Hugging Face's logo

jonasknobloch
/
gpt2-ts_cx-en_00000-00009_50k

Text Generation
Transformers
Safetensors
English
gpt2
Generated from Trainer
Eval Results
text-generation-inference
Inference Endpoints
Model card Files Community
jonasknobloch commited on
Commit
89e6e1d
•
1 Parent(s): 8f8c8d6

Initial commit

Browse files
Files changed (12) hide show
  1. README.md +96 -0
  2. all_results.json +15 -0
  3. config.json +31 -0
  4. eval_results.json +10 -0
  5. generation_config.json +6 -0
  6. model.safetensors +3 -0
  7. special_tokens_map.json +1 -0
  8. tokenizer.json +0 -0
  9. tokenizer_config.json +15 -0
  10. train_results.json +8 -0
  11. trainer_state.json +3426 -0
  12. training_args.bin +3 -0
README.md ADDED
@@ -0,0 +1,96 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - generated_from_trainer
4
+ datasets:
5
+ - uonlp/CulturaX
6
+ metrics:
7
+ - accuracy
8
+ model-index:
9
+ - name: gpt2+ts_cx-en_00000-00009_50k
10
+ results:
11
+ - task:
12
+ name: Causal Language Modeling
13
+ type: text-generation
14
+ dataset:
15
+ name: uonlp/CulturaX en
16
+ type: uonlp/CulturaX
17
+ args: en
18
+ metrics:
19
+ - name: Accuracy
20
+ type: accuracy
21
+ value: 0.3894698710798747
22
+ ---
23
+
24
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
25
+ should probably proofread and complete it, then remove this comment. -->
26
+
27
+ # gpt2+ts_cx-en_00000-00009_50k
28
+
29
+ This model is a fine-tuned version of [](https://huggingface.co/) on the uonlp/CulturaX en dataset.
30
+ It achieves the following results on the evaluation set:
31
+ - Loss: 3.4121
32
+ - Accuracy: 0.3895
33
+
34
+ ## Model description
35
+
36
+ More information needed
37
+
38
+ ## Intended uses & limitations
39
+
40
+ More information needed
41
+
42
+ ## Training and evaluation data
43
+
44
+ More information needed
45
+
46
+ ## Training procedure
47
+
48
+ ### Training hyperparameters
49
+
50
+ The following hyperparameters were used during training:
51
+ - learning_rate: 5e-05
52
+ - train_batch_size: 64
53
+ - eval_batch_size: 64
54
+ - seed: 42
55
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
56
+ - lr_scheduler_type: linear
57
+ - num_epochs: 1.0
58
+
59
+ ### Training results
60
+
61
+ | Training Loss | Epoch | Step | Validation Loss | Accuracy |
62
+ |:-------------:|:-----:|:------:|:---------------:|:--------:|
63
+ | 4.375 | 0.04 | 10000 | 4.2815 | 0.3111 |
64
+ | 4.0754 | 0.08 | 20000 | 3.9984 | 0.3341 |
65
+ | 3.9409 | 0.11 | 30000 | 3.8615 | 0.3457 |
66
+ | 3.8554 | 0.15 | 40000 | 3.7798 | 0.3531 |
67
+ | 3.7973 | 0.19 | 50000 | 3.7210 | 0.3584 |
68
+ | 3.7421 | 0.23 | 60000 | 3.6750 | 0.3630 |
69
+ | 3.7097 | 0.27 | 70000 | 3.6378 | 0.3664 |
70
+ | 3.6741 | 0.3 | 80000 | 3.6061 | 0.3694 |
71
+ | 3.6599 | 0.34 | 90000 | 3.5803 | 0.3718 |
72
+ | 3.6356 | 0.38 | 100000 | 3.5584 | 0.3741 |
73
+ | 3.6131 | 0.42 | 110000 | 3.5423 | 0.3758 |
74
+ | 3.5991 | 0.46 | 120000 | 3.5254 | 0.3776 |
75
+ | 3.591 | 0.49 | 130000 | 3.5108 | 0.3790 |
76
+ | 3.574 | 0.53 | 140000 | 3.4966 | 0.3805 |
77
+ | 3.5606 | 0.57 | 150000 | 3.4866 | 0.3815 |
78
+ | 3.5516 | 0.61 | 160000 | 3.4739 | 0.3828 |
79
+ | 3.5423 | 0.64 | 170000 | 3.4650 | 0.3838 |
80
+ | 3.5298 | 0.68 | 180000 | 3.4560 | 0.3847 |
81
+ | 3.5287 | 0.72 | 190000 | 3.4479 | 0.3857 |
82
+ | 3.5187 | 0.76 | 200000 | 3.4408 | 0.3863 |
83
+ | 3.5157 | 0.8 | 210000 | 3.4339 | 0.3870 |
84
+ | 3.5042 | 0.83 | 220000 | 3.4286 | 0.3876 |
85
+ | 3.5033 | 0.87 | 230000 | 3.4229 | 0.3883 |
86
+ | 3.501 | 0.91 | 240000 | 3.4188 | 0.3888 |
87
+ | 3.4946 | 0.95 | 250000 | 3.4149 | 0.3892 |
88
+ | 3.4971 | 0.99 | 260000 | 3.4126 | 0.3894 |
89
+
90
+
91
+ ### Framework versions
92
+
93
+ - Transformers 4.37.1
94
+ - Pytorch 2.1.2+cu121
95
+ - Datasets 2.16.1
96
+ - Tokenizers 0.15.1
all_results.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_accuracy": 0.3894698710798747,
4
+ "eval_loss": 3.4120876789093018,
5
+ "eval_runtime": 5016.6058,
6
+ "eval_samples": 443773,
7
+ "eval_samples_per_second": 88.461,
8
+ "eval_steps_per_second": 1.382,
9
+ "perplexity": 30.32849437151538,
10
+ "train_loss": 3.6950655784551065,
11
+ "train_runtime": 269467.2599,
12
+ "train_samples": 8434107,
13
+ "train_samples_per_second": 31.299,
14
+ "train_steps_per_second": 0.978
15
+ }
config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "activation_function": "gelu_new",
3
+ "architectures": [
4
+ "GPT2LMHeadModel"
5
+ ],
6
+ "attn_pdrop": 0.1,
7
+ "bos_token_id": 50256,
8
+ "embd_pdrop": 0.1,
9
+ "eos_token_id": 50256,
10
+ "initializer_range": 0.02,
11
+ "layer_norm_epsilon": 1e-05,
12
+ "model_type": "gpt2",
13
+ "n_embd": 768,
14
+ "n_head": 12,
15
+ "n_inner": null,
16
+ "n_layer": 12,
17
+ "n_positions": 1024,
18
+ "reorder_and_upcast_attn": false,
19
+ "resid_pdrop": 0.1,
20
+ "scale_attn_by_inverse_layer_idx": false,
21
+ "scale_attn_weights": true,
22
+ "summary_activation": null,
23
+ "summary_first_dropout": 0.1,
24
+ "summary_proj_to_labels": true,
25
+ "summary_type": "cls_index",
26
+ "summary_use_proj": true,
27
+ "torch_dtype": "float32",
28
+ "transformers_version": "4.37.1",
29
+ "use_cache": true,
30
+ "vocab_size": 50257
31
+ }
eval_results.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_accuracy": 0.3894698710798747,
4
+ "eval_loss": 3.4120876789093018,
5
+ "eval_runtime": 5016.6058,
6
+ "eval_samples": 443773,
7
+ "eval_samples_per_second": 88.461,
8
+ "eval_steps_per_second": 1.382,
9
+ "perplexity": 30.32849437151538
10
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 50256,
4
+ "eos_token_id": 50256,
5
+ "transformers_version": "4.37.1"
6
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35b202d2fec8227b98b48ed9e346d8eb881ffd6163c6a768999724918bf394c4
3
+ size 497774208
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {}
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "50256": {
4
+ "content": "<|endoftext|>",
5
+ "lstrip": false,
6
+ "normalized": true,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ }
11
+ },
12
+ "clean_up_tokenization_spaces": true,
13
+ "model_max_length": 1000000000000000019884624838656,
14
+ "tokenizer_class": "PreTrainedTokenizerFast"
15
+ }
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "train_loss": 3.6950655784551065,
4
+ "train_runtime": 269467.2599,
5
+ "train_samples": 8434107,
6
+ "train_samples_per_second": 31.299,
7
+ "train_steps_per_second": 0.978
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,3426 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 10000,
6
+ "global_step": 263566,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0,
13
+ "learning_rate": 4.990514709788061e-05,
14
+ "loss": 7.0943,
15
+ "step": 500
16
+ },
17
+ {
18
+ "epoch": 0.0,
19
+ "learning_rate": 4.981029419576122e-05,
20
+ "loss": 6.2317,
21
+ "step": 1000
22
+ },
23
+ {
24
+ "epoch": 0.01,
25
+ "learning_rate": 4.971544129364182e-05,
26
+ "loss": 5.9001,
27
+ "step": 1500
28
+ },
29
+ {
30
+ "epoch": 0.01,
31
+ "learning_rate": 4.9620588391522424e-05,
32
+ "loss": 5.6651,
33
+ "step": 2000
34
+ },
35
+ {
36
+ "epoch": 0.01,
37
+ "learning_rate": 4.9525735489403035e-05,
38
+ "loss": 5.4754,
39
+ "step": 2500
40
+ },
41
+ {
42
+ "epoch": 0.01,
43
+ "learning_rate": 4.9430882587283645e-05,
44
+ "loss": 5.3245,
45
+ "step": 3000
46
+ },
47
+ {
48
+ "epoch": 0.01,
49
+ "learning_rate": 4.933602968516425e-05,
50
+ "loss": 5.1815,
51
+ "step": 3500
52
+ },
53
+ {
54
+ "epoch": 0.02,
55
+ "learning_rate": 4.924117678304486e-05,
56
+ "loss": 5.0484,
57
+ "step": 4000
58
+ },
59
+ {
60
+ "epoch": 0.02,
61
+ "learning_rate": 4.914632388092546e-05,
62
+ "loss": 4.9349,
63
+ "step": 4500
64
+ },
65
+ {
66
+ "epoch": 0.02,
67
+ "learning_rate": 4.905147097880607e-05,
68
+ "loss": 4.8371,
69
+ "step": 5000
70
+ },
71
+ {
72
+ "epoch": 0.02,
73
+ "learning_rate": 4.895661807668668e-05,
74
+ "loss": 4.7574,
75
+ "step": 5500
76
+ },
77
+ {
78
+ "epoch": 0.02,
79
+ "learning_rate": 4.886176517456728e-05,
80
+ "loss": 4.6903,
81
+ "step": 6000
82
+ },
83
+ {
84
+ "epoch": 0.02,
85
+ "learning_rate": 4.876691227244789e-05,
86
+ "loss": 4.626,
87
+ "step": 6500
88
+ },
89
+ {
90
+ "epoch": 0.03,
91
+ "learning_rate": 4.8672059370328496e-05,
92
+ "loss": 4.5767,
93
+ "step": 7000
94
+ },
95
+ {
96
+ "epoch": 0.03,
97
+ "learning_rate": 4.85772064682091e-05,
98
+ "loss": 4.5388,
99
+ "step": 7500
100
+ },
101
+ {
102
+ "epoch": 0.03,
103
+ "learning_rate": 4.848235356608971e-05,
104
+ "loss": 4.4959,
105
+ "step": 8000
106
+ },
107
+ {
108
+ "epoch": 0.03,
109
+ "learning_rate": 4.838750066397032e-05,
110
+ "loss": 4.4589,
111
+ "step": 8500
112
+ },
113
+ {
114
+ "epoch": 0.03,
115
+ "learning_rate": 4.8292647761850924e-05,
116
+ "loss": 4.4324,
117
+ "step": 9000
118
+ },
119
+ {
120
+ "epoch": 0.04,
121
+ "learning_rate": 4.8197794859731535e-05,
122
+ "loss": 4.3988,
123
+ "step": 9500
124
+ },
125
+ {
126
+ "epoch": 0.04,
127
+ "learning_rate": 4.810294195761214e-05,
128
+ "loss": 4.375,
129
+ "step": 10000
130
+ },
131
+ {
132
+ "epoch": 0.04,
133
+ "eval_accuracy": 0.31109238678227563,
134
+ "eval_loss": 4.28147029876709,
135
+ "eval_runtime": 5060.877,
136
+ "eval_samples_per_second": 87.687,
137
+ "eval_steps_per_second": 1.37,
138
+ "step": 10000
139
+ },
140
+ {
141
+ "epoch": 0.04,
142
+ "learning_rate": 4.800808905549274e-05,
143
+ "loss": 4.3423,
144
+ "step": 10500
145
+ },
146
+ {
147
+ "epoch": 0.04,
148
+ "learning_rate": 4.791323615337335e-05,
149
+ "loss": 4.3204,
150
+ "step": 11000
151
+ },
152
+ {
153
+ "epoch": 0.04,
154
+ "learning_rate": 4.7818383251253956e-05,
155
+ "loss": 4.2991,
156
+ "step": 11500
157
+ },
158
+ {
159
+ "epoch": 0.05,
160
+ "learning_rate": 4.772353034913457e-05,
161
+ "loss": 4.2818,
162
+ "step": 12000
163
+ },
164
+ {
165
+ "epoch": 0.05,
166
+ "learning_rate": 4.762867744701517e-05,
167
+ "loss": 4.2627,
168
+ "step": 12500
169
+ },
170
+ {
171
+ "epoch": 0.05,
172
+ "learning_rate": 4.7533824544895774e-05,
173
+ "loss": 4.2463,
174
+ "step": 13000
175
+ },
176
+ {
177
+ "epoch": 0.05,
178
+ "learning_rate": 4.7438971642776385e-05,
179
+ "loss": 4.2265,
180
+ "step": 13500
181
+ },
182
+ {
183
+ "epoch": 0.05,
184
+ "learning_rate": 4.734411874065699e-05,
185
+ "loss": 4.2159,
186
+ "step": 14000
187
+ },
188
+ {
189
+ "epoch": 0.06,
190
+ "learning_rate": 4.72492658385376e-05,
191
+ "loss": 4.198,
192
+ "step": 14500
193
+ },
194
+ {
195
