Hugging Face's logo

benjamin
/
roberta-base-wechsel-chinese

Fill-Mask
Transformers
PyTorch
Chinese
roberta
Inference Endpoints
Model card Files Community
1
benjamin commited on
Commit
c596805
•
1 Parent(s): 1b5ce61

initial commit

Browse files
Files changed (20) hide show
  1. README.md +76 -0
  2. config.json +28 -0
  3. merges.txt +0 -0
  4. optimizer.pt +3 -0
  5. pytorch_model.bin +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. scheduler.pt +3 -0
  15. special_tokens_map.json +1 -0
  16. tokenizer.json +0 -0
  17. tokenizer_config.json +1 -0
  18. trainer_state.json +3176 -0
  19. training_args.bin +3 -0
  20. vocab.json +0 -0
README.md ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: zh
3
+ license: mit
4
+ ---
5
+
6
+ # roberta-base-wechsel-chinese
7
+
8
+ Model trained with WECHSEL: Effective initialization of subword embeddings for cross-lingual transfer of monolingual language models.
9
+
10
+ See the code here: https://github.com/CPJKU/wechsel
11
+
12
+ And the paper here: https://arxiv.org/abs/2112.06598
13
+
14
+ ## Performance
15
+
16
+ ### RoBERTa
17
+
18
+ | Model | NLI Score | NER Score | Avg Score |
19
+ |---|---|---|---|
20
+ | `roberta-base-wechsel-french` | **82.43** | **90.88** | **86.65** |
21
+ | `camembert-base` | 80.88 | 90.26 | 85.57 |
22
+
23
+
24
+ | Model | NLI Score | NER Score | Avg Score |
25
+ |---|---|---|---|
26
+ | `roberta-base-wechsel-german` | **81.79** | **89.72** | **85.76** |
27
+ | `deepset/gbert-base` | 78.64 | 89.46 | 84.05 |
28
+
29
+ | Model | NLI Score | NER Score | Avg Score |
30
+ |---|---|---|---|
31
+ | `roberta-base-wechsel-chinese` | **78.32** | 80.55 | **79.44** |
32
+ | `bert-base-chinese` | 76.55 | **82.05** | 79.30 |
33
+
34
+ | Model | NLI Score | NER Score | Avg Score |
35
+ |---|---|---|---|
36
+ | `roberta-base-wechsel-swahili` | **75.05** | **87.39** | **81.22** |
37
+ | `xlm-roberta-base` | 69.18 | 87.37 | 78.28 |
38
+
39
+ ### GPT2
40
+
41
+ | Model | PPL |
42
+ |---|---|
43
+ | `gpt2-wechsel-french` | **19.71** |
44
+ | `gpt2` (retrained from scratch) | 20.47 |
45
+
46
+ | Model | PPL |
47
+ |---|---|
48
+ | `gpt2-wechsel-german` | **26.8** |
49
+ | `gpt2` (retrained from scratch) | 27.63 |
50
+
51
+ | Model | PPL |
52
+ |---|---|
53
+ | `gpt2-wechsel-chinese` | **51.97** |
54
+ | `gpt2` (retrained from scratch) | 52.98 |
55
+
56
+ | Model | PPL |
57
+ |---|---|
58
+ | `gpt2-wechsel-swahili` | **10.14** |
59
+ | `gpt2` (retrained from scratch) | 10.58 |
60
+
61
+ See our paper for details.
62
+
63
+ ## Citation
64
+
65
+ Please cite WECHSEL as
66
+
67
+ ```
68
+ @misc{minixhofer2021wechsel,
69
+ title={WECHSEL: Effective initialization of subword embeddings for cross-lingual transfer of monolingual language models},
70
+ author={Benjamin Minixhofer and Fabian Paischer and Navid Rekabsaz},
71
+ year={2021},
72
+ eprint={2112.06598},
73
+ archivePrefix={arXiv},
74
+ primaryClass={cs.CL}
75
+ }
76
+ ```
config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "roberta-base",
3
+ "architectures": [
4
+ "RobertaForMaskedLM"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": 0,
8
+ "classifier_dropout": null,
9
+ "eos_token_id": 2,
10
+ "gradient_checkpointing": false,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 768,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 3072,
16
+ "layer_norm_eps": 1e-05,
17
+ "max_position_embeddings": 514,
18
+ "model_type": "roberta",
19
+ "num_attention_heads": 12,
20
+ "num_hidden_layers": 12,
21
+ "pad_token_id": 1,
22
+ "position_embedding_type": "absolute",
23
+ "torch_dtype": "float32",
24
+ "transformers_version": "4.12.0.dev0",
25
+ "type_vocab_size": 1,
26
+ "use_cache": true,
27
+ "vocab_size": 50265
28
+ }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e09445db4913a4cc5accc844d2829afb9c001bfc75f48c2bc9661b14135b0afa
3
+ size 997697817
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ab38287941ad0c498304b297ba22612ce6c550384462f511aa22af5a32447de4
3
+ size 498862437
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc4e5b358cbde16cbd817a521219a8e73e7bbc1fc6df5b9d99ac13dccb08a310
3
+ size 13620
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a35911494b275d7a6101b8cd2909f29e3a307728beabf86f6d2e20355ba0a89
3
+ size 13620
rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:03bf10213cb231b55052ba749f59daf593cb3766b4866c37a86c778d17e79d24
3
+ size 13619
rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eab9ca81aaf52835bf2baf801db53f77bec2238bd2b8a40e2d19c9c81fd23c21
3
+ size 13620
rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b4b04d14bd342edeadda7d8a085e8c432f4668f81df4b571a78efbb83a31d49
3
+ size 13619
rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3057f2c1791ad3f817f0b725d00b8eae4bba59a47707c8ace9719c56b129ff59
3
+ size 13620
rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d31443a34da847df7d2f93c6e0cb437f9c17c4fa63cd13379c58f4ace0214b3e
3
+ size 13619
rng_state_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:55f46944aa16d0b0bb24814e64339b9c6ae30653c66021a27a93c1f097c5d266
3
+ size 13619
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d0b4a96eb0248824e0d03cc27e17d508bd841e1a3d65ea444c3a1875e2c8a95
3
+ size 623
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "sep_token": "</s>", "pad_token": "<pad>", "cls_token": "<s>", "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": false}}
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "add_prefix_space": false, "errors": "replace", "sep_token": "</s>", "cls_token": "<s>", "pad_token": "<pad>", "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": false, "__type": "AddedToken"}, "model_max_length": 512, "special_tokens_map_file": null, "name_or_path": "models/roberta_chinese", "tokenizer_class": "RobertaTokenizer"}
trainer_state.json ADDED
@@ -0,0 +1,3176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 94.3039502215518,
5
+ "global_step": 250000,
6
+ "is_hyper_param_search": false,
7
+ "is_local_process_zero": true,
8
+ "is_world_process_zero": true,
9
+ "log_history": [
10
+ {
11
+ "epoch": 0.19,
12
+ "learning_rate": 2.0000000000000003e-06,
13
+ "loss": 8.2074,
14
+ "step": 500
15
+ },
16
+ {
17
+ "epoch": 0.38,
18
+ "learning_rate": 4.000000000000001e-06,
19
+ "loss": 7.333,
20
+ "step": 1000
21
+ },
22
+ {
23
+ "epoch": 0.57,
24
+ "learning_rate": 6e-06,
25
+ "loss": 6.7829,
26
+ "step": 1500
27
+ },
28
+ {
29
+ "epoch": 0.75,
30
+ "learning_rate": 8.000000000000001e-06,
31
+ "loss": 6.3812,
32
+ "step": 2000
33
+ },
34
+ {
35
+ "epoch": 0.94,
36
+ "learning_rate": 1e-05,
37
+ "loss": 6.1122,
38
+ "step": 2500
39
+ },
40
+ {
41
+ "epoch": 1.13,
42
+ "learning_rate": 1.2e-05,
43
+ "loss": 5.8855,
44
+ "step": 3000
45
+ },
46
+ {
47
+ "epoch": 1.32,
48
+ "learning_rate": 1.4000000000000001e-05,
49
+ "loss": 5.6587,
50
+ "step": 3500
51
+ },
52
+ {
53
+ "epoch": 1.51,
54
+ "learning_rate": 1.6000000000000003e-05,
55
+ "loss": 5.4578,
56
+ "step": 4000
57
+ },
58
+ {
59
+ "epoch": 1.7,
60
+ "learning_rate": 1.8e-05,
61
+ "loss": 5.2672,
62
+ "step": 4500
63
+ },
64
+ {
65
+ "epoch": 1.89,
66
+ "learning_rate": 2e-05,
67
+ "loss": 5.0883,
68
+ "step": 5000
69
+ },
70
+ {
71
+ "epoch": 2.07,
72
+ "learning_rate": 2.2000000000000003e-05,
73
+ "loss": 4.9455,
74
+ "step": 5500
75
+ },
76
+ {
77
+ "epoch": 2.26,
78
+ "learning_rate": 2.4e-05,
79
+ "loss": 4.7942,
80
+ "step": 6000
81
+ },
82
+ {
83
+ "epoch": 2.45,
84
+ "learning_rate": 2.6000000000000002e-05,
85
+ "loss": 4.6692,
86
+ "step": 6500
87
+ },
88
+ {
89
+ "epoch": 2.64,
90
+ "learning_rate": 2.8000000000000003e-05,
91
+ "loss": 4.5511,
92
+ "step": 7000
93
+ },
94
+ {
95
+ "epoch": 2.83,
96
+ "learning_rate": 3e-05,
97
+ "loss": 4.4469,
98
+ "step": 7500
99
+ },
100
+ {
101
+ "epoch": 3.02,
102
+ "learning_rate": 3.2000000000000005e-05,
103
+ "loss": 4.3566,
104
+ "step": 8000
105
+ },
106
+ {
107
+ "epoch": 3.21,
108
+ "learning_rate": 3.4000000000000007e-05,
109
+ "loss": 4.2571,
110
+ "step": 8500
111
+ },
112
+ {
113
+ "epoch": 3.39,
114
+ "learning_rate": 3.6e-05,
115
+ "loss": 4.1748,
116
+ "step": 9000
117
+ },
118
+ {
119
+ "epoch": 3.58,
120
+ "learning_rate": 3.8e-05,
121
+ "loss": 4.1022,
122
+ "step": 9500
123
+ },
124
+ {
125
+ "epoch": 3.77,
126
+ "learning_rate": 4e-05,
127
+ "loss": 4.027,
128
+ "step": 10000
129
+ },
130
+ {
131
+ "epoch": 3.96,
132
+ "learning_rate": 4.2e-05,
133
+ "loss": 3.9626,
134
+ "step": 10500
135
+ },
136
+ {
137
+ "epoch": 4.15,
138
+ "learning_rate": 4.4000000000000006e-05,
139
+ "loss": 3.9079,
140
+ "step": 11000
141
+ },
142
+ {
143
+ "epoch": 4.34,
144
+ "learning_rate": 4.600000000000001e-05,
145
+ "loss": 3.844,
146
+ "step": 11500
147
+ },
148
+ {
149
+ "epoch": 4.53,
150
+ "learning_rate": 4.8e-05,
151
+ "loss": 3.7907,
152
+ "step": 12000
153
+ },
154
+ {
155
+ "epoch": 4.71,
156
+ "learning_rate": 5e-05,
157
+ "loss": 3.7452,
158
+ "step": 12500
159
+ },
160
+ {
161
+ "epoch": 4.71,
162
+ "eval_loss": 3.5597941875457764,
163
+ "eval_runtime": 298.4056,
164
+ "eval_samples_per_second": 461.962,
165
+ "eval_steps_per_second": 3.609,
166
+ "step": 12500
167
+ },
168
+ {
169
+ "epoch": 4.9,
170
+ "learning_rate": 5.2000000000000004e-05,
171
+ "loss": 3.7008,
172
+ "step": 13000
173
+ },
174
+ {
175
+ "epoch": 5.09,
176
+ "learning_rate": 5.4000000000000005e-05,
177
+ "loss": 3.6604,
178
+ "step": 13500
179
+ },
180
+ {
181
+ "epoch": 5.28,
182
+ "learning_rate": 5.6000000000000006e-05,
183
+ "loss": 3.6173,
184
