Hugging Face's logo

Trelis
/
99-base

TensorBoard
Safetensors
llama
Model card Files
xet
 Metrics Community
RonanMcGovern commited on
Commit
7c65e10
·
verified ·
1 Parent(s): 1d96026

Upload folder using huggingface_hub

Browse files
Files changed (22) hide show
  1. config.json +30 -0
  2. generation_config.json +8 -0
  3. merges.txt +0 -0
  4. model.safetensors +3 -0
  5. optimizer.pt +3 -0
  6. rng_state_0.pth +3 -0
  7. rng_state_1.pth +3 -0
  8. rng_state_2.pth +3 -0
  9. rng_state_3.pth +3 -0
  10. rng_state_4.pth +3 -0
  11. rng_state_5.pth +3 -0
  12. rng_state_6.pth +3 -0
  13. rng_state_7.pth +3 -0
  14. rng_state_8.pth +3 -0
  15. rng_state_9.pth +3 -0
  16. scheduler.pt +3 -0
  17. special_tokens_map.json +34 -0
  18. tokenizer.json +0 -0
  19. tokenizer_config.json +155 -0
  20. trainer_state.json +3650 -0
  21. training_args.bin +3 -0
  22. vocab.json +0 -0
config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "Trelis/SmolLM-135M-Instruct-layer-width-pruned-90M-raw",
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
+ "hidden_act": "silu",
11
+ "hidden_size": 504,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 1344,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 9,
18
+ "num_hidden_layers": 25,
19
+ "num_key_value_heads": 3,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.2",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
generation_config.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "max_new_tokens": 40,
6
+ "pad_token_id": 2,
7
+ "transformers_version": "4.44.2"
8
+ }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d2a42cc56e07c013f73e56708f9890aeba1ef6f2507446949476e63f34240fa
3
+ size 185097216
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9dd0b2a653496b95d07793491c50539b083d1eff242426d4d9cf66c7f12edaaa
3
+ size 370329978
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72aed7659a4568e39ea9a56bdc92196603df7d730a90c6411d24926b5d12ad03
3
+ size 16433
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8ccb7aad33c882b66b79a28ece740dc71664d087d1f12ad61b65b18df1beca55
3
+ size 16433
rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf3954ab65b1da5768888e4c50598bd4c761244f7027ce8da56d21582e829948
3
+ size 16433
rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9373b1a76204778613e7dd6b7f31b34c6bb969dbb0b802f9f62ca911909492f2
3
+ size 16433
rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d1fc04b2ab887669d463a51ba4296461e18e6793bbf2221bc603bff1af03a7a6
3
+ size 16433
rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:81879e4ef695491cab2ffd4ab75ee6208dd18e8e7339d7b4aff8a08580e999d0
3
+ size 16433
rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0af269f614eb79e3ee82816f8d169e28a4fa1006f684260473e683e7ccea9d58
3
+ size 16433
rng_state_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e9f29cd3c33bc4028b8be6ed5e960831203bf34fb31356c09cdb29b4c851d3f
3
+ size 16433
rng_state_8.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62adbe536ff6dcfa7fb600a17a73fb0c108c33b017287d093401133eb29bb6d4
3
+ size 16433
rng_state_9.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b7dd71e7a0ba4defb86d5997709ee80f19a9326e37319a19a3ae8c437ad97fa
3
+ size 16433
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4729d5e48ae081b5107dc5941bf9dd080c7d24c9e46db7051ab1bfcf68e98eee
3
+ size 1064
special_tokens_map.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false
26
+ },
27
+ "unk_token": {
28
+ "content": "<|endoftext|>",
29
+ "lstrip": false,
30
+ "normalized": false,
31
+ "rstrip": false,
32
+ "single_word": false
33
+ }
34
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,155 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_eos_token": true,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "0": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "1": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "2": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "3": {
30
+ "content": "<repo_name>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "4": {
38
+ "content": "<reponame>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "5": {
46
+ "content": "<file_sep>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "6": {
54
+ "content": "<filename>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "7": {
62
+ "content": "<gh_stars>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "8": {
70
+ "content": "<issue_start>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "9": {
78
+ "content": "<issue_comment>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "10": {
86
+ "content": "<issue_closed>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "11": {
94
+ "content": "<jupyter_start>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "12": {
102
+ "content": "<jupyter_text>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "13": {
110
+ "content": "<jupyter_code>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "14": {
118
+ "content": "<jupyter_output>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": true
124
+ },
125
+ "15": {
126
+ "content": "<jupyter_script>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": true
132
+ },
133
+ "16": {
134
+ "content": "<empty_output>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": true
140
+ }
141
+ },
142
+ "additional_special_tokens": [
143
+ "<|im_start|>",
144
+ "<|im_end|>"
145
+ ],
146
+ "bos_token": "<|im_start|>",
147
+ "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
148
+ "clean_up_tokenization_spaces": false,
149
+ "eos_token": "<|im_end|>",
150
+ "model_max_length": 2048,
151
+ "pad_token": "<|im_end|>",
152
+ "tokenizer_class": "GPT2Tokenizer",
153
+ "unk_token": "<|endoftext|>",
154
+ "vocab_size": 49152
155
+ }
trainer_state.json ADDED
@@ -0,0 +1,3650 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.49979917224036463,
5
+ "eval_steps": 1431,
6
+ "global_step": 7155,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0009779438730070028,
13
+ "grad_norm": 0.52734375,
14
+ "learning_rate": 0.001,
15
+ "loss": 5.7482,
16
+ "step": 14
17
+ },
18
+ {
19
+ "epoch": 0.0019558877460140057,
20
+ "grad_norm": 0.1259765625,
21
+ "learning_rate": 0.001,
22
+ "loss": 3.6003,
23
+ "step": 28
24
+ },
25
+ {
26
+ "epoch": 0.0029338316190210085,
27
+ "grad_norm": 0.12109375,
28
+ "learning_rate": 0.001,
29
+ "loss": 3.4079,
30
+ "step": 42
31
+ },
32
+ {
33
+ "epoch": 0.003911775492028011,
34
+ "grad_norm": 0.22265625,
35
+ "learning_rate": 0.001,
36
+ "loss": 3.201,
37
+ "step": 56
38
+ },
39
+ {
40
+ "epoch": 0.004889719365035014,
41
+ "grad_norm": 0.21484375,
42
+ "learning_rate": 0.001,
43
+ "loss": 2.955,
44
+ "step": 70
45
+ },
46
+ {
47
+ "epoch": 0.005867663238042017,
48
+ "grad_norm": 0.2119140625,
49
+ "learning_rate": 0.001,
50
+ "loss": 2.8744,
51
+ "step": 84
52
+ },
53
+ {
54
+ "epoch": 0.006845607111049019,
55
+ "grad_norm": 0.15625,
56
+ "learning_rate": 0.001,
57
+ "loss": 2.8267,
58
+ "step": 98
59
+ },
60
+ {
61
+ "epoch": 0.007823550984056023,
62
+ "grad_norm": 0.1884765625,
63
+ "learning_rate": 0.001,
64
+ "loss": 2.7249,
65
+ "step": 112
66
+ },
67
+ {
68
+ "epoch": 0.008801494857063025,
69
+ "grad_norm": 2.625,
70
+ "learning_rate": 0.001,
71
+ "loss": 2.6441,
72
+ "step": 126
73
+ },
74
+ {
75
+ "epoch": 0.009779438730070028,
76
+ "grad_norm": 0.173828125,
77
+ "learning_rate": 0.001,
78
+ "loss": 2.6126,
79
+ "step": 140
80
+ },
81
+ {
82
+ "epoch": 0.01075738260307703,
83
+ "grad_norm": 0.10693359375,
84
+ "learning_rate": 0.001,
85
+ "loss": 2.5241,
86
+ "step": 154
87
+ },
88
+ {
89
+ "epoch": 0.011735326476084034,
90
+ "grad_norm": 0.158203125,
91
+ "learning_rate": 0.001,
92
+ "loss": 2.4803,
93
+ "step": 168
94
+ },
95
+ {
96
+ "epoch": 0.012713270349091036,
97
+ "grad_norm": 0.119140625,
98
+ "learning_rate": 0.001,
99
+ "loss": 2.3886,
100
+ "step": 182
101
+ },
102
+ {
103
+ "epoch": 0.013691214222098038,
104
+ "grad_norm": 0.1923828125,
105
+ "learning_rate": 0.001,
106
+ "loss": 2.3931,
107
+ "step": 196
108
+ },
109
+ {
110
+ "epoch": 0.014669158095105042,
111
+ "grad_norm": 0.283203125,
112
+ "learning_rate": 0.001,
113
+ "loss": 2.3493,
114
+ "step": 210
115
+ },
116
+ {
117
+ "epoch": 0.015647101968112045,
118
+ "grad_norm": 0.234375,
119
+ "learning_rate": 0.001,
120
+ "loss": 2.3317,
121
+ "step": 224
122
+ },
123
+ {
124
+ "epoch": 0.016625045841119047,
125
+ "grad_norm": 0.2138671875,
126
+ "learning_rate": 0.001,
127
+ "loss": 2.3284,
128
+ "step": 238
129
+ },
130
+ {
131
+ "epoch": 0.01760298971412605,
132
+ "grad_norm": 0.16796875,
133
+ "learning_rate": 0.001,
134
+ "loss": 2.2322,
135
+ "step": 252
136
+ },
137
+ {
138
+ "epoch": 0.01858093358713305,
139
+ "grad_norm": 0.1650390625,
140
+ "learning_rate": 0.001,
141
+ "loss": 2.2145,
142
+ "step": 266
143
+ },
144
+ {
145
+ "epoch": 0.019558877460140057,
146
+ "grad_norm": 0.2890625,
147
+ "learning_rate": 0.001,
148
+ "loss": 2.202,
149
+ "step": 280
150
+ },
151
+ {
152
+ "epoch": 0.02053682133314706,
153
+ "grad_norm": 0.201171875,
154
+ "learning_rate": 0.001,
155
+ "loss": 2.2156,
156
+ "step": 294
157
+ },
158
+ {
159
+ "epoch": 0.02151476520615406,
160
+ "grad_norm": 0.255859375,
161
+ "learning_rate": 0.001,
162
+ "loss": 2.1436,
163
+ "step": 308
164
+ },
165
+ {
166
+ "epoch": 0.022492709079161063,
167
+ "grad_norm": 0.390625,
168
+ "learning_rate": 0.001,
169
+ "loss": 2.1198,
170
+ "step": 322
171
+ },
172
+ {
173
+ "epoch": 0.023470652952168068,
174
+ "grad_norm": 0.234375,
175
+ "learning_rate": 0.001,
176
+ "loss": 2.1031,
177
+ "step": 336
178
+ },
179
+ {
180
+ "epoch": 0.02444859682517507,
181
+ "grad_norm": 0.1298828125,
182
+ "learning_rate": 0.001,
183
+ "loss": 2.0943,
184
+ "step": 350
185
+ },
186
+ {
187
+ "epoch": 0.025426540698182072,
188
+ "grad_norm": 0.10693359375,
189
+ "learning_rate": 0.001,
190
+ "loss": 2.0352,
191
+ "step": 364
192
+ },
193
+ {
194
+ "epoch": 0.026404484571189074,
195
+ "grad_norm": 0.158203125,
196
+ "learning_rate": 0.001,
197
+ "loss": 2.0439,
198
+ "step": 378
199
+ },
200
+ {
201
+ "epoch": 0.027382428444196076,
202
+ "grad_norm": 0.16796875,
203
+ "learning_rate": 0.001,
204
+ "loss": 2.0176,
205
+ "step": 392
206
+ },
207
+ {
208
+ "epoch": 0.02836037231720308,
209
+ "grad_norm": 0.474609375,
210
+ "learning_rate": 0.001,
211
+ "loss": 2.0212,
212
+ "step": 406
213
+ },
214
+ {
215
+ "epoch": 0.029338316190210083,
216
+ "grad_norm": 0.240234375,
217
+ "learning_rate": 0.001,
218
+ "loss": 1.9963,
219
+ "step": 420
220
+ },
221
+ {
222
+ "epoch": 0.030316260063217085,
223
+ "grad_norm": 0.1689453125,
224
+ "learning_rate": 0.001,
225
+ "loss": 1.9738,
226
+ "step": 434
227
+ },
228
+ {
229
+ "epoch": 0.03129420393622409,
230
+ "grad_norm": 0.255859375,
231
+ "learning_rate": 0.001,
232
+ "loss": 1.9362,
233
+ "step": 448
234
+ },
235
+ {
236
+ "epoch": 0.03227214780923109,
237
+ "grad_norm": 0.1982421875,
238
+ "learning_rate": 0.001,
239
+ "loss": 1.9393,
240
+ "step": 462
241
+ },
242
+ {
243
+ "epoch": 0.033250091682238095,
244
+ "grad_norm": 0.400390625,
245
+ "learning_rate": 0.001,
246
+ "loss": 1.9876,
247
+ "step": 476
248
+ },
249
+ {
250
+ "epoch": 0.0342280355552451,
251
+ "grad_norm": 0.23046875,
252
+ "learning_rate": 0.001,