+ "epoch": 0.06,
196
+ "learning_rate": 4.715441293641821e-05,
197
+ "loss": 4.1886,
198
+ "step": 15000
199
+ },
200
+ {
201
+ "epoch": 0.06,
202
+ "learning_rate": 4.7059560034298813e-05,
203
+ "loss": 4.1675,
204
+ "step": 15500
205
+ },
206
+ {
207
+ "epoch": 0.06,
208
+ "learning_rate": 4.696470713217942e-05,
209
+ "loss": 4.1553,
210
+ "step": 16000
211
+ },
212
+ {
213
+ "epoch": 0.06,
214
+ "learning_rate": 4.686985423006002e-05,
215
+ "loss": 4.1466,
216
+ "step": 16500
217
+ },
218
+ {
219
+ "epoch": 0.06,
220
+ "learning_rate": 4.677500132794063e-05,
221
+ "loss": 4.1307,
222
+ "step": 17000
223
+ },
224
+ {
225
+ "epoch": 0.07,
226
+ "learning_rate": 4.668014842582124e-05,
227
+ "loss": 4.1282,
228
+ "step": 17500
229
+ },
230
+ {
231
+ "epoch": 0.07,
232
+ "learning_rate": 4.6585295523701846e-05,
233
+ "loss": 4.1108,
234
+ "step": 18000
235
+ },
236
+ {
237
+ "epoch": 0.07,
238
+ "learning_rate": 4.6490442621582456e-05,
239
+ "loss": 4.1135,
240
+ "step": 18500
241
+ },
242
+ {
243
+ "epoch": 0.07,
244
+ "learning_rate": 4.639558971946305e-05,
245
+ "loss": 4.0876,
246
+ "step": 19000
247
+ },
248
+ {
249
+ "epoch": 0.07,
250
+ "learning_rate": 4.6300736817343664e-05,
251
+ "loss": 4.0871,
252
+ "step": 19500
253
+ },
254
+ {
255
+ "epoch": 0.08,
256
+ "learning_rate": 4.6205883915224274e-05,
257
+ "loss": 4.0754,
258
+ "step": 20000
259
+ },
260
+ {
261
+ "epoch": 0.08,
262
+ "eval_accuracy": 0.3340521142462603,
263
+ "eval_loss": 3.9983978271484375,
264
+ "eval_runtime": 5045.5203,
265
+ "eval_samples_per_second": 87.954,
266
+ "eval_steps_per_second": 1.374,
267
+ "step": 20000
268
+ },
269
+ {
270
+ "epoch": 0.08,
271
+ "learning_rate": 4.611103101310488e-05,
272
+ "loss": 4.0747,
273
+ "step": 20500
274
+ },
275
+ {
276
+ "epoch": 0.08,
277
+ "learning_rate": 4.601617811098549e-05,
278
+ "loss": 4.0582,
279
+ "step": 21000
280
+ },
281
+ {
282
+ "epoch": 0.08,
283
+ "learning_rate": 4.592132520886609e-05,
284
+ "loss": 4.0545,
285
+ "step": 21500
286
+ },
287
+ {
288
+ "epoch": 0.08,
289
+ "learning_rate": 4.5826472306746696e-05,
290
+ "loss": 4.0345,
291
+ "step": 22000
292
+ },
293
+ {
294
+ "epoch": 0.09,
295
+ "learning_rate": 4.5731619404627307e-05,
296
+ "loss": 4.0376,
297
+ "step": 22500
298
+ },
299
+ {
300
+ "epoch": 0.09,
301
+ "learning_rate": 4.563676650250791e-05,
302
+ "loss": 4.0315,
303
+ "step": 23000
304
+ },
305
+ {
306
+ "epoch": 0.09,
307
+ "learning_rate": 4.554191360038852e-05,
308
+ "loss": 4.0217,
309
+ "step": 23500
310
+ },
311
+ {
312
+ "epoch": 0.09,
313
+ "learning_rate": 4.544706069826913e-05,
314
+ "loss": 4.0135,
315
+ "step": 24000
316
+ },
317
+ {
318
+ "epoch": 0.09,
319
+ "learning_rate": 4.535220779614973e-05,
320
+ "loss": 4.0034,
321
+ "step": 24500
322
+ },
323
+ {
324
+ "epoch": 0.09,
325
+ "learning_rate": 4.525735489403034e-05,
326
+ "loss": 4.004,
327
+ "step": 25000
328
+ },
329
+ {
330
+ "epoch": 0.1,
331
+ "learning_rate": 4.516250199191095e-05,
332
+ "loss": 3.9939,
333
+ "step": 25500
334
+ },
335
+ {
336
+ "epoch": 0.1,
337
+ "learning_rate": 4.506764908979155e-05,
338
+ "loss": 3.9871,
339
+ "step": 26000
340
+ },
341
+ {
342
+ "epoch": 0.1,
343
+ "learning_rate": 4.4972796187672164e-05,
344
+ "loss": 3.9787,
345
+ "step": 26500
346
+ },
347
+ {
348
+ "epoch": 0.1,
349
+ "learning_rate": 4.487794328555277e-05,
350
+ "loss": 3.9752,
351
+ "step": 27000
352
+ },
353
+ {
354
+ "epoch": 0.1,
355
+ "learning_rate": 4.478309038343337e-05,
356
+ "loss": 3.9605,
357
+ "step": 27500
358
+ },
359
+ {
360
+ "epoch": 0.11,
361
+ "learning_rate": 4.468823748131398e-05,
362
+ "loss": 3.9542,
363
+ "step": 28000
364
+ },
365
+ {
366
+ "epoch": 0.11,
367
+ "learning_rate": 4.4593384579194585e-05,
368
+ "loss": 3.9518,
369
+ "step": 28500
370
+ },
371
+ {
372
+ "epoch": 0.11,
373
+ "learning_rate": 4.4498531677075196e-05,
374
+ "loss": 3.9479,
375
+ "step": 29000
376
+ },
377
+ {
378
+ "epoch": 0.11,
379
+ "learning_rate": 4.44036787749558e-05,
380
+ "loss": 3.9445,
381
+ "step": 29500
382
+ },
383
+ {
384
+ "epoch": 0.11,
385
+ "learning_rate": 4.430882587283641e-05,
386
+ "loss": 3.9409,
387
+ "step": 30000
388
+ },
389
+ {
390
+ "epoch": 0.11,
391
+ "eval_accuracy": 0.3456931371386037,
392
+ "eval_loss": 3.86149525642395,
393
+ "eval_runtime": 5041.4629,
394
+ "eval_samples_per_second": 88.025,
395
+ "eval_steps_per_second": 1.375,
396
+ "step": 30000
397
+ },
398
+ {
399
+ "epoch": 0.12,
400
+ "learning_rate": 4.4213972970717014e-05,
401
+ "loss": 3.9345,
402
+ "step": 30500
403
+ },
404
+ {
405
+ "epoch": 0.12,
406
+ "learning_rate": 4.411912006859762e-05,
407
+ "loss": 3.9328,
408
+ "step": 31000
409
+ },
410
+ {
411
+ "epoch": 0.12,
412
+ "learning_rate": 4.402426716647823e-05,
413
+ "loss": 3.9248,
414
+ "step": 31500
415
+ },
416
+ {
417
+ "epoch": 0.12,
418
+ "learning_rate": 4.392941426435884e-05,
419
+ "loss": 3.9126,
420
+ "step": 32000
421
+ },
422
+ {
423
+ "epoch": 0.12,
424
+ "learning_rate": 4.383456136223944e-05,
425
+ "loss": 3.9077,
426
+ "step": 32500
427
+ },
428
+ {
429
+ "epoch": 0.13,
430
+ "learning_rate": 4.3739708460120046e-05,
431
+ "loss": 3.9102,
432
+ "step": 33000
433
+ },
434
+ {
435
+ "epoch": 0.13,
436
+ "learning_rate": 4.364485555800065e-05,
437
+ "loss": 3.907,
438
+ "step": 33500
439
+ },
440
+ {
441
+ "epoch": 0.13,
442
+ "learning_rate": 4.355000265588126e-05,
443
+ "loss": 3.9017,
444
+ "step": 34000
445
+ },
446
+ {
447
+ "epoch": 0.13,
448
+ "learning_rate": 4.345514975376187e-05,
449
+ "loss": 3.8976,
450
+ "step": 34500
451
+ },
452
+ {
453
+ "epoch": 0.13,
454
+ "learning_rate": 4.3360296851642475e-05,
455
+ "loss": 3.8917,
456
+ "step": 35000
457
+ },
458
+ {
459
+ "epoch": 0.13,
460
+ "learning_rate": 4.3265443949523085e-05,
461
+ "loss": 3.8917,
462
+ "step": 35500
463
+ },
464
+ {
465
+ "epoch": 0.14,
466
+ "learning_rate": 4.317059104740369e-05,
467
+ "loss": 3.8851,
468
+ "step": 36000
469
+ },
470
+ {
471
+ "epoch": 0.14,
472
+ "learning_rate": 4.307573814528429e-05,
473
+ "loss": 3.8899,
474
+ "step": 36500
475
+ },
476
+ {
477
+ "epoch": 0.14,
478
+ "learning_rate": 4.29808852431649e-05,
479
+ "loss": 3.8667,
480
+ "step": 37000
481
+ },
482
+ {
483
+ "epoch": 0.14,
484
+ "learning_rate": 4.288603234104551e-05,
485
+ "loss": 3.8702,
486
+ "step": 37500
487
+ },
488
+ {
489
+ "epoch": 0.14,
490
+ "learning_rate": 4.279117943892612e-05,
491
+ "loss": 3.8705,
492
+ "step": 38000
493
+ },
494
+ {
495
+ "epoch": 0.15,
496
+ "learning_rate": 4.269632653680672e-05,
497
+ "loss": 3.8647,
498
+ "step": 38500
499
+ },
500
+ {
501
+ "epoch": 0.15,
502
+ "learning_rate": 4.2601473634687325e-05,
503
+ "loss": 3.8612,
504
+ "step": 39000
505
+ },
506
+ {
507
+ "epoch": 0.15,
508
+ "learning_rate": 4.2506620732567936e-05,
509
+ "loss": 3.8524,
510
+ "step": 39500
511
+ },
512
+ {
513
+ "epoch": 0.15,
514
+ "learning_rate": 4.241176783044854e-05,
515
+ "loss": 3.8554,
516
+ "step": 40000
517
+ },
518
+ {
519
+ "epoch": 0.15,
520
+ "eval_accuracy": 0.3530753183612612,
521
+ "eval_loss": 3.7798092365264893,
522
+ "eval_runtime": 5037.6086,
523
+ "eval_samples_per_second": 88.092,
524
+ "eval_steps_per_second": 1.376,
525
+ "step": 40000
526
+ },
527
+ {
528
+ "epoch": 0.15,
529
+ "learning_rate": 4.231691492832915e-05,
530
+ "loss": 3.8487,
531
+ "step": 40500
532
+ },
533
+ {
534
+ "epoch": 0.16,
535
+ "learning_rate": 4.222206202620976e-05,
536
+ "loss": 3.851,
537
+ "step": 41000
538
+ },
539
+ {
540
+ "epoch": 0.16,
541
+ "learning_rate": 4.2127209124090364e-05,
542
+ "loss": 3.8431,
543
+ "step": 41500
544
+ },
545
+ {
546
+ "epoch": 0.16,
547
+ "learning_rate": 4.203235622197097e-05,
548
+ "loss": 3.8401,
549
+ "step": 42000
550
+ },
551
+ {
552
+ "epoch": 0.16,
553
+ "learning_rate": 4.193750331985158e-05,
554
+ "loss": 3.8296,
555
+ "step": 42500
556
+ },
557
+ {
558
+ "epoch": 0.16,
559
+ "learning_rate": 4.184265041773218e-05,
560
+ "loss": 3.8338,
561
+ "step": 43000
562
+ },
563
+ {
564
+ "epoch": 0.17,
565
+ "learning_rate": 4.174779751561279e-05,
566
+ "loss": 3.8295,
567
+ "step": 43500
568
+ },
569
+ {
570
+ "epoch": 0.17,
571
+ "learning_rate": 4.1652944613493396e-05,
572
+ "loss": 3.8285,
573
+ "step": 44000
574
+ },
575
+ {
576
+ "epoch": 0.17,
577
+ "learning_rate": 4.1558091711374e-05,
578
+ "loss": 3.8217,
579
+ "step": 44500
580
+ },
581
+ {
582
+ "epoch": 0.17,
583
+ "learning_rate": 4.146323880925461e-05,
584
+ "loss": 3.8262,
585
+ "step": 45000
586
+ },
587
+ {
588
+ "epoch": 0.17,
589
+ "learning_rate": 4.1368385907135214e-05,
590
+ "loss": 3.8241,
591
+ "step": 45500
592
+ },
593
+ {
594
+ "epoch": 0.17,
595
+ "learning_rate": 4.1273533005015825e-05,
596
+ "loss": 3.8145,
597
+ "step": 46000
598
+ },
599
+ {
600
+ "epoch": 0.18,
601
+ "learning_rate": 4.117868010289643e-05,
602
+ "loss": 3.8207,
603
+ "step": 46500
604
+ },
605
+ {
606
+ "epoch": 0.18,
607
+ "learning_rate": 4.108382720077704e-05,
608
+ "loss": 3.8128,
609
+ "step": 47000
610
+ },
611
+ {
612
+ "epoch": 0.18,
613
+ "learning_rate": 4.098897429865764e-05,
614
+ "loss": 3.8053,
615
+ "step": 47500
616
+ },
617
+ {
618
+ "epoch": 0.18,
619
+ "learning_rate": 4.089412139653825e-05,
620
+ "loss": 3.8023,
621
+ "step": 48000
622
+ },
623
+ {
624
+ "epoch": 0.18,
625
+ "learning_rate": 4.079926849441886e-05,
626
+ "loss": 3.8084,
627
+ "step": 48500
628
+ },
629
+ {
630
+ "epoch": 0.19,
631
+ "learning_rate": 4.070441559229947e-05,
632
+ "loss": 3.7967,
633
+ "step": 49000
634
+ },
635
+ {
636
+ "epoch": 0.19,
637
+ "learning_rate": 4.060956269018007e-05,
638
+ "loss": 3.7947,
639
+ "step": 49500
640
+ },
641
+ {
642
+ "epoch": 0.19,
643
+ "learning_rate": 4.0514709788060675e-05,
644
+ "loss": 3.7973,
645
+ "step": 50000
646
+ },
647
+ {
648
+ "epoch": 0.19,
649
+ "eval_accuracy": 0.35837528569747157,
650
+ "eval_loss": 3.7209770679473877,
651
+ "eval_runtime": 5052.5501,
652
+ "eval_samples_per_second": 87.831,
653
+ "eval_steps_per_second": 1.372,
654
+ "step": 50000
655
+ },
656
+ {
657
+ "epoch": 0.19,
658
+ "learning_rate": 4.041985688594128e-05,
659
+ "loss": 3.7861,
660
+ "step": 50500
661
+ },
662
+ {
663
+ "epoch": 0.19,
664
+ "learning_rate": 4.032500398382189e-05,
665
+ "loss": 3.7878,
666
+ "step": 51000
667
+ },
668
+ {
669
+ "epoch": 0.2,
670
+ "learning_rate": 4.02301510817025e-05,
671
+ "loss": 3.7839,
672
+ "step": 51500
673
+ },
674
+ {
675
+ "epoch": 0.2,
676
