Hugging Face's logo

jonasknobloch
/
gpt2-morf_u0-30-50-x_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
b4320f3
•
1 Parent(s): 7009b16

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 +3474 -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+morf_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: 0.39500633343599556
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+morf_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: 3.3667
32
+ - Accuracy: 0.3950
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
+ | 4.3227 | 0.04 | 10000 | 4.2268 | 0.3161 |
64
+ | 4.0305 | 0.07 | 20000 | 3.9455 | 0.3393 |
65
+ | 3.8916 | 0.11 | 30000 | 3.8194 | 0.3502 |
66
+ | 3.8104 | 0.15 | 40000 | 3.7340 | 0.3580 |
67
+ | 3.7491 | 0.19 | 50000 | 3.6770 | 0.3633 |
68
+ | 3.7062 | 0.22 | 60000 | 3.6288 | 0.3679 |
69
+ | 3.6724 | 0.26 | 70000 | 3.5938 | 0.3714 |
70
+ | 3.6399 | 0.3 | 80000 | 3.5652 | 0.3743 |
71
+ | 3.6147 | 0.34 | 90000 | 3.5396 | 0.3768 |
72
+ | 3.5946 | 0.37 | 100000 | 3.5158 | 0.3791 |
73
+ | 3.5726 | 0.41 | 110000 | 3.4986 | 0.3809 |
74
+ | 3.5631 | 0.45 | 120000 | 3.4819 | 0.3826 |
75
+ | 3.5459 | 0.49 | 130000 | 3.4678 | 0.3842 |
76
+ | 3.5304 | 0.52 | 140000 | 3.4535 | 0.3857 |
77
+ | 3.5245 | 0.56 | 150000 | 3.4430 | 0.3867 |
78
+ | 3.5124 | 0.6 | 160000 | 3.4329 | 0.3877 |
79
+ | 3.501 | 0.63 | 170000 | 3.4223 | 0.3890 |
80
+ | 3.4934 | 0.67 | 180000 | 3.4130 | 0.3901 |
81
+ | 3.4863 | 0.71 | 190000 | 3.4042 | 0.3909 |
82
+ | 3.4799 | 0.75 | 200000 | 3.3991 | 0.3914 |
83
+ | 3.4682 | 0.78 | 210000 | 3.3909 | 0.3924 |
84
+ | 3.4667 | 0.82 | 220000 | 3.3852 | 0.3930 |
85
+ | 3.4564 | 0.86 | 230000 | 3.3790 | 0.3936 |
86
+ | 3.4581 | 0.9 | 240000 | 3.3753 | 0.3941 |
87
+ | 3.4553 | 0.93 | 250000 | 3.3710 | 0.3945 |
88
+ | 3.4508 | 0.97 | 260000 | 3.3680 | 0.3949 |
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": 0.39500633343599556,
4
+ "eval_loss": 3.3667407035827637,
5
+ "eval_runtime": 6163.948,
6
+ "eval_samples": 451185,
7
+ "eval_samples_per_second": 73.197,
8
+ "eval_steps_per_second": 1.144,
9
+ "perplexity": 28.98390580186287,
10
+ "train_loss": 3.648317173601172,
11
+ "train_runtime": 313365.2222,
12
+ "train_samples": 8574452,
13
+ "train_samples_per_second": 27.362,
14
+ "train_steps_per_second": 0.855
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": 0.39500633343599556,
4
+ "eval_loss": 3.3667407035827637,
5
+ "eval_runtime": 6163.948,
6
+ "eval_samples": 451185,
7
+ "eval_samples_per_second": 73.197,
8
+ "eval_steps_per_second": 1.144,
9
+ "perplexity": 28.98390580186287
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:d7e1b74f324dc010dc78361b9da2b2581286cc8081bc6705e54020e96014307c
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": 3.648317173601172,
4
+ "train_runtime": 313365.2222,
5
+ "train_samples": 8574452,
6
+ "train_samples_per_second": 27.362,
7
+ "train_steps_per_second": 0.855
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,3474 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 10000,
6
+ "global_step": 267952,
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": 4.9906699707410284e-05,
14
+ "loss": 7.0772,
15
+ "step": 500
16
+ },
17
+ {
18
+ "epoch": 0.0,
19
+ "learning_rate": 4.9813399414820566e-05,
20
+ "loss": 6.1824,
21
+ "step": 1000
22
+ },
23
+ {
24
+ "epoch": 0.01,
25
+ "learning_rate": 4.972009912223085e-05,
26
+ "loss": 5.8409,
27
+ "step": 1500
28
+ },
29
+ {
30
+ "epoch": 0.01,
31
+ "learning_rate": 4.962679882964113e-05,
32
+ "loss": 5.5985,
33
+ "step": 2000
34
+ },
35
+ {
36
+ "epoch": 0.01,
37
+ "learning_rate": 4.953349853705141e-05,
38
+ "loss": 5.403,
39
+ "step": 2500
40
+ },
41
+ {
42
+ "epoch": 0.01,
43
+ "learning_rate": 4.94401982444617e-05,
44
+ "loss": 5.2549,
45
+ "step": 3000
46
+ },
47
+ {
48
+ "epoch": 0.01,
49
+ "learning_rate": 4.9346897951871976e-05,
50
+ "loss": 5.1225,
51
+ "step": 3500
52
+ },
53
+ {
54
+ "epoch": 0.01,
55
+ "learning_rate": 4.9253597659282264e-05,
56
+ "loss": 4.9967,
57
+ "step": 4000
58
+ },
59
+ {
60
+ "epoch": 0.02,
61
+ "learning_rate": 4.916029736669254e-05,
62
+ "loss": 4.8795,
63
+ "step": 4500
64
+ },
65
+ {
66
+ "epoch": 0.02,
67
+ "learning_rate": 4.906699707410283e-05,
68
+ "loss": 4.7848,
69
+ "step": 5000
70
+ },
71
+ {
72
+ "epoch": 0.02,
73
+ "learning_rate": 4.897369678151311e-05,
74
+ "loss": 4.6995,
75
+ "step": 5500
76
+ },
77
+ {
78
+ "epoch": 0.02,
79
+ "learning_rate": 4.888039648892339e-05,
80
+ "loss": 4.6346,
81
+ "step": 6000
82
+ },
83
+ {
84
+ "epoch": 0.02,
85
+ "learning_rate": 4.8787096196333674e-05,
86
+ "loss": 4.5872,
87
+ "step": 6500
88
+ },
89
+ {
90
+ "epoch": 0.03,
91
+ "learning_rate": 4.8693795903743955e-05,
92
+ "loss": 4.521,
93
+ "step": 7000
94
+ },
95
+ {
96
+ "epoch": 0.03,
97
+ "learning_rate": 4.860049561115424e-05,
98
+ "loss": 4.4799,
99
+ "step": 7500
100
+ },
101
+ {
102
+ "epoch": 0.03,
103
+ "learning_rate": 4.850719531856452e-05,
104
+ "loss": 4.4502,
105
+ "step": 8000
106
+ },
107
+ {
108
+ "epoch": 0.03,
109