253
+ "loss": 1.9551,
254
+ "step": 490
255
+ },
256
+ {
257
+ "epoch": 0.0352059794282521,
258
+ "grad_norm": 0.1591796875,
259
+ "learning_rate": 0.001,
260
+ "loss": 1.9103,
261
+ "step": 504
262
+ },
263
+ {
264
+ "epoch": 0.0361839233012591,
265
+ "grad_norm": 0.1767578125,
266
+ "learning_rate": 0.001,
267
+ "loss": 1.9243,
268
+ "step": 518
269
+ },
270
+ {
271
+ "epoch": 0.0371618671742661,
272
+ "grad_norm": 0.478515625,
273
+ "learning_rate": 0.001,
274
+ "loss": 1.9207,
275
+ "step": 532
276
+ },
277
+ {
278
+ "epoch": 0.03813981104727311,
279
+ "grad_norm": 0.25390625,
280
+ "learning_rate": 0.001,
281
+ "loss": 1.9029,
282
+ "step": 546
283
+ },
284
+ {
285
+ "epoch": 0.039117754920280114,
286
+ "grad_norm": 0.2294921875,
287
+ "learning_rate": 0.001,
288
+ "loss": 1.8988,
289
+ "step": 560
290
+ },
291
+ {
292
+ "epoch": 0.040095698793287116,
293
+ "grad_norm": 0.189453125,
294
+ "learning_rate": 0.001,
295
+ "loss": 1.8653,
296
+ "step": 574
297
+ },
298
+ {
299
+ "epoch": 0.04107364266629412,
300
+ "grad_norm": 0.2158203125,
301
+ "learning_rate": 0.001,
302
+ "loss": 1.8383,
303
+ "step": 588
304
+ },
305
+ {
306
+ "epoch": 0.04205158653930112,
307
+ "grad_norm": 0.236328125,
308
+ "learning_rate": 0.001,
309
+ "loss": 1.8191,
310
+ "step": 602
311
+ },
312
+ {
313
+ "epoch": 0.04302953041230812,
314
+ "grad_norm": 0.55859375,
315
+ "learning_rate": 0.001,
316
+ "loss": 1.8837,
317
+ "step": 616
318
+ },
319
+ {
320
+ "epoch": 0.044007474285315124,
321
+ "grad_norm": 0.265625,
322
+ "learning_rate": 0.001,
323
+ "loss": 1.8838,
324
+ "step": 630
325
+ },
326
+ {
327
+ "epoch": 0.044985418158322125,
328
+ "grad_norm": 0.1513671875,
329
+ "learning_rate": 0.001,
330
+ "loss": 1.8423,
331
+ "step": 644
332
+ },
333
+ {
334
+ "epoch": 0.04596336203132913,
335
+ "grad_norm": 0.134765625,
336
+ "learning_rate": 0.001,
337
+ "loss": 1.785,
338
+ "step": 658
339
+ },
340
+ {
341
+ "epoch": 0.046941305904336136,
342
+ "grad_norm": 0.2265625,
343
+ "learning_rate": 0.001,
344
+ "loss": 1.8072,
345
+ "step": 672
346
+ },
347
+ {
348
+ "epoch": 0.04791924977734314,
349
+ "grad_norm": 0.1669921875,
350
+ "learning_rate": 0.001,
351
+ "loss": 1.7771,
352
+ "step": 686
353
+ },
354
+ {
355
+ "epoch": 0.04889719365035014,
356
+ "grad_norm": 0.33203125,
357
+ "learning_rate": 0.001,
358
+ "loss": 1.8465,
359
+ "step": 700
360
+ },
361
+ {
362
+ "epoch": 0.04987513752335714,
363
+ "grad_norm": 0.255859375,
364
+ "learning_rate": 0.001,
365
+ "loss": 1.7964,
366
+ "step": 714
367
+ },
368
+ {
369
+ "epoch": 0.050853081396364144,
370
+ "grad_norm": 0.380859375,
371
+ "learning_rate": 0.001,
372
+ "loss": 1.81,
373
+ "step": 728
374
+ },
375
+ {
376
+ "epoch": 0.051831025269371146,
377
+ "grad_norm": 0.33984375,
378
+ "learning_rate": 0.001,
379
+ "loss": 1.7867,
380
+ "step": 742
381
+ },
382
+ {
383
+ "epoch": 0.05280896914237815,
384
+ "grad_norm": 0.19140625,
385
+ "learning_rate": 0.001,
386
+ "loss": 1.7581,
387
+ "step": 756
388
+ },
389
+ {
390
+ "epoch": 0.05378691301538515,
391
+ "grad_norm": 0.2158203125,
392
+ "learning_rate": 0.001,
393
+ "loss": 1.7116,
394
+ "step": 770
395
+ },
396
+ {
397
+ "epoch": 0.05476485688839215,
398
+ "grad_norm": 0.267578125,
399
+ "learning_rate": 0.001,
400
+ "loss": 1.7431,
401
+ "step": 784
402
+ },
403
+ {
404
+ "epoch": 0.05574280076139916,
405
+ "grad_norm": 0.2255859375,
406
+ "learning_rate": 0.001,
407
+ "loss": 1.7647,
408
+ "step": 798
409
+ },
410
+ {
411
+ "epoch": 0.05672074463440616,
412
+ "grad_norm": 0.177734375,
413
+ "learning_rate": 0.001,
414
+ "loss": 1.709,
415
+ "step": 812
416
+ },
417
+ {
418
+ "epoch": 0.057698688507413165,
419
+ "grad_norm": 0.384765625,
420
+ "learning_rate": 0.001,
421
+ "loss": 1.7235,
422
+ "step": 826
423
+ },
424
+ {
425
+ "epoch": 0.05867663238042017,
426
+ "grad_norm": 0.26171875,
427
+ "learning_rate": 0.001,
428
+ "loss": 1.7161,
429
+ "step": 840
430
+ },
431
+ {
432
+ "epoch": 0.05965457625342717,
433
+ "grad_norm": 0.455078125,
434
+ "learning_rate": 0.001,
435
+ "loss": 1.812,
436
+ "step": 854
437
+ },
438
+ {
439
+ "epoch": 0.06063252012643417,
440
+ "grad_norm": 0.2099609375,
441
+ "learning_rate": 0.001,
442
+ "loss": 1.7164,
443
+ "step": 868
444
+ },
445
+ {
446
+ "epoch": 0.06161046399944117,
447
+ "grad_norm": 0.154296875,
448
+ "learning_rate": 0.001,
449
+ "loss": 1.7677,
450
+ "step": 882
451
+ },
452
+ {
453
+ "epoch": 0.06258840787244818,
454
+ "grad_norm": 0.1796875,
455
+ "learning_rate": 0.001,
456
+ "loss": 1.7001,
457
+ "step": 896
458
+ },
459
+ {
460
+ "epoch": 0.06356635174545518,
461
+ "grad_norm": 0.173828125,
462
+ "learning_rate": 0.001,
463
+ "loss": 1.7015,
464
+ "step": 910
465
+ },
466
+ {
467
+ "epoch": 0.06454429561846219,
468
+ "grad_norm": 0.203125,
469
+ "learning_rate": 0.001,
470
+ "loss": 1.6904,
471
+ "step": 924
472
+ },
473
+ {
474
+ "epoch": 0.06552223949146918,
475
+ "grad_norm": 0.2236328125,
476
+ "learning_rate": 0.001,
477
+ "loss": 1.6692,
478
+ "step": 938
479
+ },
480
+ {
481
+ "epoch": 0.06650018336447619,
482
+ "grad_norm": 0.265625,
483
+ "learning_rate": 0.001,
484
+ "loss": 1.6802,
485
+ "step": 952
486
+ },
487
+ {
488
+ "epoch": 0.06747812723748318,
489
+ "grad_norm": 0.228515625,
490
+ "learning_rate": 0.001,
491
+ "loss": 1.6566,
492
+ "step": 966
493
+ },
494
+ {
495
+ "epoch": 0.0684560711104902,
496
+ "grad_norm": 0.275390625,
497
+ "learning_rate": 0.001,
498
+ "loss": 1.7042,
499
+ "step": 980
500
+ },
501
+ {
502
+ "epoch": 0.0694340149834972,
503
+ "grad_norm": 0.2158203125,
504
+ "learning_rate": 0.001,
505
+ "loss": 1.6653,
506
+ "step": 994
507
+ },
508
+ {
509
+ "epoch": 0.0704119588565042,
510
+ "grad_norm": 0.357421875,
511
+ "learning_rate": 0.001,
512
+ "loss": 1.6729,
513
+ "step": 1008
514
+ },
515
+ {
516
+ "epoch": 0.0713899027295112,
517
+ "grad_norm": 0.28125,
518
+ "learning_rate": 0.001,
519
+ "loss": 1.6864,
520
+ "step": 1022
521
+ },
522
+ {
523
+ "epoch": 0.0723678466025182,
524
+ "grad_norm": 0.1513671875,
525
+ "learning_rate": 0.001,
526
+ "loss": 1.6541,
527
+ "step": 1036
528
+ },
529
+ {
530
+ "epoch": 0.07334579047552521,
531
+ "grad_norm": 1.0390625,
532
+ "learning_rate": 0.001,
533
+ "loss": 1.7122,
534
+ "step": 1050
535
+ },
536
+ {
537
+ "epoch": 0.0743237343485322,
538
+ "grad_norm": 0.267578125,
539
+ "learning_rate": 0.001,
540
+ "loss": 1.7231,
541
+ "step": 1064
542
+ },
543
+ {
544
+ "epoch": 0.07530167822153921,
545
+ "grad_norm": 0.251953125,
546
+ "learning_rate": 0.001,
547
+ "loss": 1.6781,
548
+ "step": 1078
549
+ },
550
+ {
551
+ "epoch": 0.07627962209454622,
552
+ "grad_norm": 0.23828125,
553
+ "learning_rate": 0.001,
554
+ "loss": 1.667,
555
+ "step": 1092
556
+ },
557
+ {
558
+ "epoch": 0.07725756596755322,
559
+ "grad_norm": 0.2333984375,
560
+ "learning_rate": 0.001,
561
+ "loss": 1.6667,
562
+ "step": 1106
563
+ },
564
+ {
565
+ "epoch": 0.07823550984056023,
566
+ "grad_norm": 0.2119140625,
567
+ "learning_rate": 0.001,
568
+ "loss": 1.622,
569
+ "step": 1120
570
+ },
571
+ {
572
+ "epoch": 0.07921345371356722,
573
+ "grad_norm": 0.1533203125,
574
+ "learning_rate": 0.001,
575
+ "loss": 1.636,
576
+ "step": 1134
577
+ },
578
+ {
579
+ "epoch": 0.08019139758657423,
580
+ "grad_norm": 0.1943359375,
581
+ "learning_rate": 0.001,
582
+ "loss": 1.6423,
583
+ "step": 1148
584
+ },
585
+ {
586
+ "epoch": 0.08116934145958123,
587
+ "grad_norm": 0.1806640625,
588
+ "learning_rate": 0.001,
589
+ "loss": 1.6136,
590
+ "step": 1162
591
+ },
592
+ {
593
+ "epoch": 0.08214728533258824,
594
+ "grad_norm": 0.224609375,
595
+ "learning_rate": 0.001,
596
+ "loss": 1.5988,
597
+ "step": 1176
598
+ },
599
+ {
600
+ "epoch": 0.08312522920559523,
601
+ "grad_norm": 0.2890625,
602
+ "learning_rate": 0.001,
603
+ "loss": 1.6216,
604
+ "step": 1190
605
+ },
606
+ {
607
+ "epoch": 0.08410317307860224,
608
+ "grad_norm": 0.2060546875,
609
+ "learning_rate": 0.001,
610
+ "loss": 1.6296,
611
+ "step": 1204
612
+ },
613
+ {
614
+ "epoch": 0.08508111695160925,
615
+ "grad_norm": 0.2451171875,
616
+ "learning_rate": 0.001,
617
+ "loss": 1.5944,
618
+ "step": 1218
619
+ },
620
+ {
621
+ "epoch": 0.08605906082461624,
622
+ "grad_norm": 0.2392578125,
623
+ "learning_rate": 0.001,
624
+ "loss": 1.6144,
625
+ "step": 1232
626
+ },
627
+ {
628
+ "epoch": 0.08703700469762325,
629
+ "grad_norm": 0.357421875,
630
+ "learning_rate": 0.001,
631
+ "loss": 1.585,
632
+ "step": 1246
633
+ },
634
+ {
635
+ "epoch": 0.08801494857063025,
636
+ "grad_norm": 0.2470703125,
637
+ "learning_rate": 0.001,
638
+ "loss": 1.6033,
639
+ "step": 1260
640
+ },
641
+ {
642
+ "epoch": 0.08899289244363726,
643
+ "grad_norm": 0.201171875,
644
+ "learning_rate": 0.001,
645
+ "loss": 1.6122,
646
+ "step": 1274
647
+ },
648
+ {
649
+ "epoch": 0.08997083631664425,
650
+ "grad_norm": 0.396484375,
651
+ "learning_rate": 0.001,
652
+ "loss": 1.5978,
653
+ "step": 1288
654
+ },
655
+ {
656
+ "epoch": 0.09094878018965126,
657
+ "grad_norm": 0.4453125,
658
+ "learning_rate": 0.001,
659
+ "loss": 1.6392,
660
+ "step": 1302
661
+ },
662
+ {
663
+ "epoch": 0.09192672406265825,
664
+ "grad_norm": 0.2470703125,
665
+ "learning_rate": 0.001,
666
+ "loss": 1.6247,
667
+ "step": 1316
668
+ },
669
+ {
670
+ "epoch": 0.09290466793566526,
671
+ "grad_norm": 0.193359375,
672
+ "learning_rate": 0.001,
673
+ "loss": 1.5888,
674
+ "step": 1330
675
+ },
676
+ {
677
+ "epoch": 0.09388261180867227,
678
+ "grad_norm": 0.2392578125,
679
+ "learning_rate": 0.001,
680
+ "loss": 1.572,
681
+ "step": 1344
682
+ },
683
+ {
684
+ "epoch": 0.09486055568167927,
685
+ "grad_norm": 0.17578125,
686
+ "learning_rate": 0.001,
687
+ "loss": 1.5725,
688
+ "step": 1358
689
+ },
690
+ {
691
+ "epoch": 0.09583849955468628,
692
+ "grad_norm": 0.314453125,
693
+ "learning_rate": 0.001,
694
+ "loss": 1.6131,
695
+ "step": 1372
696
+ },
697
+ {
698
+ "epoch": 0.09681644342769327,
699
+ "grad_norm": 0.22265625,
700
+ "learning_rate": 0.001,
701
+ "loss": 1.5467,
702
+ "step": 1386
703
+ },
704
+ {
705
+ "epoch": 0.09779438730070028,
706
+ "grad_norm": 0.296875,
707
+ "learning_rate": 0.001,
708
+ "loss": 1.5843,
709
+ "step": 1400
710
+ },
711
+ {
712
+ "epoch": 0.09877233117370728,
713
+ "grad_norm": 0.267578125,
714
+ "learning_rate": 0.001,
715
+ "loss": 1.5519,
716
+ "step": 1414
717
+ },
718
+ {
719
+ "epoch": 0.09975027504671428,
720
+ "grad_norm": 0.251953125,
721
+ "learning_rate": 0.001,
722
+ "loss": 1.5531,
723
+ "step": 1428
724
+ },
725
+ {
726
+ "epoch": 0.09995983444807292,
727
+ "eval_loss": 2.1689391136169434,
728
+ "eval_runtime": 9.1273,
729
+ "eval_samples_per_second": 109.561,
730
+ "eval_steps_per_second": 1.424,
731
+ "step": 1431
732
+ },
733
+ {
734
+ "epoch": 0.10072821891972128,
735
+ "grad_norm": 0.2333984375,
736
+ "learning_rate": 0.001,
737
+ "loss": 1.561,
738
+ "step": 1442
739
+ },
740
+ {
741
+ "epoch": 0.10170616279272829,
742
+ "grad_norm": 0.302734375,
743
+ "learning_rate": 0.001,
744
+ "loss": 1.5818,
745
+ "step": 1456
746
+ },
747
+ {
748
+ "epoch": 0.1026841066657353,
749
+ "grad_norm": 0.294921875,
750
+ "learning_rate": 0.001,
751
+ "loss": 1.5653,
752
+ "step": 1470
753
+ },
754
+ {
755
+ "epoch": 0.10366205053874229,