+ "step": 14000
185
+ },
186
+ {
187
+ "epoch": 5.47,
188
+ "learning_rate": 5.8e-05,
189
+ "loss": 3.5777,
190
+ "step": 14500
191
+ },
192
+ {
193
+ "epoch": 5.66,
194
+ "learning_rate": 6e-05,
195
+ "loss": 3.5396,
196
+ "step": 15000
197
+ },
198
+ {
199
+ "epoch": 5.85,
200
+ "learning_rate": 6.2e-05,
201
+ "loss": 3.5062,
202
+ "step": 15500
203
+ },
204
+ {
205
+ "epoch": 6.04,
206
+ "learning_rate": 6.400000000000001e-05,
207
+ "loss": 3.4766,
208
+ "step": 16000
209
+ },
210
+ {
211
+ "epoch": 6.22,
212
+ "learning_rate": 6.6e-05,
213
+ "loss": 3.4386,
214
+ "step": 16500
215
+ },
216
+ {
217
+ "epoch": 6.41,
218
+ "learning_rate": 6.800000000000001e-05,
219
+ "loss": 3.4131,
220
+ "step": 17000
221
+ },
222
+ {
223
+ "epoch": 6.6,
224
+ "learning_rate": 7e-05,
225
+ "loss": 3.388,
226
+ "step": 17500
227
+ },
228
+ {
229
+ "epoch": 6.79,
230
+ "learning_rate": 7.2e-05,
231
+ "loss": 3.3536,
232
+ "step": 18000
233
+ },
234
+ {
235
+ "epoch": 6.98,
236
+ "learning_rate": 7.4e-05,
237
+ "loss": 3.3314,
238
+ "step": 18500
239
+ },
240
+ {
241
+ "epoch": 7.17,
242
+ "learning_rate": 7.6e-05,
243
+ "loss": 3.3095,
244
+ "step": 19000
245
+ },
246
+ {
247
+ "epoch": 7.36,
248
+ "learning_rate": 7.800000000000001e-05,
249
+ "loss": 3.2802,
250
+ "step": 19500
251
+ },
252
+ {
253
+ "epoch": 7.54,
254
+ "learning_rate": 8e-05,
255
+ "loss": 3.2583,
256
+ "step": 20000
257
+ },
258
+ {
259
+ "epoch": 7.73,
260
+ "learning_rate": 8.2e-05,
261
+ "loss": 3.2374,
262
+ "step": 20500
263
+ },
264
+ {
265
+ "epoch": 7.92,
266
+ "learning_rate": 8.4e-05,
267
+ "loss": 3.2223,
268
+ "step": 21000
269
+ },
270
+ {
271
+ "epoch": 8.11,
272
+ "learning_rate": 8.6e-05,
273
+ "loss": 3.1986,
274
+ "step": 21500
275
+ },
276
+ {
277
+ "epoch": 8.3,
278
+ "learning_rate": 8.800000000000001e-05,
279
+ "loss": 3.1782,
280
+ "step": 22000
281
+ },
282
+ {
283
+ "epoch": 8.49,
284
+ "learning_rate": 9e-05,
285
+ "loss": 3.1585,
286
+ "step": 22500
287
+ },
288
+ {
289
+ "epoch": 8.68,
290
+ "learning_rate": 9.200000000000001e-05,
291
+ "loss": 3.1438,
292
+ "step": 23000
293
+ },
294
+ {
295
+ "epoch": 8.86,
296
+ "learning_rate": 9.4e-05,
297
+ "loss": 3.1313,
298
+ "step": 23500
299
+ },
300
+ {
301
+ "epoch": 9.05,
302
+ "learning_rate": 9.6e-05,
303
+ "loss": 3.1168,
304
+ "step": 24000
305
+ },
306
+ {
307
+ "epoch": 9.24,
308
+ "learning_rate": 9.8e-05,
309
+ "loss": 3.092,
310
+ "step": 24500
311
+ },
312
+ {
313
+ "epoch": 9.43,
314
+ "learning_rate": 0.0001,
315
+ "loss": 3.0788,
316
+ "step": 25000
317
+ },
318
+ {
319
+ "epoch": 9.43,
320
+ "eval_loss": 2.930635690689087,
321
+ "eval_runtime": 283.6019,
322
+ "eval_samples_per_second": 486.076,
323
+ "eval_steps_per_second": 3.798,
324
+ "step": 25000
325
+ },
326
+ {
327
+ "epoch": 9.62,
328
+ "learning_rate": 9.977777777777779e-05,
329
+ "loss": 3.0672,
330
+ "step": 25500
331
+ },
332
+ {
333
+ "epoch": 9.81,
334
+ "learning_rate": 9.955555555555556e-05,
335
+ "loss": 3.0534,
336
+ "step": 26000
337
+ },
338
+ {
339
+ "epoch": 10.0,
340
+ "learning_rate": 9.933333333333334e-05,
341
+ "loss": 3.0395,
342
+ "step": 26500
343
+ },
344
+ {
345
+ "epoch": 10.18,
346
+ "learning_rate": 9.911111111111112e-05,
347
+ "loss": 3.0276,
348
+ "step": 27000
349
+ },
350
+ {
351
+ "epoch": 10.37,
352
+ "learning_rate": 9.888888888888889e-05,
353
+ "loss": 3.0069,
354
+ "step": 27500
355
+ },
356
+ {
357
+ "epoch": 10.56,
358
+ "learning_rate": 9.866666666666668e-05,
359
+ "loss": 2.9995,
360
+ "step": 28000
361
+ },
362
+ {
363
+ "epoch": 10.75,
364
+ "learning_rate": 9.844444444444444e-05,
365
+ "loss": 2.9893,
366
+ "step": 28500
367
+ },
368
+ {
369
+ "epoch": 10.94,
370
+ "learning_rate": 9.822222222222223e-05,
371
+ "loss": 2.9776,
372
+ "step": 29000
373
+ },
374
+ {
375
+ "epoch": 11.13,
376
+ "learning_rate": 9.8e-05,
377
+ "loss": 2.9693,
378
+ "step": 29500
379
+ },
380
+ {
381
+ "epoch": 11.32,
382
+ "learning_rate": 9.777777777777778e-05,
383
+ "loss": 2.9528,
384
+ "step": 30000
385
+ },
386
+ {
387
+ "epoch": 11.5,
388
+ "learning_rate": 9.755555555555555e-05,
389
+ "loss": 2.9467,
390
+ "step": 30500
391
+ },
392
+ {
393
+ "epoch": 11.69,
394
+ "learning_rate": 9.733333333333335e-05,
395
+ "loss": 2.94,
396
+ "step": 31000
397
+ },
398
+ {
399
+ "epoch": 11.88,
400
+ "learning_rate": 9.711111111111111e-05,
401
+ "loss": 2.9276,
402
+ "step": 31500
403
+ },
404
+ {
405
+ "epoch": 12.07,
406
+ "learning_rate": 9.68888888888889e-05,
407
+ "loss": 2.9234,
408
+ "step": 32000
409
+ },
410
+ {
411
+ "epoch": 12.26,
412
+ "learning_rate": 9.666666666666667e-05,
413
+ "loss": 2.9092,
414
+ "step": 32500
415
+ },
416
+ {
417
+ "epoch": 12.45,
418
+ "learning_rate": 9.644444444444445e-05,
419
+ "loss": 2.9046,
420
+ "step": 33000
421
+ },
422
+ {
423
+ "epoch": 12.64,
424
+ "learning_rate": 9.622222222222222e-05,
425
+ "loss": 2.8948,
426
+ "step": 33500
427
+ },
428
+ {
429
+ "epoch": 12.83,
430
+ "learning_rate": 9.6e-05,
431
+ "loss": 2.8869,
432
+ "step": 34000
433
+ },
434
+ {
435
+ "epoch": 13.01,
436
+ "learning_rate": 9.577777777777777e-05,
437
+ "loss": 2.884,
438
+ "step": 34500
439
+ },
440
+ {
441
+ "epoch": 13.2,
442
+ "learning_rate": 9.555555555555557e-05,
443
+ "loss": 2.8717,
444
+ "step": 35000
445
+ },
446
+ {
447
+ "epoch": 13.39,
448
+ "learning_rate": 9.533333333333334e-05,
449
+ "loss": 2.8653,
450
+ "step": 35500
451
+ },
452
+ {
453
+ "epoch": 13.58,
454
+ "learning_rate": 9.511111111111112e-05,
455
+ "loss": 2.8565,
456
+ "step": 36000
457
+ },
458
+ {
459
+ "epoch": 13.77,
460
+ "learning_rate": 9.488888888888889e-05,
461
+ "loss": 2.8524,
462
+ "step": 36500
463
+ },
464
+ {
465
+ "epoch": 13.96,
466
+ "learning_rate": 9.466666666666667e-05,
467
+ "loss": 2.8449,
468
+ "step": 37000
469
+ },
470
+ {
471
+ "epoch": 14.15,
472
+ "learning_rate": 9.444444444444444e-05,
473
+ "loss": 2.8406,
474
+ "step": 37500
475
+ },
476
+ {
477
+ "epoch": 14.15,
478
+ "eval_loss": 2.706800937652588,
479
+ "eval_runtime": 283.8148,
480
+ "eval_samples_per_second": 485.711,
481
+ "eval_steps_per_second": 3.795,
482
+ "step": 37500
483
+ },
484
+ {
485
+ "epoch": 14.33,
486
+ "learning_rate": 9.422222222222223e-05,
487
+ "loss": 2.835,
488
+ "step": 38000
489
+ },
490
+ {
491
+ "epoch": 14.52,
492
+ "learning_rate": 9.4e-05,
493
+ "loss": 2.8281,
494
+ "step": 38500
495
+ },
496
+ {
497
+ "epoch": 14.71,
498
+ "learning_rate": 9.377777777777779e-05,
499
+ "loss": 2.8208,
500
+ "step": 39000
501
+ },
502
+ {
503
+ "epoch": 14.9,
504
+ "learning_rate": 9.355555555555556e-05,
505
+ "loss": 2.8225,
506
+ "step": 39500
507
+ },
508
+ {
509
+ "epoch": 15.09,
510
+ "learning_rate": 9.333333333333334e-05,
511
+ "loss": 2.8122,
512
+ "step": 40000
513
+ },
514
+ {
515
+ "epoch": 15.28,
516
+ "learning_rate": 9.311111111111111e-05,
517
+ "loss": 2.8051,
518
+ "step": 40500
519
+ },
520
+ {
521
+ "epoch": 15.47,
522
+ "learning_rate": 9.28888888888889e-05,
523
+ "loss": 2.8047,
524
+ "step": 41000
525
+ },
526
+ {
527
+ "epoch": 15.65,
528
+ "learning_rate": 9.266666666666666e-05,
529
+ "loss": 2.7949,
530
+ "step": 41500
531
+ },
532
+ {
533
+ "epoch": 15.84,
534
+ "learning_rate": 9.244444444444445e-05,
535
+ "loss": 2.7948,
536
+ "step": 42000
537
+ },
538
+ {
539
+ "epoch": 16.03,
540
+ "learning_rate": 9.222222222222223e-05,
541
+ "loss": 2.7895,
542
+ "step": 42500
543
+ },
544
+ {
545
+ "epoch": 16.22,
546
+ "learning_rate": 9.200000000000001e-05,
547
+ "loss": 2.784,
548
+ "step": 43000
549
+ },
550
+ {
551
+ "epoch": 16.41,
552
+ "learning_rate": 9.177777777777778e-05,
553
+ "loss": 2.7806,
554
+ "step": 43500
555
+ },
556
+ {
557
+ "epoch": 16.6,
558
+ "learning_rate": 9.155555555555557e-05,
559
+ "loss": 2.7774,
560
+ "step": 44000
561
+ },
562
+ {
563
+ "epoch": 16.79,
564
+ "learning_rate": 9.133333333333334e-05,
565
+ "loss": 2.769,
566
+ "step": 44500
567
+ },
568
+ {
569
+ "epoch": 16.97,
570
+ "learning_rate": 9.111111111111112e-05,
571
+ "loss": 2.7634,
572
+ "step": 45000
573
+ },
574
+ {
575
+ "epoch": 17.16,
576
+ "learning_rate": 9.088888888888889e-05,
577
+ "loss": 2.7645,
578
+ "step": 45500
579
+ },
580
+ {
581
+ "epoch": 17.35,
582
+ "learning_rate": 9.066666666666667e-05,
583
+ "loss": 2.7576,
584
+ "step": 46000
585
+ },
586
+ {
587
+ "epoch": 17.54,
588
+ "learning_rate": 9.044444444444445e-05,
589
+ "loss": 2.7533,
590
+ "step": 46500
591
+ },
592
+ {
593
+ "epoch": 17.73,
594
+ "learning_rate": 9.022222222222224e-05,
595
+ "loss": 2.7536,
596
+ "step": 47000
597
+ },
598
+ {
599
+ "epoch": 17.92,
600
+ "learning_rate": 9e-05,
601
+ "loss": 2.7447,
602
+ "step": 47500
603
+ },
604
+ {
605
+ "epoch": 18.11,
606
+ "learning_rate": 8.977777777777779e-05,
607
+ "loss": 2.7401,
608
+ "step": 48000
609
+ },
610
+ {
611
+ "epoch": 18.29,
612
+ "learning_rate": 8.955555555555556e-05,
613
+ "loss": 2.7365,
614
+ "step": 48500
615
+ },
616
+ {
617
+ "epoch": 18.48,
618
+ "learning_rate": 8.933333333333334e-05,
619
+ "loss": 2.7349,
620
+ "step": 49000
621
+ },
622
+ {
623
+ "epoch": 18.67,
624
+ "learning_rate": 8.911111111111111e-05,
625
+ "loss": 2.7308,
626
+ "step": 49500
627
+ },
628
+ {
629
+ "epoch": 18.86,
630
+ "learning_rate": 8.888888888888889e-05,
631
+ "loss": 2.7305,
632
+ "step": 50000
633
+ },
634
+ {
635
+ "epoch": 18.86,
636
+ "eval_loss": 2.608755111694336,
637
+ "eval_runtime": 283.7638,
638
+ "eval_samples_per_second": 485.798,
639
+ "eval_steps_per_second": 3.795,
640
+ "step": 50000
641