+ "learning_rate": 4.84138950259748e-05,
110
+ "loss": 4.4183,
111
+ "step": 8500
112
+ },
113
+ {
114
+ "epoch": 0.03,
115
+ "learning_rate": 4.832059473338509e-05,
116
+ "loss": 4.3762,
117
+ "step": 9000
118
+ },
119
+ {
120
+ "epoch": 0.04,
121
+ "learning_rate": 4.8227294440795365e-05,
122
+ "loss": 4.3529,
123
+ "step": 9500
124
+ },
125
+ {
126
+ "epoch": 0.04,
127
+ "learning_rate": 4.8133994148205653e-05,
128
+ "loss": 4.3227,
129
+ "step": 10000
130
+ },
131
+ {
132
+ "epoch": 0.04,
133
+ "eval_accuracy": 0.3160632599821231,
134
+ "eval_loss": 4.226778984069824,
135
+ "eval_runtime": 6188.29,
136
+ "eval_samples_per_second": 72.909,
137
+ "eval_steps_per_second": 1.139,
138
+ "step": 10000
139
+ },
140
+ {
141
+ "epoch": 0.04,
142
+ "learning_rate": 4.804069385561593e-05,
143
+ "loss": 4.2976,
144
+ "step": 10500
145
+ },
146
+ {
147
+ "epoch": 0.04,
148
+ "learning_rate": 4.794739356302622e-05,
149
+ "loss": 4.2731,
150
+ "step": 11000
151
+ },
152
+ {
153
+ "epoch": 0.04,
154
+ "learning_rate": 4.78540932704365e-05,
155
+ "loss": 4.248,
156
+ "step": 11500
157
+ },
158
+ {
159
+ "epoch": 0.04,
160
+ "learning_rate": 4.776079297784678e-05,
161
+ "loss": 4.2336,
162
+ "step": 12000
163
+ },
164
+ {
165
+ "epoch": 0.05,
166
+ "learning_rate": 4.766749268525706e-05,
167
+ "loss": 4.2144,
168
+ "step": 12500
169
+ },
170
+ {
171
+ "epoch": 0.05,
172
+ "learning_rate": 4.7574192392667345e-05,
173
+ "loss": 4.1986,
174
+ "step": 13000
175
+ },
176
+ {
177
+ "epoch": 0.05,
178
+ "learning_rate": 4.7480892100077627e-05,
179
+ "loss": 4.1854,
180
+ "step": 13500
181
+ },
182
+ {
183
+ "epoch": 0.05,
184
+ "learning_rate": 4.7387591807487915e-05,
185
+ "loss": 4.1669,
186
+ "step": 14000
187
+ },
188
+ {
189
+ "epoch": 0.05,
190
+ "learning_rate": 4.729429151489819e-05,
191
+ "loss": 4.1441,
192
+ "step": 14500
193
+ },
194
+ {
195
+ "epoch": 0.06,
196
+ "learning_rate": 4.720099122230848e-05,
197
+ "loss": 4.1442,
198
+ "step": 15000
199
+ },
200
+ {
201
+ "epoch": 0.06,
202
+ "learning_rate": 4.7107690929718754e-05,
203
+ "loss": 4.1226,
204
+ "step": 15500
205
+ },
206
+ {
207
+ "epoch": 0.06,
208
+ "learning_rate": 4.701439063712904e-05,
209
+ "loss": 4.1112,
210
+ "step": 16000
211
+ },
212
+ {
213
+ "epoch": 0.06,
214
+ "learning_rate": 4.692109034453932e-05,
215
+ "loss": 4.1063,
216
+ "step": 16500
217
+ },
218
+ {
219
+ "epoch": 0.06,
220
+ "learning_rate": 4.6827790051949607e-05,
221
+ "loss": 4.0869,
222
+ "step": 17000
223
+ },
224
+ {
225
+ "epoch": 0.07,
226
+ "learning_rate": 4.673448975935989e-05,
227
+ "loss": 4.0732,
228
+ "step": 17500
229
+ },
230
+ {
231
+ "epoch": 0.07,
232
+ "learning_rate": 4.664118946677017e-05,
233
+ "loss": 4.0716,
234
+ "step": 18000
235
+ },
236
+ {
237
+ "epoch": 0.07,
238
+ "learning_rate": 4.654788917418045e-05,
239
+ "loss": 4.053,
240
+ "step": 18500
241
+ },
242
+ {
243
+ "epoch": 0.07,
244
+ "learning_rate": 4.6454588881590734e-05,
245
+ "loss": 4.0442,
246
+ "step": 19000
247
+ },
248
+ {
249
+ "epoch": 0.07,
250
+ "learning_rate": 4.6361288589001016e-05,
251
+ "loss": 4.035,
252
+ "step": 19500
253
+ },
254
+ {
255
+ "epoch": 0.07,
256
+ "learning_rate": 4.6267988296411305e-05,
257
+ "loss": 4.0305,
258
+ "step": 20000
259
+ },
260
+ {
261
+ "epoch": 0.07,
262
+ "eval_accuracy": 0.33925145807254103,
263
+ "eval_loss": 3.945496082305908,
264
+ "eval_runtime": 6149.9279,
265
+ "eval_samples_per_second": 73.364,
266
+ "eval_steps_per_second": 1.146,
267
+ "step": 20000
268
+ },
269
+ {
270
+ "epoch": 0.08,
271
+ "learning_rate": 4.617468800382158e-05,
272
+ "loss": 4.0218,
273
+ "step": 20500
274
+ },
275
+ {
276
+ "epoch": 0.08,
277
+ "learning_rate": 4.608138771123187e-05,
278
+ "loss": 4.0064,
279
+ "step": 21000
280
+ },
281
+ {
282
+ "epoch": 0.08,
283
+ "learning_rate": 4.5988087418642143e-05,
284
+ "loss": 4.0014,
285
+ "step": 21500
286
+ },
287
+ {
288
+ "epoch": 0.08,
289
+ "learning_rate": 4.589478712605243e-05,
290
+ "loss": 3.993,
291
+ "step": 22000
292
+ },
293
+ {
294
+ "epoch": 0.08,
295
+ "learning_rate": 4.5801486833462714e-05,
296
+ "loss": 3.9845,
297
+ "step": 22500
298
+ },
299
+ {
300
+ "epoch": 0.09,
301
+ "learning_rate": 4.5708186540872996e-05,
302
+ "loss": 3.9814,
303
+ "step": 23000
304
+ },
305
+ {
306
+ "epoch": 0.09,
307
+ "learning_rate": 4.561488624828328e-05,
308
+ "loss": 3.9689,
309
+ "step": 23500
310
+ },
311
+ {
312
+ "epoch": 0.09,
313
+ "learning_rate": 4.552158595569356e-05,
314
+ "loss": 3.9648,
315
+ "step": 24000
316
+ },
317
+ {
318
+ "epoch": 0.09,
319
+ "learning_rate": 4.542828566310384e-05,
320
+ "loss": 3.9599,
321
+ "step": 24500
322
+ },
323
+ {
324
+ "epoch": 0.09,
325
+ "learning_rate": 4.533498537051412e-05,
326
+ "loss": 3.9514,
327
+ "step": 25000
328
+ },
329
+ {
330
+ "epoch": 0.1,
331
+ "learning_rate": 4.5241685077924405e-05,
332
+ "loss": 3.9456,
333
+ "step": 25500
334
+ },
335
+ {
336
+ "epoch": 0.1,
337
+ "learning_rate": 4.5148384785334694e-05,
338
+ "loss": 3.9438,
339
+ "step": 26000
340
+ },
341
+ {
342
+ "epoch": 0.1,