+ "learning_rate": 4.0135298179583104e-05,
677
+ "loss": 3.7898,
678
+ "step": 52000
679
+ },
680
+ {
681
+ "epoch": 0.2,
682
+ "learning_rate": 4.0040445277463714e-05,
683
+ "loss": 3.7808,
684
+ "step": 52500
685
+ },
686
+ {
687
+ "epoch": 0.2,
688
+ "learning_rate": 3.994559237534432e-05,
689
+ "loss": 3.7857,
690
+ "step": 53000
691
+ },
692
+ {
693
+ "epoch": 0.2,
694
+ "learning_rate": 3.985073947322492e-05,
695
+ "loss": 3.7754,
696
+ "step": 53500
697
+ },
698
+ {
699
+ "epoch": 0.2,
700
+ "learning_rate": 3.975588657110553e-05,
701
+ "loss": 3.769,
702
+ "step": 54000
703
+ },
704
+ {
705
+ "epoch": 0.21,
706
+ "learning_rate": 3.9661033668986136e-05,
707
+ "loss": 3.7723,
708
+ "step": 54500
709
+ },
710
+ {
711
+ "epoch": 0.21,
712
+ "learning_rate": 3.9566180766866747e-05,
713
+ "loss": 3.7719,
714
+ "step": 55000
715
+ },
716
+ {
717
+ "epoch": 0.21,
718
+ "learning_rate": 3.947132786474736e-05,
719
+ "loss": 3.7684,
720
+ "step": 55500
721
+ },
722
+ {
723
+ "epoch": 0.21,
724
+ "learning_rate": 3.9376474962627954e-05,
725
+ "loss": 3.7672,
726
+ "step": 56000
727
+ },
728
+ {
729
+ "epoch": 0.21,
730
+ "learning_rate": 3.9281622060508565e-05,
731
+ "loss": 3.7595,
732
+ "step": 56500
733
+ },
734
+ {
735
+ "epoch": 0.22,
736
+ "learning_rate": 3.918676915838917e-05,
737
+ "loss": 3.764,
738
+ "step": 57000
739
+ },
740
+ {
741
+ "epoch": 0.22,
742
+ "learning_rate": 3.909191625626978e-05,
743
+ "loss": 3.7584,
744
+ "step": 57500
745
+ },
746
+ {
747
+ "epoch": 0.22,
748
+ "learning_rate": 3.899706335415039e-05,
749
+ "loss": 3.7532,
750
+ "step": 58000
751
+ },
752
+ {
753
+ "epoch": 0.22,
754
+ "learning_rate": 3.890221045203099e-05,
755
+ "loss": 3.7476,
756
+ "step": 58500
757
+ },
758
+ {
759
+ "epoch": 0.22,
760
+ "learning_rate": 3.88073575499116e-05,
761
+ "loss": 3.7502,
762
+ "step": 59000
763
+ },
764
+ {
765
+ "epoch": 0.23,
766
+ "learning_rate": 3.871250464779221e-05,
767
+ "loss": 3.7584,
768
+ "step": 59500
769
+ },
770
+ {
771
+ "epoch": 0.23,
772
+ "learning_rate": 3.861765174567281e-05,
773
+ "loss": 3.7421,
774
+ "step": 60000
775
+ },
776
+ {
777
+ "epoch": 0.23,
778
+ "eval_accuracy": 0.3629990378932714,
779
+ "eval_loss": 3.6750495433807373,
780
+ "eval_runtime": 5050.4176,
781
+ "eval_samples_per_second": 87.869,
782
+ "eval_steps_per_second": 1.373,
783
+ "step": 60000
784
+ },
785
+ {
786
+ "epoch": 0.23,
787
+ "learning_rate": 3.852279884355342e-05,
788
+ "loss": 3.7503,
789
+ "step": 60500
790
+ },
791
+ {
792
+ "epoch": 0.23,
793
+ "learning_rate": 3.8427945941434025e-05,
794
+ "loss": 3.7485,
795
+ "step": 61000
796
+ },
797
+ {
798
+ "epoch": 0.23,
799
+ "learning_rate": 3.8333093039314636e-05,
800
+ "loss": 3.745,
801
+ "step": 61500
802
+ },
803
+ {
804
+ "epoch": 0.24,
805
+ "learning_rate": 3.823824013719524e-05,
806
+ "loss": 3.739,
807
+ "step": 62000
808
+ },
809
+ {
810
+ "epoch": 0.24,
811
+ "learning_rate": 3.8143387235075843e-05,
812
+ "loss": 3.7411,
813
+ "step": 62500
814
+ },
815
+ {
816
+ "epoch": 0.24,
817
+ "learning_rate": 3.8048534332956454e-05,
818
+ "loss": 3.739,
819
+ "step": 63000
820
+ },
821
+ {
822
+ "epoch": 0.24,
823
+ "learning_rate": 3.795368143083706e-05,
824
+ "loss": 3.7388,
825
+ "step": 63500
826
+ },
827
+ {
828
+ "epoch": 0.24,
829
+ "learning_rate": 3.785882852871767e-05,
830
+ "loss": 3.7339,
831
+ "step": 64000
832
+ },
833
+ {
834
+ "epoch": 0.24,
835
+ "learning_rate": 3.776397562659827e-05,
836
+ "loss": 3.7328,
837
+ "step": 64500
838
+ },
839
+ {
840
+ "epoch": 0.25,
841
+ "learning_rate": 3.7669122724478876e-05,
842
+ "loss": 3.7209,
843
+ "step": 65000
844
+ },
845
+ {
846
+ "epoch": 0.25,
847
+ "learning_rate": 3.7574269822359486e-05,
848
+ "loss": 3.727,
849
+ "step": 65500
850
+ },
851
+ {
852
+ "epoch": 0.25,
853
+ "learning_rate": 3.74794169202401e-05,
854
+ "loss": 3.7251,
855
+ "step": 66000
856
+ },
857
+ {
858
+ "epoch": 0.25,
859
+ "learning_rate": 3.73845640181207e-05,
860
+ "loss": 3.7199,
861
+ "step": 66500
862
+ },
863
+ {
864
+ "epoch": 0.25,
865
+ "learning_rate": 3.728971111600131e-05,
866
+ "loss": 3.7277,
867
+ "step": 67000
868
+ },
869
+ {
870
+ "epoch": 0.26,
871
+ "learning_rate": 3.719485821388191e-05,
872
+ "loss": 3.7232,
873
+ "step": 67500
874
+ },
875
+ {
876
+ "epoch": 0.26,
877
+ "learning_rate": 3.710000531176252e-05,
878
+ "loss": 3.7154,
879
+ "step": 68000
880
+ },
881
+ {
882
+ "epoch": 0.26,
883
+ "learning_rate": 3.700515240964313e-05,
884
+ "loss": 3.7159,
885
+ "step": 68500
886
+ },
887
+ {
888
+ "epoch": 0.26,
889
+ "learning_rate": 3.691029950752373e-05,
890
+ "loss": 3.711,
891
+ "step": 69000
892
+ },
893
+ {
894
+ "epoch": 0.26,
895
+ "learning_rate": 3.681544660540434e-05,
896
+ "loss": 3.7094,
897
+ "step": 69500
898
+ },
899
+ {
900
+ "epoch": 0.27,
901
+ "learning_rate": 3.672059370328495e-05,
902
+ "loss": 3.7097,
903
+ "step": 70000
904
+ },
905
+ {
906
+ "epoch": 0.27,
907
+ "eval_accuracy": 0.36635205507688484,
908
+ "eval_loss": 3.6377646923065186,
909
+ "eval_runtime": 5044.3649,
910
+ "eval_samples_per_second": 87.974,
911
+ "eval_steps_per_second": 1.375,
912
+ "step": 70000
913
+ },
914
+ {
915
+ "epoch": 0.27,
916
+ "learning_rate": 3.662574080116555e-05,
917
+ "loss": 3.7109,
918
+ "step": 70500
919
+ },
920
+ {
921
+ "epoch": 0.27,
922
+ "learning_rate": 3.653088789904616e-05,
923
+ "loss": 3.7069,
924
+ "step": 71000
925
+ },
926
+ {
927
+ "epoch": 0.27,
928
+ "learning_rate": 3.6436034996926765e-05,
929
+ "loss": 3.7025,
930
+ "step": 71500
931
+ },
932
+ {
933
+ "epoch": 0.27,
934
+ "learning_rate": 3.6341182094807376e-05,
935
+ "loss": 3.7032,
936
+ "step": 72000
937
+ },
938
+ {
939
+ "epoch": 0.28,
940
+ "learning_rate": 3.6246329192687986e-05,
941
+ "loss": 3.7005,
942
+ "step": 72500
943
+ },
944
+ {
945
+ "epoch": 0.28,
946
+ "learning_rate": 3.615147629056859e-05,
947
+ "loss": 3.704,
948
+ "step": 73000
949
+ },
950
+ {
951
+ "epoch": 0.28,
952
+ "learning_rate": 3.6056623388449194e-05,
953
+ "loss": 3.6903,
954
+ "step": 73500
955
+ },
956
+ {
957
+ "epoch": 0.28,
958
+ "learning_rate": 3.59617704863298e-05,
959
+ "loss": 3.6989,
960
+ "step": 74000
961
+ },
962
+ {
963
+ "epoch": 0.28,
964
+ "learning_rate": 3.586691758421041e-05,
965
+ "loss": 3.6947,
966
+ "step": 74500
967
+ },
968
+ {
969
+ "epoch": 0.28,
970
+ "learning_rate": 3.577206468209102e-05,
971
+ "loss": 3.6969,
972
+ "step": 75000
973
+ },
974
+ {
975
+ "epoch": 0.29,
976
+ "learning_rate": 3.567721177997162e-05,
977
+ "loss": 3.696,
978
+ "step": 75500
979
+ },
980
+ {
981
+ "epoch": 0.29,
982
+ "learning_rate": 3.5582358877852226e-05,
983
+ "loss": 3.6905,
984
+ "step": 76000
985
+ },
986
+ {
987
+ "epoch": 0.29,
988
+ "learning_rate": 3.5487505975732836e-05,
989
+ "loss": 3.6851,
990
+ "step": 76500
991
+ },
992
+ {
993
+ "epoch": 0.29,
994
+ "learning_rate": 3.539265307361344e-05,
995
+ "loss": 3.6857,
996
+ "step": 77000
997
+ },
998
+ {
999
+ "epoch": 0.29,
1000
+ "learning_rate": 3.529780017149405e-05,
1001
+ "loss": 3.6889,
1002
+ "step": 77500
1003
+ },
1004
+ {
1005
+ "epoch": 0.3,
1006
+ "learning_rate": 3.5202947269374654e-05,
1007
+ "loss": 3.6911,
1008
+ "step": 78000
1009
+ },
1010
+ {
1011
+ "epoch": 0.3,
1012
+ "learning_rate": 3.5108094367255265e-05,
1013
+ "loss": 3.687,
1014
+ "step": 78500
1015
+ },
1016
+ {
1017
+ "epoch": 0.3,
1018
+ "learning_rate": 3.501324146513587e-05,
1019
+ "loss": 3.6808,
1020
+ "step": 79000
1021
+ },
1022
+ {
1023
+ "epoch": 0.3,
1024
+ "learning_rate": 3.491838856301647e-05,
1025
+ "loss": 3.6841,
1026
+ "step": 79500
1027
+ },
1028
+ {
1029
+ "epoch": 0.3,
1030
+ "learning_rate": 3.482353566089708e-05,
1031
+ "loss": 3.6741,
1032
+ "step": 80000
1033
+ },
1034
+ {
1035
+ "epoch": 0.3,
1036
+ "eval_accuracy": 0.36940481219098525,
1037
+ "eval_loss": 3.606105089187622,
1038
+ "eval_runtime": 5043.6804,
1039
+ "eval_samples_per_second": 87.986,
1040
+ "eval_steps_per_second": 1.375,
1041
+ "step": 80000
1042
+ },
1043
+ {
1044
+ "epoch": 0.31,
1045
+ "learning_rate": 3.4728682758777694e-05,
1046
+ "loss": 3.6789,
1047
+ "step": 80500
1048
+ },
1049
+ {
1050
+ "epoch": 0.31,
1051
+ "learning_rate": 3.46338298566583e-05,
1052
+ "loss": 3.6809,
1053
+ "step": 81000
1054
+ },
1055
+ {
1056
+ "epoch": 0.31,
1057
+ "learning_rate": 3.45389769545389e-05,
1058
+ "loss": 3.6854,
1059
+ "step": 81500
1060
+ },
1061
+ {
1062
+ "epoch": 0.31,
1063
+ "learning_rate": 3.4444124052419505e-05,
1064
+ "loss": 3.6692,
1065
+ "step": 82000
1066
+ },
1067
+ {
1068
+ "epoch": 0.31,
1069
+ "learning_rate": 3.4349271150300115e-05,
1070
+ "loss": 3.6771,
1071
+ "step": 82500
1072
+ },
1073
+ {
1074
+ "epoch": 0.31,
1075
+ "learning_rate": 3.4254418248180726e-05,
1076
+ "loss": 3.6773,
1077
+ "step": 83000
1078
+ },
1079
+ {
1080
+ "epoch": 0.32,
1081
+ "learning_rate": 3.415956534606133e-05,
1082
+ "loss": 3.6701,
1083
+ "step": 83500
1084
+ },
1085
+ {
1086
+ "epoch": 0.32,
1087
+ "learning_rate": 3.406471244394194e-05,
1088
+ "loss": 3.666,
1089
+ "step": 84000
1090
+ },
1091
+ {
1092
+ "epoch": 0.32,
1093
+ "learning_rate": 3.3969859541822544e-05,
1094
+ "loss": 3.672,
1095
+ "step": 84500
1096
+ },
1097
+ {
1098
+ "epoch": 0.32,
1099
+ "learning_rate": 3.387500663970315e-05,
1100
+ "loss": 3.6738,
1101
+ "step": 85000
1102
+ },
1103
+ {
1104
+ "epoch": 0.32,
1105
+ "learning_rate": 3.378015373758376e-05,
1106
+ "loss": 3.6705,
1107
+ "step": 85500
1108
+ },
1109
+ {
1110
+ "epoch": 0.33,
1111
+ "learning_rate": 3.368530083546436e-05,
1112
+ "loss": 3.6649,
1113
+ "step": 86000
1114
+ },
1115
+ {
1116
+ "epoch": 0.33,
1117
+ "learning_rate": 3.359044793334497e-05,
1118
+ "loss": 3.6736,
1119
+ "step": 86500
1120
+ },
1121
+ {
1122
+ "epoch": 0.33,
1123
+ "learning_rate": 3.3495595031225576e-05,
1124
+ "loss": 3.6689,
1125
+ "step": 87000
1126
+ },
1127
+ {
1128
+ "epoch": 0.33,
1129
+ "learning_rate": 3.340074212910618e-05,
1130
+ "loss": 3.6665,
1131
+ "step": 87500
1132
+ },
1133
+ {
1134
+ "epoch": 0.33,
1135
+ "learning_rate": 3.330588922698679e-05,
1136
+ "loss": 3.6641,
1137
+ "step": 88000
1138
+ },
1139
+ {
1140
+ "epoch": 0.34,
1141
+ "learning_rate": 3.3211036324867394e-05,
1142
+ "loss": 3.6536,
1143
+ "step": 88500
1144
+ },
1145
+ {
1146
+ "epoch": 0.34,
1147
+ "learning_rate": 3.3116183422748005e-05,
1148
+ "loss": 3.6658,
1149
+ "step": 89000
1150
+ },
1151
+ {
1152
+ "epoch": 0.34,
1153