756
+ "grad_norm": 0.296875,
757
+ "learning_rate": 0.001,
758
+ "loss": 1.585,
759
+ "step": 1484
760
+ },
761
+ {
762
+ "epoch": 0.1046399944117493,
763
+ "grad_norm": 0.28515625,
764
+ "learning_rate": 0.001,
765
+ "loss": 1.5502,
766
+ "step": 1498
767
+ },
768
+ {
769
+ "epoch": 0.1056179382847563,
770
+ "grad_norm": 0.5703125,
771
+ "learning_rate": 0.001,
772
+ "loss": 1.5873,
773
+ "step": 1512
774
+ },
775
+ {
776
+ "epoch": 0.1065958821577633,
777
+ "grad_norm": 0.298828125,
778
+ "learning_rate": 0.001,
779
+ "loss": 1.5825,
780
+ "step": 1526
781
+ },
782
+ {
783
+ "epoch": 0.1075738260307703,
784
+ "grad_norm": 0.2578125,
785
+ "learning_rate": 0.001,
786
+ "loss": 1.5712,
787
+ "step": 1540
788
+ },
789
+ {
790
+ "epoch": 0.10855176990377731,
791
+ "grad_norm": 0.2392578125,
792
+ "learning_rate": 0.001,
793
+ "loss": 1.5443,
794
+ "step": 1554
795
+ },
796
+ {
797
+ "epoch": 0.1095297137767843,
798
+ "grad_norm": 0.95703125,
799
+ "learning_rate": 0.001,
800
+ "loss": 1.5989,
801
+ "step": 1568
802
+ },
803
+ {
804
+ "epoch": 0.11050765764979131,
805
+ "grad_norm": 0.318359375,
806
+ "learning_rate": 0.001,
807
+ "loss": 1.6003,
808
+ "step": 1582
809
+ },
810
+ {
811
+ "epoch": 0.11148560152279832,
812
+ "grad_norm": 0.2578125,
813
+ "learning_rate": 0.001,
814
+ "loss": 1.5907,
815
+ "step": 1596
816
+ },
817
+ {
818
+ "epoch": 0.11246354539580532,
819
+ "grad_norm": 0.330078125,
820
+ "learning_rate": 0.001,
821
+ "loss": 1.5373,
822
+ "step": 1610
823
+ },
824
+ {
825
+ "epoch": 0.11344148926881233,
826
+ "grad_norm": 0.2392578125,
827
+ "learning_rate": 0.001,
828
+ "loss": 1.5531,
829
+ "step": 1624
830
+ },
831
+ {
832
+ "epoch": 0.11441943314181932,
833
+ "grad_norm": 0.232421875,
834
+ "learning_rate": 0.001,
835
+ "loss": 1.5317,
836
+ "step": 1638
837
+ },
838
+ {
839
+ "epoch": 0.11539737701482633,
840
+ "grad_norm": 0.3671875,
841
+ "learning_rate": 0.001,
842
+ "loss": 1.5157,
843
+ "step": 1652
844
+ },
845
+ {
846
+ "epoch": 0.11637532088783333,
847
+ "grad_norm": 0.2001953125,
848
+ "learning_rate": 0.001,
849
+ "loss": 1.5462,
850
+ "step": 1666
851
+ },
852
+ {
853
+ "epoch": 0.11735326476084033,
854
+ "grad_norm": 0.2236328125,
855
+ "learning_rate": 0.001,
856
+ "loss": 1.5598,
857
+ "step": 1680
858
+ },
859
+ {
860
+ "epoch": 0.11833120863384734,
861
+ "grad_norm": 0.251953125,
862
+ "learning_rate": 0.001,
863
+ "loss": 1.5345,
864
+ "step": 1694
865
+ },
866
+ {
867
+ "epoch": 0.11930915250685434,
868
+ "grad_norm": 0.470703125,
869
+ "learning_rate": 0.001,
870
+ "loss": 1.5324,
871
+ "step": 1708
872
+ },
873
+ {
874
+ "epoch": 0.12028709637986135,
875
+ "grad_norm": 0.2158203125,
876
+ "learning_rate": 0.001,
877
+ "loss": 1.498,
878
+ "step": 1722
879
+ },
880
+ {
881
+ "epoch": 0.12126504025286834,
882
+ "grad_norm": 0.251953125,
883
+ "learning_rate": 0.001,
884
+ "loss": 1.5072,
885
+ "step": 1736
886
+ },
887
+ {
888
+ "epoch": 0.12224298412587535,
889
+ "grad_norm": 0.265625,
890
+ "learning_rate": 0.001,
891
+ "loss": 1.5364,
892
+ "step": 1750
893
+ },
894
+ {
895
+ "epoch": 0.12322092799888235,
896
+ "grad_norm": 0.30078125,
897
+ "learning_rate": 0.001,
898
+ "loss": 1.5,
899
+ "step": 1764
900
+ },
901
+ {
902
+ "epoch": 0.12419887187188935,
903
+ "grad_norm": 0.31640625,
904
+ "learning_rate": 0.001,
905
+ "loss": 1.5113,
906
+ "step": 1778
907
+ },
908
+ {
909
+ "epoch": 0.12517681574489636,
910
+ "grad_norm": 0.400390625,
911
+ "learning_rate": 0.001,
912
+ "loss": 1.5099,
913
+ "step": 1792
914
+ },
915
+ {
916
+ "epoch": 0.12615475961790334,
917
+ "grad_norm": 0.2294921875,
918
+ "learning_rate": 0.001,
919
+ "loss": 1.5076,
920
+ "step": 1806
921
+ },
922
+ {
923
+ "epoch": 0.12713270349091035,
924
+ "grad_norm": 0.205078125,
925
+ "learning_rate": 0.001,
926
+ "loss": 1.4994,
927
+ "step": 1820
928
+ },
929
+ {
930
+ "epoch": 0.12811064736391736,
931
+ "grad_norm": 0.349609375,
932
+ "learning_rate": 0.001,
933
+ "loss": 1.5135,
934
+ "step": 1834
935
+ },
936
+ {
937
+ "epoch": 0.12908859123692437,
938
+ "grad_norm": 0.310546875,
939
+ "learning_rate": 0.001,
940
+ "loss": 1.5214,
941
+ "step": 1848
942
+ },
943
+ {
944
+ "epoch": 0.13006653510993138,
945
+ "grad_norm": 0.326171875,
946
+ "learning_rate": 0.001,
947
+ "loss": 1.4525,
948
+ "step": 1862
949
+ },
950
+ {
951
+ "epoch": 0.13104447898293836,
952
+ "grad_norm": 0.265625,
953
+ "learning_rate": 0.001,
954
+ "loss": 1.51,
955
+ "step": 1876
956
+ },
957
+ {
958
+ "epoch": 0.13202242285594537,
959
+ "grad_norm": 0.294921875,
960
+ "learning_rate": 0.001,
961
+ "loss": 1.4965,
962
+ "step": 1890
963
+ },
964
+ {
965
+ "epoch": 0.13300036672895238,
966
+ "grad_norm": 0.1943359375,
967
+ "learning_rate": 0.001,
968
+ "loss": 1.4854,
969
+ "step": 1904
970
+ },
971
+ {
972
+ "epoch": 0.1339783106019594,
973
+ "grad_norm": 0.1943359375,
974
+ "learning_rate": 0.001,
975
+ "loss": 1.4926,
976
+ "step": 1918
977
+ },
978
+ {
979
+ "epoch": 0.13495625447496637,
980
+ "grad_norm": 0.291015625,
981
+ "learning_rate": 0.001,
982
+ "loss": 1.4897,
983
+ "step": 1932
984
+ },
985
+ {
986
+ "epoch": 0.13593419834797338,
987
+ "grad_norm": 0.365234375,
988
+ "learning_rate": 0.001,
989
+ "loss": 1.498,
990
+ "step": 1946
991
+ },
992
+ {
993
+ "epoch": 0.1369121422209804,
994
+ "grad_norm": 0.283203125,
995
+ "learning_rate": 0.001,
996
+ "loss": 1.5029,
997
+ "step": 1960
998
+ },
999
+ {
1000
+ "epoch": 0.1378900860939874,
1001
+ "grad_norm": 0.419921875,
1002
+ "learning_rate": 0.001,
1003
+ "loss": 1.482,
1004
+ "step": 1974
1005
+ },
1006
+ {
1007
+ "epoch": 0.1388680299669944,
1008
+ "grad_norm": 0.287109375,
1009
+ "learning_rate": 0.001,
1010
+ "loss": 1.5212,
1011
+ "step": 1988
1012
+ },
1013
+ {
1014
+ "epoch": 0.1398459738400014,
1015
+ "grad_norm": 0.80859375,
1016
+ "learning_rate": 0.001,
1017
+ "loss": 1.549,
1018
+ "step": 2002
1019
+ },
1020
+ {
1021
+ "epoch": 0.1408239177130084,
1022
+ "grad_norm": 0.36328125,
1023
+ "learning_rate": 0.001,
1024
+ "loss": 1.5148,
1025
+ "step": 2016
1026
+ },
1027
+ {
1028
+ "epoch": 0.1418018615860154,
1029
+ "grad_norm": 0.255859375,
1030
+ "learning_rate": 0.001,
1031
+ "loss": 1.4993,
1032
+ "step": 2030
1033
+ },
1034
+ {
1035
+ "epoch": 0.1427798054590224,
1036
+ "grad_norm": 0.220703125,
1037
+ "learning_rate": 0.001,
1038
+ "loss": 1.4985,
1039
+ "step": 2044
1040
+ },
1041
+ {
1042
+ "epoch": 0.14375774933202942,
1043
+ "grad_norm": 0.25,
1044
+ "learning_rate": 0.001,
1045
+ "loss": 1.4938,
1046
+ "step": 2058
1047
+ },
1048
+ {
1049
+ "epoch": 0.1447356932050364,
1050
+ "grad_norm": 0.177734375,
1051
+ "learning_rate": 0.001,
1052
+ "loss": 1.4777,
1053
+ "step": 2072
1054
+ },
1055
+ {
1056
+ "epoch": 0.1457136370780434,
1057
+ "grad_norm": 0.3046875,
1058
+ "learning_rate": 0.001,
1059
+ "loss": 1.4865,
1060
+ "step": 2086
1061
+ },
1062
+ {
1063
+ "epoch": 0.14669158095105042,
1064
+ "grad_norm": 0.19140625,
1065
+ "learning_rate": 0.001,
1066
+ "loss": 1.4567,
1067
+ "step": 2100
1068
+ },
1069
+ {
1070
+ "epoch": 0.14766952482405743,
1071
+ "grad_norm": 0.2099609375,
1072
+ "learning_rate": 0.001,
1073
+ "loss": 1.4706,
1074
+ "step": 2114
1075
+ },
1076
+ {
1077
+ "epoch": 0.1486474686970644,
1078
+ "grad_norm": 0.216796875,
1079
+ "learning_rate": 0.001,
1080
+ "loss": 1.4959,
1081
+ "step": 2128
1082
+ },
1083
+ {
1084
+ "epoch": 0.14962541257007142,
1085
+ "grad_norm": 0.224609375,
1086
+ "learning_rate": 0.001,
1087
+ "loss": 1.4631,
1088
+ "step": 2142
1089
+ },
1090
+ {
1091
+ "epoch": 0.15060335644307843,
1092
+ "grad_norm": 0.2490234375,
1093
+ "learning_rate": 0.001,
1094
+ "loss": 1.4669,
1095
+ "step": 2156
1096
+ },
1097
+ {
1098
+ "epoch": 0.15158130031608544,
1099
+ "grad_norm": 0.2099609375,
1100
+ "learning_rate": 0.001,
1101
+ "loss": 1.4479,
1102
+ "step": 2170
1103
+ },
1104
+ {
1105
+ "epoch": 0.15255924418909245,
1106
+ "grad_norm": 0.25,
1107
+ "learning_rate": 0.001,
1108
+ "loss": 1.4584,
1109
+ "step": 2184
1110
+ },
1111
+ {
1112
+ "epoch": 0.15353718806209943,
1113
+ "grad_norm": 0.345703125,
1114
+ "learning_rate": 0.001,
1115
+ "loss": 1.4609,
1116
+ "step": 2198
1117
+ },
1118
+ {
1119
+ "epoch": 0.15451513193510644,
1120
+ "grad_norm": 0.27734375,
1121
+ "learning_rate": 0.001,
1122
+ "loss": 1.4499,
1123
+ "step": 2212
1124
+ },
1125
+ {
1126
+ "epoch": 0.15549307580811345,
1127
+ "grad_norm": 0.2373046875,
1128
+ "learning_rate": 0.001,
1129
+ "loss": 1.4562,
1130
+ "step": 2226
1131
+ },
1132
+ {
1133
+ "epoch": 0.15647101968112045,
1134
+ "grad_norm": 0.302734375,
1135
+ "learning_rate": 0.001,
1136
+ "loss": 1.4745,
1137
+ "step": 2240
1138
+ },
1139
+ {
1140
+ "epoch": 0.15744896355412744,
1141
+ "grad_norm": 0.330078125,
1142
+ "learning_rate": 0.001,
1143
+ "loss": 1.4773,
1144
+ "step": 2254
1145
+ },
1146
+ {
1147
+ "epoch": 0.15842690742713444,
1148
+ "grad_norm": 0.4609375,
1149
+ "learning_rate": 0.001,
1150
+ "loss": 1.4501,
1151
+ "step": 2268
1152
+ },
1153
+ {
1154
+ "epoch": 0.15940485130014145,
1155
+ "grad_norm": 0.25390625,
1156
+ "learning_rate": 0.001,
1157
+ "loss": 1.4378,
1158
+ "step": 2282
1159
+ },
1160
+ {
1161
+ "epoch": 0.16038279517314846,
1162
+ "grad_norm": 0.19921875,
1163
+ "learning_rate": 0.001,
1164
+ "loss": 1.4452,
1165
+ "step": 2296
1166
+ },
1167
+ {
1168
+ "epoch": 0.16136073904615547,
1169
+ "grad_norm": 0.322265625,
1170
+ "learning_rate": 0.001,
1171
+ "loss": 1.4536,
1172
+ "step": 2310
1173
+ },
1174
+ {
1175
+ "epoch": 0.16233868291916245,
1176
+ "grad_norm": 0.302734375,
1177
+ "learning_rate": 0.001,
1178
+ "loss": 1.462,
1179
+ "step": 2324
1180
+ },
1181
+ {
1182
+ "epoch": 0.16331662679216946,
1183
+ "grad_norm": 0.2353515625,
1184
+ "learning_rate": 0.001,
1185
+ "loss": 1.451,
1186
+ "step": 2338
1187
+ },
1188
+ {
1189
+ "epoch": 0.16429457066517647,
1190
+ "grad_norm": 0.345703125,
1191
+ "learning_rate": 0.001,
1192
+ "loss": 1.4266,
1193
+ "step": 2352
1194
+ },
1195
+ {
1196
+ "epoch": 0.16527251453818348,
1197
+ "grad_norm": 0.30859375,
1198
+ "learning_rate": 0.001,
1199
+ "loss": 1.4503,
1200
+ "step": 2366
1201
+ },
1202
+ {
1203
+ "epoch": 0.16625045841119046,
1204
+ "grad_norm": 0.271484375,
1205
+ "learning_rate": 0.001,
1206
+ "loss": 1.4502,
1207
+ "step": 2380
1208
+ },
1209
+ {
1210
+ "epoch": 0.16722840228419747,
1211
+ "grad_norm": 0.279296875,
1212
+ "learning_rate": 0.001,
1213
+ "loss": 1.4587,
1214
+ "step": 2394
1215
+ },
1216
+ {
1217
+ "epoch": 0.16820634615720448,
1218
+ "grad_norm": 0.279296875,
1219
+ "learning_rate": 0.001,
1220
+ "loss": 1.4588,
1221
+ "step": 2408
1222
+ },
1223
+ {
1224
+ "epoch": 0.1691842900302115,
1225
+ "grad_norm": 0.3203125,
1226
+ "learning_rate": 0.001,
1227
+ "loss": 1.4679,
1228
+ "step": 2422
1229
+ },
1230
+ {
1231
+ "epoch": 0.1701622339032185,
1232
+ "grad_norm": 0.287109375,
1233
+ "learning_rate": 0.001,
1234
+ "loss": 1.4329,
1235
+ "step": 2436
1236
+ },
1237
+ {
1238
+ "epoch": 0.17114017777622548,
1239
+ "grad_norm": 1.109375,
1240
+ "learning_rate": 0.001,
1241
+ "loss": 1.4566,
1242
+ "step": 2450
1243
+ },
1244
+ {
1245
+ "epoch": 0.1721181216492325,
1246
+ "grad_norm": 0.357421875,
1247
+ "learning_rate": 0.001,