343
+ "learning_rate": 4.505508449274497e-05,
344
+ "loss": 3.9315,
345
+ "step": 26500
346
+ },
347
+ {
348
+ "epoch": 0.1,
349
+ "learning_rate": 4.496178420015526e-05,
350
+ "loss": 3.9294,
351
+ "step": 27000
352
+ },
353
+ {
354
+ "epoch": 0.1,
355
+ "learning_rate": 4.486848390756553e-05,
356
+ "loss": 3.9137,
357
+ "step": 27500
358
+ },
359
+ {
360
+ "epoch": 0.1,
361
+ "learning_rate": 4.477518361497582e-05,
362
+ "loss": 3.9075,
363
+ "step": 28000
364
+ },
365
+ {
366
+ "epoch": 0.11,
367
+ "learning_rate": 4.46818833223861e-05,
368
+ "loss": 3.9069,
369
+ "step": 28500
370
+ },
371
+ {
372
+ "epoch": 0.11,
373
+ "learning_rate": 4.4588583029796385e-05,
374
+ "loss": 3.9039,
375
+ "step": 29000
376
+ },
377
+ {
378
+ "epoch": 0.11,
379
+ "learning_rate": 4.449528273720667e-05,
380
+ "loss": 3.9033,
381
+ "step": 29500
382
+ },
383
+ {
384
+ "epoch": 0.11,
385
+ "learning_rate": 4.440198244461695e-05,
386
+ "loss": 3.8916,
387
+ "step": 30000
388
+ },
389
+ {
390
+ "epoch": 0.11,
391
+ "eval_accuracy": 0.3501727973835296,
392
+ "eval_loss": 3.819415807723999,
393
+ "eval_runtime": 6177.9797,
394
+ "eval_samples_per_second": 73.031,
395
+ "eval_steps_per_second": 1.141,
396
+ "step": 30000
397
+ },
398
+ {
399
+ "epoch": 0.11,
400
+ "learning_rate": 4.430868215202723e-05,
401
+ "loss": 3.8876,
402
+ "step": 30500
403
+ },
404
+ {
405
+ "epoch": 0.12,
406
+ "learning_rate": 4.421538185943751e-05,
407
+ "loss": 3.8857,
408
+ "step": 31000
409
+ },
410
+ {
411
+ "epoch": 0.12,
412
+ "learning_rate": 4.4122081566847794e-05,
413
+ "loss": 3.8756,
414
+ "step": 31500
415
+ },
416
+ {
417
+ "epoch": 0.12,
418
+ "learning_rate": 4.4028781274258076e-05,
419
+ "loss": 3.871,
420
+ "step": 32000
421
+ },
422
+ {
423
+ "epoch": 0.12,
424
+ "learning_rate": 4.393548098166836e-05,
425
+ "loss": 3.8784,
426
+ "step": 32500
427
+ },
428
+ {
429
+ "epoch": 0.12,
430
+ "learning_rate": 4.384218068907865e-05,
431
+ "loss": 3.8657,
432
+ "step": 33000
433
+ },
434
+ {
435
+ "epoch": 0.13,
436
+ "learning_rate": 4.374888039648893e-05,
437
+ "loss": 3.8643,
438
+ "step": 33500
439
+ },
440
+ {
441
+ "epoch": 0.13,
442
+ "learning_rate": 4.365558010389921e-05,
443
+ "loss": 3.8518,
444
+ "step": 34000
445
+ },
446
+ {
447
+ "epoch": 0.13,
448
+ "learning_rate": 4.356227981130949e-05,
449
+ "loss": 3.8578,
450
+ "step": 34500
451
+ },
452
+ {
453
+ "epoch": 0.13,
454
+ "learning_rate": 4.3468979518719774e-05,
455
+ "loss": 3.8479,
456
+ "step": 35000
457
+ },
458
+ {
459
+ "epoch": 0.13,
460
+ "learning_rate": 4.3375679226130056e-05,
461
+ "loss": 3.846,
462
+ "step": 35500
463
+ },
464
+ {
465
+ "epoch": 0.13,
466
+ "learning_rate": 4.328237893354034e-05,
467
+ "loss": 3.8396,
468
+ "step": 36000
469
+ },
470
+ {
471
+ "epoch": 0.14,
472
+ "learning_rate": 4.318907864095062e-05,
473
+ "loss": 3.8401,
474
+ "step": 36500
475
+ },
476
+ {
477
+ "epoch": 0.14,
478
+ "learning_rate": 4.30957783483609e-05,
479
+ "loss": 3.8269,
480
+ "step": 37000
481
+ },
482
+ {
483
+ "epoch": 0.14,
484
+ "learning_rate": 4.3002478055771184e-05,
485
+ "loss": 3.8236,
486
+ "step": 37500
487
+ },
488
+ {
489
+ "epoch": 0.14,
490
+ "learning_rate": 4.2909177763181466e-05,
491
+ "loss": 3.8246,
492
+ "step": 38000
493
+ },
494
+ {
495
+ "epoch": 0.14,
496
+ "learning_rate": 4.281587747059175e-05,
497
+ "loss": 3.8236,
498
+ "step": 38500
499
+ },
500
+ {
501
+ "epoch": 0.15,
502
+ "learning_rate": 4.272257717800203e-05,
503
+ "loss": 3.8148,
504
+ "step": 39000
505
+ },
506
+ {
507
+ "epoch": 0.15,
508
+ "learning_rate": 4.262927688541232e-05,
509
+ "loss": 3.8033,
510
+ "step": 39500
511
+ },
512
+ {
513
+ "epoch": 0.15,
514
+ "learning_rate": 4.253597659282259e-05,
515
+ "loss": 3.8104,
516
+ "step": 40000
517
+ },
518
+ {
519
+ "epoch": 0.15,
520
+ "eval_accuracy": 0.3580380029125215,
521
+ "eval_loss": 3.7339699268341064,
522
+ "eval_runtime": 6179.879,
523
+ "eval_samples_per_second": 73.009,
524
+ "eval_steps_per_second": 1.141,
525
+ "step": 40000
526
+ },
527
+ {
528
+ "epoch": 0.15,
529
+ "learning_rate": 4.244267630023288e-05,
530
+ "loss": 3.809,
531
+ "step": 40500
532
+ },
533
+ {
534
+ "epoch": 0.15,
535
+ "learning_rate": 4.234937600764316e-05,
536
+ "loss": 3.8172,
537
+ "step": 41000
538
+ },
539
+ {
540
+ "epoch": 0.15,
541
+ "learning_rate": 4.2256075715053446e-05,
542
+ "loss": 3.8018,
543
+ "step": 41500
544
+ },
545
+ {
546
+ "epoch": 0.16,
547
+ "learning_rate": 4.216277542246373e-05,
548
+ "loss": 3.7956,
549
+ "step": 42000
550
+ },
551
+ {
552
+ "epoch": 0.16,
553
+ "learning_rate": 4.206947512987401e-05,
554
+ "loss": 3.7959,
555
+ "step": 42500
556
+ },
557
+ {
558
+ "epoch": 0.16,
559
+ "learning_rate": 4.197617483728429e-05,
560
+ "loss": 3.7884,
561
+ "step": 43000
562
+ },
563
+ {
564
+ "epoch": 0.16,
565
+ "learning_rate": 4.188287454469457e-05,
566
+ "loss": 3.788,
567
+ "step": 43500
568
+ },
569
+ {
570
+ "epoch": 0.16,
571
+ "learning_rate": 4.1789574252104855e-05,
572
+ "loss": 3.7795,
573
+ "step": 44000
574
+ },
575
+ {
576