+ },
642
+ {
643
+ "epoch": 19.05,
644
+ "learning_rate": 8.866666666666668e-05,
645
+ "loss": 2.7327,
646
+ "step": 50500
647
+ },
648
+ {
649
+ "epoch": 19.24,
650
+ "learning_rate": 8.844444444444445e-05,
651
+ "loss": 2.722,
652
+ "step": 51000
653
+ },
654
+ {
655
+ "epoch": 19.43,
656
+ "learning_rate": 8.822222222222223e-05,
657
+ "loss": 2.7219,
658
+ "step": 51500
659
+ },
660
+ {
661
+ "epoch": 19.62,
662
+ "learning_rate": 8.800000000000001e-05,
663
+ "loss": 2.71,
664
+ "step": 52000
665
+ },
666
+ {
667
+ "epoch": 19.8,
668
+ "learning_rate": 8.777777777777778e-05,
669
+ "loss": 2.7107,
670
+ "step": 52500
671
+ },
672
+ {
673
+ "epoch": 19.99,
674
+ "learning_rate": 8.755555555555556e-05,
675
+ "loss": 2.7127,
676
+ "step": 53000
677
+ },
678
+ {
679
+ "epoch": 20.18,
680
+ "learning_rate": 8.733333333333333e-05,
681
+ "loss": 2.7044,
682
+ "step": 53500
683
+ },
684
+ {
685
+ "epoch": 20.37,
686
+ "learning_rate": 8.711111111111112e-05,
687
+ "loss": 2.7036,
688
+ "step": 54000
689
+ },
690
+ {
691
+ "epoch": 20.56,
692
+ "learning_rate": 8.68888888888889e-05,
693
+ "loss": 2.7002,
694
+ "step": 54500
695
+ },
696
+ {
697
+ "epoch": 20.75,
698
+ "learning_rate": 8.666666666666667e-05,
699
+ "loss": 2.6973,
700
+ "step": 55000
701
+ },
702
+ {
703
+ "epoch": 20.94,
704
+ "learning_rate": 8.644444444444445e-05,
705
+ "loss": 2.6931,
706
+ "step": 55500
707
+ },
708
+ {
709
+ "epoch": 21.12,
710
+ "learning_rate": 8.622222222222222e-05,
711
+ "loss": 2.6938,
712
+ "step": 56000
713
+ },
714
+ {
715
+ "epoch": 21.31,
716
+ "learning_rate": 8.6e-05,
717
+ "loss": 2.6837,
718
+ "step": 56500
719
+ },
720
+ {
721
+ "epoch": 21.5,
722
+ "learning_rate": 8.577777777777777e-05,
723
+ "loss": 2.6851,
724
+ "step": 57000
725
+ },
726
+ {
727
+ "epoch": 21.69,
728
+ "learning_rate": 8.555555555555556e-05,
729
+ "loss": 2.6848,
730
+ "step": 57500
731
+ },
732
+ {
733
+ "epoch": 21.88,
734
+ "learning_rate": 8.533333333333334e-05,
735
+ "loss": 2.6845,
736
+ "step": 58000
737
+ },
738
+ {
739
+ "epoch": 22.07,
740
+ "learning_rate": 8.511111111111112e-05,
741
+ "loss": 2.6839,
742
+ "step": 58500
743
+ },
744
+ {
745
+ "epoch": 22.26,
746
+ "learning_rate": 8.488888888888889e-05,
747
+ "loss": 2.6746,
748
+ "step": 59000
749
+ },
750
+ {
751
+ "epoch": 22.44,
752
+ "learning_rate": 8.466666666666667e-05,
753
+ "loss": 2.6736,
754
+ "step": 59500
755
+ },
756
+ {
757
+ "epoch": 22.63,
758
+ "learning_rate": 8.444444444444444e-05,
759
+ "loss": 2.6688,
760
+ "step": 60000
761
+ },
762
+ {
763
+ "epoch": 22.82,
764
+ "learning_rate": 8.422222222222223e-05,
765
+ "loss": 2.6699,
766
+ "step": 60500
767
+ },
768
+ {
769
+ "epoch": 23.01,
770
+ "learning_rate": 8.4e-05,
771
+ "loss": 2.673,
772
+ "step": 61000
773
+ },
774
+ {
775
+ "epoch": 23.2,
776
+ "learning_rate": 8.377777777777778e-05,
777
+ "loss": 2.6606,
778
+ "step": 61500
779
+ },
780
+ {
781
+ "epoch": 23.39,
782
+ "learning_rate": 8.355555555555556e-05,
783
+ "loss": 2.6595,
784
+ "step": 62000
785
+ },
786
+ {
787
+ "epoch": 23.58,
788
+ "learning_rate": 8.333333333333334e-05,
789
+ "loss": 2.6634,
790
+ "step": 62500
791
+ },
792
+ {
793
+ "epoch": 23.58,
794
+ "eval_loss": 2.541694164276123,
795
+ "eval_runtime": 283.9564,
796
+ "eval_samples_per_second": 485.469,
797
+ "eval_steps_per_second": 3.793,
798
+ "step": 62500
799
+ },
800
+ {
801
+ "epoch": 23.76,
802
+ "learning_rate": 8.311111111111111e-05,
803
+ "loss": 2.6599,
804
+ "step": 63000
805
+ },
806
+ {
807
+ "epoch": 23.95,
808
+ "learning_rate": 8.28888888888889e-05,
809
+ "loss": 2.6509,
810
+ "step": 63500
811
+ },
812
+ {
813
+ "epoch": 24.14,
814
+ "learning_rate": 8.266666666666667e-05,
815
+ "loss": 2.6573,
816
+ "step": 64000
817
+ },
818
+ {
819
+ "epoch": 24.33,
820
+ "learning_rate": 8.244444444444445e-05,
821
+ "loss": 2.6497,
822
+ "step": 64500
823
+ },
824
+ {
825
+ "epoch": 24.52,
826
+ "learning_rate": 8.222222222222222e-05,
827
+ "loss": 2.6492,
828
+ "step": 65000
829
+ },
830
+ {
831
+ "epoch": 24.71,
832
+ "learning_rate": 8.2e-05,
833
+ "loss": 2.646,
834
+ "step": 65500
835
+ },
836
+ {
837
+ "epoch": 24.9,
838
+ "learning_rate": 8.177777777777778e-05,
839
+ "loss": 2.6444,
840
+ "step": 66000
841
+ },
842
+ {
843
+ "epoch": 25.08,
844
+ "learning_rate": 8.155555555555557e-05,
845
+ "loss": 2.6465,
846
+ "step": 66500
847
+ },
848
+ {
849
+ "epoch": 25.27,
850
+ "learning_rate": 8.133333333333334e-05,
851
+ "loss": 2.6412,
852
+ "step": 67000
853
+ },
854
+ {
855
+ "epoch": 25.46,
856
+ "learning_rate": 8.111111111111112e-05,
857
+ "loss": 2.6364,
858
+ "step": 67500
859
+ },
860
+ {
861
+ "epoch": 25.65,
862
+ "learning_rate": 8.088888888888889e-05,
863
+ "loss": 2.6377,
864
+ "step": 68000
865
+ },
866
+ {
867
+ "epoch": 25.84,
868
+ "learning_rate": 8.066666666666667e-05,
869
+ "loss": 2.6331,
870
+ "step": 68500
871
+ },
872
+ {
873
+ "epoch": 26.03,
874
+ "learning_rate": 8.044444444444444e-05,
875
+ "loss": 2.6374,
876
+ "step": 69000
877
+ },
878
+ {
879
+ "epoch": 26.22,
880
+ "learning_rate": 8.022222222222222e-05,
881
+ "loss": 2.6269,
882
+ "step": 69500
883
+ },
884
+ {
885
+ "epoch": 26.41,
886
+ "learning_rate": 8e-05,
887
+ "loss": 2.625,
888
+ "step": 70000
889
+ },
890
+ {
891
+ "epoch": 26.59,
892
+ "learning_rate": 7.977777777777779e-05,
893
+ "loss": 2.6279,
894
+ "step": 70500
895
+ },
896
+ {
897
+ "epoch": 26.78,
898
+ "learning_rate": 7.955555555555556e-05,
899
+ "loss": 2.6278,
900
+ "step": 71000
901
+ },
902
+ {
903
+ "epoch": 26.97,
904
+ "learning_rate": 7.933333333333334e-05,
905
+ "loss": 2.6267,
906
+ "step": 71500
907
+ },
908
+ {
909
+ "epoch": 27.16,
910
+ "learning_rate": 7.911111111111111e-05,
911
+ "loss": 2.6273,
912
+ "step": 72000
913
+ },
914
+ {
915
+ "epoch": 27.35,
916
+ "learning_rate": 7.88888888888889e-05,
917
+ "loss": 2.6175,
918
+ "step": 72500
919
+ },
920
+ {
921
+ "epoch": 27.54,
922
+ "learning_rate": 7.866666666666666e-05,
923
+ "loss": 2.6149,
924
+ "step": 73000
925
+ },
926
+ {
927
+ "epoch": 27.73,
928
+ "learning_rate": 7.844444444444446e-05,
929
+ "loss": 2.6183,
930
+ "step": 73500
931
+ },
932
+ {
933
+ "epoch": 27.91,
934
+ "learning_rate": 7.822222222222223e-05,
935
+ "loss": 2.6107,
936
+ "step": 74000
937
+ },
938
+ {
939
+ "epoch": 28.1,
940
+ "learning_rate": 7.800000000000001e-05,
941
+ "loss": 2.6181,
942
+ "step": 74500
943
+ },
944
+ {
945
+ "epoch": 28.29,
946
+ "learning_rate": 7.777777777777778e-05,
947
+ "loss": 2.6134,
948
+ "step": 75000
949
+ },
950
+ {
951
+ "epoch": 28.29,
952
+ "eval_loss": 2.4989469051361084,
953
+ "eval_runtime": 283.5232,
954
+ "eval_samples_per_second": 486.211,
955
+ "eval_steps_per_second": 3.799,
956
+ "step": 75000
957
+ },
958
+ {
959
+ "epoch": 28.48,
960
+ "learning_rate": 7.755555555555556e-05,
961
+ "loss": 2.6136,
962
+ "step": 75500
963
+ },
964
+ {
965
+ "epoch": 28.67,
966
+ "learning_rate": 7.733333333333333e-05,
967
+ "loss": 2.6061,
968
+ "step": 76000
969
+ },
970
+ {
971
+ "epoch": 28.86,
972
+ "learning_rate": 7.711111111111112e-05,
973
+ "loss": 2.6038,
974
+ "step": 76500
975
+ },
976
+ {
977
+ "epoch": 29.05,
978
+ "learning_rate": 7.688888888888889e-05,
979
+ "loss": 2.6056,
980
+ "step": 77000
981
+ },
982
+ {
983
+ "epoch": 29.23,
984
+ "learning_rate": 7.666666666666667e-05,
985
+ "loss": 2.6044,
986
+ "step": 77500
987
+ },
988
+ {
989
+ "epoch": 29.42,
990
+ "learning_rate": 7.644444444444445e-05,
991
+ "loss": 2.6012,
992
+ "step": 78000
993
+ },
994
+ {
995
+ "epoch": 29.61,
996
+ "learning_rate": 7.622222222222223e-05,
997
+ "loss": 2.5976,
998
+ "step": 78500
999
+ },
1000
+ {
1001
+ "epoch": 29.8,
1002
+ "learning_rate": 7.6e-05,
1003
+ "loss": 2.5993,
1004
+ "step": 79000
1005
+ },
1006
+ {
1007
+ "epoch": 29.99,
1008
+ "learning_rate": 7.577777777777779e-05,
1009
+ "loss": 2.5964,
1010
+ "step": 79500
1011
+ },
1012
+ {
1013
+ "epoch": 30.18,
1014
+ "learning_rate": 7.555555555555556e-05,
1015
+ "loss": 2.5953,
1016
+ "step": 80000
1017
+ },
1018
+ {
1019
+ "epoch": 30.37,
1020
+ "learning_rate": 7.533333333333334e-05,
1021
+ "loss": 2.5916,
1022
+ "step": 80500
1023
+ },
1024
+ {
1025
+ "epoch": 30.55,
1026
+ "learning_rate": 7.511111111111111e-05,
1027
+ "loss": 2.6004,
1028
+ "step": 81000
1029
+ },
1030
+ {
1031
+ "epoch": 30.74,
1032
+ "learning_rate": 7.488888888888889e-05,
1033
+ "loss": 2.5902,
1034
+ "step": 81500
1035
+ },
1036
+ {
1037
+ "epoch": 30.93,
1038
+ "learning_rate": 7.466666666666667e-05,
1039
+ "loss": 2.5936,
1040
+ "step": 82000
1041
+ },
1042
+ {
1043
+ "epoch": 31.12,
1044
+ "learning_rate": 7.444444444444444e-05,
1045
+ "loss": 2.5939,
1046
+ "step": 82500
1047
+ },
1048
+ {
1049
+ "epoch": 31.31,
1050
+ "learning_rate": 7.422222222222223e-05,
1051
+ "loss": 2.5893,
1052
+ "step": 83000
1053
+ },
1054
+ {
1055
+ "epoch": 31.5,
1056
+ "learning_rate": 7.4e-05,
1057
+ "loss": 2.5834,
1058
+ "step": 83500
1059
+ },
1060
+ {
1061
+ "epoch": 31.69,
1062
+ "learning_rate": 7.377777777777778e-05,
1063
+ "loss": 2.5853,
1064
+ "step": 84000
1065
+ },
1066
+ {
1067
+ "epoch": 31.87,
1068
+ "learning_rate": 7.355555555555556e-05,
1069
+ "loss": 2.5825,
1070
+ "step": 84500
1071
+ },
1072
+ {
1073
+ "epoch": 32.06,
1074
+ "learning_rate": 7.333333333333333e-05,
1075
+ "loss": 2.5865,
1076
+ "step": 85000
1077
+ },
1078
+ {
1079
+ "epoch": 32.25,
1080
+ "learning_rate": 7.311111111111111e-05,
1081
+ "loss": 2.5756,
1082
+ "step": 85500
1083
+ },
1084
+ {
1085