1248
+ "loss": 1.4728,
1249
+ "step": 2464
1250
+ },
1251
+ {
1252
+ "epoch": 0.1730960655222395,
1253
+ "grad_norm": 0.2734375,
1254
+ "learning_rate": 0.001,
1255
+ "loss": 1.4518,
1256
+ "step": 2478
1257
+ },
1258
+ {
1259
+ "epoch": 0.1740740093952465,
1260
+ "grad_norm": 0.3203125,
1261
+ "learning_rate": 0.001,
1262
+ "loss": 1.4507,
1263
+ "step": 2492
1264
+ },
1265
+ {
1266
+ "epoch": 0.17505195326825349,
1267
+ "grad_norm": 0.2412109375,
1268
+ "learning_rate": 0.001,
1269
+ "loss": 1.4242,
1270
+ "step": 2506
1271
+ },
1272
+ {
1273
+ "epoch": 0.1760298971412605,
1274
+ "grad_norm": 0.197265625,
1275
+ "learning_rate": 0.001,
1276
+ "loss": 1.4116,
1277
+ "step": 2520
1278
+ },
1279
+ {
1280
+ "epoch": 0.1770078410142675,
1281
+ "grad_norm": 0.3671875,
1282
+ "learning_rate": 0.001,
1283
+ "loss": 1.4294,
1284
+ "step": 2534
1285
+ },
1286
+ {
1287
+ "epoch": 0.1779857848872745,
1288
+ "grad_norm": 0.451171875,
1289
+ "learning_rate": 0.001,
1290
+ "loss": 1.4448,
1291
+ "step": 2548
1292
+ },
1293
+ {
1294
+ "epoch": 0.17896372876028152,
1295
+ "grad_norm": 0.291015625,
1296
+ "learning_rate": 0.001,
1297
+ "loss": 1.4468,
1298
+ "step": 2562
1299
+ },
1300
+ {
1301
+ "epoch": 0.1799416726332885,
1302
+ "grad_norm": 0.419921875,
1303
+ "learning_rate": 0.001,
1304
+ "loss": 1.4319,
1305
+ "step": 2576
1306
+ },
1307
+ {
1308
+ "epoch": 0.1809196165062955,
1309
+ "grad_norm": 0.2275390625,
1310
+ "learning_rate": 0.001,
1311
+ "loss": 1.4309,
1312
+ "step": 2590
1313
+ },
1314
+ {
1315
+ "epoch": 0.18189756037930252,
1316
+ "grad_norm": 0.236328125,
1317
+ "learning_rate": 0.001,
1318
+ "loss": 1.4324,
1319
+ "step": 2604
1320
+ },
1321
+ {
1322
+ "epoch": 0.18287550425230953,
1323
+ "grad_norm": 0.1875,
1324
+ "learning_rate": 0.001,
1325
+ "loss": 1.4143,
1326
+ "step": 2618
1327
+ },
1328
+ {
1329
+ "epoch": 0.1838534481253165,
1330
+ "grad_norm": 0.197265625,
1331
+ "learning_rate": 0.001,
1332
+ "loss": 1.417,
1333
+ "step": 2632
1334
+ },
1335
+ {
1336
+ "epoch": 0.18483139199832352,
1337
+ "grad_norm": 0.2099609375,
1338
+ "learning_rate": 0.001,
1339
+ "loss": 1.4318,
1340
+ "step": 2646
1341
+ },
1342
+ {
1343
+ "epoch": 0.18580933587133053,
1344
+ "grad_norm": 0.255859375,
1345
+ "learning_rate": 0.001,
1346
+ "loss": 1.4151,
1347
+ "step": 2660
1348
+ },
1349
+ {
1350
+ "epoch": 0.18678727974433754,
1351
+ "grad_norm": 0.275390625,
1352
+ "learning_rate": 0.001,
1353
+ "loss": 1.4169,
1354
+ "step": 2674
1355
+ },
1356
+ {
1357
+ "epoch": 0.18776522361734455,
1358
+ "grad_norm": 0.302734375,
1359
+ "learning_rate": 0.001,
1360
+ "loss": 1.4352,
1361
+ "step": 2688
1362
+ },
1363
+ {
1364
+ "epoch": 0.18874316749035153,
1365
+ "grad_norm": 0.4296875,
1366
+ "learning_rate": 0.001,
1367
+ "loss": 1.4346,
1368
+ "step": 2702
1369
+ },
1370
+ {
1371
+ "epoch": 0.18972111136335854,
1372
+ "grad_norm": 0.310546875,
1373
+ "learning_rate": 0.001,
1374
+ "loss": 1.4253,
1375
+ "step": 2716
1376
+ },
1377
+ {
1378
+ "epoch": 0.19069905523636554,
1379
+ "grad_norm": 0.31640625,
1380
+ "learning_rate": 0.001,
1381
+ "loss": 1.4272,
1382
+ "step": 2730
1383
+ },
1384
+ {
1385
+ "epoch": 0.19167699910937255,
1386
+ "grad_norm": 0.8984375,
1387
+ "learning_rate": 0.001,
1388
+ "loss": 1.4618,
1389
+ "step": 2744
1390
+ },
1391
+ {
1392
+ "epoch": 0.19265494298237953,
1393
+ "grad_norm": 0.322265625,
1394
+ "learning_rate": 0.001,
1395
+ "loss": 1.4616,
1396
+ "step": 2758
1397
+ },
1398
+ {
1399
+ "epoch": 0.19363288685538654,
1400
+ "grad_norm": 0.287109375,
1401
+ "learning_rate": 0.001,
1402
+ "loss": 1.4231,
1403
+ "step": 2772
1404
+ },
1405
+ {
1406
+ "epoch": 0.19461083072839355,
1407
+ "grad_norm": 0.20703125,
1408
+ "learning_rate": 0.001,
1409
+ "loss": 1.4185,
1410
+ "step": 2786
1411
+ },
1412
+ {
1413
+ "epoch": 0.19558877460140056,
1414
+ "grad_norm": 0.283203125,
1415
+ "learning_rate": 0.001,
1416
+ "loss": 1.4578,
1417
+ "step": 2800
1418
+ },
1419
+ {
1420
+ "epoch": 0.19656671847440757,
1421
+ "grad_norm": 0.26953125,
1422
+ "learning_rate": 0.001,
1423
+ "loss": 1.4134,
1424
+ "step": 2814
1425
+ },
1426
+ {
1427
+ "epoch": 0.19754466234741455,
1428
+ "grad_norm": 0.2099609375,
1429
+ "learning_rate": 0.001,
1430
+ "loss": 1.4253,
1431
+ "step": 2828
1432
+ },
1433
+ {
1434
+ "epoch": 0.19852260622042156,
1435
+ "grad_norm": 0.208984375,
1436
+ "learning_rate": 0.001,
1437
+ "loss": 1.414,
1438
+ "step": 2842
1439
+ },
1440
+ {
1441
+ "epoch": 0.19950055009342857,
1442
+ "grad_norm": 0.337890625,
1443
+ "learning_rate": 0.001,
1444
+ "loss": 1.4069,
1445
+ "step": 2856
1446
+ },
1447
+ {
1448
+ "epoch": 0.19991966889614585,
1449
+ "eval_loss": 1.8960140943527222,
1450
+ "eval_runtime": 9.2129,
1451
+ "eval_samples_per_second": 108.543,
1452
+ "eval_steps_per_second": 1.411,
1453
+ "step": 2862
1454
+ },
1455
+ {
1456
+ "epoch": 0.20047849396643558,
1457
+ "grad_norm": 0.31640625,
1458
+ "learning_rate": 0.001,
1459
+ "loss": 1.4267,
1460
+ "step": 2870
1461
+ },
1462
+ {
1463
+ "epoch": 0.20145643783944256,
1464
+ "grad_norm": 0.2373046875,
1465
+ "learning_rate": 0.001,
1466
+ "loss": 1.4186,
1467
+ "step": 2884
1468
+ },
1469
+ {
1470
+ "epoch": 0.20243438171244957,
1471
+ "grad_norm": 0.2236328125,
1472
+ "learning_rate": 0.001,
1473
+ "loss": 1.4105,
1474
+ "step": 2898
1475
+ },
1476
+ {
1477
+ "epoch": 0.20341232558545658,
1478
+ "grad_norm": 0.271484375,
1479
+ "learning_rate": 0.001,
1480
+ "loss": 1.4012,
1481
+ "step": 2912
1482
+ },
1483
+ {
1484
+ "epoch": 0.20439026945846359,
1485
+ "grad_norm": 0.234375,
1486
+ "learning_rate": 0.001,
1487
+ "loss": 1.3917,
1488
+ "step": 2926
1489
+ },
1490
+ {
1491
+ "epoch": 0.2053682133314706,
1492
+ "grad_norm": 0.259765625,
1493
+ "learning_rate": 0.001,
1494
+ "loss": 1.3952,
1495
+ "step": 2940
1496
+ },
1497
+ {
1498
+ "epoch": 0.20634615720447758,
1499
+ "grad_norm": 0.1923828125,
1500
+ "learning_rate": 0.001,
1501
+ "loss": 1.396,
1502
+ "step": 2954
1503
+ },
1504
+ {
1505
+ "epoch": 0.20732410107748458,
1506
+ "grad_norm": 0.453125,
1507
+ "learning_rate": 0.001,
1508
+ "loss": 1.4556,
1509
+ "step": 2968
1510
+ },
1511
+ {
1512
+ "epoch": 0.2083020449504916,
1513
+ "grad_norm": 0.3046875,
1514
+ "learning_rate": 0.001,
1515
+ "loss": 1.4215,
1516
+ "step": 2982
1517
+ },
1518
+ {
1519
+ "epoch": 0.2092799888234986,
1520
+ "grad_norm": 0.2158203125,
1521
+ "learning_rate": 0.001,
1522
+ "loss": 1.4166,
1523
+ "step": 2996
1524
+ },
1525
+ {
1526
+ "epoch": 0.21025793269650558,
1527
+ "grad_norm": 0.25,
1528
+ "learning_rate": 0.001,
1529
+ "loss": 1.3975,
1530
+ "step": 3010
1531
+ },
1532
+ {
1533
+ "epoch": 0.2112358765695126,
1534
+ "grad_norm": 0.2392578125,
1535
+ "learning_rate": 0.001,
1536
+ "loss": 1.4019,
1537
+ "step": 3024
1538
+ },
1539
+ {
1540
+ "epoch": 0.2122138204425196,
1541
+ "grad_norm": 0.2578125,
1542
+ "learning_rate": 0.001,
1543
+ "loss": 1.4016,
1544
+ "step": 3038
1545
+ },
1546
+ {
1547
+ "epoch": 0.2131917643155266,
1548
+ "grad_norm": 0.271484375,
1549
+ "learning_rate": 0.001,
1550
+ "loss": 1.4069,
1551
+ "step": 3052
1552
+ },
1553
+ {
1554
+ "epoch": 0.21416970818853362,
1555
+ "grad_norm": 0.275390625,
1556
+ "learning_rate": 0.001,
1557
+ "loss": 1.4214,
1558
+ "step": 3066
1559
+ },
1560
+ {
1561
+ "epoch": 0.2151476520615406,
1562
+ "grad_norm": 0.23046875,
1563
+ "learning_rate": 0.001,
1564
+ "loss": 1.4123,
1565
+ "step": 3080
1566
+ },
1567
+ {
1568
+ "epoch": 0.2161255959345476,
1569
+ "grad_norm": 0.298828125,
1570
+ "learning_rate": 0.001,
1571
+ "loss": 1.3981,
1572
+ "step": 3094
1573
+ },
1574
+ {
1575
+ "epoch": 0.21710353980755462,
1576
+ "grad_norm": 0.2451171875,
1577
+ "learning_rate": 0.001,
1578
+ "loss": 1.4067,
1579
+ "step": 3108
1580
+ },
1581
+ {
1582
+ "epoch": 0.21808148368056163,
1583
+ "grad_norm": 0.2373046875,
1584
+ "learning_rate": 0.001,
1585
+ "loss": 1.3893,
1586
+ "step": 3122
1587
+ },
1588
+ {
1589
+ "epoch": 0.2190594275535686,
1590
+ "grad_norm": 0.298828125,
1591
+ "learning_rate": 0.001,
1592
+ "loss": 1.4263,
1593
+ "step": 3136
1594
+ },
1595
+ {
1596
+ "epoch": 0.22003737142657562,
1597
+ "grad_norm": 0.318359375,
1598
+ "learning_rate": 0.001,
1599
+ "loss": 1.3931,
1600
+ "step": 3150
1601
+ },
1602
+ {
1603
+ "epoch": 0.22101531529958263,
1604
+ "grad_norm": 0.24609375,
1605
+ "learning_rate": 0.001,
1606
+ "loss": 1.3901,
1607
+ "step": 3164
1608
+ },
1609
+ {
1610
+ "epoch": 0.22199325917258964,
1611
+ "grad_norm": 0.259765625,
1612
+ "learning_rate": 0.001,
1613
+ "loss": 1.3805,
1614
+ "step": 3178
1615
+ },
1616
+ {
1617
+ "epoch": 0.22297120304559664,
1618
+ "grad_norm": 0.259765625,
1619
+ "learning_rate": 0.001,
1620
+ "loss": 1.4044,
1621
+ "step": 3192
1622
+ },
1623
+ {
1624
+ "epoch": 0.22394914691860363,
1625
+ "grad_norm": 0.30859375,
1626
+ "learning_rate": 0.001,
1627
+ "loss": 1.4046,
1628
+ "step": 3206
1629
+ },
1630
+ {
1631
+ "epoch": 0.22492709079161063,
1632
+ "grad_norm": 0.369140625,
1633
+ "learning_rate": 0.001,
1634
+ "loss": 1.4084,
1635
+ "step": 3220
1636
+ },
1637
+ {
1638
+ "epoch": 0.22590503466461764,
1639
+ "grad_norm": 0.310546875,
1640
+ "learning_rate": 0.001,
1641
+ "loss": 1.3929,
1642
+ "step": 3234
1643
+ },
1644
+ {
1645
+ "epoch": 0.22688297853762465,
1646
+ "grad_norm": 0.251953125,
1647
+ "learning_rate": 0.001,
1648
+ "loss": 1.3462,
1649
+ "step": 3248
1650
+ },
1651
+ {
1652
+ "epoch": 0.22786092241063163,
1653
+ "grad_norm": 0.29296875,
1654
+ "learning_rate": 0.001,
1655
+ "loss": 1.3721,
1656
+ "step": 3262
1657
+ },
1658
+ {
1659
+ "epoch": 0.22883886628363864,
1660
+ "grad_norm": 0.3359375,
1661
+ "learning_rate": 0.001,
1662
+ "loss": 1.4027,
1663
+ "step": 3276
1664
+ },
1665
+ {
1666
+ "epoch": 0.22981681015664565,
1667
+ "grad_norm": 0.2890625,
1668
+ "learning_rate": 0.001,
1669
+ "loss": 1.4081,
1670
+ "step": 3290
1671
+ },
1672
+ {
1673
+ "epoch": 0.23079475402965266,
1674
+ "grad_norm": 0.34765625,
1675
+ "learning_rate": 0.001,
1676
+ "loss": 1.3706,
1677
+ "step": 3304
1678
+ },
1679
+ {
1680
+ "epoch": 0.23177269790265967,
1681
+ "grad_norm": 0.3046875,
1682
+ "learning_rate": 0.001,
1683
+ "loss": 1.3652,
1684
+ "step": 3318
1685
+ },
1686
+ {
1687
+ "epoch": 0.23275064177566665,
1688
+ "grad_norm": 0.4296875,
1689
+ "learning_rate": 0.001,
1690
+ "loss": 1.3623,
1691
+ "step": 3332
1692
+ },
1693
+ {
1694
+ "epoch": 0.23372858564867366,
1695
+ "grad_norm": 0.236328125,
1696
+ "learning_rate": 0.001,
1697
+ "loss": 1.37,
1698
+ "step": 3346
1699
+ },
1700
+ {
1701
+ "epoch": 0.23470652952168067,
1702
+ "grad_norm": 0.306640625,
1703
+ "learning_rate": 0.001,
1704
+ "loss": 1.3627,
1705
+ "step": 3360
1706
+ },
1707
+ {
1708
+ "epoch": 0.23568447339468768,
1709
+ "grad_norm": 0.25390625,
1710
+ "learning_rate": 0.001,
1711
+ "loss": 1.3979,
1712
+ "step": 3374
1713
+ },
1714
+ {
1715
+ "epoch": 0.23666241726769469,
1716
+ "grad_norm": 0.28515625,
1717
+ "learning_rate": 0.001,
1718
+ "loss": 1.3967,
1719
+ "step": 3388
1720
+ },
1721
+ {
1722
+ "epoch": 0.23764036114070167,
1723
+ "grad_norm": 0.306640625,
1724
+ "learning_rate": 0.001,
1725
+ "loss": 1.3696,
1726
+ "step": 3402
1727
+ },
1728
+ {
1729
+ "epoch": 0.23861830501370868,