+ "epoch": 0.17,
577
+ "learning_rate": 4.169627395951514e-05,
578
+ "loss": 3.783,
579
+ "step": 44500
580
+ },
581
+ {
582
+ "epoch": 0.17,
583
+ "learning_rate": 4.160297366692542e-05,
584
+ "loss": 3.7811,
585
+ "step": 45000
586
+ },
587
+ {
588
+ "epoch": 0.17,
589
+ "learning_rate": 4.150967337433571e-05,
590
+ "loss": 3.7804,
591
+ "step": 45500
592
+ },
593
+ {
594
+ "epoch": 0.17,
595
+ "learning_rate": 4.141637308174598e-05,
596
+ "loss": 3.7765,
597
+ "step": 46000
598
+ },
599
+ {
600
+ "epoch": 0.17,
601
+ "learning_rate": 4.132307278915627e-05,
602
+ "loss": 3.7656,
603
+ "step": 46500
604
+ },
605
+ {
606
+ "epoch": 0.18,
607
+ "learning_rate": 4.1229772496566546e-05,
608
+ "loss": 3.7627,
609
+ "step": 47000
610
+ },
611
+ {
612
+ "epoch": 0.18,
613
+ "learning_rate": 4.1136472203976835e-05,
614
+ "loss": 3.7595,
615
+ "step": 47500
616
+ },
617
+ {
618
+ "epoch": 0.18,
619
+ "learning_rate": 4.104317191138712e-05,
620
+ "loss": 3.7624,
621
+ "step": 48000
622
+ },
623
+ {
624
+ "epoch": 0.18,
625
+ "learning_rate": 4.09498716187974e-05,
626
+ "loss": 3.7571,
627
+ "step": 48500
628
+ },
629
+ {
630
+ "epoch": 0.18,
631
+ "learning_rate": 4.085657132620768e-05,
632
+ "loss": 3.7583,
633
+ "step": 49000
634
+ },
635
+ {
636
+ "epoch": 0.18,
637
+ "learning_rate": 4.076327103361796e-05,
638
+ "loss": 3.7583,
639
+ "step": 49500
640
+ },
641
+ {
642
+ "epoch": 0.19,
643
+ "learning_rate": 4.0669970741028244e-05,
644
+ "loss": 3.7491,
645
+ "step": 50000
646
+ },
647
+ {
648
+ "epoch": 0.19,
649
+ "eval_accuracy": 0.36325465564769804,
650
+ "eval_loss": 3.6769518852233887,
651
+ "eval_runtime": 6174.8828,
652
+ "eval_samples_per_second": 73.068,
653
+ "eval_steps_per_second": 1.142,
654
+ "step": 50000
655
+ },
656
+ {
657
+ "epoch": 0.19,
658
+ "learning_rate": 4.057667044843853e-05,
659
+ "loss": 3.7497,
660
+ "step": 50500
661
+ },
662
+ {
663
+ "epoch": 0.19,
664
+ "learning_rate": 4.048337015584881e-05,
665
+ "loss": 3.7401,
666
+ "step": 51000
667
+ },
668
+ {
669
+ "epoch": 0.19,
670
+ "learning_rate": 4.0390069863259097e-05,
671
+ "loss": 3.7423,
672
+ "step": 51500
673
+ },
674
+ {
675
+ "epoch": 0.19,
676
+ "learning_rate": 4.029676957066937e-05,
677
+ "loss": 3.7368,
678
+ "step": 52000
679
+ },
680
+ {
681
+ "epoch": 0.2,
682
+ "learning_rate": 4.020346927807966e-05,
683
+ "loss": 3.7358,
684
+ "step": 52500
685
+ },
686
+ {
687
+ "epoch": 0.2,
688
+ "learning_rate": 4.011016898548994e-05,
689
+ "loss": 3.7352,
690
+ "step": 53000
691
+ },
692
+ {
693
+ "epoch": 0.2,
694
+ "learning_rate": 4.0016868692900224e-05,
695
+ "loss": 3.7311,
696
+ "step": 53500
697
+ },
698
+ {
699
+ "epoch": 0.2,
700
+ "learning_rate": 3.9923568400310506e-05,
701
+ "loss": 3.7334,
702
+ "step": 54000
703
+ },
704
+ {
705
+ "epoch": 0.2,
706
+ "learning_rate": 3.983026810772079e-05,
707
+ "loss": 3.7347,
708
+ "step": 54500
709
+ },
710
+ {
711
+ "epoch": 0.21,
712
+ "learning_rate": 3.973696781513107e-05,
713
+ "loss": 3.7229,
714
+ "step": 55000
715
+ },
716
+ {
717
+ "epoch": 0.21,
718
+ "learning_rate": 3.964366752254135e-05,
719
+ "loss": 3.7265,
720
+ "step": 55500
721
+ },
722
+ {
723
+ "epoch": 0.21,
724
+ "learning_rate": 3.9550367229951633e-05,
725
+ "loss": 3.7286,
726
+ "step": 56000
727
+ },
728
+ {
729
+ "epoch": 0.21,
730
+ "learning_rate": 3.945706693736192e-05,
731
+ "loss": 3.7247,
732
+ "step": 56500
733
+ },
734
+ {
735
+ "epoch": 0.21,
736
+ "learning_rate": 3.93637666447722e-05,
737
+ "loss": 3.7186,
738
+ "step": 57000
739
+ },
740
+ {
741
+ "epoch": 0.21,
742
+ "learning_rate": 3.9270466352182486e-05,
743
+ "loss": 3.7152,
744
+ "step": 57500
745
+ },
746
+ {
747
+ "epoch": 0.22,
748
+ "learning_rate": 3.917716605959276e-05,
749
+ "loss": 3.713,
750
+ "step": 58000
751
+ },
752
+ {
753
+ "epoch": 0.22,
754
+ "learning_rate": 3.908386576700305e-05,
755
+ "loss": 3.7133,
756
+ "step": 58500
757
+ },
758
+ {
759
+ "epoch": 0.22,
760
+ "learning_rate": 3.899056547441333e-05,
761
+ "loss": 3.7076,
762
+ "step": 59000
763
+ },
764
+ {
765
+ "epoch": 0.22,
766
+ "learning_rate": 3.8897265181823613e-05,
767
+ "loss": 3.7082,
768
+ "step": 59500
769
+ },
770
+ {
771
+ "epoch": 0.22,
772
+ "learning_rate": 3.8803964889233895e-05,
773
+ "loss": 3.7062,
774
+ "step": 60000
775
+ },
776
+ {
777
+ "epoch": 0.22,
778
+ "eval_accuracy": 0.36794739205873755,
779
+ "eval_loss": 3.6288115978240967,
780
+ "eval_runtime": 6164.6808,
781
+ "eval_samples_per_second": 73.189,
782
+ "eval_steps_per_second": 1.144,
783
+ "step": 60000
784
+ },
785
+ {
786
+ "epoch": 0.23,
787
+ "learning_rate": 3.871066459664418e-05,
788
+ "loss": 3.7008,
789
+ "step": 60500
790
+ },
791
+ {
792
+ "epoch": 0.23,
793
+ "learning_rate": 3.861736430405446e-05,
794
+ "loss": 3.7013,
795
+ "step": 61000
796
+ },
797
+ {
798
+ "epoch": 0.23,
799
+ "learning_rate": 3.852406401146475e-05,
800
+ "loss": 3.6952,
801
+ "step": 61500
802
+ },
803
+ {
804
+ "epoch": 0.23,
805
+ "learning_rate": 3.843076371887502e-05,
806
+ "loss": 3.7065,
807
+ "step": 62000