+ "learning_rate": 3.3021330520628615e-05,
1154
+ "loss": 3.6544,
1155
+ "step": 89500
1156
+ },
1157
+ {
1158
+ "epoch": 0.34,
1159
+ "learning_rate": 3.292647761850922e-05,
1160
+ "loss": 3.6599,
1161
+ "step": 90000
1162
+ },
1163
+ {
1164
+ "epoch": 0.34,
1165
+ "eval_accuracy": 0.37180447854264453,
1166
+ "eval_loss": 3.5803401470184326,
1167
+ "eval_runtime": 5042.0712,
1168
+ "eval_samples_per_second": 88.014,
1169
+ "eval_steps_per_second": 1.375,
1170
+ "step": 90000
1171
+ },
1172
+ {
1173
+ "epoch": 0.34,
1174
+ "learning_rate": 3.283162471638982e-05,
1175
+ "loss": 3.6605,
1176
+ "step": 90500
1177
+ },
1178
+ {
1179
+ "epoch": 0.35,
1180
+ "learning_rate": 3.2736771814270426e-05,
1181
+ "loss": 3.6603,
1182
+ "step": 91000
1183
+ },
1184
+ {
1185
+ "epoch": 0.35,
1186
+ "learning_rate": 3.264191891215104e-05,
1187
+ "loss": 3.6576,
1188
+ "step": 91500
1189
+ },
1190
+ {
1191
+ "epoch": 0.35,
1192
+ "learning_rate": 3.254706601003165e-05,
1193
+ "loss": 3.6511,
1194
+ "step": 92000
1195
+ },
1196
+ {
1197
+ "epoch": 0.35,
1198
+ "learning_rate": 3.245221310791225e-05,
1199
+ "loss": 3.6518,
1200
+ "step": 92500
1201
+ },
1202
+ {
1203
+ "epoch": 0.35,
1204
+ "learning_rate": 3.2357360205792855e-05,
1205
+ "loss": 3.6522,
1206
+ "step": 93000
1207
+ },
1208
+ {
1209
+ "epoch": 0.35,
1210
+ "learning_rate": 3.2262507303673465e-05,
1211
+ "loss": 3.646,
1212
+ "step": 93500
1213
+ },
1214
+ {
1215
+ "epoch": 0.36,
1216
+ "learning_rate": 3.216765440155407e-05,
1217
+ "loss": 3.6494,
1218
+ "step": 94000
1219
+ },
1220
+ {
1221
+ "epoch": 0.36,
1222
+ "learning_rate": 3.207280149943468e-05,
1223
+ "loss": 3.6388,
1224
+ "step": 94500
1225
+ },
1226
+ {
1227
+ "epoch": 0.36,
1228
+ "learning_rate": 3.1977948597315284e-05,
1229
+ "loss": 3.6456,
1230
+ "step": 95000
1231
+ },
1232
+ {
1233
+ "epoch": 0.36,
1234
+ "learning_rate": 3.1883095695195894e-05,
1235
+ "loss": 3.6398,
1236
+ "step": 95500
1237
+ },
1238
+ {
1239
+ "epoch": 0.36,
1240
+ "learning_rate": 3.17882427930765e-05,
1241
+ "loss": 3.6476,
1242
+ "step": 96000
1243
+ },
1244
+ {
1245
+ "epoch": 0.37,
1246
+ "learning_rate": 3.16933898909571e-05,
1247
+ "loss": 3.6364,
1248
+ "step": 96500
1249
+ },
1250
+ {
1251
+ "epoch": 0.37,
1252
+ "learning_rate": 3.159853698883771e-05,
1253
+ "loss": 3.6456,
1254
+ "step": 97000
1255
+ },
1256
+ {
1257
+ "epoch": 0.37,
1258
+ "learning_rate": 3.150368408671832e-05,
1259
+ "loss": 3.645,
1260
+ "step": 97500
1261
+ },
1262
+ {
1263
+ "epoch": 0.37,
1264
+ "learning_rate": 3.1408831184598926e-05,
1265
+ "loss": 3.6357,
1266
+ "step": 98000
1267
+ },
1268
+ {
1269
+ "epoch": 0.37,
1270
+ "learning_rate": 3.131397828247954e-05,
1271
+ "loss": 3.6415,
1272
+ "step": 98500
1273
+ },
1274
+ {
1275
+ "epoch": 0.38,
1276
+ "learning_rate": 3.1219125380360134e-05,
1277
+ "loss": 3.6314,
1278
+ "step": 99000
1279
+ },
1280
+ {
1281
+ "epoch": 0.38,
1282
+ "learning_rate": 3.1124272478240744e-05,
1283
+ "loss": 3.64,
1284
+ "step": 99500
1285
+ },
1286
+ {
1287
+ "epoch": 0.38,
1288
+ "learning_rate": 3.1029419576121355e-05,
1289
+ "loss": 3.6356,
1290
+ "step": 100000
1291
+ },
1292
+ {
1293
+ "epoch": 0.38,
1294
+ "eval_accuracy": 0.37408231347678594,
1295
+ "eval_loss": 3.558403253555298,
1296
+ "eval_runtime": 5044.122,
1297
+ "eval_samples_per_second": 87.978,
1298
+ "eval_steps_per_second": 1.375,
1299
+ "step": 100000
1300
+ },
1301
+ {
1302
+ "epoch": 0.38,
1303
+ "learning_rate": 3.093456667400196e-05,
1304
+ "loss": 3.637,
1305
+ "step": 100500
1306
+ },
1307
+ {
1308
+ "epoch": 0.38,
1309
+ "learning_rate": 3.083971377188257e-05,
1310
+ "loss": 3.6353,
1311
+ "step": 101000
1312
+ },
1313
+ {
1314
+ "epoch": 0.39,
1315
+ "learning_rate": 3.074486086976317e-05,
1316
+ "loss": 3.6331,
1317
+ "step": 101500
1318
+ },
1319
+ {
1320
+ "epoch": 0.39,
1321
+ "learning_rate": 3.065000796764378e-05,
1322
+ "loss": 3.6288,
1323
+ "step": 102000
1324
+ },
1325
+ {
1326
+ "epoch": 0.39,
1327
+ "learning_rate": 3.055515506552439e-05,
1328
+ "loss": 3.6273,
1329
+ "step": 102500
1330
+ },
1331
+ {
1332
+ "epoch": 0.39,
1333
+ "learning_rate": 3.046030216340499e-05,
1334
+ "loss": 3.6351,
1335
+ "step": 103000
1336
+ },
1337
+ {
1338
+ "epoch": 0.39,
1339
+ "learning_rate": 3.03654492612856e-05,
1340
+ "loss": 3.6285,
1341
+ "step": 103500
1342
+ },
1343
+ {
1344
+ "epoch": 0.39,
1345
+ "learning_rate": 3.027059635916621e-05,
1346
+ "loss": 3.6256,
1347
+ "step": 104000
1348
+ },
1349
+ {
1350
+ "epoch": 0.4,
1351
+ "learning_rate": 3.0175743457046812e-05,
1352
+ "loss": 3.6248,
1353
+ "step": 104500
1354
+ },
1355
+ {
1356
+ "epoch": 0.4,
1357
+ "learning_rate": 3.0080890554927423e-05,
1358
+ "loss": 3.6182,
1359
+ "step": 105000
1360
+ },
1361
+ {
1362
+ "epoch": 0.4,
1363
+ "learning_rate": 2.9986037652808023e-05,
1364
+ "loss": 3.6242,
1365
+ "step": 105500
1366
+ },
1367
+ {
1368
+ "epoch": 0.4,
1369
+ "learning_rate": 2.9891184750688634e-05,
1370
+ "loss": 3.625,
1371
+ "step": 106000
1372
+ },
1373
+ {
1374
+ "epoch": 0.4,
1375
+ "learning_rate": 2.979633184856924e-05,
1376
+ "loss": 3.6191,
1377
+ "step": 106500
1378
+ },
1379
+ {
1380
+ "epoch": 0.41,
1381
+ "learning_rate": 2.9701478946449845e-05,
1382
+ "loss": 3.6267,
1383
+ "step": 107000
1384
+ },
1385
+ {
1386
+ "epoch": 0.41,
1387
+ "learning_rate": 2.9606626044330455e-05,
1388
+ "loss": 3.6227,
1389
+ "step": 107500
1390
+ },
1391
+ {
1392
+ "epoch": 0.41,
1393
+ "learning_rate": 2.9511773142211062e-05,
1394
+ "loss": 3.6217,
1395
+ "step": 108000
1396
+ },
1397
+ {
1398
+ "epoch": 0.41,
1399
+ "learning_rate": 2.9416920240091666e-05,
1400
+ "loss": 3.6168,
1401
+ "step": 108500
1402
+ },
1403
+ {
1404
+ "epoch": 0.41,
1405
+ "learning_rate": 2.9322067337972277e-05,
1406
+ "loss": 3.6204,
1407
+ "step": 109000
1408
+ },
1409
+ {
1410
+ "epoch": 0.42,
1411
+ "learning_rate": 2.922721443585288e-05,
1412
+ "loss": 3.6217,
1413
+ "step": 109500
1414
+ },
1415
+ {
1416
+ "epoch": 0.42,
1417
+ "learning_rate": 2.9132361533733487e-05,
1418
+ "loss": 3.6131,
1419
+ "step": 110000
1420
+ },
1421
+ {
1422
+ "epoch": 0.42,
1423
+ "eval_accuracy": 0.37584134336520747,
1424
+ "eval_loss": 3.542346715927124,
1425
+ "eval_runtime": 5051.4787,
1426
+ "eval_samples_per_second": 87.85,
1427
+ "eval_steps_per_second": 1.373,
1428
+ "step": 110000
1429
+ },
1430
+ {
1431
+ "epoch": 0.42,
1432
+ "learning_rate": 2.9037508631614098e-05,
1433
+ "loss": 3.6165,
1434
+ "step": 110500
1435
+ },
1436
+ {
1437
+ "epoch": 0.42,
1438
+ "learning_rate": 2.8942655729494698e-05,
1439
+ "loss": 3.6129,
1440
+ "step": 111000
1441
+ },
1442
+ {
1443
+ "epoch": 0.42,
1444
+ "learning_rate": 2.884780282737531e-05,
1445
+ "loss": 3.6177,
1446
+ "step": 111500
1447
+ },
1448
+ {
1449
+ "epoch": 0.42,
1450
+ "learning_rate": 2.8752949925255913e-05,
1451
+ "loss": 3.6187,
1452
+ "step": 112000
1453
+ },
1454
+ {
1455
+ "epoch": 0.43,
1456
+ "learning_rate": 2.865809702313652e-05,
1457
+ "loss": 3.6112,
1458
+ "step": 112500
1459
+ },
1460
+ {
1461
+ "epoch": 0.43,
1462
+ "learning_rate": 2.856324412101713e-05,
1463
+ "loss": 3.6103,
1464
+ "step": 113000
1465
+ },
1466
+ {
1467
+ "epoch": 0.43,
1468
+ "learning_rate": 2.8468391218897734e-05,
1469
+ "loss": 3.6103,
1470
+ "step": 113500
1471
+ },
1472
+ {
1473
+ "epoch": 0.43,
1474
+ "learning_rate": 2.837353831677834e-05,
1475
+ "loss": 3.615,
1476
+ "step": 114000
1477
+ },
1478
+ {
1479
+ "epoch": 0.43,
1480
+ "learning_rate": 2.827868541465895e-05,
1481
+ "loss": 3.6151,
1482
+ "step": 114500
1483
+ },
1484
+ {
1485
+ "epoch": 0.44,
1486
+ "learning_rate": 2.8183832512539555e-05,
1487
+ "loss": 3.6039,
1488
+ "step": 115000
1489
+ },
1490
+ {
1491
+ "epoch": 0.44,
1492
+ "learning_rate": 2.8088979610420162e-05,
1493
+ "loss": 3.6133,
1494
+ "step": 115500
1495
+ },
1496
+ {
1497
+ "epoch": 0.44,
1498
+ "learning_rate": 2.7994126708300766e-05,
1499
+ "loss": 3.6063,
1500
+ "step": 116000
1501
+ },
1502
+ {
1503
+ "epoch": 0.44,
1504
+ "learning_rate": 2.7899273806181377e-05,
1505
+ "loss": 3.6029,
1506
+ "step": 116500
1507
+ },
1508
+ {
1509
+ "epoch": 0.44,
1510
+ "learning_rate": 2.7804420904061984e-05,
1511
+ "loss": 3.6082,
1512
+ "step": 117000
1513
+ },
1514
+ {
1515
+ "epoch": 0.45,
1516
+ "learning_rate": 2.7709568001942588e-05,
1517
+ "loss": 3.6099,
1518
+ "step": 117500
1519
+ },
1520
+ {
1521
+ "epoch": 0.45,
1522
+ "learning_rate": 2.7614715099823195e-05,
1523
+ "loss": 3.6035,
1524
+ "step": 118000
1525
+ },
1526
+ {
1527
+ "epoch": 0.45,
1528
+ "learning_rate": 2.75198621977038e-05,
1529
+ "loss": 3.5997,
1530
+ "step": 118500
1531
+ },
1532
+ {
1533
+ "epoch": 0.45,
1534
+ "learning_rate": 2.742500929558441e-05,
1535
+ "loss": 3.6032,
1536
+ "step": 119000
1537
+ },
1538
+ {
1539
+ "epoch": 0.45,
1540
+ "learning_rate": 2.7330156393465016e-05,
1541
+ "loss": 3.5998,
1542
+ "step": 119500
1543
+ },
1544
+ {
1545
+ "epoch": 0.46,
1546
+ "learning_rate": 2.723530349134562e-05,
1547
+ "loss": 3.5991,
1548
+ "step": 120000
1549
+ },
1550
+ {
1551
+ "epoch": 0.46,
1552
+ "eval_accuracy": 0.3775510164297428,
1553
+ "eval_loss": 3.525380849838257,
1554
+ "eval_runtime": 5041.3032,
1555
+ "eval_samples_per_second": 88.027,
1556
+ "eval_steps_per_second": 1.375,
1557
+ "step": 120000
1558
+ },
1559
+ {
1560
+ "epoch": 0.46,
1561
+ "learning_rate": 2.714045058922623e-05,
1562
+ "loss": 3.6027,
1563
+ "step": 120500
1564
+ },
1565
+ {
1566
+ "epoch": 0.46,
1567
+ "learning_rate": 2.7045597687106838e-05,
1568
+ "loss": 3.5947,
1569
+ "step": 121000
1570
+ },
1571
+ {
1572
+ "epoch": 0.46,
1573
+ "learning_rate": 2.695074478498744e-05,
1574
+ "loss": 3.6033,
1575
+ "step": 121500
1576
+ },
1577
+ {
1578
+ "epoch": 0.46,
1579
+ "learning_rate": 2.6855891882868052e-05,
1580
+ "loss": 3.5933,
1581
+ "step": 122000
1582
+ },
1583
+ {
1584
+ "epoch": 0.46,
1585
+ "learning_rate": 2.6761038980748652e-05,
1586
+ "loss": 3.594,
1587
+ "step": 122500
1588
+ },
1589
+ {
1590
+ "epoch": 0.47,
1591
+ "learning_rate": 2.6666186078629263e-05,
1592
+ "loss": 3.599,
1593
+ "step": 123000
1594
+ },
1595
+ {
1596
+ "epoch": 0.47,
1597
+ "learning_rate": 2.6571333176509873e-05,
1598
+ "loss": 3.6013,
1599
+ "step": 123500
1600
+ },
1601
+ {
1602
+ "epoch": 0.47,
1603
+ "learning_rate": 2.6476480274390474e-05,
1604
+ "loss": 3.5982,
1605
+ "step": 124000
1606
+ },
1607
+ {
1608
+ "epoch": 0.47,
1609
+ "learning_rate": 2.6381627372271084e-05,