+ "epoch": 32.44,
1086
+ "learning_rate": 7.28888888888889e-05,
1087
+ "loss": 2.5774,
1088
+ "step": 86000
1089
+ },
1090
+ {
1091
+ "epoch": 32.63,
1092
+ "learning_rate": 7.266666666666667e-05,
1093
+ "loss": 2.5792,
1094
+ "step": 86500
1095
+ },
1096
+ {
1097
+ "epoch": 32.82,
1098
+ "learning_rate": 7.244444444444445e-05,
1099
+ "loss": 2.5753,
1100
+ "step": 87000
1101
+ },
1102
+ {
1103
+ "epoch": 33.01,
1104
+ "learning_rate": 7.222222222222222e-05,
1105
+ "loss": 2.5812,
1106
+ "step": 87500
1107
+ },
1108
+ {
1109
+ "epoch": 33.01,
1110
+ "eval_loss": 2.469001531600952,
1111
+ "eval_runtime": 283.5082,
1112
+ "eval_samples_per_second": 486.236,
1113
+ "eval_steps_per_second": 3.799,
1114
+ "step": 87500
1115
+ },
1116
+ {
1117
+ "epoch": 33.19,
1118
+ "learning_rate": 7.2e-05,
1119
+ "loss": 2.571,
1120
+ "step": 88000
1121
+ },
1122
+ {
1123
+ "epoch": 33.38,
1124
+ "learning_rate": 7.177777777777777e-05,
1125
+ "loss": 2.5736,
1126
+ "step": 88500
1127
+ },
1128
+ {
1129
+ "epoch": 33.57,
1130
+ "learning_rate": 7.155555555555555e-05,
1131
+ "loss": 2.5712,
1132
+ "step": 89000
1133
+ },
1134
+ {
1135
+ "epoch": 33.76,
1136
+ "learning_rate": 7.133333333333334e-05,
1137
+ "loss": 2.5728,
1138
+ "step": 89500
1139
+ },
1140
+ {
1141
+ "epoch": 33.95,
1142
+ "learning_rate": 7.111111111111112e-05,
1143
+ "loss": 2.5676,
1144
+ "step": 90000
1145
+ },
1146
+ {
1147
+ "epoch": 34.14,
1148
+ "learning_rate": 7.088888888888889e-05,
1149
+ "loss": 2.5682,
1150
+ "step": 90500
1151
+ },
1152
+ {
1153
+ "epoch": 34.33,
1154
+ "learning_rate": 7.066666666666667e-05,
1155
+ "loss": 2.5662,
1156
+ "step": 91000
1157
+ },
1158
+ {
1159
+ "epoch": 34.52,
1160
+ "learning_rate": 7.044444444444444e-05,
1161
+ "loss": 2.5671,
1162
+ "step": 91500
1163
+ },
1164
+ {
1165
+ "epoch": 34.7,
1166
+ "learning_rate": 7.022222222222222e-05,
1167
+ "loss": 2.5667,
1168
+ "step": 92000
1169
+ },
1170
+ {
1171
+ "epoch": 34.89,
1172
+ "learning_rate": 7e-05,
1173
+ "loss": 2.5619,
1174
+ "step": 92500
1175
+ },
1176
+ {
1177
+ "epoch": 35.08,
1178
+ "learning_rate": 6.977777777777779e-05,
1179
+ "loss": 2.5639,
1180
+ "step": 93000
1181
+ },
1182
+ {
1183
+ "epoch": 35.27,
1184
+ "learning_rate": 6.955555555555556e-05,
1185
+ "loss": 2.5581,
1186
+ "step": 93500
1187
+ },
1188
+ {
1189
+ "epoch": 35.46,
1190
+ "learning_rate": 6.933333333333334e-05,
1191
+ "loss": 2.5586,
1192
+ "step": 94000
1193
+ },
1194
+ {
1195
+ "epoch": 35.65,
1196
+ "learning_rate": 6.911111111111111e-05,
1197
+ "loss": 2.5612,
1198
+ "step": 94500
1199
+ },
1200
+ {
1201
+ "epoch": 35.84,
1202
+ "learning_rate": 6.88888888888889e-05,
1203
+ "loss": 2.5552,
1204
+ "step": 95000
1205
+ },
1206
+ {
1207
+ "epoch": 36.02,
1208
+ "learning_rate": 6.866666666666666e-05,
1209
+ "loss": 2.5598,
1210
+ "step": 95500
1211
+ },
1212
+ {
1213
+ "epoch": 36.21,
1214
+ "learning_rate": 6.844444444444445e-05,
1215
+ "loss": 2.5479,
1216
+ "step": 96000
1217
+ },
1218
+ {
1219
+ "epoch": 36.4,
1220
+ "learning_rate": 6.822222222222222e-05,
1221
+ "loss": 2.5545,
1222
+ "step": 96500
1223
+ },
1224
+ {
1225
+ "epoch": 36.59,
1226
+ "learning_rate": 6.800000000000001e-05,
1227
+ "loss": 2.5506,
1228
+ "step": 97000
1229
+ },
1230
+ {
1231
+ "epoch": 36.78,
1232
+ "learning_rate": 6.777777777777778e-05,
1233
+ "loss": 2.5526,
1234
+ "step": 97500
1235
+ },
1236
+ {
1237
+ "epoch": 36.97,
1238
+ "learning_rate": 6.755555555555557e-05,
1239
+ "loss": 2.552,
1240
+ "step": 98000
1241
+ },
1242
+ {
1243
+ "epoch": 37.16,
1244
+ "learning_rate": 6.733333333333333e-05,
1245
+ "loss": 2.5493,
1246
+ "step": 98500
1247
+ },
1248
+ {
1249
+ "epoch": 37.34,
1250
+ "learning_rate": 6.711111111111112e-05,
1251
+ "loss": 2.5474,
1252
+ "step": 99000
1253
+ },
1254
+ {
1255
+ "epoch": 37.53,
1256
+ "learning_rate": 6.688888888888889e-05,
1257
+ "loss": 2.5475,
1258
+ "step": 99500
1259
+ },
1260
+ {
1261
+ "epoch": 37.72,
1262
+ "learning_rate": 6.666666666666667e-05,
1263
+ "loss": 2.5447,
1264
+ "step": 100000
1265
+ },
1266
+ {
1267
+ "epoch": 37.72,
1268
+ "eval_loss": 2.4395639896392822,
1269
+ "eval_runtime": 283.6209,
1270
+ "eval_samples_per_second": 486.043,
1271
+ "eval_steps_per_second": 3.797,
1272
+ "step": 100000
1273
+ },
1274
+ {
1275
+ "epoch": 37.91,
1276
+ "learning_rate": 6.644444444444444e-05,
1277
+ "loss": 2.5444,
1278
+ "step": 100500
1279
+ },
1280
+ {
1281
+ "epoch": 38.1,
1282
+ "learning_rate": 6.622222222222224e-05,
1283
+ "loss": 2.5467,
1284
+ "step": 101000
1285
+ },
1286
+ {
1287
+ "epoch": 38.29,
1288
+ "learning_rate": 6.6e-05,
1289
+ "loss": 2.537,
1290
+ "step": 101500
1291
+ },
1292
+ {
1293
+ "epoch": 38.48,
1294
+ "learning_rate": 6.577777777777779e-05,
1295
+ "loss": 2.5344,
1296
+ "step": 102000
1297
+ },
1298
+ {
1299
+ "epoch": 38.66,
1300
+ "learning_rate": 6.555555555555556e-05,
1301
+ "loss": 2.5431,
1302
+ "step": 102500
1303
+ },
1304
+ {
1305
+ "epoch": 38.85,
1306
+ "learning_rate": 6.533333333333334e-05,
1307
+ "loss": 2.5435,
1308
+ "step": 103000
1309
+ },
1310
+ {
1311
+ "epoch": 39.04,
1312
+ "learning_rate": 6.511111111111111e-05,
1313
+ "loss": 2.5434,
1314
+ "step": 103500
1315
+ },
1316
+ {
1317
+ "epoch": 39.23,
1318
+ "learning_rate": 6.488888888888889e-05,
1319
+ "loss": 2.5297,
1320
+ "step": 104000
1321
+ },
1322
+ {
1323
+ "epoch": 39.42,
1324
+ "learning_rate": 6.466666666666666e-05,
1325
+ "loss": 2.5408,
1326
+ "step": 104500
1327
+ },
1328
+ {
1329
+ "epoch": 39.61,
1330
+ "learning_rate": 6.444444444444446e-05,
1331
+ "loss": 2.5298,
1332
+ "step": 105000
1333
+ },
1334
+ {
1335
+ "epoch": 39.8,
1336
+ "learning_rate": 6.422222222222223e-05,
1337
+ "loss": 2.5358,
1338
+ "step": 105500
1339
+ },
1340
+ {
1341
+ "epoch": 39.98,
1342
+ "learning_rate": 6.400000000000001e-05,
1343
+ "loss": 2.5379,
1344
+ "step": 106000
1345
+ },
1346
+ {
1347
+ "epoch": 40.17,
1348
+ "learning_rate": 6.377777777777778e-05,
1349
+ "loss": 2.5357,
1350
+ "step": 106500
1351
+ },
1352
+ {
1353
+ "epoch": 40.36,
1354
+ "learning_rate": 6.355555555555556e-05,
1355
+ "loss": 2.5268,
1356
+ "step": 107000
1357
+ },
1358
+ {
1359
+ "epoch": 40.55,
1360
+ "learning_rate": 6.333333333333333e-05,
1361
+ "loss": 2.5263,
1362
+ "step": 107500
1363
+ },
1364
+ {
1365
+ "epoch": 40.74,
1366
+ "learning_rate": 6.311111111111112e-05,
1367
+ "loss": 2.5329,
1368
+ "step": 108000
1369
+ },
1370
+ {
1371
+ "epoch": 40.93,
1372
+ "learning_rate": 6.28888888888889e-05,
1373
+ "loss": 2.533,
1374
+ "step": 108500
1375
+ },
1376
+ {
1377
+ "epoch": 41.12,
1378
+ "learning_rate": 6.266666666666667e-05,
1379
+ "loss": 2.5282,
1380
+ "step": 109000
1381
+ },
1382
+ {
1383
+ "epoch": 41.31,
1384
+ "learning_rate": 6.244444444444445e-05,
1385
+ "loss": 2.5227,
1386
+ "step": 109500
1387
+ },
1388
+ {
1389
+ "epoch": 41.49,
1390
+ "learning_rate": 6.222222222222222e-05,
1391
+ "loss": 2.5279,
1392
+ "step": 110000
1393
+ },
1394
+ {
1395
+ "epoch": 41.68,
1396
+ "learning_rate": 6.2e-05,
1397
+ "loss": 2.5272,
1398
+ "step": 110500
1399
+ },
1400
+ {
1401
+ "epoch": 41.87,
1402
+ "learning_rate": 6.177777777777779e-05,
1403
+ "loss": 2.522,
1404
+ "step": 111000
1405
+ },
1406
+ {
1407
+ "epoch": 42.06,
1408
+ "learning_rate": 6.155555555555555e-05,
1409
+ "loss": 2.5253,
1410
+ "step": 111500
1411
+ },
1412
+ {
1413
+ "epoch": 42.25,
1414
+ "learning_rate": 6.133333333333334e-05,
1415
+ "loss": 2.5215,
1416
+ "step": 112000
1417
+ },
1418
+ {
1419
+ "epoch": 42.44,
1420
+ "learning_rate": 6.111111111111112e-05,
1421
+ "loss": 2.5205,
1422
+ "step": 112500
1423
+ },
1424
+ {
1425
+ "epoch": 42.44,
1426
+ "eval_loss": 2.419991970062256,
1427
+ "eval_runtime": 283.6709,
1428
+ "eval_samples_per_second": 485.957,
1429
+ "eval_steps_per_second": 3.797,
1430
+ "step": 112500
1431
+ },
1432
+ {
1433
+ "epoch": 42.63,
1434
+ "learning_rate": 6.08888888888889e-05,
1435
+ "loss": 2.5194,
1436
+ "step": 113000
1437
+ },
1438
+ {
1439
+ "epoch": 42.81,
1440
+ "learning_rate": 6.066666666666667e-05,
1441
+ "loss": 2.5179,
1442
+ "step": 113500
1443
+ },
1444
+ {
1445
+ "epoch": 43.0,
1446
+ "learning_rate": 6.044444444444445e-05,
1447
+ "loss": 2.526,
1448
+ "step": 114000
1449
+ },
1450
+ {
1451
+ "epoch": 43.19,
1452
+ "learning_rate": 6.0222222222222225e-05,
1453
+ "loss": 2.5172,
1454
+ "step": 114500
1455
+ },
1456
+ {
1457
+ "epoch": 43.38,
1458
+ "learning_rate": 6e-05,
1459
+ "loss": 2.5114,
1460
+ "step": 115000
1461
+ },
1462
+ {
1463
+ "epoch": 43.57,
1464
+ "learning_rate": 5.977777777777778e-05,
1465
+ "loss": 2.5179,
1466
+ "step": 115500
1467
+ },
1468
+ {
1469
+ "epoch": 43.76,
1470
+ "learning_rate": 5.9555555555555554e-05,
1471
+ "loss": 2.5189,
1472
+ "step": 116000
1473
+ },
1474
+ {
1475
+ "epoch": 43.95,
1476
+ "learning_rate": 5.9333333333333343e-05,
1477
+ "loss": 2.5202,
1478
+ "step": 116500
1479
+ },
1480
+ {
1481
+ "epoch": 44.13,
1482
+ "learning_rate": 5.911111111111112e-05,
1483
+ "loss": 2.517,
1484
+ "step": 117000
1485
+ },
1486
+ {
1487
+ "epoch": 44.32,
1488
+ "learning_rate": 5.8888888888888896e-05,
1489
+ "loss": 2.5151,
1490
+ "step": 117500
1491
+ },
1492
+ {
1493
+ "epoch": 44.51,
1494
+ "learning_rate": 5.866666666666667e-05,
1495
+ "loss": 2.5101,
1496
+ "step": 118000
1497
+ },
1498
+ {
1499
+ "epoch": 44.7,
1500
+ "learning_rate": 5.844444444444445e-05,
1501
+ "loss": 2.511,
1502
+ "step": 118500
1503
+ },
1504
+ {
1505
+ "epoch": 44.89,
1506
+ "learning_rate": 5.8222222222222224e-05,