808
+ },
809
+ {
810
+ "epoch": 0.23,
811
+ "learning_rate": 3.833746342628531e-05,
812
+ "loss": 3.692,
813
+ "step": 62500
814
+ },
815
+ {
816
+ "epoch": 0.24,
817
+ "learning_rate": 3.8244163133695587e-05,
818
+ "loss": 3.6942,
819
+ "step": 63000
820
+ },
821
+ {
822
+ "epoch": 0.24,
823
+ "learning_rate": 3.8150862841105875e-05,
824
+ "loss": 3.6906,
825
+ "step": 63500
826
+ },
827
+ {
828
+ "epoch": 0.24,
829
+ "learning_rate": 3.805756254851615e-05,
830
+ "loss": 3.6879,
831
+ "step": 64000
832
+ },
833
+ {
834
+ "epoch": 0.24,
835
+ "learning_rate": 3.796426225592644e-05,
836
+ "loss": 3.6885,
837
+ "step": 64500
838
+ },
839
+ {
840
+ "epoch": 0.24,
841
+ "learning_rate": 3.787096196333672e-05,
842
+ "loss": 3.6833,
843
+ "step": 65000
844
+ },
845
+ {
846
+ "epoch": 0.24,
847
+ "learning_rate": 3.7777661670747e-05,
848
+ "loss": 3.6834,
849
+ "step": 65500
850
+ },
851
+ {
852
+ "epoch": 0.25,
853
+ "learning_rate": 3.7684361378157285e-05,
854
+ "loss": 3.683,
855
+ "step": 66000
856
+ },
857
+ {
858
+ "epoch": 0.25,
859
+ "learning_rate": 3.7591061085567566e-05,
860
+ "loss": 3.6844,
861
+ "step": 66500
862
+ },
863
+ {
864
+ "epoch": 0.25,
865
+ "learning_rate": 3.749776079297785e-05,
866
+ "loss": 3.6726,
867
+ "step": 67000
868
+ },
869
+ {
870
+ "epoch": 0.25,
871
+ "learning_rate": 3.740446050038813e-05,
872
+ "loss": 3.6783,
873
+ "step": 67500
874
+ },
875
+ {
876
+ "epoch": 0.25,
877
+ "learning_rate": 3.731116020779841e-05,
878
+ "loss": 3.6768,
879
+ "step": 68000
880
+ },
881
+ {
882
+ "epoch": 0.26,
883
+ "learning_rate": 3.7217859915208694e-05,
884
+ "loss": 3.674,
885
+ "step": 68500
886
+ },
887
+ {
888
+ "epoch": 0.26,
889
+ "learning_rate": 3.7124559622618976e-05,
890
+ "loss": 3.6769,
891
+ "step": 69000
892
+ },
893
+ {
894
+ "epoch": 0.26,
895
+ "learning_rate": 3.703125933002926e-05,
896
+ "loss": 3.6695,
897
+ "step": 69500
898
+ },
899
+ {
900
+ "epoch": 0.26,
901
+ "learning_rate": 3.6937959037439546e-05,
902
+ "loss": 3.6724,
903
+ "step": 70000
904
+ },
905
+ {
906
+ "epoch": 0.26,
907
+ "eval_accuracy": 0.37135094159724996,
908
+ "eval_loss": 3.5937752723693848,
909
+ "eval_runtime": 6186.7336,
910
+ "eval_samples_per_second": 72.928,
911
+ "eval_steps_per_second": 1.14,
912
+ "step": 70000
913
+ },
914
+ {
915
+ "epoch": 0.26,
916
+ "learning_rate": 3.684465874484983e-05,
917
+ "loss": 3.6657,
918
+ "step": 70500
919
+ },
920
+ {
921
+ "epoch": 0.26,
922
+ "learning_rate": 3.675135845226011e-05,
923
+ "loss": 3.6618,
924
+ "step": 71000
925
+ },
926
+ {
927
+ "epoch": 0.27,
928
+ "learning_rate": 3.665805815967039e-05,
929
+ "loss": 3.6659,
930
+ "step": 71500
931
+ },
932
+ {
933
+ "epoch": 0.27,
934
+ "learning_rate": 3.6564757867080674e-05,
935
+ "loss": 3.6638,
936
+ "step": 72000
937
+ },
938
+ {
939
+ "epoch": 0.27,
940
+ "learning_rate": 3.6471457574490956e-05,
941
+ "loss": 3.6641,
942
+ "step": 72500
943
+ },
944
+ {
945
+ "epoch": 0.27,
946
+ "learning_rate": 3.637815728190124e-05,
947
+ "loss": 3.6576,
948
+ "step": 73000
949
+ },
950
+ {
951
+ "epoch": 0.27,
952
+ "learning_rate": 3.628485698931152e-05,
953
+ "loss": 3.6601,
954
+ "step": 73500
955
+ },
956
+ {
957
+ "epoch": 0.28,
958
+ "learning_rate": 3.61915566967218e-05,
959
+ "loss": 3.6542,
960
+ "step": 74000
961
+ },
962
+ {
963
+ "epoch": 0.28,
964
+ "learning_rate": 3.609825640413208e-05,
965
+ "loss": 3.6542,
966
+ "step": 74500
967
+ },
968
+ {
969
+ "epoch": 0.28,
970
+ "learning_rate": 3.6004956111542365e-05,
971
+ "loss": 3.658,
972
+ "step": 75000
973
+ },
974
+ {
975
+ "epoch": 0.28,
976
+ "learning_rate": 3.591165581895265e-05,
977
+ "loss": 3.6544,
978
+ "step": 75500
979
+ },
980
+ {
981
+ "epoch": 0.28,
982
+ "learning_rate": 3.5818355526362936e-05,
983
+ "loss": 3.6568,
984
+ "step": 76000
985
+ },
986
+ {
987
+ "epoch": 0.29,
988
+ "learning_rate": 3.572505523377321e-05,
989
+ "loss": 3.6491,
990
+ "step": 76500
991
+ },
992
+ {
993
+ "epoch": 0.29,
994
+ "learning_rate": 3.56317549411835e-05,
995
+ "loss": 3.6549,
996
+ "step": 77000
997
+ },
998
+ {
999
+ "epoch": 0.29,
1000
+ "learning_rate": 3.5538454648593774e-05,
1001
+ "loss": 3.6428,
1002
+ "step": 77500
1003
+ },
1004
+ {
1005
+ "epoch": 0.29,
1006
+ "learning_rate": 3.544515435600406e-05,
1007
+ "loss": 3.6446,
1008
+ "step": 78000
1009
+ },
1010
+ {
1011
+ "epoch": 0.29,
1012
+ "learning_rate": 3.5351854063414345e-05,
1013
+ "loss": 3.6421,
1014
+ "step": 78500
1015
+ },
1016
+ {
1017
+ "epoch": 0.29,
1018
+ "learning_rate": 3.525855377082463e-05,
1019
+ "loss": 3.6414,
1020
+ "step": 79000
1021
+ },
1022
+ {
1023
+ "epoch": 0.3,
1024
+ "learning_rate": 3.516525347823491e-05,
1025
+ "loss": 3.6386,
1026
+ "step": 79500
1027
+ },
1028
+ {
1029
+ "epoch": 0.3,
1030
+ "learning_rate": 3.507195318564519e-05,
1031
+ "loss": 3.6399,
1032
+ "step": 80000
1033
+ },
1034
+ {
1035
+ "epoch": 0.3,
1036
+ "eval_accuracy": 0.37431053802265524,
1037
+ "eval_loss": 3.5651655197143555,
1038
+ "eval_runtime": 6163.5869,
1039
+ "eval_samples_per_second": 73.202,