1730
+ "grad_norm": 0.291015625,
1731
+ "learning_rate": 0.001,
1732
+ "loss": 1.3867,
1733
+ "step": 3416
1734
+ },
1735
+ {
1736
+ "epoch": 0.23959624888671568,
1737
+ "grad_norm": 0.291015625,
1738
+ "learning_rate": 0.001,
1739
+ "loss": 1.4115,
1740
+ "step": 3430
1741
+ },
1742
+ {
1743
+ "epoch": 0.2405741927597227,
1744
+ "grad_norm": 0.271484375,
1745
+ "learning_rate": 0.001,
1746
+ "loss": 1.3501,
1747
+ "step": 3444
1748
+ },
1749
+ {
1750
+ "epoch": 0.24155213663272967,
1751
+ "grad_norm": 0.337890625,
1752
+ "learning_rate": 0.001,
1753
+ "loss": 1.377,
1754
+ "step": 3458
1755
+ },
1756
+ {
1757
+ "epoch": 0.24253008050573668,
1758
+ "grad_norm": 0.205078125,
1759
+ "learning_rate": 0.001,
1760
+ "loss": 1.3585,
1761
+ "step": 3472
1762
+ },
1763
+ {
1764
+ "epoch": 0.2435080243787437,
1765
+ "grad_norm": 0.259765625,
1766
+ "learning_rate": 0.001,
1767
+ "loss": 1.36,
1768
+ "step": 3486
1769
+ },
1770
+ {
1771
+ "epoch": 0.2444859682517507,
1772
+ "grad_norm": 0.296875,
1773
+ "learning_rate": 0.001,
1774
+ "loss": 1.3759,
1775
+ "step": 3500
1776
+ },
1777
+ {
1778
+ "epoch": 0.2454639121247577,
1779
+ "grad_norm": 0.24609375,
1780
+ "learning_rate": 0.001,
1781
+ "loss": 1.3782,
1782
+ "step": 3514
1783
+ },
1784
+ {
1785
+ "epoch": 0.2464418559977647,
1786
+ "grad_norm": 0.294921875,
1787
+ "learning_rate": 0.001,
1788
+ "loss": 1.3489,
1789
+ "step": 3528
1790
+ },
1791
+ {
1792
+ "epoch": 0.2474197998707717,
1793
+ "grad_norm": 1.203125,
1794
+ "learning_rate": 0.001,
1795
+ "loss": 1.4092,
1796
+ "step": 3542
1797
+ },
1798
+ {
1799
+ "epoch": 0.2483977437437787,
1800
+ "grad_norm": 0.38671875,
1801
+ "learning_rate": 0.001,
1802
+ "loss": 1.3806,
1803
+ "step": 3556
1804
+ },
1805
+ {
1806
+ "epoch": 0.24937568761678572,
1807
+ "grad_norm": 0.353515625,
1808
+ "learning_rate": 0.001,
1809
+ "loss": 1.3973,
1810
+ "step": 3570
1811
+ },
1812
+ {
1813
+ "epoch": 0.2503536314897927,
1814
+ "grad_norm": 0.287109375,
1815
+ "learning_rate": 0.001,
1816
+ "loss": 1.3725,
1817
+ "step": 3584
1818
+ },
1819
+ {
1820
+ "epoch": 0.2513315753627997,
1821
+ "grad_norm": 0.28125,
1822
+ "learning_rate": 0.001,
1823
+ "loss": 1.3523,
1824
+ "step": 3598
1825
+ },
1826
+ {
1827
+ "epoch": 0.2523095192358067,
1828
+ "grad_norm": 0.41796875,
1829
+ "learning_rate": 0.001,
1830
+ "loss": 1.3929,
1831
+ "step": 3612
1832
+ },
1833
+ {
1834
+ "epoch": 0.2532874631088137,
1835
+ "grad_norm": 0.359375,
1836
+ "learning_rate": 0.001,
1837
+ "loss": 1.3541,
1838
+ "step": 3626
1839
+ },
1840
+ {
1841
+ "epoch": 0.2542654069818207,
1842
+ "grad_norm": 0.2197265625,
1843
+ "learning_rate": 0.001,
1844
+ "loss": 1.3471,
1845
+ "step": 3640
1846
+ },
1847
+ {
1848
+ "epoch": 0.25524335085482774,
1849
+ "grad_norm": 0.248046875,
1850
+ "learning_rate": 0.001,
1851
+ "loss": 1.3887,
1852
+ "step": 3654
1853
+ },
1854
+ {
1855
+ "epoch": 0.2562212947278347,
1856
+ "grad_norm": 0.38671875,
1857
+ "learning_rate": 0.001,
1858
+ "loss": 1.3675,
1859
+ "step": 3668
1860
+ },
1861
+ {
1862
+ "epoch": 0.2571992386008417,
1863
+ "grad_norm": 0.259765625,
1864
+ "learning_rate": 0.001,
1865
+ "loss": 1.3591,
1866
+ "step": 3682
1867
+ },
1868
+ {
1869
+ "epoch": 0.25817718247384874,
1870
+ "grad_norm": 0.462890625,
1871
+ "learning_rate": 0.001,
1872
+ "loss": 1.3813,
1873
+ "step": 3696
1874
+ },
1875
+ {
1876
+ "epoch": 0.2591551263468557,
1877
+ "grad_norm": 0.2099609375,
1878
+ "learning_rate": 0.001,
1879
+ "loss": 1.3555,
1880
+ "step": 3710
1881
+ },
1882
+ {
1883
+ "epoch": 0.26013307021986276,
1884
+ "grad_norm": 0.95703125,
1885
+ "learning_rate": 0.001,
1886
+ "loss": 1.3931,
1887
+ "step": 3724
1888
+ },
1889
+ {
1890
+ "epoch": 0.26111101409286974,
1891
+ "grad_norm": 0.60546875,
1892
+ "learning_rate": 0.001,
1893
+ "loss": 1.4361,
1894
+ "step": 3738
1895
+ },
1896
+ {
1897
+ "epoch": 0.2620889579658767,
1898
+ "grad_norm": 0.259765625,
1899
+ "learning_rate": 0.001,
1900
+ "loss": 1.3841,
1901
+ "step": 3752
1902
+ },
1903
+ {
1904
+ "epoch": 0.26306690183888376,
1905
+ "grad_norm": 0.2490234375,
1906
+ "learning_rate": 0.001,
1907
+ "loss": 1.3808,
1908
+ "step": 3766
1909
+ },
1910
+ {
1911
+ "epoch": 0.26404484571189074,
1912
+ "grad_norm": 0.263671875,
1913
+ "learning_rate": 0.001,
1914
+ "loss": 1.3792,
1915
+ "step": 3780
1916
+ },
1917
+ {
1918
+ "epoch": 0.2650227895848978,
1919
+ "grad_norm": 0.2109375,
1920
+ "learning_rate": 0.001,
1921
+ "loss": 1.3488,
1922
+ "step": 3794
1923
+ },
1924
+ {
1925
+ "epoch": 0.26600073345790476,
1926
+ "grad_norm": 0.3359375,
1927
+ "learning_rate": 0.001,
1928
+ "loss": 1.3644,
1929
+ "step": 3808
1930
+ },
1931
+ {
1932
+ "epoch": 0.26697867733091174,
1933
+ "grad_norm": 0.359375,
1934
+ "learning_rate": 0.001,
1935
+ "loss": 1.3707,
1936
+ "step": 3822
1937
+ },
1938
+ {
1939
+ "epoch": 0.2679566212039188,
1940
+ "grad_norm": 0.2236328125,
1941
+ "learning_rate": 0.001,
1942
+ "loss": 1.3584,
1943
+ "step": 3836
1944
+ },
1945
+ {
1946
+ "epoch": 0.26893456507692576,
1947
+ "grad_norm": 0.259765625,
1948
+ "learning_rate": 0.001,
1949
+ "loss": 1.3605,
1950
+ "step": 3850
1951
+ },
1952
+ {
1953
+ "epoch": 0.26991250894993274,
1954
+ "grad_norm": 0.251953125,
1955
+ "learning_rate": 0.001,
1956
+ "loss": 1.3703,
1957
+ "step": 3864
1958
+ },
1959
+ {
1960
+ "epoch": 0.2708904528229398,
1961
+ "grad_norm": 0.353515625,
1962
+ "learning_rate": 0.001,
1963
+ "loss": 1.3507,
1964
+ "step": 3878
1965
+ },
1966
+ {
1967
+ "epoch": 0.27186839669594676,
1968
+ "grad_norm": 0.291015625,
1969
+ "learning_rate": 0.001,
1970
+ "loss": 1.3601,
1971
+ "step": 3892
1972
+ },
1973
+ {
1974
+ "epoch": 0.2728463405689538,
1975
+ "grad_norm": 0.259765625,
1976
+ "learning_rate": 0.001,
1977
+ "loss": 1.3663,
1978
+ "step": 3906
1979
+ },
1980
+ {
1981
+ "epoch": 0.2738242844419608,
1982
+ "grad_norm": 0.2490234375,
1983
+ "learning_rate": 0.001,
1984
+ "loss": 1.3625,
1985
+ "step": 3920
1986
+ },
1987
+ {
1988
+ "epoch": 0.27480222831496776,
1989
+ "grad_norm": 0.333984375,
1990
+ "learning_rate": 0.001,
1991
+ "loss": 1.3722,
1992
+ "step": 3934
1993
+ },
1994
+ {
1995
+ "epoch": 0.2757801721879748,
1996
+ "grad_norm": 0.3046875,
1997
+ "learning_rate": 0.001,
1998
+ "loss": 1.342,
1999
+ "step": 3948
2000
+ },
2001
+ {
2002
+ "epoch": 0.2767581160609818,
2003
+ "grad_norm": 0.283203125,
2004
+ "learning_rate": 0.001,
2005
+ "loss": 1.3682,
2006
+ "step": 3962
2007
+ },
2008
+ {
2009
+ "epoch": 0.2777360599339888,
2010
+ "grad_norm": 0.2373046875,
2011
+ "learning_rate": 0.001,
2012
+ "loss": 1.3662,
2013
+ "step": 3976
2014
+ },
2015
+ {
2016
+ "epoch": 0.2787140038069958,
2017
+ "grad_norm": 0.24609375,
2018
+ "learning_rate": 0.001,
2019
+ "loss": 1.3528,
2020
+ "step": 3990
2021
+ },
2022
+ {
2023
+ "epoch": 0.2796919476800028,
2024
+ "grad_norm": 0.27734375,
2025
+ "learning_rate": 0.001,
2026
+ "loss": 1.3576,
2027
+ "step": 4004
2028
+ },
2029
+ {
2030
+ "epoch": 0.2806698915530098,
2031
+ "grad_norm": 0.265625,
2032
+ "learning_rate": 0.001,
2033
+ "loss": 1.3744,
2034
+ "step": 4018
2035
+ },
2036
+ {
2037
+ "epoch": 0.2816478354260168,
2038
+ "grad_norm": 0.2333984375,
2039
+ "learning_rate": 0.001,
2040
+ "loss": 1.3554,
2041
+ "step": 4032
2042
+ },
2043
+ {
2044
+ "epoch": 0.2826257792990238,
2045
+ "grad_norm": 0.380859375,
2046
+ "learning_rate": 0.001,
2047
+ "loss": 1.3518,
2048
+ "step": 4046
2049
+ },
2050
+ {
2051
+ "epoch": 0.2836037231720308,
2052
+ "grad_norm": 0.25,
2053
+ "learning_rate": 0.001,
2054
+ "loss": 1.3595,
2055
+ "step": 4060
2056
+ },
2057
+ {
2058
+ "epoch": 0.2845816670450378,
2059
+ "grad_norm": 0.267578125,
2060
+ "learning_rate": 0.001,
2061
+ "loss": 1.3361,
2062
+ "step": 4074
2063
+ },
2064
+ {
2065
+ "epoch": 0.2855596109180448,
2066
+ "grad_norm": 0.275390625,
2067
+ "learning_rate": 0.001,
2068
+ "loss": 1.3664,
2069
+ "step": 4088
2070
+ },
2071
+ {
2072
+ "epoch": 0.2865375547910518,
2073
+ "grad_norm": 0.310546875,
2074
+ "learning_rate": 0.001,
2075
+ "loss": 1.3633,
2076
+ "step": 4102
2077
+ },
2078
+ {
2079
+ "epoch": 0.28751549866405884,
2080
+ "grad_norm": 0.2275390625,
2081
+ "learning_rate": 0.001,
2082
+ "loss": 1.3437,
2083
+ "step": 4116
2084
+ },
2085
+ {
2086
+ "epoch": 0.2884934425370658,
2087
+ "grad_norm": 0.1943359375,
2088
+ "learning_rate": 0.001,
2089
+ "loss": 1.3401,
2090
+ "step": 4130
2091
+ },
2092
+ {
2093
+ "epoch": 0.2894713864100728,
2094
+ "grad_norm": 0.2392578125,
2095
+ "learning_rate": 0.001,
2096
+ "loss": 1.3465,
2097
+ "step": 4144
2098
+ },
2099
+ {
2100
+ "epoch": 0.29044933028307984,
2101
+ "grad_norm": 0.5,
2102
+ "learning_rate": 0.001,
2103
+ "loss": 1.3517,
2104
+ "step": 4158
2105
+ },
2106
+ {
2107
+ "epoch": 0.2914272741560868,
2108
+ "grad_norm": 0.36328125,
2109
+ "learning_rate": 0.001,
2110
+ "loss": 1.3272,
2111
+ "step": 4172
2112
+ },
2113
+ {
2114
+ "epoch": 0.2924052180290938,
2115
+ "grad_norm": 0.333984375,
2116
+ "learning_rate": 0.001,
2117
+ "loss": 1.3676,
2118
+ "step": 4186
2119
+ },
2120
+ {
2121
+ "epoch": 0.29338316190210084,
2122
+ "grad_norm": 0.216796875,
2123
+ "learning_rate": 0.001,
2124
+ "loss": 1.3501,
2125
+ "step": 4200
2126
+ },
2127
+ {
2128
+ "epoch": 0.2943611057751078,
2129
+ "grad_norm": 0.24609375,
2130
+ "learning_rate": 0.001,
2131
+ "loss": 1.3423,
2132
+ "step": 4214
2133
+ },
2134
+ {
2135
+ "epoch": 0.29533904964811486,
2136
+ "grad_norm": 0.3671875,
2137
+ "learning_rate": 0.001,
2138
+ "loss": 1.3396,
2139
+ "step": 4228
2140
+ },
2141
+ {
2142
+ "epoch": 0.29631699352112184,
2143
+ "grad_norm": 0.328125,
2144
+ "learning_rate": 0.001,
2145
+ "loss": 1.3416,
2146
+ "step": 4242
2147
+ },
2148
+ {
2149
+ "epoch": 0.2972949373941288,
2150
+ "grad_norm": 0.326171875,
2151
+ "learning_rate": 0.001,
2152
+ "loss": 1.3419,
2153
+ "step": 4256
2154
+ },
2155
+ {
2156
+ "epoch": 0.29827288126713586,
2157
+ "grad_norm": 0.3671875,
2158
+ "learning_rate": 0.001,
2159
+ "loss": 1.3128,
2160
+ "step": 4270
2161
+ },
2162
+ {
2163
+ "epoch": 0.29925082514014284,
2164
+ "grad_norm": 0.28125,
2165
+ "learning_rate": 0.001,
2166
+ "loss": 1.3528,
2167
+ "step": 4284
2168
+ },
2169
+ {
2170
+ "epoch": 0.29987950334421876,
2171
+ "eval_loss": 1.7873083353042603,
2172
+ "eval_runtime": 9.1065,
2173
+ "eval_samples_per_second": 109.812,
2174
+ "eval_steps_per_second": 1.428,
2175
+ "step": 4293
2176
+ },
2177
+ {
2178
+ "epoch": 0.3002287690131499,
2179
+ "grad_norm": 0.330078125,
2180
+ "learning_rate": 0.001,
2181
+ "loss": 1.3496,
2182
+ "step": 4298
2183
+ },
2184
+ {
2185
+ "epoch": 0.30120671288615686,
2186
+ "grad_norm": 0.2451171875,
2187
+ "learning_rate": 0.001,
2188
+ "loss": 1.3369,
2189
+ "step": 4312
2190
+ },
2191
+ {
2192
+ "epoch": 0.30218465675916384,
2193
+ "grad_norm": 0.34765625,
2194
+ "learning_rate": 0.001,
2195
+ "loss": 1.3803,
2196
+ "step": 4326
2197
+ },
2198
+ {
2199
+ "epoch": 0.3031626006321709,
2200
+ "grad_norm": 0.41796875,
2201
+ "learning_rate": 0.001,
2202
+ "loss": 1.324,
2203
+ "step": 4340
2204
+ },
2205
+ {
2206
+ "epoch": 0.30414054450517786,
2207
+ "grad_norm": 0.42578125,
2208
+ "learning_rate": 0.001,
2209
+ "loss": 1.3559,
2210
+ "step": 4354