1610
+ "loss": 3.5937,
1611
+ "step": 124500
1612
+ },
1613
+ {
1614
+ "epoch": 0.47,
1615
+ "learning_rate": 2.628677447015169e-05,
1616
+ "loss": 3.5945,
1617
+ "step": 125000
1618
+ },
1619
+ {
1620
+ "epoch": 0.48,
1621
+ "learning_rate": 2.6191921568032295e-05,
1622
+ "loss": 3.5926,
1623
+ "step": 125500
1624
+ },
1625
+ {
1626
+ "epoch": 0.48,
1627
+ "learning_rate": 2.6097068665912906e-05,
1628
+ "loss": 3.5927,
1629
+ "step": 126000
1630
+ },
1631
+ {
1632
+ "epoch": 0.48,
1633
+ "learning_rate": 2.600221576379351e-05,
1634
+ "loss": 3.5892,
1635
+ "step": 126500
1636
+ },
1637
+ {
1638
+ "epoch": 0.48,
1639
+ "learning_rate": 2.5907362861674116e-05,
1640
+ "loss": 3.5938,
1641
+ "step": 127000
1642
+ },
1643
+ {
1644
+ "epoch": 0.48,
1645
+ "learning_rate": 2.5812509959554727e-05,
1646
+ "loss": 3.5867,
1647
+ "step": 127500
1648
+ },
1649
+ {
1650
+ "epoch": 0.49,
1651
+ "learning_rate": 2.571765705743533e-05,
1652
+ "loss": 3.5879,
1653
+ "step": 128000
1654
+ },
1655
+ {
1656
+ "epoch": 0.49,
1657
+ "learning_rate": 2.5622804155315938e-05,
1658
+ "loss": 3.5909,
1659
+ "step": 128500
1660
+ },
1661
+ {
1662
+ "epoch": 0.49,
1663
+ "learning_rate": 2.552795125319654e-05,
1664
+ "loss": 3.5861,
1665
+ "step": 129000
1666
+ },
1667
+ {
1668
+ "epoch": 0.49,
1669
+ "learning_rate": 2.543309835107715e-05,
1670
+ "loss": 3.5913,
1671
+ "step": 129500
1672
+ },
1673
+ {
1674
+ "epoch": 0.49,
1675
+ "learning_rate": 2.533824544895776e-05,
1676
+ "loss": 3.591,
1677
+ "step": 130000
1678
+ },
1679
+ {
1680
+ "epoch": 0.49,
1681
+ "eval_accuracy": 0.37901210353247916,
1682
+ "eval_loss": 3.510841131210327,
1683
+ "eval_runtime": 5053.0574,
1684
+ "eval_samples_per_second": 87.823,
1685
+ "eval_steps_per_second": 1.372,
1686
+ "step": 130000
1687
+ },
1688
+ {
1689
+ "epoch": 0.5,
1690
+ "learning_rate": 2.5243392546838363e-05,
1691
+ "loss": 3.5849,
1692
+ "step": 130500
1693
+ },
1694
+ {
1695
+ "epoch": 0.5,
1696
+ "learning_rate": 2.514853964471897e-05,
1697
+ "loss": 3.5868,
1698
+ "step": 131000
1699
+ },
1700
+ {
1701
+ "epoch": 0.5,
1702
+ "learning_rate": 2.505368674259958e-05,
1703
+ "loss": 3.5838,
1704
+ "step": 131500
1705
+ },
1706
+ {
1707
+ "epoch": 0.5,
1708
+ "learning_rate": 2.4958833840480184e-05,
1709
+ "loss": 3.5848,
1710
+ "step": 132000
1711
+ },
1712
+ {
1713
+ "epoch": 0.5,
1714
+ "learning_rate": 2.486398093836079e-05,
1715
+ "loss": 3.5818,
1716
+ "step": 132500
1717
+ },
1718
+ {
1719
+ "epoch": 0.5,
1720
+ "learning_rate": 2.47691280362414e-05,
1721
+ "loss": 3.5842,
1722
+ "step": 133000
1723
+ },
1724
+ {
1725
+ "epoch": 0.51,
1726
+ "learning_rate": 2.4674275134122006e-05,
1727
+ "loss": 3.584,
1728
+ "step": 133500
1729
+ },
1730
+ {
1731
+ "epoch": 0.51,
1732
+ "learning_rate": 2.457942223200261e-05,
1733
+ "loss": 3.5817,
1734
+ "step": 134000
1735
+ },
1736
+ {
1737
+ "epoch": 0.51,
1738
+ "learning_rate": 2.448456932988322e-05,
1739
+ "loss": 3.5781,
1740
+ "step": 134500
1741
+ },
1742
+ {
1743
+ "epoch": 0.51,
1744
+ "learning_rate": 2.4389716427763827e-05,
1745
+ "loss": 3.5761,
1746
+ "step": 135000
1747
+ },
1748
+ {
1749
+ "epoch": 0.51,
1750
+ "learning_rate": 2.429486352564443e-05,
1751
+ "loss": 3.5755,
1752
+ "step": 135500
1753
+ },
1754
+ {
1755
+ "epoch": 0.52,
1756
+ "learning_rate": 2.4200010623525038e-05,
1757
+ "loss": 3.5822,
1758
+ "step": 136000
1759
+ },
1760
+ {
1761
+ "epoch": 0.52,
1762
+ "learning_rate": 2.4105157721405645e-05,
1763
+ "loss": 3.5749,
1764
+ "step": 136500
1765
+ },
1766
+ {
1767
+ "epoch": 0.52,
1768
+ "learning_rate": 2.4010304819286252e-05,
1769
+ "loss": 3.5812,
1770
+ "step": 137000
1771
+ },
1772
+ {
1773
+ "epoch": 0.52,
1774
+ "learning_rate": 2.391545191716686e-05,
1775
+ "loss": 3.5786,
1776
+ "step": 137500
1777
+ },
1778
+ {
1779
+ "epoch": 0.52,
1780
+ "learning_rate": 2.3820599015047467e-05,
1781
+ "loss": 3.5826,
1782
+ "step": 138000
1783
+ },
1784
+ {
1785
+ "epoch": 0.53,
1786
+ "learning_rate": 2.372574611292807e-05,
1787
+ "loss": 3.5731,
1788
+ "step": 138500
1789
+ },
1790
+ {
1791
+ "epoch": 0.53,
1792
+ "learning_rate": 2.363089321080868e-05,
1793
+ "loss": 3.5784,
1794
+ "step": 139000
1795
+ },
1796
+ {
1797
+ "epoch": 0.53,
1798
+ "learning_rate": 2.3536040308689285e-05,
1799
+ "loss": 3.5738,
1800
+ "step": 139500
1801
+ },
1802
+ {
1803
+ "epoch": 0.53,
1804
+ "learning_rate": 2.3441187406569892e-05,
1805
+ "loss": 3.574,
1806
+ "step": 140000
1807
+ },
1808
+ {
1809
+ "epoch": 0.53,
1810
+ "eval_accuracy": 0.3804677520669924,
1811
+ "eval_loss": 3.4966471195220947,
1812
+ "eval_runtime": 5044.6959,
1813
+ "eval_samples_per_second": 87.968,
1814
+ "eval_steps_per_second": 1.375,
1815
+ "step": 140000
1816
+ },
1817
+ {
1818
+ "epoch": 0.53,
1819
+ "learning_rate": 2.33463345044505e-05,
1820
+ "loss": 3.5722,
1821
+ "step": 140500
1822
+ },
1823
+ {
1824
+ "epoch": 0.53,
1825
+ "learning_rate": 2.3251481602331106e-05,
1826
+ "loss": 3.5778,
1827
+ "step": 141000
1828
+ },
1829
+ {
1830
+ "epoch": 0.54,
1831
+ "learning_rate": 2.3156628700211713e-05,
1832
+ "loss": 3.5722,
1833
+ "step": 141500
1834
+ },
1835
+ {
1836
+ "epoch": 0.54,
1837
+ "learning_rate": 2.306177579809232e-05,
1838
+ "loss": 3.5658,
1839
+ "step": 142000
1840
+ },
1841
+ {
1842
+ "epoch": 0.54,
1843
+ "learning_rate": 2.2966922895972924e-05,
1844
+ "loss": 3.5671,
1845
+ "step": 142500
1846
+ },
1847
+ {
1848
+ "epoch": 0.54,
1849
+ "learning_rate": 2.2872069993853535e-05,
1850
+ "loss": 3.5696,
1851
+ "step": 143000
1852
+ },
1853
+ {
1854
+ "epoch": 0.54,
1855
+ "learning_rate": 2.2777217091734142e-05,
1856
+ "loss": 3.5691,
1857
+ "step": 143500
1858
+ },
1859
+ {
1860
+ "epoch": 0.55,
1861
+ "learning_rate": 2.2682364189614745e-05,
1862
+ "loss": 3.5737,
1863
+ "step": 144000
1864
+ },
1865
+ {
1866
+ "epoch": 0.55,
1867
+ "learning_rate": 2.2587511287495353e-05,
1868
+ "loss": 3.5693,
1869
+ "step": 144500
1870
+ },
1871
+ {
1872
+ "epoch": 0.55,
1873
+ "learning_rate": 2.249265838537596e-05,
1874
+ "loss": 3.5728,
1875
+ "step": 145000
1876
+ },
1877
+ {
1878
+ "epoch": 0.55,
1879
+ "learning_rate": 2.2397805483256567e-05,
1880
+ "loss": 3.569,
1881
+ "step": 145500
1882
+ },
1883
+ {
1884
+ "epoch": 0.55,
1885
+ "learning_rate": 2.2302952581137174e-05,
1886
+ "loss": 3.5559,
1887
+ "step": 146000
1888
+ },
1889
+ {
1890
+ "epoch": 0.56,
1891
+ "learning_rate": 2.220809967901778e-05,
1892
+ "loss": 3.5673,
1893
+ "step": 146500
1894
+ },
1895
+ {
1896
+ "epoch": 0.56,
1897
+ "learning_rate": 2.2113246776898385e-05,
1898
+ "loss": 3.567,
1899
+ "step": 147000
1900
+ },
1901
+ {
1902
+ "epoch": 0.56,
1903
+ "learning_rate": 2.2018393874778995e-05,
1904
+ "loss": 3.5692,
1905
+ "step": 147500
1906
+ },
1907
+ {
1908
+ "epoch": 0.56,
1909
+ "learning_rate": 2.19235409726596e-05,
1910
+ "loss": 3.5651,
1911
+ "step": 148000
1912
+ },
1913
+ {
1914
+ "epoch": 0.56,
1915
+ "learning_rate": 2.1828688070540206e-05,
1916
+ "loss": 3.559,
1917
+ "step": 148500
1918
+ },
1919
+ {
1920
+ "epoch": 0.57,
1921
+ "learning_rate": 2.1733835168420813e-05,
1922
+ "loss": 3.5657,
1923
+ "step": 149000
1924
+ },
1925
+ {
1926
+ "epoch": 0.57,
1927
+ "learning_rate": 2.163898226630142e-05,
1928
+ "loss": 3.5632,
1929
+ "step": 149500
1930
+ },
1931
+ {
1932
+ "epoch": 0.57,
1933
+ "learning_rate": 2.1544129364182028e-05,
1934
+ "loss": 3.5606,
1935
+ "step": 150000
1936
+ },
1937
+ {
1938
+ "epoch": 0.57,
1939
+ "eval_accuracy": 0.38146521499584235,
1940
+ "eval_loss": 3.486565113067627,
1941
+ "eval_runtime": 5047.1623,
1942
+ "eval_samples_per_second": 87.925,
1943
+ "eval_steps_per_second": 1.374,
1944
+ "step": 150000
1945
+ },
1946
+ {
1947
+ "epoch": 0.57,
1948
+ "learning_rate": 2.1449276462062635e-05,
1949
+ "loss": 3.5635,
1950
+ "step": 150500
1951
+ },
1952
+ {
1953
+ "epoch": 0.57,
1954
+ "learning_rate": 2.135442355994324e-05,
1955
+ "loss": 3.569,
1956
+ "step": 151000
1957
+ },
1958
+ {
1959
+ "epoch": 0.57,
1960
+ "learning_rate": 2.125957065782385e-05,
1961
+ "loss": 3.551,
1962
+ "step": 151500
1963
+ },
1964
+ {
1965
+ "epoch": 0.58,
1966
+ "learning_rate": 2.1164717755704456e-05,
1967
+ "loss": 3.5543,
1968
+ "step": 152000
1969
+ },
1970
+ {
1971
+ "epoch": 0.58,
1972
+ "learning_rate": 2.106986485358506e-05,
1973
+ "loss": 3.5556,
1974
+ "step": 152500
1975
+ },
1976
+ {
1977
+ "epoch": 0.58,
1978
+ "learning_rate": 2.0975011951465667e-05,
1979
+ "loss": 3.5598,
1980
+ "step": 153000
1981
+ },
1982
+ {
1983
+ "epoch": 0.58,
1984
+ "learning_rate": 2.0880159049346274e-05,
1985
+ "loss": 3.5592,
1986
+ "step": 153500
1987
+ },
1988
+ {
1989
+ "epoch": 0.58,
1990
+ "learning_rate": 2.078530614722688e-05,
1991
+ "loss": 3.5562,
1992
+ "step": 154000
1993
+ },
1994
+ {
1995
+ "epoch": 0.59,
1996
+ "learning_rate": 2.069045324510749e-05,
1997
+ "loss": 3.5561,
1998
+ "step": 154500
1999
+ },
2000
+ {
2001
+ "epoch": 0.59,
2002
+ "learning_rate": 2.0595600342988096e-05,
2003
+ "loss": 3.5573,
2004
+ "step": 155000
2005
+ },
2006
+ {
2007
+ "epoch": 0.59,
2008
+ "learning_rate": 2.05007474408687e-05,
2009
+ "loss": 3.5585,
2010
+ "step": 155500
2011
+ },
2012
+ {
2013
+ "epoch": 0.59,
2014
+ "learning_rate": 2.040589453874931e-05,
2015
+ "loss": 3.5576,
2016
+ "step": 156000
2017
+ },
2018
+ {
2019
+ "epoch": 0.59,
2020
+ "learning_rate": 2.0311041636629917e-05,
2021
+ "loss": 3.5529,
2022
+ "step": 156500
2023
+ },
2024
+ {
2025
+ "epoch": 0.6,
2026
+ "learning_rate": 2.021618873451052e-05,
2027
+ "loss": 3.5575,
2028
+ "step": 157000
2029
+ },
2030
+ {
2031
+ "epoch": 0.6,
2032
+ "learning_rate": 2.0121335832391128e-05,
2033
+ "loss": 3.5569,
2034
+ "step": 157500
2035
+ },
2036
+ {
2037
+ "epoch": 0.6,
2038
+ "learning_rate": 2.0026482930271735e-05,
2039
+ "loss": 3.5537,
2040
+ "step": 158000
2041
+ },
2042
+ {
2043
+ "epoch": 0.6,
2044
+ "learning_rate": 1.9931630028152342e-05,
2045
+ "loss": 3.5553,
2046
+ "step": 158500
2047
+ },
2048
+ {
2049
+ "epoch": 0.6,
2050
+ "learning_rate": 1.983677712603295e-05,
2051
+ "loss": 3.5524,
2052
+ "step": 159000
2053
+ },
2054
+ {
2055
+ "epoch": 0.61,
2056
+ "learning_rate": 1.9741924223913557e-05,
2057
+ "loss": 3.5562,
2058
+ "step": 159500
2059
+ },
2060
+ {
2061
+ "epoch": 0.61,
2062
+ "learning_rate": 1.9647071321794164e-05,
2063
+ "loss": 3.5516,
2064
+ "step": 160000
2065
+ },
2066
+ {
2067
+ "epoch": 0.61,
2068