1040
+ "eval_steps_per_second": 1.144,
1041
+ "step": 80000
1042
+ },
1043
+ {
1044
+ "epoch": 0.3,
1045
+ "learning_rate": 3.497865289305547e-05,
1046
+ "loss": 3.6369,
1047
+ "step": 80500
1048
+ },
1049
+ {
1050
+ "epoch": 0.3,
1051
+ "learning_rate": 3.488535260046576e-05,
1052
+ "loss": 3.6385,
1053
+ "step": 81000
1054
+ },
1055
+ {
1056
+ "epoch": 0.3,
1057
+ "learning_rate": 3.4792052307876036e-05,
1058
+ "loss": 3.6342,
1059
+ "step": 81500
1060
+ },
1061
+ {
1062
+ "epoch": 0.31,
1063
+ "learning_rate": 3.4698752015286325e-05,
1064
+ "loss": 3.6351,
1065
+ "step": 82000
1066
+ },
1067
+ {
1068
+ "epoch": 0.31,
1069
+ "learning_rate": 3.46054517226966e-05,
1070
+ "loss": 3.6348,
1071
+ "step": 82500
1072
+ },
1073
+ {
1074
+ "epoch": 0.31,
1075
+ "learning_rate": 3.451215143010689e-05,
1076
+ "loss": 3.6289,
1077
+ "step": 83000
1078
+ },
1079
+ {
1080
+ "epoch": 0.31,
1081
+ "learning_rate": 3.4418851137517164e-05,
1082
+ "loss": 3.6314,
1083
+ "step": 83500
1084
+ },
1085
+ {
1086
+ "epoch": 0.31,
1087
+ "learning_rate": 3.432555084492745e-05,
1088
+ "loss": 3.6277,
1089
+ "step": 84000
1090
+ },
1091
+ {
1092
+ "epoch": 0.32,
1093
+ "learning_rate": 3.4232250552337734e-05,
1094
+ "loss": 3.6296,
1095
+ "step": 84500
1096
+ },
1097
+ {
1098
+ "epoch": 0.32,
1099
+ "learning_rate": 3.4138950259748016e-05,
1100
+ "loss": 3.6274,
1101
+ "step": 85000
1102
+ },
1103
+ {
1104
+ "epoch": 0.32,
1105
+ "learning_rate": 3.40456499671583e-05,
1106
+ "loss": 3.62,
1107
+ "step": 85500
1108
+ },
1109
+ {
1110
+ "epoch": 0.32,
1111
+ "learning_rate": 3.395234967456858e-05,
1112
+ "loss": 3.6217,
1113
+ "step": 86000
1114
+ },
1115
+ {
1116
+ "epoch": 0.32,
1117
+ "learning_rate": 3.385904938197886e-05,
1118
+ "loss": 3.6205,
1119
+ "step": 86500
1120
+ },
1121
+ {
1122
+ "epoch": 0.32,
1123
+ "learning_rate": 3.376574908938915e-05,
1124
+ "loss": 3.6182,
1125
+ "step": 87000
1126
+ },
1127
+ {
1128
+ "epoch": 0.33,
1129
+ "learning_rate": 3.3672448796799426e-05,
1130
+ "loss": 3.6196,
1131
+ "step": 87500
1132
+ },
1133
+ {
1134
+ "epoch": 0.33,
1135
+ "learning_rate": 3.3579148504209714e-05,
1136
+ "loss": 3.6154,
1137
+ "step": 88000
1138
+ },
1139
+ {
1140
+ "epoch": 0.33,
1141
+ "learning_rate": 3.348584821161999e-05,
1142
+ "loss": 3.6242,
1143
+ "step": 88500
1144
+ },
1145
+ {
1146
+ "epoch": 0.33,
1147
+ "learning_rate": 3.339254791903028e-05,
1148
+ "loss": 3.6162,
1149
+ "step": 89000
1150
+ },
1151
+ {
1152
+ "epoch": 0.33,
1153
+ "learning_rate": 3.329924762644056e-05,
1154
+ "loss": 3.6205,
1155
+ "step": 89500
1156
+ },
1157
+ {
1158
+ "epoch": 0.34,
1159
+ "learning_rate": 3.320594733385084e-05,
1160
+ "loss": 3.6147,
1161
+ "step": 90000
1162
+ },
1163
+ {
1164
+ "epoch": 0.34,
1165
+ "eval_accuracy": 0.3768022387359209,
1166
+ "eval_loss": 3.539621591567993,
1167
+ "eval_runtime": 6175.1379,
1168
+ "eval_samples_per_second": 73.065,
1169
+ "eval_steps_per_second": 1.142,
1170
+ "step": 90000
1171
+ },
1172
+ {
1173
+ "epoch": 0.34,
1174
+ "learning_rate": 3.3112647041261124e-05,
1175
+ "loss": 3.6185,
1176
+ "step": 90500
1177
+ },
1178
+ {
1179
+ "epoch": 0.34,
1180
+ "learning_rate": 3.3019346748671405e-05,
1181
+ "loss": 3.6125,
1182
+ "step": 91000
1183
+ },
1184
+ {
1185
+ "epoch": 0.34,
1186
+ "learning_rate": 3.292604645608169e-05,
1187
+ "loss": 3.6112,
1188
+ "step": 91500
1189
+ },
1190
+ {
1191
+ "epoch": 0.34,
1192
+ "learning_rate": 3.283274616349197e-05,
1193
+ "loss": 3.618,
1194
+ "step": 92000
1195
+ },
1196
+ {
1197
+ "epoch": 0.35,
1198
+ "learning_rate": 3.273944587090225e-05,
1199
+ "loss": 3.6135,
1200
+ "step": 92500
1201
+ },
1202
+ {
1203
+ "epoch": 0.35,
1204
+ "learning_rate": 3.264614557831254e-05,
1205
+ "loss": 3.6084,
1206
+ "step": 93000
1207
+ },
1208
+ {
1209
+ "epoch": 0.35,
1210
+ "learning_rate": 3.2552845285722815e-05,
1211
+ "loss": 3.606,
1212
+ "step": 93500
1213
+ },
1214
+ {
1215
+ "epoch": 0.35,
1216
+ "learning_rate": 3.2459544993133103e-05,
1217
+ "loss": 3.6125,
1218
+ "step": 94000
1219
+ },
1220
+ {
1221
+ "epoch": 0.35,
1222
+ "learning_rate": 3.236624470054338e-05,
1223
+ "loss": 3.6065,
1224
+ "step": 94500
1225
+ },
1226
+ {
1227
+ "epoch": 0.35,
1228
+ "learning_rate": 3.227294440795367e-05,
1229
+ "loss": 3.6035,
1230
+ "step": 95000
1231
+ },
1232
+ {
1233
+ "epoch": 0.36,
1234
+ "learning_rate": 3.217964411536395e-05,
1235
+ "loss": 3.5971,
1236
+ "step": 95500
1237
+ },
1238
+ {
1239
+ "epoch": 0.36,
1240
+ "learning_rate": 3.208634382277423e-05,
1241
+ "loss": 3.6011,
1242
+ "step": 96000
1243
+ },
1244
+ {
1245
+ "epoch": 0.36,
1246
+ "learning_rate": 3.199304353018451e-05,
1247
+ "loss": 3.601,
1248
+ "step": 96500
1249
+ },
1250
+ {
1251
+ "epoch": 0.36,
1252
+ "learning_rate": 3.1899743237594795e-05,
1253
+ "loss": 3.5967,
1254
+ "step": 97000
1255
+ },
1256
+ {
1257
+ "epoch": 0.36,
1258
+ "learning_rate": 3.180644294500508e-05,
1259
+ "loss": 3.6038,
1260
+ "step": 97500
1261
+ },
1262
+ {
1263
+ "epoch": 0.37,
1264
+ "learning_rate": 3.171314265241536e-05,