2211
+ },
2212
+ {
2213
+ "epoch": 0.3051184883781849,
2214
+ "grad_norm": 0.373046875,
2215
+ "learning_rate": 0.001,
2216
+ "loss": 1.3166,
2217
+ "step": 4368
2218
+ },
2219
+ {
2220
+ "epoch": 0.3060964322511919,
2221
+ "grad_norm": 0.25390625,
2222
+ "learning_rate": 0.001,
2223
+ "loss": 1.3376,
2224
+ "step": 4382
2225
+ },
2226
+ {
2227
+ "epoch": 0.30707437612419886,
2228
+ "grad_norm": 0.2197265625,
2229
+ "learning_rate": 0.001,
2230
+ "loss": 1.3155,
2231
+ "step": 4396
2232
+ },
2233
+ {
2234
+ "epoch": 0.3080523199972059,
2235
+ "grad_norm": 0.337890625,
2236
+ "learning_rate": 0.001,
2237
+ "loss": 1.3278,
2238
+ "step": 4410
2239
+ },
2240
+ {
2241
+ "epoch": 0.3090302638702129,
2242
+ "grad_norm": 0.25390625,
2243
+ "learning_rate": 0.001,
2244
+ "loss": 1.34,
2245
+ "step": 4424
2246
+ },
2247
+ {
2248
+ "epoch": 0.31000820774321985,
2249
+ "grad_norm": 0.2431640625,
2250
+ "learning_rate": 0.001,
2251
+ "loss": 1.313,
2252
+ "step": 4438
2253
+ },
2254
+ {
2255
+ "epoch": 0.3109861516162269,
2256
+ "grad_norm": 0.359375,
2257
+ "learning_rate": 0.001,
2258
+ "loss": 1.3421,
2259
+ "step": 4452
2260
+ },
2261
+ {
2262
+ "epoch": 0.3119640954892339,
2263
+ "grad_norm": 0.423828125,
2264
+ "learning_rate": 0.001,
2265
+ "loss": 1.3327,
2266
+ "step": 4466
2267
+ },
2268
+ {
2269
+ "epoch": 0.3129420393622409,
2270
+ "grad_norm": 0.30078125,
2271
+ "learning_rate": 0.001,
2272
+ "loss": 1.3345,
2273
+ "step": 4480
2274
+ },
2275
+ {
2276
+ "epoch": 0.3139199832352479,
2277
+ "grad_norm": 0.30859375,
2278
+ "learning_rate": 0.001,
2279
+ "loss": 1.3157,
2280
+ "step": 4494
2281
+ },
2282
+ {
2283
+ "epoch": 0.31489792710825487,
2284
+ "grad_norm": 0.19921875,
2285
+ "learning_rate": 0.001,
2286
+ "loss": 1.334,
2287
+ "step": 4508
2288
+ },
2289
+ {
2290
+ "epoch": 0.3158758709812619,
2291
+ "grad_norm": 0.26171875,
2292
+ "learning_rate": 0.001,
2293
+ "loss": 1.3526,
2294
+ "step": 4522
2295
+ },
2296
+ {
2297
+ "epoch": 0.3168538148542689,
2298
+ "grad_norm": 0.35546875,
2299
+ "learning_rate": 0.001,
2300
+ "loss": 1.3565,
2301
+ "step": 4536
2302
+ },
2303
+ {
2304
+ "epoch": 0.3178317587272759,
2305
+ "grad_norm": 0.33984375,
2306
+ "learning_rate": 0.001,
2307
+ "loss": 1.3343,
2308
+ "step": 4550
2309
+ },
2310
+ {
2311
+ "epoch": 0.3188097026002829,
2312
+ "grad_norm": 0.28125,
2313
+ "learning_rate": 0.001,
2314
+ "loss": 1.3372,
2315
+ "step": 4564
2316
+ },
2317
+ {
2318
+ "epoch": 0.3197876464732899,
2319
+ "grad_norm": 0.328125,
2320
+ "learning_rate": 0.001,
2321
+ "loss": 1.3105,
2322
+ "step": 4578
2323
+ },
2324
+ {
2325
+ "epoch": 0.3207655903462969,
2326
+ "grad_norm": 0.34765625,
2327
+ "learning_rate": 0.001,
2328
+ "loss": 1.3297,
2329
+ "step": 4592
2330
+ },
2331
+ {
2332
+ "epoch": 0.3217435342193039,
2333
+ "grad_norm": 0.341796875,
2334
+ "learning_rate": 0.001,
2335
+ "loss": 1.3549,
2336
+ "step": 4606
2337
+ },
2338
+ {
2339
+ "epoch": 0.32272147809231094,
2340
+ "grad_norm": 0.52734375,
2341
+ "learning_rate": 0.001,
2342
+ "loss": 1.3499,
2343
+ "step": 4620
2344
+ },
2345
+ {
2346
+ "epoch": 0.3236994219653179,
2347
+ "grad_norm": 0.58984375,
2348
+ "learning_rate": 0.001,
2349
+ "loss": 1.3642,
2350
+ "step": 4634
2351
+ },
2352
+ {
2353
+ "epoch": 0.3246773658383249,
2354
+ "grad_norm": 0.58984375,
2355
+ "learning_rate": 0.001,
2356
+ "loss": 1.3611,
2357
+ "step": 4648
2358
+ },
2359
+ {
2360
+ "epoch": 0.32565530971133194,
2361
+ "grad_norm": 0.3359375,
2362
+ "learning_rate": 0.001,
2363
+ "loss": 1.3255,
2364
+ "step": 4662
2365
+ },
2366
+ {
2367
+ "epoch": 0.3266332535843389,
2368
+ "grad_norm": 0.265625,
2369
+ "learning_rate": 0.001,
2370
+ "loss": 1.3255,
2371
+ "step": 4676
2372
+ },
2373
+ {
2374
+ "epoch": 0.3276111974573459,
2375
+ "grad_norm": 0.330078125,
2376
+ "learning_rate": 0.001,
2377
+ "loss": 1.3431,
2378
+ "step": 4690
2379
+ },
2380
+ {
2381
+ "epoch": 0.32858914133035294,
2382
+ "grad_norm": 0.326171875,
2383
+ "learning_rate": 0.001,
2384
+ "loss": 1.322,
2385
+ "step": 4704
2386
+ },
2387
+ {
2388
+ "epoch": 0.3295670852033599,
2389
+ "grad_norm": 0.26171875,
2390
+ "learning_rate": 0.001,
2391
+ "loss": 1.3271,
2392
+ "step": 4718
2393
+ },
2394
+ {
2395
+ "epoch": 0.33054502907636696,
2396
+ "grad_norm": 0.353515625,
2397
+ "learning_rate": 0.001,
2398
+ "loss": 1.3364,
2399
+ "step": 4732
2400
+ },
2401
+ {
2402
+ "epoch": 0.33152297294937394,
2403
+ "grad_norm": 0.25390625,
2404
+ "learning_rate": 0.001,
2405
+ "loss": 1.3311,
2406
+ "step": 4746
2407
+ },
2408
+ {
2409
+ "epoch": 0.3325009168223809,
2410
+ "grad_norm": 0.33984375,
2411
+ "learning_rate": 0.001,
2412
+ "loss": 1.3555,
2413
+ "step": 4760
2414
+ },
2415
+ {
2416
+ "epoch": 0.33347886069538796,
2417
+ "grad_norm": 0.267578125,
2418
+ "learning_rate": 0.001,
2419
+ "loss": 1.329,
2420
+ "step": 4774
2421
+ },
2422
+ {
2423
+ "epoch": 0.33445680456839494,
2424
+ "grad_norm": 0.330078125,
2425
+ "learning_rate": 0.001,
2426
+ "loss": 1.3264,
2427
+ "step": 4788
2428
+ },
2429
+ {
2430
+ "epoch": 0.335434748441402,
2431
+ "grad_norm": 0.279296875,
2432
+ "learning_rate": 0.001,
2433
+ "loss": 1.3142,
2434
+ "step": 4802
2435
+ },
2436
+ {
2437
+ "epoch": 0.33641269231440896,
2438
+ "grad_norm": 0.205078125,
2439
+ "learning_rate": 0.001,
2440
+ "loss": 1.3202,
2441
+ "step": 4816
2442
+ },
2443
+ {
2444
+ "epoch": 0.33739063618741594,
2445
+ "grad_norm": 0.234375,
2446
+ "learning_rate": 0.001,
2447
+ "loss": 1.3516,
2448
+ "step": 4830
2449
+ },
2450
+ {
2451
+ "epoch": 0.338368580060423,
2452
+ "grad_norm": 0.259765625,
2453
+ "learning_rate": 0.001,
2454
+ "loss": 1.3365,
2455
+ "step": 4844
2456
+ },
2457
+ {
2458
+ "epoch": 0.33934652393342996,
2459
+ "grad_norm": 0.259765625,
2460
+ "learning_rate": 0.001,
2461
+ "loss": 1.3334,
2462
+ "step": 4858
2463
+ },
2464
+ {
2465
+ "epoch": 0.340324467806437,
2466
+ "grad_norm": 0.421875,
2467
+ "learning_rate": 0.001,
2468
+ "loss": 1.3485,
2469
+ "step": 4872
2470
+ },
2471
+ {
2472
+ "epoch": 0.341302411679444,
2473
+ "grad_norm": 0.265625,
2474
+ "learning_rate": 0.001,
2475
+ "loss": 1.3343,
2476
+ "step": 4886
2477
+ },
2478
+ {
2479
+ "epoch": 0.34228035555245095,
2480
+ "grad_norm": 0.380859375,
2481
+ "learning_rate": 0.001,
2482
+ "loss": 1.3395,
2483
+ "step": 4900
2484
+ },
2485
+ {
2486
+ "epoch": 0.343258299425458,
2487
+ "grad_norm": 0.359375,
2488
+ "learning_rate": 0.001,
2489
+ "loss": 1.3193,
2490
+ "step": 4914
2491
+ },
2492
+ {
2493
+ "epoch": 0.344236243298465,
2494
+ "grad_norm": 0.333984375,
2495
+ "learning_rate": 0.001,
2496
+ "loss": 1.327,
2497
+ "step": 4928
2498
+ },
2499
+ {
2500
+ "epoch": 0.34521418717147195,
2501
+ "grad_norm": 0.27734375,
2502
+ "learning_rate": 0.001,
2503
+ "loss": 1.2993,
2504
+ "step": 4942
2505
+ },
2506
+ {
2507
+ "epoch": 0.346192131044479,
2508
+ "grad_norm": 0.3828125,
2509
+ "learning_rate": 0.001,
2510
+ "loss": 1.3182,
2511
+ "step": 4956
2512
+ },
2513
+ {
2514
+ "epoch": 0.34717007491748597,
2515
+ "grad_norm": 0.2333984375,
2516
+ "learning_rate": 0.001,
2517
+ "loss": 1.3409,
2518
+ "step": 4970
2519
+ },
2520
+ {
2521
+ "epoch": 0.348148018790493,
2522
+ "grad_norm": 0.345703125,
2523
+ "learning_rate": 0.001,
2524
+ "loss": 1.3354,
2525
+ "step": 4984
2526
+ },
2527
+ {
2528
+ "epoch": 0.3491259626635,
2529
+ "grad_norm": 0.28125,
2530
+ "learning_rate": 0.001,
2531
+ "loss": 1.305,
2532
+ "step": 4998
2533
+ },
2534
+ {
2535
+ "epoch": 0.35010390653650697,
2536
+ "grad_norm": 0.2216796875,
2537
+ "learning_rate": 0.001,
2538
+ "loss": 1.314,
2539
+ "step": 5012
2540
+ },
2541
+ {
2542
+ "epoch": 0.351081850409514,
2543
+ "grad_norm": 0.328125,
2544
+ "learning_rate": 0.001,
2545
+ "loss": 1.3267,
2546
+ "step": 5026
2547
+ },
2548
+ {
2549
+ "epoch": 0.352059794282521,
2550
+ "grad_norm": 0.42578125,
2551
+ "learning_rate": 0.001,
2552
+ "loss": 1.3327,
2553
+ "step": 5040
2554
+ },
2555
+ {
2556
+ "epoch": 0.353037738155528,
2557
+ "grad_norm": 0.30078125,
2558
+ "learning_rate": 0.001,
2559
+ "loss": 1.3113,
2560
+ "step": 5054
2561
+ },
2562
+ {
2563
+ "epoch": 0.354015682028535,
2564
+ "grad_norm": 0.3984375,
2565
+ "learning_rate": 0.001,
2566
+ "loss": 1.3293,
2567
+ "step": 5068
2568
+ },
2569
+ {
2570
+ "epoch": 0.354993625901542,
2571
+ "grad_norm": 0.294921875,
2572
+ "learning_rate": 0.001,
2573
+ "loss": 1.3094,
2574
+ "step": 5082
2575
+ },
2576
+ {
2577
+ "epoch": 0.355971569774549,
2578
+ "grad_norm": 0.259765625,
2579
+ "learning_rate": 0.001,
2580
+ "loss": 1.3053,
2581
+ "step": 5096
2582
+ },
2583
+ {
2584
+ "epoch": 0.356949513647556,
2585
+ "grad_norm": 0.2109375,
2586
+ "learning_rate": 0.001,
2587
+ "loss": 1.325,
2588
+ "step": 5110
2589
+ },
2590
+ {
2591
+ "epoch": 0.35792745752056304,
2592
+ "grad_norm": 0.337890625,
2593
+ "learning_rate": 0.001,
2594
+ "loss": 1.3392,
2595
+ "step": 5124
2596
+ },
2597
+ {
2598
+ "epoch": 0.35890540139357,
2599
+ "grad_norm": 0.412109375,
2600
+ "learning_rate": 0.001,
2601
+ "loss": 1.3363,
2602
+ "step": 5138
2603
+ },
2604
+ {
2605
+ "epoch": 0.359883345266577,
2606
+ "grad_norm": 0.59375,
2607
+ "learning_rate": 0.001,
2608
+ "loss": 1.3693,
2609
+ "step": 5152
2610
+ },
2611
+ {
2612
+ "epoch": 0.36086128913958404,
2613
+ "grad_norm": 0.390625,
2614
+ "learning_rate": 0.001,
2615
+ "loss": 1.3341,
2616
+ "step": 5166
2617
+ },
2618
+ {
2619
+ "epoch": 0.361839233012591,
2620
+ "grad_norm": 0.255859375,
2621
+ "learning_rate": 0.001,
2622
+ "loss": 1.3104,
2623
+ "step": 5180
2624
+ },
2625
+ {
2626
+ "epoch": 0.362817176885598,
2627
+ "grad_norm": 0.328125,
2628
+ "learning_rate": 0.001,
2629
+ "loss": 1.3277,
2630
+ "step": 5194
2631
+ },
2632
+ {
2633
+ "epoch": 0.36379512075860504,
2634
+ "grad_norm": 0.2275390625,
2635
+ "learning_rate": 0.001,
2636
+ "loss": 1.3274,
2637
+ "step": 5208
2638
+ },
2639
+ {
2640
+ "epoch": 0.364773064631612,
2641
+ "grad_norm": 0.26953125,
2642
+ "learning_rate": 0.001,
2643
+ "loss": 1.3295,
2644
+ "step": 5222
2645
+ },
2646
+ {
2647
+ "epoch": 0.36575100850461906,
2648
+ "grad_norm": 0.2490234375,
2649
+ "learning_rate": 0.001,
2650
+ "loss": 1.334,
2651
+ "step": 5236
2652
+ },
2653
+ {
2654
+ "epoch": 0.36672895237762604,
2655
+ "grad_norm": 0.318359375,
2656
+ "learning_rate": 0.001,
2657
+ "loss": 1.3081,
2658
+ "step": 5250
2659
+ },
2660
+ {
2661
+ "epoch": 0.367706896250633,
2662
+ "grad_norm": 0.232421875,
2663
+ "learning_rate": 0.001,
2664
+ "loss": 1.2691,
2665
+ "step": 5264
2666
+ },
2667
+ {
2668
+ "epoch": 0.36868484012364006,
2669
+ "grad_norm": 0.47265625,
2670
+ "learning_rate": 0.001,
2671
+ "loss": 1.3241,
2672
+ "step": 5278
2673
+ },
2674
+ {
2675
+ "epoch": 0.36966278399664704,
2676
+ "grad_norm": 0.25390625,
2677
+ "learning_rate": 0.001,
2678
+ "loss": 1.3015,
2679
+ "step": 5292
2680
+ },
2681
+ {
2682
+ "epoch": 0.3706407278696541,
2683
+ "grad_norm": 0.30078125,
2684
+ "learning_rate": 0.001,
2685
+ "loss": 1.3269,
2686
+ "step": 5306
2687
+ },
2688
+ {
2689
+ "epoch": 0.37161867174266106,
2690
+ "grad_norm": 0.23046875,
2691
+ "learning_rate": 0.001,
2692
+ "loss": 1.3249,
2693
+ "step": 5320
2694
+ },
2695
+ {
2696
+ "epoch": 0.37259661561566804,
2697