+ "eval_accuracy": 0.3828251390025017,
2069
+ "eval_loss": 3.4739012718200684,
2070
+ "eval_runtime": 5050.1987,
2071
+ "eval_samples_per_second": 87.872,
2072
+ "eval_steps_per_second": 1.373,
2073
+ "step": 160000
2074
+ },
2075
+ {
2076
+ "epoch": 0.61,
2077
+ "learning_rate": 1.955221841967477e-05,
2078
+ "loss": 3.5508,
2079
+ "step": 160500
2080
+ },
2081
+ {
2082
+ "epoch": 0.61,
2083
+ "learning_rate": 1.9457365517555375e-05,
2084
+ "loss": 3.5424,
2085
+ "step": 161000
2086
+ },
2087
+ {
2088
+ "epoch": 0.61,
2089
+ "learning_rate": 1.936251261543598e-05,
2090
+ "loss": 3.5526,
2091
+ "step": 161500
2092
+ },
2093
+ {
2094
+ "epoch": 0.61,
2095
+ "learning_rate": 1.9267659713316592e-05,
2096
+ "loss": 3.5469,
2097
+ "step": 162000
2098
+ },
2099
+ {
2100
+ "epoch": 0.62,
2101
+ "learning_rate": 1.9172806811197196e-05,
2102
+ "loss": 3.5401,
2103
+ "step": 162500
2104
+ },
2105
+ {
2106
+ "epoch": 0.62,
2107
+ "learning_rate": 1.9077953909077803e-05,
2108
+ "loss": 3.5525,
2109
+ "step": 163000
2110
+ },
2111
+ {
2112
+ "epoch": 0.62,
2113
+ "learning_rate": 1.898310100695841e-05,
2114
+ "loss": 3.5494,
2115
+ "step": 163500
2116
+ },
2117
+ {
2118
+ "epoch": 0.62,
2119
+ "learning_rate": 1.8888248104839014e-05,
2120
+ "loss": 3.5527,
2121
+ "step": 164000
2122
+ },
2123
+ {
2124
+ "epoch": 0.62,
2125
+ "learning_rate": 1.8793395202719624e-05,
2126
+ "loss": 3.5477,
2127
+ "step": 164500
2128
+ },
2129
+ {
2130
+ "epoch": 0.63,
2131
+ "learning_rate": 1.869854230060023e-05,
2132
+ "loss": 3.548,
2133
+ "step": 165000
2134
+ },
2135
+ {
2136
+ "epoch": 0.63,
2137
+ "learning_rate": 1.8603689398480835e-05,
2138
+ "loss": 3.5466,
2139
+ "step": 165500
2140
+ },
2141
+ {
2142
+ "epoch": 0.63,
2143
+ "learning_rate": 1.8508836496361442e-05,
2144
+ "loss": 3.5491,
2145
+ "step": 166000
2146
+ },
2147
+ {
2148
+ "epoch": 0.63,
2149
+ "learning_rate": 1.8413983594242053e-05,
2150
+ "loss": 3.5431,
2151
+ "step": 166500
2152
+ },
2153
+ {
2154
+ "epoch": 0.63,
2155
+ "learning_rate": 1.8319130692122657e-05,
2156
+ "loss": 3.5462,
2157
+ "step": 167000
2158
+ },
2159
+ {
2160
+ "epoch": 0.64,
2161
+ "learning_rate": 1.8224277790003264e-05,
2162
+ "loss": 3.5453,
2163
+ "step": 167500
2164
+ },
2165
+ {
2166
+ "epoch": 0.64,
2167
+ "learning_rate": 1.812942488788387e-05,
2168
+ "loss": 3.5408,
2169
+ "step": 168000
2170
+ },
2171
+ {
2172
+ "epoch": 0.64,
2173
+ "learning_rate": 1.8034571985764478e-05,
2174
+ "loss": 3.5465,
2175
+ "step": 168500
2176
+ },
2177
+ {
2178
+ "epoch": 0.64,
2179
+ "learning_rate": 1.7939719083645085e-05,
2180
+ "loss": 3.5437,
2181
+ "step": 169000
2182
+ },
2183
+ {
2184
+ "epoch": 0.64,
2185
+ "learning_rate": 1.784486618152569e-05,
2186
+ "loss": 3.533,
2187
+ "step": 169500
2188
+ },
2189
+ {
2190
+ "epoch": 0.64,
2191
+ "learning_rate": 1.7750013279406296e-05,
2192
+ "loss": 3.5423,
2193
+ "step": 170000
2194
+ },
2195
+ {
2196
+ "epoch": 0.64,
2197
+ "eval_accuracy": 0.38377248075624093,
2198
+ "eval_loss": 3.4649875164031982,
2199
+ "eval_runtime": 5024.5121,
2200
+ "eval_samples_per_second": 88.322,
2201
+ "eval_steps_per_second": 1.38,
2202
+ "step": 170000
2203
+ },
2204
+ {
2205
+ "epoch": 0.65,
2206
+ "learning_rate": 1.7655160377286907e-05,
2207
+ "loss": 3.5446,
2208
+ "step": 170500
2209
+ },
2210
+ {
2211
+ "epoch": 0.65,
2212
+ "learning_rate": 1.756030747516751e-05,
2213
+ "loss": 3.5374,
2214
+ "step": 171000
2215
+ },
2216
+ {
2217
+ "epoch": 0.65,
2218
+ "learning_rate": 1.7465454573048118e-05,
2219
+ "loss": 3.5426,
2220
+ "step": 171500
2221
+ },
2222
+ {
2223
+ "epoch": 0.65,
2224
+ "learning_rate": 1.7370601670928725e-05,
2225
+ "loss": 3.5391,
2226
+ "step": 172000
2227
+ },
2228
+ {
2229
+ "epoch": 0.65,
2230
+ "learning_rate": 1.727574876880933e-05,
2231
+ "loss": 3.544,
2232
+ "step": 172500
2233
+ },
2234
+ {
2235
+ "epoch": 0.66,
2236
+ "learning_rate": 1.718089586668994e-05,
2237
+ "loss": 3.5396,
2238
+ "step": 173000
2239
+ },
2240
+ {
2241
+ "epoch": 0.66,
2242
+ "learning_rate": 1.7086042964570546e-05,
2243
+ "loss": 3.5385,
2244
+ "step": 173500
2245
+ },
2246
+ {
2247
+ "epoch": 0.66,
2248
+ "learning_rate": 1.699119006245115e-05,
2249
+ "loss": 3.534,
2250
+ "step": 174000
2251
+ },
2252
+ {
2253
+ "epoch": 0.66,
2254
+ "learning_rate": 1.6896337160331757e-05,
2255
+ "loss": 3.5374,
2256
+ "step": 174500
2257
+ },
2258
+ {
2259
+ "epoch": 0.66,
2260
+ "learning_rate": 1.6801484258212368e-05,
2261
+ "loss": 3.5351,
2262
+ "step": 175000
2263
+ },
2264
+ {
2265
+ "epoch": 0.67,
2266
+ "learning_rate": 1.670663135609297e-05,
2267
+ "loss": 3.5391,
2268
+ "step": 175500
2269
+ },
2270
+ {
2271
+ "epoch": 0.67,
2272
+ "learning_rate": 1.661177845397358e-05,
2273
+ "loss": 3.5351,
2274
+ "step": 176000
2275
+ },
2276
+ {
2277
+ "epoch": 0.67,
2278
+ "learning_rate": 1.6516925551854186e-05,
2279
+ "loss": 3.5315,
2280
+ "step": 176500
2281
+ },
2282
+ {
2283
+ "epoch": 0.67,
2284
+ "learning_rate": 1.6422072649734793e-05,
2285
+ "loss": 3.5324,
2286
+ "step": 177000
2287
+ },
2288
+ {
2289
+ "epoch": 0.67,
2290
+ "learning_rate": 1.63272197476154e-05,
2291
+ "loss": 3.5379,
2292
+ "step": 177500
2293
+ },
2294
+ {
2295
+ "epoch": 0.68,
2296
+ "learning_rate": 1.6232366845496007e-05,
2297
+ "loss": 3.534,
2298
+ "step": 178000
2299
+ },
2300
+ {
2301
+ "epoch": 0.68,
2302
+ "learning_rate": 1.613751394337661e-05,
2303
+ "loss": 3.5366,
2304
+ "step": 178500
2305
+ },
2306
+ {
2307
+ "epoch": 0.68,
2308
+ "learning_rate": 1.604266104125722e-05,
2309
+ "loss": 3.5364,
2310
+ "step": 179000
2311
+ },
2312
+ {
2313
+ "epoch": 0.68,
2314
+ "learning_rate": 1.5947808139137825e-05,
2315
+ "loss": 3.5398,
2316
+ "step": 179500
2317
+ },
2318
+ {
2319
+ "epoch": 0.68,
2320
+ "learning_rate": 1.5852955237018432e-05,
2321
+ "loss": 3.5298,
2322
+ "step": 180000
2323
+ },
2324
+ {
2325
+ "epoch": 0.68,
2326
+ "eval_accuracy": 0.3846720031995081,
2327
+ "eval_loss": 3.455994129180908,
2328
+ "eval_runtime": 5033.5621,
2329
+ "eval_samples_per_second": 88.163,
2330
+ "eval_steps_per_second": 1.378,
2331
+ "step": 180000
2332
+ },
2333
+ {
2334
+ "epoch": 0.68,
2335
+ "learning_rate": 1.575810233489904e-05,
2336
+ "loss": 3.5337,
2337
+ "step": 180500
2338
+ },
2339
+ {
2340
+ "epoch": 0.69,
2341
+ "learning_rate": 1.5663249432779646e-05,
2342
+ "loss": 3.5324,
2343
+ "step": 181000
2344
+ },
2345
+ {
2346
+ "epoch": 0.69,
2347
+ "learning_rate": 1.5568396530660254e-05,
2348
+ "loss": 3.5307,
2349
+ "step": 181500
2350
+ },
2351
+ {
2352
+ "epoch": 0.69,
2353
+ "learning_rate": 1.547354362854086e-05,
2354
+ "loss": 3.5273,
2355
+ "step": 182000
2356
+ },
2357
+ {
2358
+ "epoch": 0.69,
2359
+ "learning_rate": 1.5378690726421464e-05,
2360
+ "loss": 3.5301,
2361
+ "step": 182500
2362
+ },
2363
+ {
2364
+ "epoch": 0.69,
2365
+ "learning_rate": 1.528383782430207e-05,
2366
+ "loss": 3.5321,
2367
+ "step": 183000
2368
+ },
2369
+ {
2370
+ "epoch": 0.7,
2371
+ "learning_rate": 1.518898492218268e-05,
2372
+ "loss": 3.5283,
2373
+ "step": 183500
2374
+ },
2375
+ {
2376
+ "epoch": 0.7,
2377
+ "learning_rate": 1.5094132020063287e-05,
2378
+ "loss": 3.534,
2379
+ "step": 184000
2380
+ },
2381
+ {
2382
+ "epoch": 0.7,
2383
+ "learning_rate": 1.4999279117943893e-05,
2384
+ "loss": 3.5346,
2385
+ "step": 184500
2386
+ },
2387
+ {
2388
+ "epoch": 0.7,
2389
+ "learning_rate": 1.4904426215824498e-05,
2390
+ "loss": 3.5234,
2391
+ "step": 185000
2392
+ },
2393
+ {
2394
+ "epoch": 0.7,
2395
+ "learning_rate": 1.4809573313705107e-05,
2396
+ "loss": 3.5243,
2397
+ "step": 185500
2398
+ },
2399
+ {
2400
+ "epoch": 0.71,
2401
+ "learning_rate": 1.4714720411585714e-05,
2402
+ "loss": 3.5292,
2403
+ "step": 186000
2404
+ },
2405
+ {
2406
+ "epoch": 0.71,
2407
+ "learning_rate": 1.461986750946632e-05,
2408
+ "loss": 3.5241,
2409
+ "step": 186500
2410
+ },
2411
+ {
2412
+ "epoch": 0.71,
2413
+ "learning_rate": 1.4525014607346927e-05,
2414
+ "loss": 3.5258,
2415
+ "step": 187000
2416
+ },
2417
+ {
2418
+ "epoch": 0.71,
2419
+ "learning_rate": 1.4430161705227536e-05,
2420
+ "loss": 3.5241,
2421
+ "step": 187500
2422
+ },
2423
+ {
2424
+ "epoch": 0.71,
2425
+ "learning_rate": 1.4335308803108141e-05,
2426
+ "loss": 3.5281,
2427
+ "step": 188000
2428
+ },
2429
+ {
2430
+ "epoch": 0.72,
2431
+ "learning_rate": 1.4240455900988747e-05,
2432
+ "loss": 3.5271,
2433
+ "step": 188500
2434
+ },
2435
+ {
2436
+ "epoch": 0.72,
2437
+ "learning_rate": 1.4145602998869354e-05,
2438
+ "loss": 3.5222,
2439
+ "step": 189000
2440
+ },
2441
+ {
2442
+ "epoch": 0.72,
2443
+ "learning_rate": 1.405075009674996e-05,
2444
+ "loss": 3.5197,
2445
+ "step": 189500
2446
+ },
2447
+ {
2448
+ "epoch": 0.72,
2449
+ "learning_rate": 1.3955897194630568e-05,
2450
+ "loss": 3.5287,
2451
+ "step": 190000
2452
+ },
2453
+ {
2454
+ "epoch": 0.72,
2455
+ "eval_accuracy": 0.38565153361158844,
2456
+ "eval_loss": 3.447903871536255,
2457
+ "eval_runtime": 5039.8049,
2458
+ "eval_samples_per_second": 88.054,
2459
+ "eval_steps_per_second": 1.376,
2460
+ "step": 190000
2461
+ },
2462
+ {
2463
+ "epoch": 0.72,
2464
+ "learning_rate": 1.3861044292511175e-05,
2465
+ "loss": 3.5309,
2466
+ "step": 190500
2467
+ },
2468
+ {
2469
+ "epoch": 0.72,
2470
+ "learning_rate": 1.376619139039178e-05,
2471
+ "loss": 3.5236,
2472
+ "step": 191000
2473
+ },
2474
+ {
2475
+ "epoch": 0.73,
2476
+ "learning_rate": 1.3671338488272386e-05,
2477
+ "loss": 3.5268,
2478
+ "step": 191500
2479
+ },
2480
+ {
2481
+ "epoch": 0.73,
2482
+ "learning_rate": 1.3576485586152995e-05,
2483
+ "loss": 3.5261,
2484
+ "step": 192000
2485
+ },
2486
+ {
2487
+ "epoch": 0.73,
2488
+ "learning_rate": 1.3481632684033602e-05,
2489
+ "loss": 3.5242,
2490
+ "step": 192500
2491
+ },
2492
+ {
2493
+ "epoch": 0.73,
2494
+ "learning_rate": 1.3386779781914207e-05,
2495
+ "loss": 3.5268,
2496
+ "step": 193000
2497
+ },
2498
+ {
2499
+ "epoch": 0.73,
2500
+ "learning_rate": 1.3291926879794815e-05,
2501
+ "loss": 3.5302,
2502
+ "step": 193500
2503
+ },
2504
+ {
2505
+ "epoch": 0.74,
2506
+ "learning_rate": 1.3197073977675423e-05,
2507
+ "loss": 3.525,
2508
+ "step": 194000
2509
+ },
2510
+ {
2511
+ "epoch": 0.74,
2512
+ "learning_rate": 1.3102221075556029e-05,
2513
+ "loss": 3.5278,
2514
+ "step": 194500
2515
+ },
2516
+ {
2517
+ "epoch": 0.74,
2518
+ "learning_rate": 1.3007368173436634e-05,
2519
+ "loss": 3.5221,
2520
+ "step": 195000
2521
+ },
2522
+ {
2523
+ "epoch": 0.74,