1265
+ "loss": 3.599,
1266
+ "step": 98000
1267
+ },
1268
+ {
1269
+ "epoch": 0.37,
1270
+ "learning_rate": 3.161984235982564e-05,
1271
+ "loss": 3.593,
1272
+ "step": 98500
1273
+ },
1274
+ {
1275
+ "epoch": 0.37,
1276
+ "learning_rate": 3.152654206723593e-05,
1277
+ "loss": 3.6019,
1278
+ "step": 99000
1279
+ },
1280
+ {
1281
+ "epoch": 0.37,
1282
+ "learning_rate": 3.1433241774646204e-05,
1283
+ "loss": 3.5927,
1284
+ "step": 99500
1285
+ },
1286
+ {
1287
+ "epoch": 0.37,
1288
+ "learning_rate": 3.133994148205649e-05,
1289
+ "loss": 3.5946,
1290
+ "step": 100000
1291
+ },
1292
+ {
1293
+ "epoch": 0.37,
1294
+ "eval_accuracy": 0.3791379626568468,
1295
+ "eval_loss": 3.515845775604248,
1296
+ "eval_runtime": 6163.9176,
1297
+ "eval_samples_per_second": 73.198,
1298
+ "eval_steps_per_second": 1.144,
1299
+ "step": 100000
1300
+ },
1301
+ {
1302
+ "epoch": 0.38,
1303
+ "learning_rate": 3.1246641189466775e-05,
1304
+ "loss": 3.5983,
1305
+ "step": 100500
1306
+ },
1307
+ {
1308
+ "epoch": 0.38,
1309
+ "learning_rate": 3.1153340896877057e-05,
1310
+ "loss": 3.5947,
1311
+ "step": 101000
1312
+ },
1313
+ {
1314
+ "epoch": 0.38,
1315
+ "learning_rate": 3.106004060428734e-05,
1316
+ "loss": 3.5877,
1317
+ "step": 101500
1318
+ },
1319
+ {
1320
+ "epoch": 0.38,
1321
+ "learning_rate": 3.096674031169762e-05,
1322
+ "loss": 3.5896,
1323
+ "step": 102000
1324
+ },
1325
+ {
1326
+ "epoch": 0.38,
1327
+ "learning_rate": 3.08734400191079e-05,
1328
+ "loss": 3.5859,
1329
+ "step": 102500
1330
+ },
1331
+ {
1332
+ "epoch": 0.38,
1333
+ "learning_rate": 3.0780139726518184e-05,
1334
+ "loss": 3.5884,
1335
+ "step": 103000
1336
+ },
1337
+ {
1338
+ "epoch": 0.39,
1339
+ "learning_rate": 3.0686839433928466e-05,
1340
+ "loss": 3.5824,
1341
+ "step": 103500
1342
+ },
1343
+ {
1344
+ "epoch": 0.39,
1345
+ "learning_rate": 3.059353914133875e-05,
1346
+ "loss": 3.582,
1347
+ "step": 104000
1348
+ },
1349
+ {
1350
+ "epoch": 0.39,
1351
+ "learning_rate": 3.050023884874903e-05,
1352
+ "loss": 3.5859,
1353
+ "step": 104500
1354
+ },
1355
+ {
1356
+ "epoch": 0.39,
1357
+ "learning_rate": 3.0406938556159315e-05,
1358
+ "loss": 3.5828,
1359
+ "step": 105000
1360
+ },
1361
+ {
1362
+ "epoch": 0.39,
1363
+ "learning_rate": 3.0313638263569593e-05,
1364
+ "loss": 3.5827,
1365
+ "step": 105500
1366
+ },
1367
+ {
1368
+ "epoch": 0.4,
1369
+ "learning_rate": 3.022033797097988e-05,
1370
+ "loss": 3.58,
1371
+ "step": 106000
1372
+ },
1373
+ {
1374
+ "epoch": 0.4,
1375
+ "learning_rate": 3.0127037678390164e-05,
1376
+ "loss": 3.5793,
1377
+ "step": 106500
1378
+ },
1379
+ {
1380
+ "epoch": 0.4,
1381
+ "learning_rate": 3.0033737385800442e-05,
1382
+ "loss": 3.5823,
1383
+ "step": 107000
1384
+ },
1385
+ {
1386
+ "epoch": 0.4,
1387
+ "learning_rate": 2.9940437093210728e-05,
1388
+ "loss": 3.584,
1389
+ "step": 107500
1390
+ },
1391
+ {
1392
+ "epoch": 0.4,
1393
+ "learning_rate": 2.9847136800621006e-05,
1394
+ "loss": 3.5731,
1395
+ "step": 108000
1396
+ },
1397
+ {
1398
+ "epoch": 0.4,
1399
+ "learning_rate": 2.975383650803129e-05,
1400
+ "loss": 3.5804,
1401
+ "step": 108500
1402
+ },
1403
+ {
1404
+ "epoch": 0.41,
1405
+ "learning_rate": 2.9660536215441577e-05,
1406
+ "loss": 3.5809,
1407
+ "step": 109000
1408
+ },
1409
+ {
1410
+ "epoch": 0.41,
1411
+ "learning_rate": 2.9567235922851855e-05,
1412
+ "loss": 3.5782,
1413
+ "step": 109500
1414
+ },
1415
+ {
1416
+ "epoch": 0.41,
1417
+ "learning_rate": 2.947393563026214e-05,
1418
+ "loss": 3.5726,
1419
+ "step": 110000
1420
+ },
1421
+ {
1422
+ "epoch": 0.41,
1423
+ "eval_accuracy": 0.3808981412485733,
1424
+ "eval_loss": 3.4986438751220703,
1425
+ "eval_runtime": 6162.0171,
1426
+ "eval_samples_per_second": 73.22,
1427
+ "eval_steps_per_second": 1.144,
1428
+ "step": 110000
1429
+ },
1430
+ {
1431
+ "epoch": 0.41,
1432
+ "learning_rate": 2.938063533767242e-05,
1433
+ "loss": 3.571,
1434
+ "step": 110500
1435
+ },
1436
+ {
1437
+ "epoch": 0.41,
1438
+ "learning_rate": 2.9287335045082704e-05,
1439
+ "loss": 3.5747,
1440
+ "step": 111000
1441
+ },
1442
+ {
1443
+ "epoch": 0.42,
1444
+ "learning_rate": 2.9194034752492983e-05,
1445
+ "loss": 3.5711,
1446
+ "step": 111500
1447
+ },
1448
+ {
1449
+ "epoch": 0.42,
1450
+ "learning_rate": 2.9100734459903268e-05,
1451
+ "loss": 3.574,
1452
+ "step": 112000
1453
+ },
1454
+ {
1455
+ "epoch": 0.42,
1456
+ "learning_rate": 2.9007434167313553e-05,
1457
+ "loss": 3.5739,
1458
+ "step": 112500
1459
+ },
1460
+ {
1461
+ "epoch": 0.42,
1462
+ "learning_rate": 2.8914133874723832e-05,
1463
+ "loss": 3.5682,
1464
+ "step": 113000
1465
+ },
1466
+ {
1467
+ "epoch": 0.42,
1468
+ "learning_rate": 2.8820833582134117e-05,
1469
+ "loss": 3.5678,
1470
+ "step": 113500
1471
+ },
1472
+ {
1473
+ "epoch": 0.43,
1474
+ "learning_rate": 2.8727533289544395e-05,
1475
+ "loss": 3.5692,
1476
+ "step": 114000
1477
+ },
1478
+ {
1479
+ "epoch": 0.43,
1480
+ "learning_rate": 2.863423299695468e-05,
1481
+ "loss": 3.5606,
1482
+ "step": 114500
1483
+ },
1484
+ {
1485
+ "epoch": 0.43,
1486