+ "grad_norm": 0.384765625,
2698
+ "learning_rate": 0.001,
2699
+ "loss": 1.3091,
2700
+ "step": 5334
2701
+ },
2702
+ {
2703
+ "epoch": 0.3735745594886751,
2704
+ "grad_norm": 0.283203125,
2705
+ "learning_rate": 0.001,
2706
+ "loss": 1.3251,
2707
+ "step": 5348
2708
+ },
2709
+ {
2710
+ "epoch": 0.37455250336168205,
2711
+ "grad_norm": 0.263671875,
2712
+ "learning_rate": 0.001,
2713
+ "loss": 1.2866,
2714
+ "step": 5362
2715
+ },
2716
+ {
2717
+ "epoch": 0.3755304472346891,
2718
+ "grad_norm": 0.29296875,
2719
+ "learning_rate": 0.001,
2720
+ "loss": 1.2987,
2721
+ "step": 5376
2722
+ },
2723
+ {
2724
+ "epoch": 0.37650839110769607,
2725
+ "grad_norm": 0.306640625,
2726
+ "learning_rate": 0.001,
2727
+ "loss": 1.3095,
2728
+ "step": 5390
2729
+ },
2730
+ {
2731
+ "epoch": 0.37748633498070305,
2732
+ "grad_norm": 0.236328125,
2733
+ "learning_rate": 0.001,
2734
+ "loss": 1.304,
2735
+ "step": 5404
2736
+ },
2737
+ {
2738
+ "epoch": 0.3784642788537101,
2739
+ "grad_norm": 0.21484375,
2740
+ "learning_rate": 0.001,
2741
+ "loss": 1.3158,
2742
+ "step": 5418
2743
+ },
2744
+ {
2745
+ "epoch": 0.37944222272671707,
2746
+ "grad_norm": 0.3046875,
2747
+ "learning_rate": 0.001,
2748
+ "loss": 1.3164,
2749
+ "step": 5432
2750
+ },
2751
+ {
2752
+ "epoch": 0.3804201665997241,
2753
+ "grad_norm": 0.291015625,
2754
+ "learning_rate": 0.001,
2755
+ "loss": 1.3078,
2756
+ "step": 5446
2757
+ },
2758
+ {
2759
+ "epoch": 0.3813981104727311,
2760
+ "grad_norm": 0.365234375,
2761
+ "learning_rate": 0.001,
2762
+ "loss": 1.3603,
2763
+ "step": 5460
2764
+ },
2765
+ {
2766
+ "epoch": 0.38237605434573807,
2767
+ "grad_norm": 0.3046875,
2768
+ "learning_rate": 0.001,
2769
+ "loss": 1.3174,
2770
+ "step": 5474
2771
+ },
2772
+ {
2773
+ "epoch": 0.3833539982187451,
2774
+ "grad_norm": 0.291015625,
2775
+ "learning_rate": 0.001,
2776
+ "loss": 1.3204,
2777
+ "step": 5488
2778
+ },
2779
+ {
2780
+ "epoch": 0.3843319420917521,
2781
+ "grad_norm": 0.33203125,
2782
+ "learning_rate": 0.001,
2783
+ "loss": 1.3005,
2784
+ "step": 5502
2785
+ },
2786
+ {
2787
+ "epoch": 0.38530988596475907,
2788
+ "grad_norm": 0.3828125,
2789
+ "learning_rate": 0.001,
2790
+ "loss": 1.3396,
2791
+ "step": 5516
2792
+ },
2793
+ {
2794
+ "epoch": 0.3862878298377661,
2795
+ "grad_norm": 0.287109375,
2796
+ "learning_rate": 0.001,
2797
+ "loss": 1.3231,
2798
+ "step": 5530
2799
+ },
2800
+ {
2801
+ "epoch": 0.3872657737107731,
2802
+ "grad_norm": 0.375,
2803
+ "learning_rate": 0.001,
2804
+ "loss": 1.3353,
2805
+ "step": 5544
2806
+ },
2807
+ {
2808
+ "epoch": 0.3882437175837801,
2809
+ "grad_norm": 0.33203125,
2810
+ "learning_rate": 0.001,
2811
+ "loss": 1.3314,
2812
+ "step": 5558
2813
+ },
2814
+ {
2815
+ "epoch": 0.3892216614567871,
2816
+ "grad_norm": 0.35546875,
2817
+ "learning_rate": 0.001,
2818
+ "loss": 1.3228,
2819
+ "step": 5572
2820
+ },
2821
+ {
2822
+ "epoch": 0.3901996053297941,
2823
+ "grad_norm": 0.294921875,
2824
+ "learning_rate": 0.001,
2825
+ "loss": 1.3034,
2826
+ "step": 5586
2827
+ },
2828
+ {
2829
+ "epoch": 0.3911775492028011,
2830
+ "grad_norm": 0.349609375,
2831
+ "learning_rate": 0.001,
2832
+ "loss": 1.3307,
2833
+ "step": 5600
2834
+ },
2835
+ {
2836
+ "epoch": 0.3921554930758081,
2837
+ "grad_norm": 0.26171875,
2838
+ "learning_rate": 0.001,
2839
+ "loss": 1.2888,
2840
+ "step": 5614
2841
+ },
2842
+ {
2843
+ "epoch": 0.39313343694881514,
2844
+ "grad_norm": 0.25390625,
2845
+ "learning_rate": 0.001,
2846
+ "loss": 1.3107,
2847
+ "step": 5628
2848
+ },
2849
+ {
2850
+ "epoch": 0.3941113808218221,
2851
+ "grad_norm": 0.3984375,
2852
+ "learning_rate": 0.001,
2853
+ "loss": 1.32,
2854
+ "step": 5642
2855
+ },
2856
+ {
2857
+ "epoch": 0.3950893246948291,
2858
+ "grad_norm": 0.310546875,
2859
+ "learning_rate": 0.001,
2860
+ "loss": 1.3383,
2861
+ "step": 5656
2862
+ },
2863
+ {
2864
+ "epoch": 0.39606726856783614,
2865
+ "grad_norm": 0.291015625,
2866
+ "learning_rate": 0.001,
2867
+ "loss": 1.2903,
2868
+ "step": 5670
2869
+ },
2870
+ {
2871
+ "epoch": 0.3970452124408431,
2872
+ "grad_norm": 0.330078125,
2873
+ "learning_rate": 0.001,
2874
+ "loss": 1.3338,
2875
+ "step": 5684
2876
+ },
2877
+ {
2878
+ "epoch": 0.39802315631385016,
2879
+ "grad_norm": 0.267578125,
2880
+ "learning_rate": 0.001,
2881
+ "loss": 1.3371,
2882
+ "step": 5698
2883
+ },
2884
+ {
2885
+ "epoch": 0.39900110018685714,
2886
+ "grad_norm": 0.2265625,
2887
+ "learning_rate": 0.001,
2888
+ "loss": 1.3169,
2889
+ "step": 5712
2890
+ },
2891
+ {
2892
+ "epoch": 0.3998393377922917,
2893
+ "eval_loss": 1.7229478359222412,
2894
+ "eval_runtime": 9.1252,
2895
+ "eval_samples_per_second": 109.587,
2896
+ "eval_steps_per_second": 1.425,
2897
+ "step": 5724
2898
+ },
2899
+ {
2900
+ "epoch": 0.3999790440598641,
2901
+ "grad_norm": 0.287109375,
2902
+ "learning_rate": 0.001,
2903
+ "loss": 1.3098,
2904
+ "step": 5726
2905
+ },
2906
+ {
2907
+ "epoch": 0.40095698793287116,
2908
+ "grad_norm": 0.283203125,
2909
+ "learning_rate": 0.001,
2910
+ "loss": 1.3131,
2911
+ "step": 5740
2912
+ },
2913
+ {
2914
+ "epoch": 0.40193493180587814,
2915
+ "grad_norm": 0.408203125,
2916
+ "learning_rate": 0.001,
2917
+ "loss": 1.3199,
2918
+ "step": 5754
2919
+ },
2920
+ {
2921
+ "epoch": 0.4029128756788851,
2922
+ "grad_norm": 0.46875,
2923
+ "learning_rate": 0.001,
2924
+ "loss": 1.3238,
2925
+ "step": 5768
2926
+ },
2927
+ {
2928
+ "epoch": 0.40389081955189216,
2929
+ "grad_norm": 0.46484375,
2930
+ "learning_rate": 0.001,
2931
+ "loss": 1.3192,
2932
+ "step": 5782
2933
+ },
2934
+ {
2935
+ "epoch": 0.40486876342489914,
2936
+ "grad_norm": 0.31640625,
2937
+ "learning_rate": 0.001,
2938
+ "loss": 1.3274,
2939
+ "step": 5796
2940
+ },
2941
+ {
2942
+ "epoch": 0.4058467072979062,
2943
+ "grad_norm": 0.318359375,
2944
+ "learning_rate": 0.001,
2945
+ "loss": 1.2836,
2946
+ "step": 5810
2947
+ },
2948
+ {
2949
+ "epoch": 0.40682465117091315,
2950
+ "grad_norm": 0.353515625,
2951
+ "learning_rate": 0.001,
2952
+ "loss": 1.3021,
2953
+ "step": 5824
2954
+ },
2955
+ {
2956
+ "epoch": 0.40780259504392014,
2957
+ "grad_norm": 0.333984375,
2958
+ "learning_rate": 0.001,
2959
+ "loss": 1.2903,
2960
+ "step": 5838
2961
+ },
2962
+ {
2963
+ "epoch": 0.40878053891692717,
2964
+ "grad_norm": 0.267578125,
2965
+ "learning_rate": 0.001,
2966
+ "loss": 1.2965,
2967
+ "step": 5852
2968
+ },
2969
+ {
2970
+ "epoch": 0.40975848278993415,
2971
+ "grad_norm": 0.2421875,
2972
+ "learning_rate": 0.001,
2973
+ "loss": 1.2816,
2974
+ "step": 5866
2975
+ },
2976
+ {
2977
+ "epoch": 0.4107364266629412,
2978
+ "grad_norm": 0.306640625,
2979
+ "learning_rate": 0.001,
2980
+ "loss": 1.3082,
2981
+ "step": 5880
2982
+ },
2983
+ {
2984
+ "epoch": 0.41171437053594817,
2985
+ "grad_norm": 0.255859375,
2986
+ "learning_rate": 0.001,
2987
+ "loss": 1.3229,
2988
+ "step": 5894
2989
+ },
2990
+ {
2991
+ "epoch": 0.41269231440895515,
2992
+ "grad_norm": 0.291015625,
2993
+ "learning_rate": 0.001,
2994
+ "loss": 1.3056,
2995
+ "step": 5908
2996
+ },
2997
+ {
2998
+ "epoch": 0.4136702582819622,
2999
+ "grad_norm": 0.310546875,
3000
+ "learning_rate": 0.001,
3001
+ "loss": 1.2618,
3002
+ "step": 5922
3003
+ },
3004
+ {
3005
+ "epoch": 0.41464820215496917,
3006
+ "grad_norm": 0.24609375,
3007
+ "learning_rate": 0.001,
3008
+ "loss": 1.29,
3009
+ "step": 5936
3010
+ },
3011
+ {
3012
+ "epoch": 0.4156261460279762,
3013
+ "grad_norm": 0.36328125,
3014
+ "learning_rate": 0.001,
3015
+ "loss": 1.3132,
3016
+ "step": 5950
3017
+ },
3018
+ {
3019
+ "epoch": 0.4166040899009832,
3020
+ "grad_norm": 0.271484375,
3021
+ "learning_rate": 0.001,
3022
+ "loss": 1.3083,
3023
+ "step": 5964
3024
+ },
3025
+ {
3026
+ "epoch": 0.41758203377399017,
3027
+ "grad_norm": 0.326171875,
3028
+ "learning_rate": 0.001,
3029
+ "loss": 1.3118,
3030
+ "step": 5978
3031
+ },
3032
+ {
3033
+ "epoch": 0.4185599776469972,
3034
+ "grad_norm": 0.30078125,
3035
+ "learning_rate": 0.001,
3036
+ "loss": 1.3105,
3037
+ "step": 5992
3038
+ },
3039
+ {
3040
+ "epoch": 0.4195379215200042,
3041
+ "grad_norm": 0.25,
3042
+ "learning_rate": 0.001,
3043
+ "loss": 1.3032,
3044
+ "step": 6006
3045
+ },
3046
+ {
3047
+ "epoch": 0.42051586539301117,
3048
+ "grad_norm": 0.232421875,
3049
+ "learning_rate": 0.001,
3050
+ "loss": 1.2815,
3051
+ "step": 6020
3052
+ },
3053
+ {
3054
+ "epoch": 0.4214938092660182,
3055
+ "grad_norm": 0.263671875,
3056
+ "learning_rate": 0.001,
3057
+ "loss": 1.2913,
3058
+ "step": 6034
3059
+ },
3060
+ {
3061
+ "epoch": 0.4224717531390252,
3062
+ "grad_norm": 0.2353515625,
3063
+ "learning_rate": 0.001,
3064
+ "loss": 1.3015,
3065
+ "step": 6048
3066
+ },
3067
+ {
3068
+ "epoch": 0.4234496970120322,
3069
+ "grad_norm": 0.291015625,
3070
+ "learning_rate": 0.001,
3071
+ "loss": 1.3072,
3072
+ "step": 6062
3073
+ },
3074
+ {
3075
+ "epoch": 0.4244276408850392,
3076
+ "grad_norm": 0.251953125,
3077
+ "learning_rate": 0.001,
3078
+ "loss": 1.3412,
3079
+ "step": 6076
3080
+ },
3081
+ {
3082
+ "epoch": 0.4254055847580462,
3083
+ "grad_norm": 0.30859375,
3084
+ "learning_rate": 0.001,
3085
+ "loss": 1.3024,
3086
+ "step": 6090
3087
+ },
3088
+ {
3089
+ "epoch": 0.4263835286310532,
3090
+ "grad_norm": 0.2255859375,
3091
+ "learning_rate": 0.001,
3092
+ "loss": 1.3124,
3093
+ "step": 6104
3094
+ },
3095
+ {
3096
+ "epoch": 0.4273614725040602,
3097
+ "grad_norm": 0.55078125,
3098
+ "learning_rate": 0.001,
3099
+ "loss": 1.3066,
3100
+ "step": 6118
3101
+ },
3102
+ {
3103
+ "epoch": 0.42833941637706724,
3104
+ "grad_norm": 0.353515625,
3105
+ "learning_rate": 0.001,
3106
+ "loss": 1.3194,
3107
+ "step": 6132
3108
+ },
3109
+ {
3110
+ "epoch": 0.4293173602500742,
3111
+ "grad_norm": 0.314453125,
3112
+ "learning_rate": 0.001,
3113
+ "loss": 1.2943,
3114
+ "step": 6146
3115
+ },
3116
+ {
3117
+ "epoch": 0.4302953041230812,
3118
+ "grad_norm": 0.279296875,
3119
+ "learning_rate": 0.001,
3120
+ "loss": 1.2726,
3121
+ "step": 6160
3122
+ },
3123
+ {
3124
+ "epoch": 0.43127324799608824,
3125
+ "grad_norm": 0.2158203125,
3126
+ "learning_rate": 0.001,
3127
+ "loss": 1.2693,
3128
+ "step": 6174
3129
+ },
3130
+ {
3131
+ "epoch": 0.4322511918690952,
3132
+ "grad_norm": 0.240234375,
3133
+ "learning_rate": 0.001,
3134
+ "loss": 1.2966,
3135
+ "step": 6188
3136
+ },
3137
+ {
3138
+ "epoch": 0.43322913574210226,
3139
+ "grad_norm": 0.345703125,
3140
+ "learning_rate": 0.001,
3141
+ "loss": 1.3144,
3142
+ "step": 6202
3143
+ },
3144
+ {
3145
+ "epoch": 0.43420707961510924,
3146
+ "grad_norm": 0.255859375,
3147
+ "learning_rate": 0.001,
3148
+ "loss": 1.2774,
3149
+ "step": 6216
3150
+ },
3151
+ {
3152
+ "epoch": 0.4351850234881162,
3153
+ "grad_norm": 0.3046875,
3154
+ "learning_rate": 0.001,
3155
+ "loss": 1.3109,
3156
+ "step": 6230
3157
+ },
3158
+ {
3159
+ "epoch": 0.43616296736112325,
3160
+ "grad_norm": 0.265625,
3161
+ "learning_rate": 0.001,
3162
+ "loss": 1.2897,
3163
+ "step": 6244
3164
+ },
3165
+ {
3166
+ "epoch": 0.43714091123413024,
3167
+ "grad_norm": 0.216796875,
3168
+ "learning_rate": 0.001,
3169
+ "loss": 1.2796,
3170
+ "step": 6258
3171
+ },
3172
+ {
3173
+ "epoch": 0.4381188551071372,
3174
+ "grad_norm": 0.3125,
3175
+ "learning_rate": 0.001,
3176
+ "loss": 1.3235,
3177
+ "step": 6272
3178
+ },
3179
+ {
3180