2524
+ "learning_rate": 1.2912515271317241e-05,
2525
+ "loss": 3.5231,
2526
+ "step": 195500
2527
+ },
2528
+ {
2529
+ "epoch": 0.74,
2530
+ "learning_rate": 1.281766236919785e-05,
2531
+ "loss": 3.5237,
2532
+ "step": 196000
2533
+ },
2534
+ {
2535
+ "epoch": 0.75,
2536
+ "learning_rate": 1.2722809467078456e-05,
2537
+ "loss": 3.5236,
2538
+ "step": 196500
2539
+ },
2540
+ {
2541
+ "epoch": 0.75,
2542
+ "learning_rate": 1.2627956564959063e-05,
2543
+ "loss": 3.5201,
2544
+ "step": 197000
2545
+ },
2546
+ {
2547
+ "epoch": 0.75,
2548
+ "learning_rate": 1.2533103662839668e-05,
2549
+ "loss": 3.5221,
2550
+ "step": 197500
2551
+ },
2552
+ {
2553
+ "epoch": 0.75,
2554
+ "learning_rate": 1.2438250760720275e-05,
2555
+ "loss": 3.5216,
2556
+ "step": 198000
2557
+ },
2558
+ {
2559
+ "epoch": 0.75,
2560
+ "learning_rate": 1.2343397858600881e-05,
2561
+ "loss": 3.5202,
2562
+ "step": 198500
2563
+ },
2564
+ {
2565
+ "epoch": 0.76,
2566
+ "learning_rate": 1.224854495648149e-05,
2567
+ "loss": 3.5207,
2568
+ "step": 199000
2569
+ },
2570
+ {
2571
+ "epoch": 0.76,
2572
+ "learning_rate": 1.2153692054362095e-05,
2573
+ "loss": 3.5172,
2574
+ "step": 199500
2575
+ },
2576
+ {
2577
+ "epoch": 0.76,
2578
+ "learning_rate": 1.2058839152242702e-05,
2579
+ "loss": 3.5187,
2580
+ "step": 200000
2581
+ },
2582
+ {
2583
+ "epoch": 0.76,
2584
+ "eval_accuracy": 0.3863054900513532,
2585
+ "eval_loss": 3.440758466720581,
2586
+ "eval_runtime": 5028.552,
2587
+ "eval_samples_per_second": 88.251,
2588
+ "eval_steps_per_second": 1.379,
2589
+ "step": 200000
2590
+ },
2591
+ {
2592
+ "epoch": 0.76,
2593
+ "learning_rate": 1.196398625012331e-05,
2594
+ "loss": 3.5153,
2595
+ "step": 200500
2596
+ },
2597
+ {
2598
+ "epoch": 0.76,
2599
+ "learning_rate": 1.1869133348003917e-05,
2600
+ "loss": 3.5164,
2601
+ "step": 201000
2602
+ },
2603
+ {
2604
+ "epoch": 0.76,
2605
+ "learning_rate": 1.1774280445884522e-05,
2606
+ "loss": 3.5225,
2607
+ "step": 201500
2608
+ },
2609
+ {
2610
+ "epoch": 0.77,
2611
+ "learning_rate": 1.1679427543765129e-05,
2612
+ "loss": 3.5142,
2613
+ "step": 202000
2614
+ },
2615
+ {
2616
+ "epoch": 0.77,
2617
+ "learning_rate": 1.1584574641645736e-05,
2618
+ "loss": 3.519,
2619
+ "step": 202500
2620
+ },
2621
+ {
2622
+ "epoch": 0.77,
2623
+ "learning_rate": 1.1489721739526343e-05,
2624
+ "loss": 3.522,
2625
+ "step": 203000
2626
+ },
2627
+ {
2628
+ "epoch": 0.77,
2629
+ "learning_rate": 1.1394868837406949e-05,
2630
+ "loss": 3.5142,
2631
+ "step": 203500
2632
+ },
2633
+ {
2634
+ "epoch": 0.77,
2635
+ "learning_rate": 1.1300015935287558e-05,
2636
+ "loss": 3.5101,
2637
+ "step": 204000
2638
+ },
2639
+ {
2640
+ "epoch": 0.78,
2641
+ "learning_rate": 1.1205163033168163e-05,
2642
+ "loss": 3.5151,
2643
+ "step": 204500
2644
+ },
2645
+ {
2646
+ "epoch": 0.78,
2647
+ "learning_rate": 1.111031013104877e-05,
2648
+ "loss": 3.5112,
2649
+ "step": 205000
2650
+ },
2651
+ {
2652
+ "epoch": 0.78,
2653
+ "learning_rate": 1.1015457228929377e-05,
2654
+ "loss": 3.5161,
2655
+ "step": 205500
2656
+ },
2657
+ {
2658
+ "epoch": 0.78,
2659
+ "learning_rate": 1.0920604326809984e-05,
2660
+ "loss": 3.5135,
2661
+ "step": 206000
2662
+ },
2663
+ {
2664
+ "epoch": 0.78,
2665
+ "learning_rate": 1.082575142469059e-05,
2666
+ "loss": 3.5143,
2667
+ "step": 206500
2668
+ },
2669
+ {
2670
+ "epoch": 0.79,
2671
+ "learning_rate": 1.0730898522571197e-05,
2672
+ "loss": 3.5134,
2673
+ "step": 207000
2674
+ },
2675
+ {
2676
+ "epoch": 0.79,
2677
+ "learning_rate": 1.0636045620451804e-05,
2678
+ "loss": 3.5162,
2679
+ "step": 207500
2680
+ },
2681
+ {
2682
+ "epoch": 0.79,
2683
+ "learning_rate": 1.054119271833241e-05,
2684
+ "loss": 3.5202,
2685
+ "step": 208000
2686
+ },
2687
+ {
2688
+ "epoch": 0.79,
2689
+ "learning_rate": 1.0446339816213017e-05,
2690
+ "loss": 3.507,
2691
+ "step": 208500
2692
+ },
2693
+ {
2694
+ "epoch": 0.79,
2695
+ "learning_rate": 1.0351486914093624e-05,
2696
+ "loss": 3.5108,
2697
+ "step": 209000
2698
+ },
2699
+ {
2700
+ "epoch": 0.79,
2701
+ "learning_rate": 1.0256634011974231e-05,
2702
+ "loss": 3.5144,
2703
+ "step": 209500
2704
+ },
2705
+ {
2706
+ "epoch": 0.8,
2707
+ "learning_rate": 1.0161781109854836e-05,
2708
+ "loss": 3.5157,
2709
+ "step": 210000
2710
+ },
2711
+ {
2712
+ "epoch": 0.8,
2713
+ "eval_accuracy": 0.3870160062789933,
2714
+ "eval_loss": 3.4338622093200684,
2715
+ "eval_runtime": 5031.2535,
2716
+ "eval_samples_per_second": 88.203,
2717
+ "eval_steps_per_second": 1.378,
2718
+ "step": 210000
2719
+ },
2720
+ {
2721
+ "epoch": 0.8,
2722
+ "learning_rate": 1.0066928207735445e-05,
2723
+ "loss": 3.5167,
2724
+ "step": 210500
2725
+ },
2726
+ {
2727
+ "epoch": 0.8,
2728
+ "learning_rate": 9.97207530561605e-06,
2729
+ "loss": 3.5107,
2730
+ "step": 211000
2731
+ },
2732
+ {
2733
+ "epoch": 0.8,
2734
+ "learning_rate": 9.877222403496658e-06,
2735
+ "loss": 3.5114,
2736
+ "step": 211500
2737
+ },
2738
+ {
2739
+ "epoch": 0.8,
2740
+ "learning_rate": 9.782369501377265e-06,
2741
+ "loss": 3.5121,
2742
+ "step": 212000
2743
+ },
2744
+ {
2745
+ "epoch": 0.81,
2746
+ "learning_rate": 9.687516599257872e-06,
2747
+ "loss": 3.5084,
2748
+ "step": 212500
2749
+ },
2750
+ {
2751
+ "epoch": 0.81,
2752
+ "learning_rate": 9.592663697138478e-06,
2753
+ "loss": 3.5178,
2754
+ "step": 213000
2755
+ },
2756
+ {
2757
+ "epoch": 0.81,
2758
+ "learning_rate": 9.497810795019085e-06,
2759
+ "loss": 3.5076,
2760
+ "step": 213500
2761
+ },
2762
+ {
2763
+ "epoch": 0.81,
2764
+ "learning_rate": 9.402957892899692e-06,
2765
+ "loss": 3.5102,
2766
+ "step": 214000
2767
+ },
2768
+ {
2769
+ "epoch": 0.81,
2770
+ "learning_rate": 9.308104990780299e-06,
2771
+ "loss": 3.5069,
2772
+ "step": 214500
2773
+ },
2774
+ {
2775
+ "epoch": 0.82,
2776
+ "learning_rate": 9.213252088660904e-06,
2777
+ "loss": 3.5108,
2778
+ "step": 215000
2779
+ },
2780
+ {
2781
+ "epoch": 0.82,
2782
+ "learning_rate": 9.118399186541513e-06,
2783
+ "loss": 3.5087,
2784
+ "step": 215500
2785
+ },
2786
+ {
2787
+ "epoch": 0.82,
2788
+ "learning_rate": 9.023546284422119e-06,
2789
+ "loss": 3.5041,
2790
+ "step": 216000
2791
+ },
2792
+ {
2793
+ "epoch": 0.82,
2794
+ "learning_rate": 8.928693382302724e-06,
2795
+ "loss": 3.5125,
2796
+ "step": 216500
2797
+ },
2798
+ {
2799
+ "epoch": 0.82,
2800
+ "learning_rate": 8.833840480183333e-06,
2801
+ "loss": 3.5084,
2802
+ "step": 217000
2803
+ },
2804
+ {
2805
+ "epoch": 0.83,
2806
+ "learning_rate": 8.738987578063938e-06,
2807
+ "loss": 3.5034,
2808
+ "step": 217500
2809
+ },
2810
+ {
2811
+ "epoch": 0.83,
2812
+ "learning_rate": 8.644134675944546e-06,
2813
+ "loss": 3.5077,
2814
+ "step": 218000
2815
+ },
2816
+ {
2817
+ "epoch": 0.83,
2818
+ "learning_rate": 8.549281773825151e-06,
2819
+ "loss": 3.505,
2820
+ "step": 218500
2821
+ },
2822
+ {
2823
+ "epoch": 0.83,
2824
+ "learning_rate": 8.45442887170576e-06,
2825
+ "loss": 3.5072,
2826
+ "step": 219000
2827
+ },
2828
+ {
2829
+ "epoch": 0.83,
2830
+ "learning_rate": 8.359575969586365e-06,
2831
+ "loss": 3.5,
2832
+ "step": 219500
2833
+ },
2834
+ {
2835
+ "epoch": 0.83,
2836
+ "learning_rate": 8.264723067466972e-06,
2837
+ "loss": 3.5042,
2838
+ "step": 220000
2839
+ },
2840
+ {
2841
+ "epoch": 0.83,
2842
+ "eval_accuracy": 0.3876448779010485,
2843
+ "eval_loss": 3.4285898208618164,
2844
+ "eval_runtime": 5039.4169,
2845
+ "eval_samples_per_second": 88.06,
2846
+ "eval_steps_per_second": 1.376,
2847
+ "step": 220000
2848
+ },
2849
+ {
2850
+ "epoch": 0.84,
2851
+ "learning_rate": 8.16987016534758e-06,
2852
+ "loss": 3.5075,
2853
+ "step": 220500
2854
+ },
2855
+ {
2856
+ "epoch": 0.84,
2857
+ "learning_rate": 8.075017263228187e-06,
2858
+ "loss": 3.5056,
2859
+ "step": 221000
2860
+ },
2861
+ {
2862
+ "epoch": 0.84,
2863
+ "learning_rate": 7.980164361108792e-06,
2864
+ "loss": 3.5089,
2865
+ "step": 221500
2866
+ },
2867
+ {
2868
+ "epoch": 0.84,
2869
+ "learning_rate": 7.8853114589894e-06,
2870
+ "loss": 3.5114,
2871
+ "step": 222000
2872
+ },
2873
+ {
2874
+ "epoch": 0.84,
2875
+ "learning_rate": 7.790458556870006e-06,
2876
+ "loss": 3.5076,
2877
+ "step": 222500
2878
+ },
2879
+ {
2880
+ "epoch": 0.85,
2881
+ "learning_rate": 7.695605654750614e-06,
2882
+ "loss": 3.5053,
2883
+ "step": 223000
2884
+ },
2885
+ {
2886
+ "epoch": 0.85,
2887
+ "learning_rate": 7.60075275263122e-06,
2888
+ "loss": 3.5058,
2889
+ "step": 223500
2890
+ },
2891
+ {
2892
+ "epoch": 0.85,
2893
+ "learning_rate": 7.505899850511827e-06,
2894
+ "loss": 3.5078,
2895
+ "step": 224000
2896
+ },
2897
+ {
2898
+ "epoch": 0.85,
2899
+ "learning_rate": 7.411046948392433e-06,
2900
+ "loss": 3.5084,
2901
+ "step": 224500
2902
+ },
2903
+ {
2904
+ "epoch": 0.85,
2905
+ "learning_rate": 7.31619404627304e-06,
2906
+ "loss": 3.5038,
2907
+ "step": 225000
2908
+ },
2909
+ {
2910
+ "epoch": 0.86,
2911
+ "learning_rate": 7.221341144153647e-06,
2912
+ "loss": 3.5015,
2913
+ "step": 225500
2914
+ },
2915
+ {
2916
+ "epoch": 0.86,
2917
+ "learning_rate": 7.126488242034253e-06,
2918
+ "loss": 3.5034,
2919
+ "step": 226000
2920
+ },
2921
+ {
2922
+ "epoch": 0.86,
2923
+ "learning_rate": 7.03163533991486e-06,
2924
+ "loss": 3.5098,
2925
+ "step": 226500
2926
+ },
2927
+ {
2928
+ "epoch": 0.86,
2929
+ "learning_rate": 6.936782437795466e-06,
2930
+ "loss": 3.5001,
2931
+ "step": 227000
2932
+ },
2933
+ {
2934
+ "epoch": 0.86,
2935
+ "learning_rate": 6.841929535676074e-06,
2936
+ "loss": 3.5055,
2937
+ "step": 227500
2938
+ },
2939
+ {
2940
+ "epoch": 0.87,
2941
+ "learning_rate": 6.74707663355668e-06,
2942
+ "loss": 3.5018,
2943
+ "step": 228000
2944
+ },
2945
+ {
2946
+ "epoch": 0.87,
2947
+ "learning_rate": 6.652223731437288e-06,
2948
+ "loss": 3.503,
2949
+ "step": 228500
2950
+ },
2951
+ {
2952
+ "epoch": 0.87,
2953
+ "learning_rate": 6.557370829317894e-06,
2954
+ "loss": 3.503,
2955
+ "step": 229000
2956
+ },
2957
+ {
2958
+ "epoch": 0.87,
2959
+ "learning_rate": 6.462517927198501e-06,
2960
+ "loss": 3.498,
2961
+ "step": 229500
2962
+ },
2963
+ {
2964
+ "epoch": 0.87,
2965
+ "learning_rate": 6.3676650250791075e-06,
2966
+ "loss": 3.5033,
2967
+ "step": 230000
2968
+ },
2969
+ {
2970
+ "epoch": 0.87,
2971
+ "eval_accuracy": 0.38827686860475785,
2972
+ "eval_loss": 3.422902822494507,
2973
+ "eval_runtime": 5026.5934,
2974
+ "eval_samples_per_second": 88.285,
2975
+ "eval_steps_per_second": 1.379,
2976
+ "step": 230000
2977
+ },
2978
+ {
2979
+ "epoch": 0.87,
2980