1507
+ "loss": 2.5105,
1508
+ "step": 119000
1509
+ },
1510
+ {
1511
+ "epoch": 45.08,
1512
+ "learning_rate": 5.8e-05,
1513
+ "loss": 2.5107,
1514
+ "step": 119500
1515
+ },
1516
+ {
1517
+ "epoch": 45.27,
1518
+ "learning_rate": 5.7777777777777776e-05,
1519
+ "loss": 2.5089,
1520
+ "step": 120000
1521
+ },
1522
+ {
1523
+ "epoch": 45.45,
1524
+ "learning_rate": 5.755555555555556e-05,
1525
+ "loss": 2.5051,
1526
+ "step": 120500
1527
+ },
1528
+ {
1529
+ "epoch": 45.64,
1530
+ "learning_rate": 5.7333333333333336e-05,
1531
+ "loss": 2.5083,
1532
+ "step": 121000
1533
+ },
1534
+ {
1535
+ "epoch": 45.83,
1536
+ "learning_rate": 5.711111111111112e-05,
1537
+ "loss": 2.5072,
1538
+ "step": 121500
1539
+ },
1540
+ {
1541
+ "epoch": 46.02,
1542
+ "learning_rate": 5.6888888888888895e-05,
1543
+ "loss": 2.5107,
1544
+ "step": 122000
1545
+ },
1546
+ {
1547
+ "epoch": 46.21,
1548
+ "learning_rate": 5.666666666666667e-05,
1549
+ "loss": 2.5036,
1550
+ "step": 122500
1551
+ },
1552
+ {
1553
+ "epoch": 46.4,
1554
+ "learning_rate": 5.644444444444445e-05,
1555
+ "loss": 2.5051,
1556
+ "step": 123000
1557
+ },
1558
+ {
1559
+ "epoch": 46.59,
1560
+ "learning_rate": 5.622222222222222e-05,
1561
+ "loss": 2.5039,
1562
+ "step": 123500
1563
+ },
1564
+ {
1565
+ "epoch": 46.77,
1566
+ "learning_rate": 5.6000000000000006e-05,
1567
+ "loss": 2.4972,
1568
+ "step": 124000
1569
+ },
1570
+ {
1571
+ "epoch": 46.96,
1572
+ "learning_rate": 5.577777777777778e-05,
1573
+ "loss": 2.5073,
1574
+ "step": 124500
1575
+ },
1576
+ {
1577
+ "epoch": 47.15,
1578
+ "learning_rate": 5.555555555555556e-05,
1579
+ "loss": 2.5052,
1580
+ "step": 125000
1581
+ },
1582
+ {
1583
+ "epoch": 47.15,
1584
+ "eval_loss": 2.4018545150756836,
1585
+ "eval_runtime": 283.8972,
1586
+ "eval_samples_per_second": 485.57,
1587
+ "eval_steps_per_second": 3.794,
1588
+ "step": 125000
1589
+ },
1590
+ {
1591
+ "epoch": 47.34,
1592
+ "learning_rate": 5.5333333333333334e-05,
1593
+ "loss": 2.5015,
1594
+ "step": 125500
1595
+ },
1596
+ {
1597
+ "epoch": 47.53,
1598
+ "learning_rate": 5.511111111111111e-05,
1599
+ "loss": 2.4983,
1600
+ "step": 126000
1601
+ },
1602
+ {
1603
+ "epoch": 47.72,
1604
+ "learning_rate": 5.488888888888889e-05,
1605
+ "loss": 2.5022,
1606
+ "step": 126500
1607
+ },
1608
+ {
1609
+ "epoch": 47.91,
1610
+ "learning_rate": 5.466666666666666e-05,
1611
+ "loss": 2.4932,
1612
+ "step": 127000
1613
+ },
1614
+ {
1615
+ "epoch": 48.1,
1616
+ "learning_rate": 5.4444444444444446e-05,
1617
+ "loss": 2.5007,
1618
+ "step": 127500
1619
+ },
1620
+ {
1621
+ "epoch": 48.28,
1622
+ "learning_rate": 5.422222222222223e-05,
1623
+ "loss": 2.4944,
1624
+ "step": 128000
1625
+ },
1626
+ {
1627
+ "epoch": 48.47,
1628
+ "learning_rate": 5.4000000000000005e-05,
1629
+ "loss": 2.4964,
1630
+ "step": 128500
1631
+ },
1632
+ {
1633
+ "epoch": 48.66,
1634
+ "learning_rate": 5.377777777777778e-05,
1635
+ "loss": 2.4976,
1636
+ "step": 129000
1637
+ },
1638
+ {
1639
+ "epoch": 48.85,
1640
+ "learning_rate": 5.355555555555556e-05,
1641
+ "loss": 2.4948,
1642
+ "step": 129500
1643
+ },
1644
+ {
1645
+ "epoch": 49.04,
1646
+ "learning_rate": 5.333333333333333e-05,
1647
+ "loss": 2.5018,
1648
+ "step": 130000
1649
+ },
1650
+ {
1651
+ "epoch": 49.23,
1652
+ "learning_rate": 5.311111111111111e-05,
1653
+ "loss": 2.4919,
1654
+ "step": 130500
1655
+ },
1656
+ {
1657
+ "epoch": 49.42,
1658
+ "learning_rate": 5.2888888888888885e-05,
1659
+ "loss": 2.4925,
1660
+ "step": 131000
1661
+ },
1662
+ {
1663
+ "epoch": 49.6,
1664
+ "learning_rate": 5.266666666666666e-05,
1665
+ "loss": 2.491,
1666
+ "step": 131500
1667
+ },
1668
+ {
1669
+ "epoch": 49.79,
1670
+ "learning_rate": 5.244444444444445e-05,
1671
+ "loss": 2.4932,
1672
+ "step": 132000
1673
+ },
1674
+ {
1675
+ "epoch": 49.98,
1676
+ "learning_rate": 5.222222222222223e-05,
1677
+ "loss": 2.4892,
1678
+ "step": 132500
1679
+ },
1680
+ {
1681
+ "epoch": 50.17,
1682
+ "learning_rate": 5.2000000000000004e-05,
1683
+ "loss": 2.4877,
1684
+ "step": 133000
1685
+ },
1686
+ {
1687
+ "epoch": 50.36,
1688
+ "learning_rate": 5.177777777777778e-05,
1689
+ "loss": 2.4899,
1690
+ "step": 133500
1691
+ },
1692
+ {
1693
+ "epoch": 50.55,
1694
+ "learning_rate": 5.1555555555555556e-05,
1695
+ "loss": 2.49,
1696
+ "step": 134000
1697
+ },
1698
+ {
1699
+ "epoch": 50.74,
1700
+ "learning_rate": 5.133333333333333e-05,
1701
+ "loss": 2.4843,
1702
+ "step": 134500
1703
+ },
1704
+ {
1705
+ "epoch": 50.92,
1706
+ "learning_rate": 5.111111111111111e-05,
1707
+ "loss": 2.4901,
1708
+ "step": 135000
1709
+ },
1710
+ {
1711
+ "epoch": 51.11,
1712
+ "learning_rate": 5.0888888888888884e-05,
1713
+ "loss": 2.4919,
1714
+ "step": 135500
1715
+ },
1716
+ {
1717
+ "epoch": 51.3,
1718
+ "learning_rate": 5.0666666666666674e-05,
1719
+ "loss": 2.4862,
1720
+ "step": 136000
1721
+ },
1722
+ {
1723
+ "epoch": 51.49,
1724
+ "learning_rate": 5.044444444444445e-05,
1725
+ "loss": 2.4813,
1726
+ "step": 136500
1727
+ },
1728
+ {
1729
+ "epoch": 51.68,
1730
+ "learning_rate": 5.0222222222222226e-05,
1731
+ "loss": 2.4818,
1732
+ "step": 137000
1733
+ },
1734
+ {
1735
+ "epoch": 51.87,
1736
+ "learning_rate": 5e-05,
1737
+ "loss": 2.4834,
1738
+ "step": 137500
1739
+ },
1740
+ {
1741
+ "epoch": 51.87,
1742
+ "eval_loss": 2.3841142654418945,
1743
+ "eval_runtime": 283.7333,
1744
+ "eval_samples_per_second": 485.851,
1745
+ "eval_steps_per_second": 3.796,
1746
+ "step": 137500
1747
+ },
1748
+ {
1749
+ "epoch": 52.06,
1750
+ "learning_rate": 4.977777777777778e-05,
1751
+ "loss": 2.4823,
1752
+ "step": 138000
1753
+ },
1754
+ {
1755
+ "epoch": 52.24,
1756
+ "learning_rate": 4.955555555555556e-05,
1757
+ "loss": 2.4773,
1758
+ "step": 138500
1759
+ },
1760
+ {
1761
+ "epoch": 52.43,
1762
+ "learning_rate": 4.933333333333334e-05,
1763
+ "loss": 2.4818,
1764
+ "step": 139000
1765
+ },
1766
+ {
1767
+ "epoch": 52.62,
1768
+ "learning_rate": 4.9111111111111114e-05,
1769
+ "loss": 2.4807,
1770
+ "step": 139500
1771
+ },
1772
+ {
1773
+ "epoch": 52.81,
1774
+ "learning_rate": 4.888888888888889e-05,
1775
+ "loss": 2.4801,
1776
+ "step": 140000
1777
+ },
1778
+ {
1779
+ "epoch": 53.0,
1780
+ "learning_rate": 4.866666666666667e-05,
1781
+ "loss": 2.4808,
1782
+ "step": 140500
1783
+ },
1784
+ {
1785
+ "epoch": 53.19,
1786
+ "learning_rate": 4.844444444444445e-05,
1787
+ "loss": 2.4788,
1788
+ "step": 141000
1789
+ },
1790
+ {
1791
+ "epoch": 53.38,
1792
+ "learning_rate": 4.8222222222222225e-05,
1793
+ "loss": 2.478,
1794
+ "step": 141500
1795
+ },
1796
+ {
1797
+ "epoch": 53.56,
1798
+ "learning_rate": 4.8e-05,
1799
+ "loss": 2.4758,
1800
+ "step": 142000
1801
+ },
1802
+ {
1803
+ "epoch": 53.75,
1804
+ "learning_rate": 4.7777777777777784e-05,
1805
+ "loss": 2.4792,
1806
+ "step": 142500
1807
+ },
1808
+ {
1809
+ "epoch": 53.94,
1810
+ "learning_rate": 4.755555555555556e-05,
1811
+ "loss": 2.4781,
1812
+ "step": 143000
1813
+ },
1814
+ {
1815
+ "epoch": 54.13,
1816
+ "learning_rate": 4.7333333333333336e-05,
1817
+ "loss": 2.4773,
1818
+ "step": 143500
1819
+ },
1820
+ {
1821
+ "epoch": 54.32,
1822
+ "learning_rate": 4.711111111111111e-05,
1823
+ "loss": 2.4725,
1824
+ "step": 144000
1825
+ },
1826
+ {
1827
+ "epoch": 54.51,
1828
+ "learning_rate": 4.6888888888888895e-05,
1829
+ "loss": 2.4748,
1830
+ "step": 144500
1831
+ },
1832
+ {
1833
+ "epoch": 54.7,
1834
+ "learning_rate": 4.666666666666667e-05,
1835
+ "loss": 2.4751,
1836
+ "step": 145000
1837
+ },
1838
+ {
1839
+ "epoch": 54.88,
1840
+ "learning_rate": 4.644444444444445e-05,
1841
+ "loss": 2.47,
1842
+ "step": 145500
1843
+ },
1844
+ {
1845
+ "epoch": 55.07,
1846
+ "learning_rate": 4.6222222222222224e-05,
1847
+ "loss": 2.4796,
1848
+ "step": 146000
1849
+ },
1850
+ {
1851
+ "epoch": 55.26,
1852
+ "learning_rate": 4.600000000000001e-05,
1853
+ "loss": 2.4717,
1854
+ "step": 146500
1855
+ },
1856
+ {
1857
+ "epoch": 55.45,
1858
+ "learning_rate": 4.577777777777778e-05,
1859
+ "loss": 2.4759,
1860
+ "step": 147000
1861
+ },
1862
+ {
1863
+ "epoch": 55.64,
1864
+ "learning_rate": 4.555555555555556e-05,
1865
+ "loss": 2.4696,
1866
+ "step": 147500
1867
+ },
1868
+ {
1869
+ "epoch": 55.83,
1870
+ "learning_rate": 4.5333333333333335e-05,
1871
+ "loss": 2.4707,
1872
+ "step": 148000
1873
+ },
1874
+ {
1875
+ "epoch": 56.02,
1876
+ "learning_rate": 4.511111111111112e-05,
1877
+ "loss": 2.4664,
1878
+ "step": 148500
1879
+ },
1880
+ {
1881
+ "epoch": 56.21,
1882
+ "learning_rate": 4.4888888888888894e-05,
1883
+ "loss": 2.4663,
1884
+ "step": 149000
1885
+ },
1886
+ {
1887
+ "epoch": 56.39,
1888
+ "learning_rate": 4.466666666666667e-05,
1889
+ "loss": 2.4679,
1890
+ "step": 149500
1891
+ },
1892
+ {
1893
+ "epoch": 56.58,
1894
+ "learning_rate": 4.4444444444444447e-05,
1895
+ "loss": 2.4694,
1896
+ "step": 150000
1897
+ },
1898
+ {
1899
+ "epoch": 56.58,
1900
+ "eval_loss": 2.371152877807617,
1901
+ "eval_runtime": 283.5132,
1902
+ "eval_samples_per_second": 486.228,
1903
+ "eval_steps_per_second": 3.799,
1904
+ "step": 150000
1905
+ },
1906
+ {
1907
+ "epoch": 56.77,
1908
+ "learning_rate": 4.422222222222222e-05,
1909
+ "loss": 2.4619,
1910
+ "step": 150500
1911
+ },
1912
+ {
1913
+ "epoch": 56.96,
1914
+ "learning_rate": 4.4000000000000006e-05,
1915
+ "loss": 2.4686,
1916
+ "step": 151000
1917
+ },
1918
+ {
1919
+ "epoch": 57.15,
1920
+ "learning_rate": 4.377777777777778e-05,
1921
+ "loss": 2.4715,
1922
+ "step": 151500
1923
+ },
1924
+ {
1925
+ "epoch": 57.34,
1926
+ "learning_rate": 4.355555555555556e-05,