+ "epoch": 0.43909679898014425,
3181
+ "grad_norm": 0.265625,
3182
+ "learning_rate": 0.001,
3183
+ "loss": 1.2996,
3184
+ "step": 6286
3185
+ },
3186
+ {
3187
+ "epoch": 0.44007474285315124,
3188
+ "grad_norm": 0.32421875,
3189
+ "learning_rate": 0.001,
3190
+ "loss": 1.3074,
3191
+ "step": 6300
3192
+ },
3193
+ {
3194
+ "epoch": 0.44105268672615827,
3195
+ "grad_norm": 0.34375,
3196
+ "learning_rate": 0.001,
3197
+ "loss": 1.2851,
3198
+ "step": 6314
3199
+ },
3200
+ {
3201
+ "epoch": 0.44203063059916525,
3202
+ "grad_norm": 0.345703125,
3203
+ "learning_rate": 0.001,
3204
+ "loss": 1.3315,
3205
+ "step": 6328
3206
+ },
3207
+ {
3208
+ "epoch": 0.44300857447217223,
3209
+ "grad_norm": 0.2421875,
3210
+ "learning_rate": 0.001,
3211
+ "loss": 1.3049,
3212
+ "step": 6342
3213
+ },
3214
+ {
3215
+ "epoch": 0.44398651834517927,
3216
+ "grad_norm": 0.263671875,
3217
+ "learning_rate": 0.001,
3218
+ "loss": 1.3186,
3219
+ "step": 6356
3220
+ },
3221
+ {
3222
+ "epoch": 0.44496446221818625,
3223
+ "grad_norm": 0.392578125,
3224
+ "learning_rate": 0.001,
3225
+ "loss": 1.276,
3226
+ "step": 6370
3227
+ },
3228
+ {
3229
+ "epoch": 0.4459424060911933,
3230
+ "grad_norm": 0.3828125,
3231
+ "learning_rate": 0.001,
3232
+ "loss": 1.2827,
3233
+ "step": 6384
3234
+ },
3235
+ {
3236
+ "epoch": 0.44692034996420027,
3237
+ "grad_norm": 0.310546875,
3238
+ "learning_rate": 0.001,
3239
+ "loss": 1.2925,
3240
+ "step": 6398
3241
+ },
3242
+ {
3243
+ "epoch": 0.44789829383720725,
3244
+ "grad_norm": 0.2294921875,
3245
+ "learning_rate": 0.001,
3246
+ "loss": 1.2788,
3247
+ "step": 6412
3248
+ },
3249
+ {
3250
+ "epoch": 0.4488762377102143,
3251
+ "grad_norm": 0.32421875,
3252
+ "learning_rate": 0.001,
3253
+ "loss": 1.2982,
3254
+ "step": 6426
3255
+ },
3256
+ {
3257
+ "epoch": 0.44985418158322127,
3258
+ "grad_norm": 0.265625,
3259
+ "learning_rate": 0.001,
3260
+ "loss": 1.2977,
3261
+ "step": 6440
3262
+ },
3263
+ {
3264
+ "epoch": 0.4508321254562283,
3265
+ "grad_norm": 0.349609375,
3266
+ "learning_rate": 0.001,
3267
+ "loss": 1.3041,
3268
+ "step": 6454
3269
+ },
3270
+ {
3271
+ "epoch": 0.4518100693292353,
3272
+ "grad_norm": 0.30859375,
3273
+ "learning_rate": 0.001,
3274
+ "loss": 1.2828,
3275
+ "step": 6468
3276
+ },
3277
+ {
3278
+ "epoch": 0.45278801320224227,
3279
+ "grad_norm": 0.291015625,
3280
+ "learning_rate": 0.001,
3281
+ "loss": 1.2942,
3282
+ "step": 6482
3283
+ },
3284
+ {
3285
+ "epoch": 0.4537659570752493,
3286
+ "grad_norm": 0.26953125,
3287
+ "learning_rate": 0.001,
3288
+ "loss": 1.2626,
3289
+ "step": 6496
3290
+ },
3291
+ {
3292
+ "epoch": 0.4547439009482563,
3293
+ "grad_norm": 0.32421875,
3294
+ "learning_rate": 0.001,
3295
+ "loss": 1.2788,
3296
+ "step": 6510
3297
+ },
3298
+ {
3299
+ "epoch": 0.45572184482126327,
3300
+ "grad_norm": 0.349609375,
3301
+ "learning_rate": 0.001,
3302
+ "loss": 1.2733,
3303
+ "step": 6524
3304
+ },
3305
+ {
3306
+ "epoch": 0.4566997886942703,
3307
+ "grad_norm": 0.271484375,
3308
+ "learning_rate": 0.001,
3309
+ "loss": 1.3146,
3310
+ "step": 6538
3311
+ },
3312
+ {
3313
+ "epoch": 0.4576777325672773,
3314
+ "grad_norm": 0.302734375,
3315
+ "learning_rate": 0.001,
3316
+ "loss": 1.3027,
3317
+ "step": 6552
3318
+ },
3319
+ {
3320
+ "epoch": 0.4586556764402843,
3321
+ "grad_norm": 0.28125,
3322
+ "learning_rate": 0.001,
3323
+ "loss": 1.2753,
3324
+ "step": 6566
3325
+ },
3326
+ {
3327
+ "epoch": 0.4596336203132913,
3328
+ "grad_norm": 0.439453125,
3329
+ "learning_rate": 0.001,
3330
+ "loss": 1.2955,
3331
+ "step": 6580
3332
+ },
3333
+ {
3334
+ "epoch": 0.4606115641862983,
3335
+ "grad_norm": 0.353515625,
3336
+ "learning_rate": 0.001,
3337
+ "loss": 1.3105,
3338
+ "step": 6594
3339
+ },
3340
+ {
3341
+ "epoch": 0.4615895080593053,
3342
+ "grad_norm": 0.28515625,
3343
+ "learning_rate": 0.001,
3344
+ "loss": 1.2798,
3345
+ "step": 6608
3346
+ },
3347
+ {
3348
+ "epoch": 0.4625674519323123,
3349
+ "grad_norm": 0.28515625,
3350
+ "learning_rate": 0.001,
3351
+ "loss": 1.2983,
3352
+ "step": 6622
3353
+ },
3354
+ {
3355
+ "epoch": 0.46354539580531934,
3356
+ "grad_norm": 0.52734375,
3357
+ "learning_rate": 0.001,
3358
+ "loss": 1.3166,
3359
+ "step": 6636
3360
+ },
3361
+ {
3362
+ "epoch": 0.4645233396783263,
3363
+ "grad_norm": 0.287109375,
3364
+ "learning_rate": 0.001,
3365
+ "loss": 1.287,
3366
+ "step": 6650
3367
+ },
3368
+ {
3369
+ "epoch": 0.4655012835513333,
3370
+ "grad_norm": 0.298828125,
3371
+ "learning_rate": 0.001,
3372
+ "loss": 1.28,
3373
+ "step": 6664
3374
+ },
3375
+ {
3376
+ "epoch": 0.46647922742434034,
3377
+ "grad_norm": 0.380859375,
3378
+ "learning_rate": 0.001,
3379
+ "loss": 1.2896,
3380
+ "step": 6678
3381
+ },
3382
+ {
3383
+ "epoch": 0.4674571712973473,
3384
+ "grad_norm": 0.298828125,
3385
+ "learning_rate": 0.001,
3386
+ "loss": 1.2933,
3387
+ "step": 6692
3388
+ },
3389
+ {
3390
+ "epoch": 0.46843511517035435,
3391
+ "grad_norm": 0.34375,
3392
+ "learning_rate": 0.001,
3393
+ "loss": 1.2602,
3394
+ "step": 6706
3395
+ },
3396
+ {
3397
+ "epoch": 0.46941305904336134,
3398
+ "grad_norm": 0.25,
3399
+ "learning_rate": 0.001,
3400
+ "loss": 1.267,
3401
+ "step": 6720
3402
+ },
3403
+ {
3404
+ "epoch": 0.4703910029163683,
3405
+ "grad_norm": 0.287109375,
3406
+ "learning_rate": 0.001,
3407
+ "loss": 1.2745,
3408
+ "step": 6734
3409
+ },
3410
+ {
3411
+ "epoch": 0.47136894678937535,
3412
+ "grad_norm": 0.453125,
3413
+ "learning_rate": 0.001,
3414
+ "loss": 1.3102,
3415
+ "step": 6748
3416
+ },
3417
+ {
3418
+ "epoch": 0.47234689066238233,
3419
+ "grad_norm": 0.36328125,
3420
+ "learning_rate": 0.001,
3421
+ "loss": 1.2844,
3422
+ "step": 6762
3423
+ },
3424
+ {
3425
+ "epoch": 0.47332483453538937,
3426
+ "grad_norm": 0.404296875,
3427
+ "learning_rate": 0.001,
3428
+ "loss": 1.2787,
3429
+ "step": 6776
3430
+ },
3431
+ {
3432
+ "epoch": 0.47430277840839635,
3433
+ "grad_norm": 0.41796875,
3434
+ "learning_rate": 0.001,
3435
+ "loss": 1.2921,
3436
+ "step": 6790
3437
+ },
3438
+ {
3439
+ "epoch": 0.47528072228140333,
3440
+ "grad_norm": 0.447265625,
3441
+ "learning_rate": 0.001,
3442
+ "loss": 1.3157,
3443
+ "step": 6804
3444
+ },
3445
+ {
3446
+ "epoch": 0.47625866615441037,
3447
+ "grad_norm": 0.380859375,
3448
+ "learning_rate": 0.001,
3449
+ "loss": 1.2932,
3450
+ "step": 6818
3451
+ },
3452
+ {
3453
+ "epoch": 0.47723661002741735,
3454
+ "grad_norm": 0.296875,
3455
+ "learning_rate": 0.001,
3456
+ "loss": 1.3011,
3457
+ "step": 6832
3458
+ },
3459
+ {
3460
+ "epoch": 0.47821455390042433,
3461
+ "grad_norm": 0.333984375,
3462
+ "learning_rate": 0.001,
3463
+ "loss": 1.2883,
3464
+ "step": 6846
3465
+ },
3466
+ {
3467
+ "epoch": 0.47919249777343137,
3468
+ "grad_norm": 0.2431640625,
3469
+ "learning_rate": 0.001,
3470
+ "loss": 1.3087,
3471
+ "step": 6860
3472
+ },
3473
+ {
3474
+ "epoch": 0.48017044164643835,
3475
+ "grad_norm": 0.259765625,
3476
+ "learning_rate": 0.001,
3477
+ "loss": 1.2855,
3478
+ "step": 6874
3479
+ },
3480
+ {
3481
+ "epoch": 0.4811483855194454,
3482
+ "grad_norm": 0.267578125,
3483
+ "learning_rate": 0.001,
3484
+ "loss": 1.3106,
3485
+ "step": 6888
3486
+ },
3487
+ {
3488
+ "epoch": 0.48212632939245237,
3489
+ "grad_norm": 0.29296875,
3490
+ "learning_rate": 0.001,
3491
+ "loss": 1.2869,
3492
+ "step": 6902
3493
+ },
3494
+ {
3495
+ "epoch": 0.48310427326545935,
3496
+ "grad_norm": 0.53125,
3497
+ "learning_rate": 0.001,
3498
+ "loss": 1.2852,
3499
+ "step": 6916
3500
+ },
3501
+ {
3502
+ "epoch": 0.4840822171384664,
3503
+ "grad_norm": 0.46484375,
3504
+ "learning_rate": 0.001,
3505
+ "loss": 1.2995,
3506
+ "step": 6930
3507
+ },
3508
+ {
3509
+ "epoch": 0.48506016101147337,
3510
+ "grad_norm": 0.3203125,
3511
+ "learning_rate": 0.001,
3512
+ "loss": 1.3098,
3513
+ "step": 6944
3514
+ },
3515
+ {
3516
+ "epoch": 0.4860381048844804,
3517
+ "grad_norm": 0.322265625,
3518
+ "learning_rate": 0.001,
3519
+ "loss": 1.2687,
3520
+ "step": 6958
3521
+ },
3522
+ {
3523
+ "epoch": 0.4870160487574874,
3524
+ "grad_norm": 0.400390625,
3525
+ "learning_rate": 0.001,
3526
+ "loss": 1.2885,
3527
+ "step": 6972
3528
+ },
3529
+ {
3530
+ "epoch": 0.48799399263049437,
3531
+ "grad_norm": 0.302734375,
3532
+ "learning_rate": 0.001,
3533
+ "loss": 1.3135,
3534
+ "step": 6986
3535
+ },
3536
+ {
3537
+ "epoch": 0.4889719365035014,
3538
+ "grad_norm": 0.314453125,
3539
+ "learning_rate": 0.001,
3540
+ "loss": 1.2776,
3541
+ "step": 7000
3542
+ },
3543
+ {
3544
+ "epoch": 0.4899498803765084,
3545
+ "grad_norm": 0.271484375,
3546
+ "learning_rate": 0.001,
3547
+ "loss": 1.2761,
3548
+ "step": 7014
3549
+ },
3550
+ {
3551
+ "epoch": 0.4909278242495154,
3552
+ "grad_norm": 0.326171875,
3553
+ "learning_rate": 0.001,
3554
+ "loss": 1.2805,
3555
+ "step": 7028
3556
+ },
3557
+ {
3558
+ "epoch": 0.4919057681225224,
3559
+ "grad_norm": 0.310546875,
3560
+ "learning_rate": 0.001,
3561
+ "loss": 1.2836,
3562
+ "step": 7042
3563
+ },
3564
+ {
3565
+ "epoch": 0.4928837119955294,
3566
+ "grad_norm": 0.3203125,
3567
+ "learning_rate": 0.001,
3568
+ "loss": 1.3029,
3569
+ "step": 7056
3570
+ },
3571
+ {
3572
+ "epoch": 0.4938616558685364,
3573
+ "grad_norm": 0.306640625,
3574
+ "learning_rate": 0.001,
3575
+ "loss": 1.2929,
3576
+ "step": 7070
3577
+ },
3578
+ {
3579
+ "epoch": 0.4948395997415434,
3580
+ "grad_norm": 0.80859375,
3581
+ "learning_rate": 0.001,
3582
+ "loss": 1.2995,
3583
+ "step": 7084
3584
+ },
3585
+ {
3586
+ "epoch": 0.4958175436145504,
3587
+ "grad_norm": 0.4609375,
3588
+ "learning_rate": 0.001,
3589
+ "loss": 1.2788,
3590
+ "step": 7098
3591
+ },
3592
+ {
3593
+ "epoch": 0.4967954874875574,
3594
+ "grad_norm": 0.279296875,
3595
+ "learning_rate": 0.001,
3596
+ "loss": 1.276,
3597
+ "step": 7112
3598
+ },
3599
+ {
3600
+ "epoch": 0.4977734313605644,
3601
+ "grad_norm": 0.28515625,
3602
+ "learning_rate": 0.001,
3603
+ "loss": 1.2883,
3604
+ "step": 7126
3605
+ },
3606
+ {
3607
+ "epoch": 0.49875137523357144,
3608
+ "grad_norm": 0.267578125,
3609
+ "learning_rate": 0.001,
3610
+ "loss": 1.2803,
3611
+ "step": 7140
3612
+ },
3613
+ {
3614
+ "epoch": 0.4997293191065784,
3615
+ "grad_norm": 0.22265625,
3616
+ "learning_rate": 0.001,
3617
+ "loss": 1.2576,
3618
+ "step": 7154
3619
+ },
3620
+ {
3621
+ "epoch": 0.49979917224036463,
3622
+ "eval_loss": 1.6829583644866943,
3623
+ "eval_runtime": 9.1256,
3624
+ "eval_samples_per_second": 109.582,
3625
+ "eval_steps_per_second": 1.425,
3626
+ "step": 7155
3627
+ }
3628
+ ],
3629
+ "logging_steps": 14,
3630
+ "max_steps": 14315,
3631
+ "num_input_tokens_seen": 0,
3632
+ "num_train_epochs": 1,
3633
+ "save_steps": 1431,
3634
+ "stateful_callbacks": {
3635
+ "TrainerControl": {
3636
+ "args": {
3637
+ "should_epoch_stop": false,
3638
+ "should_evaluate": false,
3639
+ "should_log": false,
3640
+ "should_save": true,
3641
+ "should_training_stop": false
3642
+ },
3643
+ "attributes": {}
3644
+ }
3645
+ },
3646
+ "total_flos": 1.906493869248217e+18,
3647
+ "train_batch_size": 8,
3648
+ "trial_name": null,
3649
+ "trial_params": null
3650
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa27b488e0cf856b98999f39376ab09cec60242a7aeed9d5c299aa11d2ab25e3
3
+ size 5496
vocab.json ADDED
The diff for this file is too large to render. See raw diff