,
3583
+ "loss": 1.0677,
3584
+ "step": 2485
3585
+ },
3586
+ {
3587
+ "epoch": 0.5996387718242023,
3588
+ "grad_norm": 0.7159491777420044,
3589
+ "learning_rate": 8.005780346820809e-06,
3590
+ "loss": 1.1407,
3591
+ "step": 2490
3592
+ },
3593
+ {
3594
+ "epoch": 0.600842865743528,
3595
+ "grad_norm": 0.7194905877113342,
3596
+ "learning_rate": 7.981695568400771e-06,
3597
+ "loss": 1.1851,
3598
+ "step": 2495
3599
+ },
3600
+ {
3601
+ "epoch": 0.6020469596628537,
3602
+ "grad_norm": 0.7550024390220642,
3603
+ "learning_rate": 7.957610789980734e-06,
3604
+ "loss": 1.1766,
3605
+ "step": 2500
3606
+ }
3607
+ ],
3608
+ "logging_steps": 5,
3609
+ "max_steps": 4152,
3610
+ "num_input_tokens_seen": 0,
3611
+ "num_train_epochs": 1,
3612
+ "save_steps": 500,
3613
+ "stateful_callbacks": {
3614
+ "TrainerControl": {
3615
+ "args": {
3616
+ "should_epoch_stop": false,
3617
+ "should_evaluate": false,
3618
+ "should_log": false,
3619
+ "should_save": true,
3620
+ "should_training_stop": false
3621
+ },
3622
+ "attributes": {}
3623
+ }
3624
+ },
3625
+ "total_flos": 1.2188715544418688e+16,
3626
+ "train_batch_size": 1,
3627
+ "trial_name": null,
3628
+ "trial_params": null
3629
+ }