1927
+ "loss": 2.4619,
1928
+ "step": 152000
1929
+ },
1930
+ {
1931
+ "epoch": 57.53,
1932
+ "learning_rate": 4.3333333333333334e-05,
1933
+ "loss": 2.463,
1934
+ "step": 152500
1935
+ },
1936
+ {
1937
+ "epoch": 57.71,
1938
+ "learning_rate": 4.311111111111111e-05,
1939
+ "loss": 2.4609,
1940
+ "step": 153000
1941
+ },
1942
+ {
1943
+ "epoch": 57.9,
1944
+ "learning_rate": 4.2888888888888886e-05,
1945
+ "loss": 2.4646,
1946
+ "step": 153500
1947
+ },
1948
+ {
1949
+ "epoch": 58.09,
1950
+ "learning_rate": 4.266666666666667e-05,
1951
+ "loss": 2.4658,
1952
+ "step": 154000
1953
+ },
1954
+ {
1955
+ "epoch": 58.28,
1956
+ "learning_rate": 4.2444444444444445e-05,
1957
+ "loss": 2.4614,
1958
+ "step": 154500
1959
+ },
1960
+ {
1961
+ "epoch": 58.47,
1962
+ "learning_rate": 4.222222222222222e-05,
1963
+ "loss": 2.4646,
1964
+ "step": 155000
1965
+ },
1966
+ {
1967
+ "epoch": 58.66,
1968
+ "learning_rate": 4.2e-05,
1969
+ "loss": 2.4629,
1970
+ "step": 155500
1971
+ },
1972
+ {
1973
+ "epoch": 58.85,
1974
+ "learning_rate": 4.177777777777778e-05,
1975
+ "loss": 2.4587,
1976
+ "step": 156000
1977
+ },
1978
+ {
1979
+ "epoch": 59.03,
1980
+ "learning_rate": 4.155555555555556e-05,
1981
+ "loss": 2.466,
1982
+ "step": 156500
1983
+ },
1984
+ {
1985
+ "epoch": 59.22,
1986
+ "learning_rate": 4.133333333333333e-05,
1987
+ "loss": 2.452,
1988
+ "step": 157000
1989
+ },
1990
+ {
1991
+ "epoch": 59.41,
1992
+ "learning_rate": 4.111111111111111e-05,
1993
+ "loss": 2.4608,
1994
+ "step": 157500
1995
+ },
1996
+ {
1997
+ "epoch": 59.6,
1998
+ "learning_rate": 4.088888888888889e-05,
1999
+ "loss": 2.4629,
2000
+ "step": 158000
2001
+ },
2002
+ {
2003
+ "epoch": 59.79,
2004
+ "learning_rate": 4.066666666666667e-05,
2005
+ "loss": 2.4605,
2006
+ "step": 158500
2007
+ },
2008
+ {
2009
+ "epoch": 59.98,
2010
+ "learning_rate": 4.0444444444444444e-05,
2011
+ "loss": 2.4576,
2012
+ "step": 159000
2013
+ },
2014
+ {
2015
+ "epoch": 60.17,
2016
+ "learning_rate": 4.022222222222222e-05,
2017
+ "loss": 2.4599,
2018
+ "step": 159500
2019
+ },
2020
+ {
2021
+ "epoch": 60.35,
2022
+ "learning_rate": 4e-05,
2023
+ "loss": 2.4567,
2024
+ "step": 160000
2025
+ },
2026
+ {
2027
+ "epoch": 60.54,
2028
+ "learning_rate": 3.977777777777778e-05,
2029
+ "loss": 2.4604,
2030
+ "step": 160500
2031
+ },
2032
+ {
2033
+ "epoch": 60.73,
2034
+ "learning_rate": 3.9555555555555556e-05,
2035
+ "loss": 2.4565,
2036
+ "step": 161000
2037
+ },
2038
+ {
2039
+ "epoch": 60.92,
2040
+ "learning_rate": 3.933333333333333e-05,
2041
+ "loss": 2.4559,
2042
+ "step": 161500
2043
+ },
2044
+ {
2045
+ "epoch": 61.11,
2046
+ "learning_rate": 3.9111111111111115e-05,
2047
+ "loss": 2.4587,
2048
+ "step": 162000
2049
+ },
2050
+ {
2051
+ "epoch": 61.3,
2052
+ "learning_rate": 3.888888888888889e-05,
2053
+ "loss": 2.4509,
2054
+ "step": 162500
2055
+ },
2056
+ {
2057
+ "epoch": 61.3,
2058
+ "eval_loss": 2.3613131046295166,
2059
+ "eval_runtime": 283.6567,
2060
+ "eval_samples_per_second": 485.982,
2061
+ "eval_steps_per_second": 3.797,
2062
+ "step": 162500
2063
+ },
2064
+ {
2065
+ "epoch": 61.49,
2066
+ "learning_rate": 3.866666666666667e-05,
2067
+ "loss": 2.4554,
2068
+ "step": 163000
2069
+ },
2070
+ {
2071
+ "epoch": 61.67,
2072
+ "learning_rate": 3.844444444444444e-05,
2073
+ "loss": 2.4562,
2074
+ "step": 163500
2075
+ },
2076
+ {
2077
+ "epoch": 61.86,
2078
+ "learning_rate": 3.8222222222222226e-05,
2079
+ "loss": 2.4511,
2080
+ "step": 164000
2081
+ },
2082
+ {
2083
+ "epoch": 62.05,
2084
+ "learning_rate": 3.8e-05,
2085
+ "loss": 2.4578,
2086
+ "step": 164500
2087
+ },
2088
+ {
2089
+ "epoch": 62.24,
2090
+ "learning_rate": 3.777777777777778e-05,
2091
+ "loss": 2.4513,
2092
+ "step": 165000
2093
+ },
2094
+ {
2095
+ "epoch": 62.43,
2096
+ "learning_rate": 3.7555555555555554e-05,
2097
+ "loss": 2.4509,
2098
+ "step": 165500
2099
+ },
2100
+ {
2101
+ "epoch": 62.62,
2102
+ "learning_rate": 3.733333333333334e-05,
2103
+ "loss": 2.4498,
2104
+ "step": 166000
2105
+ },
2106
+ {
2107
+ "epoch": 62.81,
2108
+ "learning_rate": 3.7111111111111113e-05,
2109
+ "loss": 2.4499,
2110
+ "step": 166500
2111
+ },
2112
+ {
2113
+ "epoch": 62.99,
2114
+ "learning_rate": 3.688888888888889e-05,
2115
+ "loss": 2.4543,
2116
+ "step": 167000
2117
+ },
2118
+ {
2119
+ "epoch": 63.18,
2120
+ "learning_rate": 3.6666666666666666e-05,
2121
+ "loss": 2.4498,
2122
+ "step": 167500
2123
+ },
2124
+ {
2125
+ "epoch": 63.37,
2126
+ "learning_rate": 3.644444444444445e-05,
2127
+ "loss": 2.4463,
2128
+ "step": 168000
2129
+ },
2130
+ {
2131
+ "epoch": 63.56,
2132
+ "learning_rate": 3.6222222222222225e-05,
2133
+ "loss": 2.4491,
2134
+ "step": 168500
2135
+ },
2136
+ {
2137
+ "epoch": 63.75,
2138
+ "learning_rate": 3.6e-05,
2139
+ "loss": 2.4532,
2140
+ "step": 169000
2141
+ },
2142
+ {
2143
+ "epoch": 63.94,
2144
+ "learning_rate": 3.577777777777778e-05,
2145
+ "loss": 2.4441,
2146
+ "step": 169500
2147
+ },
2148
+ {
2149
+ "epoch": 64.13,
2150
+ "learning_rate": 3.555555555555556e-05,
2151
+ "loss": 2.45,
2152
+ "step": 170000
2153
+ },
2154
+ {
2155
+ "epoch": 64.32,
2156
+ "learning_rate": 3.5333333333333336e-05,
2157
+ "loss": 2.4453,
2158
+ "step": 170500
2159
+ },
2160
+ {
2161
+ "epoch": 64.5,
2162
+ "learning_rate": 3.511111111111111e-05,
2163
+ "loss": 2.4463,
2164
+ "step": 171000
2165
+ },
2166
+ {
2167
+ "epoch": 64.69,
2168
+ "learning_rate": 3.4888888888888895e-05,
2169
+ "loss": 2.4457,
2170
+ "step": 171500
2171
+ },
2172
+ {
2173
+ "epoch": 64.88,
2174
+ "learning_rate": 3.466666666666667e-05,
2175
+ "loss": 2.4477,
2176
+ "step": 172000
2177
+ },
2178
+ {
2179
+ "epoch": 65.07,
2180
+ "learning_rate": 3.444444444444445e-05,
2181
+ "loss": 2.4467,
2182
+ "step": 172500
2183
+ },
2184
+ {
2185
+ "epoch": 65.26,
2186
+ "learning_rate": 3.4222222222222224e-05,
2187
+ "loss": 2.447,
2188
+ "step": 173000
2189
+ },
2190
+ {
2191
+ "epoch": 65.45,
2192
+ "learning_rate": 3.4000000000000007e-05,
2193
+ "loss": 2.443,
2194
+ "step": 173500
2195
+ },
2196
+ {
2197
+ "epoch": 65.64,
2198
+ "learning_rate": 3.377777777777778e-05,
2199
+ "loss": 2.443,
2200
+ "step": 174000
2201
+ },
2202
+ {
2203
+ "epoch": 65.82,
2204
+ "learning_rate": 3.355555555555556e-05,
2205
+ "loss": 2.4441,
2206
+ "step": 174500
2207
+ },
2208
+ {
2209
+ "epoch": 66.01,
2210
+ "learning_rate": 3.3333333333333335e-05,
2211
+ "loss": 2.4469,
2212
+ "step": 175000
2213
+ },
2214
+ {
2215
+ "epoch": 66.01,
2216
+ "eval_loss": 2.3445606231689453,
2217
+ "eval_runtime": 283.6225,
2218
+ "eval_samples_per_second": 486.04,
2219
+ "eval_steps_per_second": 3.797,
2220
+ "step": 175000
2221
+ },
2222
+ {
2223
+ "epoch": 66.2,
2224
+ "learning_rate": 3.311111111111112e-05,
2225
+ "loss": 2.4367,
2226
+ "step": 175500
2227
+ },
2228
+ {
2229
+ "epoch": 66.39,
2230
+ "learning_rate": 3.2888888888888894e-05,
2231
+ "loss": 2.4414,
2232
+ "step": 176000
2233
+ },
2234
+ {
2235
+ "epoch": 66.58,
2236
+ "learning_rate": 3.266666666666667e-05,
2237
+ "loss": 2.4424,
2238
+ "step": 176500
2239
+ },
2240
+ {
2241
+ "epoch": 66.77,
2242
+ "learning_rate": 3.2444444444444446e-05,
2243
+ "loss": 2.4405,
2244
+ "step": 177000
2245
+ },
2246
+ {
2247
+ "epoch": 66.96,
2248
+ "learning_rate": 3.222222222222223e-05,
2249
+ "loss": 2.4415,
2250
+ "step": 177500
2251
+ },
2252
+ {
2253
+ "epoch": 67.14,
2254
+ "learning_rate": 3.2000000000000005e-05,
2255
+ "loss": 2.4398,
2256
+ "step": 178000
2257
+ },
2258
+ {
2259
+ "epoch": 67.33,
2260
+ "learning_rate": 3.177777777777778e-05,
2261
+ "loss": 2.4406,
2262
+ "step": 178500
2263
+ },
2264
+ {
2265
+ "epoch": 67.52,
2266
+ "learning_rate": 3.155555555555556e-05,
2267
+ "loss": 2.4375,
2268
+ "step": 179000
2269
+ },
2270
+ {
2271
+ "epoch": 67.71,
2272
+ "learning_rate": 3.1333333333333334e-05,
2273
+ "loss": 2.4365,
2274
+ "step": 179500
2275
+ },
2276
+ {
2277
+ "epoch": 67.9,
2278
+ "learning_rate": 3.111111111111111e-05,
2279
+ "loss": 2.4415,
2280
+ "step": 180000
2281
+ },
2282
+ {
2283
+ "epoch": 68.09,
2284
+ "learning_rate": 3.088888888888889e-05,
2285
+ "loss": 2.4391,
2286
+ "step": 180500
2287
+ },
2288
+ {
2289
+ "epoch": 68.28,
2290
+ "learning_rate": 3.066666666666667e-05,
2291
+ "loss": 2.437,
2292
+ "step": 181000
2293
+ },
2294
+ {
2295
+ "epoch": 68.46,
2296
+ "learning_rate": 3.044444444444445e-05,
2297
+ "loss": 2.4357,
2298
+ "step": 181500
2299
+ },
2300
+ {
2301
+ "epoch": 68.65,
2302
+ "learning_rate": 3.0222222222222225e-05,
2303
+ "loss": 2.4375,
2304
+ "step": 182000
2305
+ },
2306
+ {
2307
+ "epoch": 68.84,
2308
+ "learning_rate": 3e-05,
2309
+ "loss": 2.4356,
2310
+ "step": 182500
2311
+ },
2312
+ {
2313
+ "epoch": 69.03,
2314
+ "learning_rate": 2.9777777777777777e-05,
2315
+ "loss": 2.4364,
2316
+ "step": 183000
2317
+ },
2318
+ {
2319
+ "epoch": 69.22,
2320
+ "learning_rate": 2.955555555555556e-05,
2321
+ "loss": 2.4335,
2322
+ "step": 183500
2323
+ },
2324
+ {
2325
+ "epoch": 69.41,
2326
+ "learning_rate": 2.9333333333333336e-05,
2327
+ "loss": 2.4325,
2328
+ "step": 184000
2329
+ },
2330
+ {
2331
+ "epoch": 69.6,
2332
+ "learning_rate": 2.9111111111111112e-05,
2333
+ "loss": 2.434,
2334
+ "step": 184500
2335
+ },
2336
+ {
2337
+ "epoch": 69.78,
2338
+ "learning_rate": 2.8888888888888888e-05,
2339
+ "loss": 2.4397,
2340
+ "step": 185000
2341
+ },
2342
+ {
2343
+ "epoch": 69.97,
2344
+ "learning_rate": 2.8666666666666668e-05,
2345
+ "loss": 2.4375,
2346
+ "step": 185500
2347
+ },
2348
+ {
2349