+ "learning_rate": 6.272812122959715e-06,
2981
+ "loss": 3.5024,
2982
+ "step": 230500
2983
+ },
2984
+ {
2985
+ "epoch": 0.88,
2986
+ "learning_rate": 6.177959220840321e-06,
2987
+ "loss": 3.4986,
2988
+ "step": 231000
2989
+ },
2990
+ {
2991
+ "epoch": 0.88,
2992
+ "learning_rate": 6.083106318720928e-06,
2993
+ "loss": 3.5025,
2994
+ "step": 231500
2995
+ },
2996
+ {
2997
+ "epoch": 0.88,
2998
+ "learning_rate": 5.988253416601534e-06,
2999
+ "loss": 3.5065,
3000
+ "step": 232000
3001
+ },
3002
+ {
3003
+ "epoch": 0.88,
3004
+ "learning_rate": 5.8934005144821415e-06,
3005
+ "loss": 3.5047,
3006
+ "step": 232500
3007
+ },
3008
+ {
3009
+ "epoch": 0.88,
3010
+ "learning_rate": 5.798547612362748e-06,
3011
+ "loss": 3.502,
3012
+ "step": 233000
3013
+ },
3014
+ {
3015
+ "epoch": 0.89,
3016
+ "learning_rate": 5.703694710243355e-06,
3017
+ "loss": 3.5061,
3018
+ "step": 233500
3019
+ },
3020
+ {
3021
+ "epoch": 0.89,
3022
+ "learning_rate": 5.608841808123961e-06,
3023
+ "loss": 3.5036,
3024
+ "step": 234000
3025
+ },
3026
+ {
3027
+ "epoch": 0.89,
3028
+ "learning_rate": 5.513988906004568e-06,
3029
+ "loss": 3.4966,
3030
+ "step": 234500
3031
+ },
3032
+ {
3033
+ "epoch": 0.89,
3034
+ "learning_rate": 5.4191360038851754e-06,
3035
+ "loss": 3.5048,
3036
+ "step": 235000
3037
+ },
3038
+ {
3039
+ "epoch": 0.89,
3040
+ "learning_rate": 5.324283101765782e-06,
3041
+ "loss": 3.5024,
3042
+ "step": 235500
3043
+ },
3044
+ {
3045
+ "epoch": 0.9,
3046
+ "learning_rate": 5.229430199646389e-06,
3047
+ "loss": 3.495,
3048
+ "step": 236000
3049
+ },
3050
+ {
3051
+ "epoch": 0.9,
3052
+ "learning_rate": 5.134577297526995e-06,
3053
+ "loss": 3.5035,
3054
+ "step": 236500
3055
+ },
3056
+ {
3057
+ "epoch": 0.9,
3058
+ "learning_rate": 5.039724395407602e-06,
3059
+ "loss": 3.5026,
3060
+ "step": 237000
3061
+ },
3062
+ {
3063
+ "epoch": 0.9,
3064
+ "learning_rate": 4.9448714932882094e-06,
3065
+ "loss": 3.4997,
3066
+ "step": 237500
3067
+ },
3068
+ {
3069
+ "epoch": 0.9,
3070
+ "learning_rate": 4.850018591168816e-06,
3071
+ "loss": 3.5035,
3072
+ "step": 238000
3073
+ },
3074
+ {
3075
+ "epoch": 0.9,
3076
+ "learning_rate": 4.755165689049423e-06,
3077
+ "loss": 3.4975,
3078
+ "step": 238500
3079
+ },
3080
+ {
3081
+ "epoch": 0.91,
3082
+ "learning_rate": 4.660312786930029e-06,
3083
+ "loss": 3.5003,
3084
+ "step": 239000
3085
+ },
3086
+ {
3087
+ "epoch": 0.91,
3088
+ "learning_rate": 4.5654598848106354e-06,
3089
+ "loss": 3.4994,
3090
+ "step": 239500
3091
+ },
3092
+ {
3093
+ "epoch": 0.91,
3094
+ "learning_rate": 4.4706069826912426e-06,
3095
+ "loss": 3.501,
3096
+ "step": 240000
3097
+ },
3098
+ {
3099
+ "epoch": 0.91,
3100
+ "eval_accuracy": 0.38875705078485445,
3101
+ "eval_loss": 3.4187748432159424,
3102
+ "eval_runtime": 5045.601,
3103
+ "eval_samples_per_second": 87.952,
3104
+ "eval_steps_per_second": 1.374,
3105
+ "step": 240000
3106
+ },
3107
+ {
3108
+ "epoch": 0.91,
3109
+ "learning_rate": 4.375754080571849e-06,
3110
+ "loss": 3.4917,
3111
+ "step": 240500
3112
+ },
3113
+ {
3114
+ "epoch": 0.91,
3115
+ "learning_rate": 4.280901178452456e-06,
3116
+ "loss": 3.4986,
3117
+ "step": 241000
3118
+ },
3119
+ {
3120
+ "epoch": 0.92,
3121
+ "learning_rate": 4.186048276333063e-06,
3122
+ "loss": 3.496,
3123
+ "step": 241500
3124
+ },
3125
+ {
3126
+ "epoch": 0.92,
3127
+ "learning_rate": 4.091195374213669e-06,
3128
+ "loss": 3.4978,
3129
+ "step": 242000
3130
+ },
3131
+ {
3132
+ "epoch": 0.92,
3133
+ "learning_rate": 3.9963424720942765e-06,
3134
+ "loss": 3.491,
3135
+ "step": 242500
3136
+ },
3137
+ {
3138
+ "epoch": 0.92,
3139
+ "learning_rate": 3.901489569974883e-06,
3140
+ "loss": 3.4953,
3141
+ "step": 243000
3142
+ },
3143
+ {
3144
+ "epoch": 0.92,
3145
+ "learning_rate": 3.80663666785549e-06,
3146
+ "loss": 3.4962,
3147
+ "step": 243500
3148
+ },
3149
+ {
3150
+ "epoch": 0.93,
3151
+ "learning_rate": 3.7117837657360967e-06,
3152
+ "loss": 3.4982,
3153
+ "step": 244000
3154
+ },
3155
+ {
3156
+ "epoch": 0.93,
3157
+ "learning_rate": 3.6169308636167034e-06,
3158
+ "loss": 3.4973,
3159
+ "step": 244500
3160
+ },
3161
+ {
3162
+ "epoch": 0.93,
3163
+ "learning_rate": 3.52207796149731e-06,
3164
+ "loss": 3.4922,
3165
+ "step": 245000
3166
+ },
3167
+ {
3168
+ "epoch": 0.93,
3169
+ "learning_rate": 3.4272250593779172e-06,
3170
+ "loss": 3.4921,
3171
+ "step": 245500
3172
+ },
3173
+ {
3174
+ "epoch": 0.93,
3175
+ "learning_rate": 3.332372157258524e-06,
3176
+ "loss": 3.5039,
3177
+ "step": 246000
3178
+ },
3179
+ {
3180
+ "epoch": 0.94,
3181
+ "learning_rate": 3.2375192551391307e-06,
3182
+ "loss": 3.5025,
3183
+ "step": 246500
3184
+ },
3185
+ {
3186
+ "epoch": 0.94,
3187
+ "learning_rate": 3.1426663530197374e-06,
3188
+ "loss": 3.4964,
3189
+ "step": 247000
3190
+ },
3191
+ {
3192
+ "epoch": 0.94,
3193
+ "learning_rate": 3.0478134509003437e-06,
3194
+ "loss": 3.4936,
3195
+ "step": 247500
3196
+ },
3197
+ {
3198
+ "epoch": 0.94,
3199
+ "learning_rate": 2.952960548780951e-06,
3200
+ "loss": 3.4959,
3201
+ "step": 248000
3202
+ },
3203
+ {
3204
+ "epoch": 0.94,
3205
+ "learning_rate": 2.8581076466615575e-06,
3206
+ "loss": 3.4963,
3207
+ "step": 248500
3208
+ },
3209
+ {
3210
+ "epoch": 0.94,
3211
+ "learning_rate": 2.7632547445421642e-06,
3212
+ "loss": 3.5014,
3213
+ "step": 249000
3214
+ },
3215
+ {
3216
+ "epoch": 0.95,
3217
+ "learning_rate": 2.668401842422771e-06,
3218
+ "loss": 3.4948,
3219
+ "step": 249500
3220
+ },
3221
+ {
3222
+ "epoch": 0.95,
3223
+ "learning_rate": 2.5735489403033776e-06,
3224
+ "loss": 3.4946,
3225
+ "step": 250000
3226
+ },
3227
+ {
3228
+ "epoch": 0.95,
3229
+ "eval_accuracy": 0.38918558969561506,
3230
+ "eval_loss": 3.4148616790771484,
3231
+ "eval_runtime": 5028.0262,
3232
+ "eval_samples_per_second": 88.26,
3233
+ "eval_steps_per_second": 1.379,
3234
+ "step": 250000
3235
+ },
3236
+ {
3237
+ "epoch": 0.95,
3238
+ "learning_rate": 2.4786960381839844e-06,
3239
+ "loss": 3.4959,
3240
+ "step": 250500
3241
+ },
3242
+ {
3243
+ "epoch": 0.95,
3244
+ "learning_rate": 2.383843136064591e-06,
3245
+ "loss": 3.4955,
3246
+ "step": 251000
3247
+ },
3248
+ {
3249
+ "epoch": 0.95,
3250
+ "learning_rate": 2.2889902339451978e-06,
3251
+ "loss": 3.4869,
3252
+ "step": 251500
3253
+ },
3254
+ {
3255
+ "epoch": 0.96,
3256
+ "learning_rate": 2.1941373318258045e-06,
3257
+ "loss": 3.4975,
3258
+ "step": 252000
3259
+ },
3260
+ {
3261
+ "epoch": 0.96,
3262
+ "learning_rate": 2.099284429706411e-06,
3263
+ "loss": 3.4871,
3264
+ "step": 252500
3265
+ },
3266
+ {
3267
+ "epoch": 0.96,
3268
+ "learning_rate": 2.0044315275870183e-06,
3269
+ "loss": 3.4889,
3270
+ "step": 253000
3271
+ },
3272
+ {
3273
+ "epoch": 0.96,
3274
+ "learning_rate": 1.909578625467625e-06,
3275
+ "loss": 3.4928,
3276
+ "step": 253500
3277
+ },
3278
+ {
3279
+ "epoch": 0.96,
3280
+ "learning_rate": 1.8147257233482318e-06,
3281
+ "loss": 3.4921,
3282
+ "step": 254000
3283
+ },
3284
+ {
3285
+ "epoch": 0.97,
3286
+ "learning_rate": 1.7198728212288385e-06,
3287
+ "loss": 3.494,
3288
+ "step": 254500
3289
+ },
3290
+ {
3291
+ "epoch": 0.97,
3292
+ "learning_rate": 1.625019919109445e-06,
3293
+ "loss": 3.4918,
3294
+ "step": 255000
3295
+ },
3296
+ {
3297
+ "epoch": 0.97,
3298
+ "learning_rate": 1.5301670169900519e-06,
3299
+ "loss": 3.4919,
3300
+ "step": 255500
3301
+ },
3302
+ {
3303
+ "epoch": 0.97,
3304
+ "learning_rate": 1.4353141148706586e-06,
3305
+ "loss": 3.4907,
3306
+ "step": 256000
3307
+ },
3308
+ {
3309
+ "epoch": 0.97,
3310
+ "learning_rate": 1.3404612127512653e-06,
3311
+ "loss": 3.4937,
3312
+ "step": 256500
3313
+ },
3314
+ {
3315
+ "epoch": 0.98,
3316
+ "learning_rate": 1.2456083106318722e-06,
3317
+ "loss": 3.4893,
3318
+ "step": 257000
3319
+ },
3320
+ {
3321
+ "epoch": 0.98,
3322
+ "learning_rate": 1.150755408512479e-06,
3323
+ "loss": 3.4944,
3324
+ "step": 257500
3325
+ },
3326
+ {
3327
+ "epoch": 0.98,
3328
+ "learning_rate": 1.0559025063930857e-06,
3329
+ "loss": 3.4914,
3330
+ "step": 258000
3331
+ },
3332
+ {
3333
+ "epoch": 0.98,
3334
+ "learning_rate": 9.610496042736924e-07,
3335
+ "loss": 3.4947,
3336
+ "step": 258500
3337
+ },
3338
+ {
3339
+ "epoch": 0.98,
3340
+ "learning_rate": 8.661967021542991e-07,
3341
+ "loss": 3.4916,
3342
+ "step": 259000
3343
+ },
3344
+ {
3345
+ "epoch": 0.98,
3346
+ "learning_rate": 7.713438000349059e-07,
3347
+ "loss": 3.4884,
3348
+ "step": 259500
3349
+ },
3350
+ {
3351
+ "epoch": 0.99,
3352
+ "learning_rate": 6.764908979155126e-07,
3353
+ "loss": 3.4971,
3354
+ "step": 260000
3355
+ },
3356
+ {
3357
+ "epoch": 0.99,
3358
+ "eval_accuracy": 0.3894276158938788,
3359
+ "eval_loss": 3.412609100341797,
3360
+ "eval_runtime": 5032.8837,
3361
+ "eval_samples_per_second": 88.175,
3362
+ "eval_steps_per_second": 1.378,
3363
+ "step": 260000
3364
+ },
3365
+ {
3366
+ "epoch": 0.99,
3367
+ "learning_rate": 5.816379957961194e-07,
3368
+ "loss": 3.4929,
3369
+ "step": 260500
3370
+ },
3371
+ {
3372
+ "epoch": 0.99,
3373
+ "learning_rate": 4.867850936767261e-07,
3374
+ "loss": 3.4937,
3375
+ "step": 261000
3376
+ },
3377
+ {
3378
+ "epoch": 0.99,
3379
+ "learning_rate": 3.919321915573329e-07,
3380
+ "loss": 3.4859,
3381
+ "step": 261500
3382
+ },
3383
+ {
3384
+ "epoch": 0.99,
3385
+ "learning_rate": 2.9707928943793967e-07,
3386
+ "loss": 3.4796,
3387
+ "step": 262000
3388
+ },
3389
+ {
3390
+ "epoch": 1.0,
3391
+ "learning_rate": 2.022263873185464e-07,
3392
+ "loss": 3.4871,
3393
+ "step": 262500
3394
+ },
3395
+ {
3396
+ "epoch": 1.0,
3397
+ "learning_rate": 1.0737348519915316e-07,
3398
+ "loss": 3.4949,
3399
+ "step": 263000
3400
+ },
3401
+ {
3402
+ "epoch": 1.0,
3403
+ "learning_rate": 1.252058307975991e-08,
3404
+ "loss": 3.4884,
3405
+ "step": 263500
3406
+ },
3407
+ {
3408
+ "epoch": 1.0,
3409
+ "step": 263566,
3410
+ "total_flos": 4.407529912270848e+18,
3411
+ "train_loss": 3.6950655784551065,
3412
+ "train_runtime": 269467.2599,
3413
+ "train_samples_per_second": 31.299,
3414
+ "train_steps_per_second": 0.978
3415
+ }
3416
+ ],
3417
+ "logging_steps": 500,
3418
+ "max_steps": 263566,
3419
+ "num_input_tokens_seen": 0,
3420
+ "num_train_epochs": 1,
3421
+ "save_steps": 10000,
3422
+ "total_flos": 4.407529912270848e+18,
3423
+ "train_batch_size": 32,
3424
+ "trial_name": null,
3425
+ "trial_params": null
3426
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3fdab7f7855ae6a771f26388f2d1505d0aef9de137fd2b12132902757927b853
3
+ size 4728