+ "epoch": 70.16,
2350
+ "learning_rate": 2.8444444444444447e-05,
2351
+ "loss": 2.4338,
2352
+ "step": 186000
2353
+ },
2354
+ {
2355
+ "epoch": 70.35,
2356
+ "learning_rate": 2.8222222222222223e-05,
2357
+ "loss": 2.4317,
2358
+ "step": 186500
2359
+ },
2360
+ {
2361
+ "epoch": 70.54,
2362
+ "learning_rate": 2.8000000000000003e-05,
2363
+ "loss": 2.4321,
2364
+ "step": 187000
2365
+ },
2366
+ {
2367
+ "epoch": 70.73,
2368
+ "learning_rate": 2.777777777777778e-05,
2369
+ "loss": 2.4296,
2370
+ "step": 187500
2371
+ },
2372
+ {
2373
+ "epoch": 70.73,
2374
+ "eval_loss": 2.3378496170043945,
2375
+ "eval_runtime": 283.6193,
2376
+ "eval_samples_per_second": 486.046,
2377
+ "eval_steps_per_second": 3.797,
2378
+ "step": 187500
2379
+ },
2380
+ {
2381
+ "epoch": 70.92,
2382
+ "learning_rate": 2.7555555555555555e-05,
2383
+ "loss": 2.4349,
2384
+ "step": 188000
2385
+ },
2386
+ {
2387
+ "epoch": 71.11,
2388
+ "learning_rate": 2.733333333333333e-05,
2389
+ "loss": 2.4326,
2390
+ "step": 188500
2391
+ },
2392
+ {
2393
+ "epoch": 71.29,
2394
+ "learning_rate": 2.7111111111111114e-05,
2395
+ "loss": 2.4284,
2396
+ "step": 189000
2397
+ },
2398
+ {
2399
+ "epoch": 71.48,
2400
+ "learning_rate": 2.688888888888889e-05,
2401
+ "loss": 2.4299,
2402
+ "step": 189500
2403
+ },
2404
+ {
2405
+ "epoch": 71.67,
2406
+ "learning_rate": 2.6666666666666667e-05,
2407
+ "loss": 2.4317,
2408
+ "step": 190000
2409
+ },
2410
+ {
2411
+ "epoch": 71.86,
2412
+ "learning_rate": 2.6444444444444443e-05,
2413
+ "loss": 2.4288,
2414
+ "step": 190500
2415
+ },
2416
+ {
2417
+ "epoch": 72.05,
2418
+ "learning_rate": 2.6222222222222226e-05,
2419
+ "loss": 2.4322,
2420
+ "step": 191000
2421
+ },
2422
+ {
2423
+ "epoch": 72.24,
2424
+ "learning_rate": 2.6000000000000002e-05,
2425
+ "loss": 2.4317,
2426
+ "step": 191500
2427
+ },
2428
+ {
2429
+ "epoch": 72.43,
2430
+ "learning_rate": 2.5777777777777778e-05,
2431
+ "loss": 2.4262,
2432
+ "step": 192000
2433
+ },
2434
+ {
2435
+ "epoch": 72.61,
2436
+ "learning_rate": 2.5555555555555554e-05,
2437
+ "loss": 2.4239,
2438
+ "step": 192500
2439
+ },
2440
+ {
2441
+ "epoch": 72.8,
2442
+ "learning_rate": 2.5333333333333337e-05,
2443
+ "loss": 2.4279,
2444
+ "step": 193000
2445
+ },
2446
+ {
2447
+ "epoch": 72.99,
2448
+ "learning_rate": 2.5111111111111113e-05,
2449
+ "loss": 2.4271,
2450
+ "step": 193500
2451
+ },
2452
+ {
2453
+ "epoch": 73.18,
2454
+ "learning_rate": 2.488888888888889e-05,
2455
+ "loss": 2.4349,
2456
+ "step": 194000
2457
+ },
2458
+ {
2459
+ "epoch": 73.37,
2460
+ "learning_rate": 2.466666666666667e-05,
2461
+ "loss": 2.4285,
2462
+ "step": 194500
2463
+ },
2464
+ {
2465
+ "epoch": 73.56,
2466
+ "learning_rate": 2.4444444444444445e-05,
2467
+ "loss": 2.4217,
2468
+ "step": 195000
2469
+ },
2470
+ {
2471
+ "epoch": 73.75,
2472
+ "learning_rate": 2.4222222222222224e-05,
2473
+ "loss": 2.4215,
2474
+ "step": 195500
2475
+ },
2476
+ {
2477
+ "epoch": 73.93,
2478
+ "learning_rate": 2.4e-05,
2479
+ "loss": 2.4267,
2480
+ "step": 196000
2481
+ },
2482
+ {
2483
+ "epoch": 74.12,
2484
+ "learning_rate": 2.377777777777778e-05,
2485
+ "loss": 2.4281,
2486
+ "step": 196500
2487
+ },
2488
+ {
2489
+ "epoch": 74.31,
2490
+ "learning_rate": 2.3555555555555556e-05,
2491
+ "loss": 2.4238,
2492
+ "step": 197000
2493
+ },
2494
+ {
2495
+ "epoch": 74.5,
2496
+ "learning_rate": 2.3333333333333336e-05,
2497
+ "loss": 2.4243,
2498
+ "step": 197500
2499
+ },
2500
+ {
2501
+ "epoch": 74.69,
2502
+ "learning_rate": 2.3111111111111112e-05,
2503
+ "loss": 2.4238,
2504
+ "step": 198000
2505
+ },
2506
+ {
2507
+ "epoch": 74.88,
2508
+ "learning_rate": 2.288888888888889e-05,
2509
+ "loss": 2.4231,
2510
+ "step": 198500
2511
+ },
2512
+ {
2513
+ "epoch": 75.07,
2514
+ "learning_rate": 2.2666666666666668e-05,
2515
+ "loss": 2.4231,
2516
+ "step": 199000
2517
+ },
2518
+ {
2519
+ "epoch": 75.25,
2520
+ "learning_rate": 2.2444444444444447e-05,
2521
+ "loss": 2.4238,
2522
+ "step": 199500
2523
+ },
2524
+ {
2525
+ "epoch": 75.44,
2526
+ "learning_rate": 2.2222222222222223e-05,
2527
+ "loss": 2.4213,
2528
+ "step": 200000
2529
+ },
2530
+ {
2531
+ "epoch": 75.44,
2532
+ "eval_loss": 2.328047513961792,
2533
+ "eval_runtime": 283.8545,
2534
+ "eval_samples_per_second": 485.643,
2535
+ "eval_steps_per_second": 3.794,
2536
+ "step": 200000
2537
+ },
2538
+ {
2539
+ "epoch": 75.63,
2540
+ "learning_rate": 2.2000000000000003e-05,
2541
+ "loss": 2.4195,
2542
+ "step": 200500
2543
+ },
2544
+ {
2545
+ "epoch": 75.82,
2546
+ "learning_rate": 2.177777777777778e-05,
2547
+ "loss": 2.427,
2548
+ "step": 201000
2549
+ },
2550
+ {
2551
+ "epoch": 76.01,
2552
+ "learning_rate": 2.1555555555555555e-05,
2553
+ "loss": 2.4211,
2554
+ "step": 201500
2555
+ },
2556
+ {
2557
+ "epoch": 76.2,
2558
+ "learning_rate": 2.1333333333333335e-05,
2559
+ "loss": 2.421,
2560
+ "step": 202000
2561
+ },
2562
+ {
2563
+ "epoch": 76.39,
2564
+ "learning_rate": 2.111111111111111e-05,
2565
+ "loss": 2.4212,
2566
+ "step": 202500
2567
+ },
2568
+ {
2569
+ "epoch": 76.57,
2570
+ "learning_rate": 2.088888888888889e-05,
2571
+ "loss": 2.4211,
2572
+ "step": 203000
2573
+ },
2574
+ {
2575
+ "epoch": 76.76,
2576
+ "learning_rate": 2.0666666666666666e-05,
2577
+ "loss": 2.4172,
2578
+ "step": 203500
2579
+ },
2580
+ {
2581
+ "epoch": 76.95,
2582
+ "learning_rate": 2.0444444444444446e-05,
2583
+ "loss": 2.4197,
2584
+ "step": 204000
2585
+ },
2586
+ {
2587
+ "epoch": 77.14,
2588
+ "learning_rate": 2.0222222222222222e-05,
2589
+ "loss": 2.4202,
2590
+ "step": 204500
2591
+ },
2592
+ {
2593
+ "epoch": 77.33,
2594
+ "learning_rate": 2e-05,
2595
+ "loss": 2.4198,
2596
+ "step": 205000
2597
+ },
2598
+ {
2599
+ "epoch": 77.52,
2600
+ "learning_rate": 1.9777777777777778e-05,
2601
+ "loss": 2.4182,
2602
+ "step": 205500
2603
+ },
2604
+ {
2605
+ "epoch": 77.71,
2606
+ "learning_rate": 1.9555555555555557e-05,
2607
+ "loss": 2.4151,
2608
+ "step": 206000
2609
+ },
2610
+ {
2611
+ "epoch": 77.89,
2612
+ "learning_rate": 1.9333333333333333e-05,
2613
+ "loss": 2.417,
2614
+ "step": 206500
2615
+ },
2616
+ {
2617
+ "epoch": 78.08,
2618
+ "learning_rate": 1.9111111111111113e-05,
2619
+ "loss": 2.4166,
2620
+ "step": 207000
2621
+ },
2622
+ {
2623
+ "epoch": 78.27,
2624
+ "learning_rate": 1.888888888888889e-05,
2625
+ "loss": 2.4173,
2626
+ "step": 207500
2627
+ },
2628
+ {
2629
+ "epoch": 78.46,
2630
+ "learning_rate": 1.866666666666667e-05,
2631
+ "loss": 2.413,
2632
+ "step": 208000
2633
+ },
2634
+ {
2635
+ "epoch": 78.65,
2636
+ "learning_rate": 1.8444444444444445e-05,
2637
+ "loss": 2.4116,
2638
+ "step": 208500
2639
+ },
2640
+ {
2641
+ "epoch": 78.84,
2642
+ "learning_rate": 1.8222222222222224e-05,
2643
+ "loss": 2.4129,
2644
+ "step": 209000
2645
+ },
2646
+ {
2647
+ "epoch": 79.03,
2648
+ "learning_rate": 1.8e-05,
2649
+ "loss": 2.4186,
2650
+ "step": 209500
2651
+ },
2652
+ {
2653
+ "epoch": 79.22,
2654
+ "learning_rate": 1.777777777777778e-05,
2655
+ "loss": 2.4105,
2656
+ "step": 210000
2657
+ },
2658
+ {
2659
+ "epoch": 79.4,
2660
+ "learning_rate": 1.7555555555555556e-05,
2661
+ "loss": 2.4136,
2662
+ "step": 210500
2663
+ },
2664
+ {
2665
+ "epoch": 79.59,
2666
+ "learning_rate": 1.7333333333333336e-05,
2667
+ "loss": 2.4117,
2668
+ "step": 211000
2669
+ },
2670
+ {
2671
+ "epoch": 79.78,
2672
+ "learning_rate": 1.7111111111111112e-05,
2673
+ "loss": 2.412,
2674
+ "step": 211500
2675
+ },
2676
+ {
2677
+ "epoch": 79.97,
2678
+ "learning_rate": 1.688888888888889e-05,
2679
+ "loss": 2.414,
2680
+ "step": 212000
2681
+ },
2682
+ {
2683
+ "epoch": 80.16,
2684
+ "learning_rate": 1.6666666666666667e-05,
2685
+ "loss": 2.4166,
2686
+ "step": 212500
2687
+ },
2688
+ {
2689
+ "epoch": 80.16,
2690
+ "eval_loss": 2.32003116607666,
2691
+ "eval_runtime": 283.7234,
2692
+ "eval_samples_per_second": 485.867,
2693
+ "eval_steps_per_second": 3.796,
2694
+ "step": 212500
2695
+ },
2696
+ {
2697
+ "epoch": 80.35,
2698
+ "learning_rate": 1.6444444444444447e-05,
2699
+ "loss": 2.4142,
2700
+ "step": 213000
2701
+ },
2702
+ {
2703
+ "epoch": 80.54,
2704
+ "learning_rate": 1.6222222222222223e-05,
2705
+ "loss": 2.4085,
2706
+ "step": 213500
2707
+ },
2708
+ {
2709
+ "epoch": 80.72,
2710
+ "learning_rate": 1.6000000000000003e-05,
2711
+ "loss": 2.4121,
2712
+ "step": 214000
2713
+ },
2714
+ {
2715
+ "epoch": 80.91,
2716
+ "learning_rate": 1.577777777777778e-05,
2717
+ "loss": 2.4077,
2718
+ "step": 214500
2719
+ },
2720
+ {
2721
+ "epoch": 81.1,
2722
+ "learning_rate": 1.5555555555555555e-05,
2723
+ "loss": 2.4129,
2724
+ "step": 215000
2725
+ },
2726
+ {
2727
+ "epoch": 81.29,
2728
+ "learning_rate": 1.5333333333333334e-05,
2729
+ "loss": 2.4103,
2730
+ "step": 215500
2731
+ },
2732
+ {
2733
+ "epoch": 81.48,
2734
+ "learning_rate": 1.5111111111111112e-05,
2735
+ "loss": 2.4116,
2736
+ "step": 216000
2737
+ },
2738
+ {
2739
+ "epoch": 81.67,
2740
+ "learning_rate": 1.4888888888888888e-05,
2741
+ "loss": 2.4073,
2742
+ "step": 216500
2743
+ },
2744
+ {
2745
+ "epoch": 81.86,
2746
+ "learning_rate": 1.4666666666666668e-05,
2747
+ "loss": 2.4111,
2748
+ "step": 217000
2749
+ },
2750
+ {
2751
+ "epoch": 82.04,
2752
+ "learning_rate": 1.4444444444444444e-05,
2753
+ "loss": 2.4099,
2754
+ "step": 217500
2755
+ },
2756
+ {
2757
+ "epoch": 82.23,
2758
+ "learning_rate": 1.4222222222222224e-05,
2759
+ "loss": 2.4051,
2760
+ "step": 218000
2761
+ },
2762
+ {
2763
+ "epoch": 82.42,
2764
+ "learning_rate": 1.4000000000000001e-05,
2765
+ "loss": 2.406,
2766
+ "step": 218500
2767
+ },
2768
+ {
2769
+ "epoch": 82.61,
2770
+ "learning_rate": 1.3777777777777778e-05,
2771
+ "loss": 2.4115,
2772
+ "step": 219000
2773
+ },
2774
+ {
2775
+ "epoch": 82.8,
2776
+ "learning_rate": 1.3555555555555557e-05,
2777
+ "loss": 2.4027,
2778
+ "step": 219500
2779
+ },
2780
+ {
2781
+ "epoch": 82.99,
2782
+ "learning_rate": 1.3333333333333333e-05,
2783
+ "loss": 2.4124,
2784
+ "step": 220000
2785
+ },
2786
+ {
2787
+ "epoch": 83.18,
2788
+ "learning_rate": 1.3111111111111113e-05,
2789
+ "loss": 2.4104,
2790
+ "step": 220500
2791
+ },
2792
+ {
2793
+ "epoch": 83.36,
2794
+ "learning_rate": 1.2888888888888889e-05,
2795
+ "loss": 2.405,
2796
+ "step": 221000
2797
+ },
2798
+ {
2799
+ "epoch": 83.55,
2800
+ "learning_rate": 1.2666666666666668e-05,
2801
+ "loss": 2.4078,
2802
+ "step": 221500
2803
+ },
2804
+ {
2805
+ "epoch": 83.74,
2806
+ "learning_rate": 1.2444444444444445e-05,
2807
+ "loss": 2.4091,
2808
+ "step": 222000
2809
+ },
2810
+ {
2811
+ "epoch": 83.93,
2812
+ "learning_rate": 1.2222222222222222e-05,
2813
+ "loss": 2.4013,
2814
+ "step": 222500
2815
+ },
2816
+ {
2817
+ "epoch": 84.12,
2818
+ "learning_rate": 1.2e-05,
2819
+ "loss": 2.4117,
2820
+ "step": 223000
2821
+ },
2822
+ {
2823
+ "epoch": 84.31,
2824
+ "learning_rate": 1.1777777777777778e-05,
2825
+ "loss": 2.408,
2826
+ "step": 223500
2827
+ },
2828
+ {
2829
+ "epoch": 84.5,
2830
+ "learning_rate": 1.1555555555555556e-05,
2831
+ "loss": 2.4035,
2832
+ "step": 224000
2833
+ },
2834
+ {
2835
+ "epoch": 84.68,
2836
+ "learning_rate": 1.1333333333333334e-05,
2837
+ "loss": 2.4005,
2838
+ "step": 224500
2839
+ },
2840
+ {
2841
+ "epoch": 84.87,
2842
+ "learning_rate": 1.1111111111111112e-05,
2843
+ "loss": 2.3995,
2844
+ "step": 225000
2845
+ },
2846
+ {
2847
+ "epoch": 84.87,
2848
+ "eval_loss": 2.3128256797790527,
2849
+ "eval_runtime": 283.7082,
2850
+ "eval_samples_per_second": 485.894,
2851
+ "eval_steps_per_second": 3.796,
2852
+ "step": 225000
2853
+ },
2854
+ {
2855
+ "epoch": 85.06,
2856
+ "learning_rate": 1.088888888888889e-05,
2857
+ "loss": 2.41,
2858
+ "step": 225500
2859
+ },
2860
+ {
2861
+ "epoch": 85.25,
2862
+ "learning_rate": 1.0666666666666667e-05,
2863
+ "loss": 2.4006,
2864
+ "step": 226000
2865
+ },
2866
+ {
2867
+ "epoch": 85.44,
2868
+ "learning_rate": 1.0444444444444445e-05,
2869
+ "loss": 2.4071,
2870
+ "step": 226500
2871
+ },
2872
+ {
2873
+ "epoch": 85.63,
2874
+ "learning_rate": 1.0222222222222223e-05,
2875
+ "loss": 2.4026,
2876
+ "step": 227000
2877
+ },
2878
+ {
2879
+ "epoch": 85.82,
2880
+ "learning_rate": 1e-05,
2881
+ "loss": 2.4054,
2882
+ "step": 227500
2883
+ },
2884
+ {
2885
+ "epoch": 86.01,
2886
+ "learning_rate": 9.777777777777779e-06,
2887
+ "loss": 2.406,
2888
+ "step": 228000
2889
+ },
2890
+ {
2891
+ "epoch": 86.19,
2892
+ "learning_rate": 9.555555555555556e-06,
2893
+ "loss": 2.4008,
2894
+ "step": 228500
2895
+ },
2896
+ {
2897
+ "epoch": 86.38,
2898
+ "learning_rate": 9.333333333333334e-06,
2899
+ "loss": 2.4,
2900
+ "step": 229000
2901
+ },
2902
+ {
2903
+ "epoch": 86.57,
2904
+ "learning_rate": 9.111111111111112e-06,
2905
+ "loss": 2.4014,
2906
+ "step": 229500
2907
+ },
2908
+ {
2909
+ "epoch": 86.76,
2910
+ "learning_rate": 8.88888888888889e-06,
2911
+ "loss": 2.4026,
2912
+ "step": 230000
2913
+ },
2914
+ {
2915
+ "epoch": 86.95,
2916
+ "learning_rate": 8.666666666666668e-06,
2917
+ "loss": 2.4038,
2918
+ "step": 230500
2919
+ },
2920
+ {
2921
+ "epoch": 87.14,
2922
+ "learning_rate": 8.444444444444446e-06,
2923
+ "loss": 2.4036,
2924
+ "step": 231000
2925
+ },
2926
+ {
2927
+ "epoch": 87.33,
2928
+ "learning_rate": 8.222222222222223e-06,
2929
+ "loss": 2.4024,
2930
+ "step": 231500
2931
+ },
2932
+ {
2933
+ "epoch": 87.51,
2934
+ "learning_rate": 8.000000000000001e-06,
2935
+ "loss": 2.4045,
2936
+ "step": 232000
2937
+ },
2938
+ {
2939
+ "epoch": 87.7,
2940
+ "learning_rate": 7.777777777777777e-06,
2941
+ "loss": 2.4001,
2942
+ "step": 232500
2943
+ },
2944
+ {
2945
+ "epoch": 87.89,
2946
+ "learning_rate": 7.555555555555556e-06,
2947
+ "loss": 2.4001,
2948
+ "step": 233000
2949
+ },
2950
+ {
2951
+ "epoch": 88.08,
2952
+ "learning_rate": 7.333333333333334e-06,
2953
+ "loss": 2.4025,
2954
+ "step": 233500
2955
+ },
2956
+ {
2957
+ "epoch": 88.27,
2958
+ "learning_rate": 7.111111111111112e-06,
2959
+ "loss": 2.3957,
2960
+ "step": 234000
2961
+ },
2962
+ {
2963
+ "epoch": 88.46,
2964
+ "learning_rate": 6.888888888888889e-06,
2965
+ "loss": 2.3971,
2966
+ "step": 234500
2967
+ },
2968
+ {
2969
+ "epoch": 88.65,
2970
+ "learning_rate": 6.666666666666667e-06,
2971
+ "loss": 2.4007,
2972
+ "step": 235000
2973
+ },
2974
+ {
2975
+ "epoch": 88.83,
2976
+ "learning_rate": 6.4444444444444445e-06,
2977
+ "loss": 2.4002,
2978
+ "step": 235500
2979
+ },
2980
+ {
2981
+ "epoch": 89.02,
2982
+ "learning_rate": 6.222222222222222e-06,
2983
+ "loss": 2.3995,
2984
+ "step": 236000
2985
+ },
2986
+ {
2987
+ "epoch": 89.21,
2988
+ "learning_rate": 6e-06,
2989
+ "loss": 2.3985,
2990
+ "step": 236500
2991
+ },
2992
+ {
2993
+ "epoch": 89.4,
2994
+ "learning_rate": 5.777777777777778e-06,
2995
+ "loss": 2.4028,
2996
+ "step": 237000
2997
+ },
2998
+ {
2999
+ "epoch": 89.59,
3000
+ "learning_rate": 5.555555555555556e-06,
3001
+ "loss": 2.395,
3002
+ "step": 237500
3003
+ },
3004
+ {
3005
+ "epoch": 89.59,
3006
+ "eval_loss": 2.3064496517181396,
3007
+ "eval_runtime": 283.9176,
3008
+ "eval_samples_per_second": 485.535,
3009
+ "eval_steps_per_second": 3.793,
3010
+ "step": 237500
3011
+ },
3012
+ {
3013
+ "epoch": 89.78,
3014
+ "learning_rate": 5.333333333333334e-06,
3015
+ "loss": 2.3968,
3016
+ "step": 238000
3017
+ },
3018
+ {
3019
+ "epoch": 89.97,
3020
+ "learning_rate": 5.1111111111111115e-06,
3021
+ "loss": 2.3952,
3022
+ "step": 238500
3023
+ },
3024
+ {
3025
+ "epoch": 90.15,
3026
+ "learning_rate": 4.888888888888889e-06,
3027
+ "loss": 2.4006,
3028
+ "step": 239000
3029
+ },
3030
+ {
3031
+ "epoch": 90.34,
3032
+ "learning_rate": 4.666666666666667e-06,
3033
+ "loss": 2.3952,
3034
+ "step": 239500
3035
+ },
3036
+ {
3037
+ "epoch": 90.53,
3038
+ "learning_rate": 4.444444444444445e-06,
3039
+ "loss": 2.3949,
3040
+ "step": 240000
3041
+ },
3042
+ {
3043
+ "epoch": 90.72,
3044
+ "learning_rate": 4.222222222222223e-06,
3045
+ "loss": 2.3966,
3046
+ "step": 240500
3047
+ },
3048
+ {
3049
+ "epoch": 90.91,
3050
+ "learning_rate": 4.000000000000001e-06,
3051
+ "loss": 2.3955,
3052
+ "step": 241000
3053
+ },
3054
+ {
3055
+ "epoch": 91.1,
3056
+ "learning_rate": 3.777777777777778e-06,
3057
+ "loss": 2.4001,
3058
+ "step": 241500
3059
+ },
3060
+ {
3061
+ "epoch": 91.29,
3062
+ "learning_rate": 3.555555555555556e-06,
3063
+ "loss": 2.3956,
3064
+ "step": 242000
3065
+ },
3066
+ {
3067
+ "epoch": 91.47,
3068
+ "learning_rate": 3.3333333333333333e-06,
3069
+ "loss": 2.3951,
3070
+ "step": 242500
3071
+ },
3072
+ {
3073
+ "epoch": 91.66,
3074
+ "learning_rate": 3.111111111111111e-06,
3075
+ "loss": 2.4001,
3076
+ "step": 243000
3077
+ },
3078
+ {
3079
+ "epoch": 91.85,
3080
+ "learning_rate": 2.888888888888889e-06,
3081
+ "loss": 2.3945,
3082
+ "step": 243500
3083
+ },
3084
+ {
3085
+ "epoch": 92.04,
3086
+ "learning_rate": 2.666666666666667e-06,
3087
+ "loss": 2.395,
3088
+ "step": 244000
3089
+ },
3090
+ {
3091
+ "epoch": 92.23,
3092
+ "learning_rate": 2.4444444444444447e-06,
3093
+ "loss": 2.3963,
3094
+ "step": 244500
3095
+ },
3096
+ {
3097
+ "epoch": 92.42,
3098
+ "learning_rate": 2.2222222222222225e-06,
3099
+ "loss": 2.3944,
3100
+ "step": 245000
3101
+ },
3102
+ {
3103
+ "epoch": 92.61,
3104
+ "learning_rate": 2.0000000000000003e-06,
3105
+ "loss": 2.3951,
3106
+ "step": 245500
3107
+ },
3108
+ {
3109
+ "epoch": 92.79,
3110
+ "learning_rate": 1.777777777777778e-06,
3111
+ "loss": 2.3897,
3112
+ "step": 246000
3113
+ },
3114
+ {
3115
+ "epoch": 92.98,
3116
+ "learning_rate": 1.5555555555555556e-06,
3117
+ "loss": 2.3926,
3118
+ "step": 246500
3119
+ },
3120
+ {
3121
+ "epoch": 93.17,
3122
+ "learning_rate": 1.3333333333333334e-06,
3123
+ "loss": 2.3968,
3124
+ "step": 247000
3125
+ },
3126
+ {
3127
+ "epoch": 93.36,
3128
+ "learning_rate": 1.1111111111111112e-06,
3129
+ "loss": 2.3924,
3130
+ "step": 247500
3131
+ },
3132
+ {
3133
+ "epoch": 93.55,
3134
+ "learning_rate": 8.88888888888889e-07,
3135
+ "loss": 2.3908,
3136
+ "step": 248000
3137
+ },
3138
+ {
3139
+ "epoch": 93.74,
3140
+ "learning_rate": 6.666666666666667e-07,
3141
+ "loss": 2.3932,
3142
+ "step": 248500
3143
+ },
3144
+ {
3145
+ "epoch": 93.93,
3146
+ "learning_rate": 4.444444444444445e-07,
3147
+ "loss": 2.3942,
3148
+ "step": 249000
3149
+ },
3150
+ {
3151
+ "epoch": 94.12,
3152
+ "learning_rate": 2.2222222222222224e-07,
3153
+ "loss": 2.398,
3154
+ "step": 249500
3155
+ },
3156
+ {
3157
+ "epoch": 94.3,
3158
+ "learning_rate": 0.0,
3159
+ "loss": 2.3932,
3160
+ "step": 250000
3161
+ },
3162
+ {
3163
+ "epoch": 94.3,
3164
+ "eval_loss": 2.302946090698242,
3165
+ "eval_runtime": 283.7177,
3166
+ "eval_samples_per_second": 485.877,
3167
+ "eval_steps_per_second": 3.796,
3168
+ "step": 250000
3169
+ }
3170
+ ],
3171
+ "max_steps": 250000,
3172
+ "num_train_epochs": 95,
3173
+ "total_flos": 4.2134352665063916e+18,
3174
+ "trial_name": null,
3175
+ "trial_params": null
3176
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f59ef38e343b6167de80144ed6d7d8e716282d43ed98908d70aad132c770e990
3
+ size 2927
vocab.json ADDED
The diff for this file is too large to render. See raw diff