+ "learning_rate": 2.8540932704364966e-05,
1487
+ "loss": 3.5713,
1488
+ "step": 115000
1489
+ },
1490
+ {
1491
+ "epoch": 0.43,
1492
+ "learning_rate": 2.8447632411775244e-05,
1493
+ "loss": 3.5642,
1494
+ "step": 115500
1495
+ },
1496
+ {
1497
+ "epoch": 0.43,
1498
+ "learning_rate": 2.835433211918553e-05,
1499
+ "loss": 3.5661,
1500
+ "step": 116000
1501
+ },
1502
+ {
1503
+ "epoch": 0.43,
1504
+ "learning_rate": 2.8261031826595808e-05,
1505
+ "loss": 3.5661,
1506
+ "step": 116500
1507
+ },
1508
+ {
1509
+ "epoch": 0.44,
1510
+ "learning_rate": 2.8167731534006093e-05,
1511
+ "loss": 3.5653,
1512
+ "step": 117000
1513
+ },
1514
+ {
1515
+ "epoch": 0.44,
1516
+ "learning_rate": 2.8074431241416375e-05,
1517
+ "loss": 3.5709,
1518
+ "step": 117500
1519
+ },
1520
+ {
1521
+ "epoch": 0.44,
1522
+ "learning_rate": 2.7981130948826657e-05,
1523
+ "loss": 3.5604,
1524
+ "step": 118000
1525
+ },
1526
+ {
1527
+ "epoch": 0.44,
1528
+ "learning_rate": 2.788783065623694e-05,
1529
+ "loss": 3.5622,
1530
+ "step": 118500
1531
+ },
1532
+ {
1533
+ "epoch": 0.44,
1534
+ "learning_rate": 2.779453036364722e-05,
1535
+ "loss": 3.5606,
1536
+ "step": 119000
1537
+ },
1538
+ {
1539
+ "epoch": 0.45,
1540
+ "learning_rate": 2.7701230071057503e-05,
1541
+ "loss": 3.5591,
1542
+ "step": 119500
1543
+ },
1544
+ {
1545
+ "epoch": 0.45,
1546
+ "learning_rate": 2.7607929778467788e-05,
1547
+ "loss": 3.5631,
1548
+ "step": 120000
1549
+ },
1550
+ {
1551
+ "epoch": 0.45,
1552
+ "eval_accuracy": 0.38260092346589303,
1553
+ "eval_loss": 3.4819467067718506,
1554
+ "eval_runtime": 6173.7982,
1555
+ "eval_samples_per_second": 73.081,
1556
+ "eval_steps_per_second": 1.142,
1557
+ "step": 120000
1558
+ },
1559
+ {
1560
+ "epoch": 0.45,
1561
+ "learning_rate": 2.751462948587807e-05,
1562
+ "loss": 3.5558,
1563
+ "step": 120500
1564
+ },
1565
+ {
1566
+ "epoch": 0.45,
1567
+ "learning_rate": 2.7421329193288352e-05,
1568
+ "loss": 3.5569,
1569
+ "step": 121000
1570
+ },
1571
+ {
1572
+ "epoch": 0.45,
1573
+ "learning_rate": 2.7328028900698634e-05,
1574
+ "loss": 3.559,
1575
+ "step": 121500
1576
+ },
1577
+ {
1578
+ "epoch": 0.46,
1579
+ "learning_rate": 2.7234728608108916e-05,
1580
+ "loss": 3.5596,
1581
+ "step": 122000
1582
+ },
1583
+ {
1584
+ "epoch": 0.46,
1585
+ "learning_rate": 2.7141428315519198e-05,
1586
+ "loss": 3.5582,
1587
+ "step": 122500
1588
+ },
1589
+ {
1590
+ "epoch": 0.46,
1591
+ "learning_rate": 2.704812802292948e-05,
1592
+ "loss": 3.553,
1593
+ "step": 123000
1594
+ },
1595
+ {
1596
+ "epoch": 0.46,
1597
+ "learning_rate": 2.6954827730339765e-05,
1598
+ "loss": 3.5566,
1599
+ "step": 123500
1600
+ },
1601
+ {
1602
+ "epoch": 0.46,
1603
+ "learning_rate": 2.6861527437750043e-05,
1604
+ "loss": 3.555,
1605
+ "step": 124000
1606
+ },
1607
+ {
1608
+ "epoch": 0.46,
1609
+ "learning_rate": 2.676822714516033e-05,
1610
+ "loss": 3.551,
1611
+ "step": 124500
1612
+ },
1613
+ {
1614
+ "epoch": 0.47,
1615
+ "learning_rate": 2.667492685257061e-05,
1616
+ "loss": 3.5519,
1617
+ "step": 125000
1618
+ },
1619
+ {
1620
+ "epoch": 0.47,
1621
+ "learning_rate": 2.6581626559980892e-05,
1622
+ "loss": 3.5491,
1623
+ "step": 125500
1624
+ },
1625
+ {
1626
+ "epoch": 0.47,
1627
+ "learning_rate": 2.6488326267391177e-05,
1628
+ "loss": 3.5511,
1629
+ "step": 126000
1630
+ },
1631
+ {
1632
+ "epoch": 0.47,
1633
+ "learning_rate": 2.6395025974801456e-05,
1634
+ "loss": 3.5442,
1635
+ "step": 126500
1636
+ },
1637
+ {
1638
+ "epoch": 0.47,
1639
+ "learning_rate": 2.630172568221174e-05,
1640
+ "loss": 3.5535,
1641
+ "step": 127000
1642
+ },
1643
+ {
1644
+ "epoch": 0.48,
1645
+ "learning_rate": 2.620842538962202e-05,
1646
+ "loss": 3.5475,
1647
+ "step": 127500
1648
+ },
1649
+ {
1650
+ "epoch": 0.48,
1651
+ "learning_rate": 2.6115125097032305e-05,
1652
+ "loss": 3.5421,
1653
+ "step": 128000
1654
+ },
1655
+ {
1656
+ "epoch": 0.48,
1657
+ "learning_rate": 2.602182480444259e-05,
1658
+ "loss": 3.5486,
1659
+ "step": 128500
1660
+ },
1661
+ {
1662
+ "epoch": 0.48,
1663
+ "learning_rate": 2.592852451185287e-05,
1664
+ "loss": 3.546,
1665
+ "step": 129000
1666
+ },
1667
+ {
1668
+ "epoch": 0.48,
1669
+ "learning_rate": 2.5835224219263154e-05,
1670
+ "loss": 3.5397,
1671
+ "step": 129500
1672
+ },
1673
+ {
1674
+ "epoch": 0.49,
1675
+ "learning_rate": 2.5741923926673432e-05,
1676
+ "loss": 3.5459,
1677
+ "step": 130000
1678
+ },
1679
+ {
1680
+ "epoch": 0.49,
1681
+ "eval_accuracy": 0.384199481389569,
1682
+ "eval_loss": 3.467806816101074,
1683
+ "eval_runtime": 6189.2425,
1684
+ "eval_samples_per_second": 72.898,
1685
+ "eval_steps_per_second": 1.139,
1686
+ "step": 130000
1687
+ },
1688
+ {
1689
+ "epoch": 0.49,
1690
+ "learning_rate": 2.5648623634083718e-05,
1691
+ "loss": 3.5437,
1692
+ "step": 130500
1693
+ },
1694
+ {
1695
+ "epoch": 0.49,
1696
+ "learning_rate": 2.5555323341493996e-05,
1697
+ "loss": 3.5432,
1698
+ "step": 131000
1699
+ },
1700
+ {
1701
+ "epoch": 0.49,
1702
+ "learning_rate": 2.546202304890428e-05,
1703
+ "loss": 3.5416,
1704
+ "step": 131500
1705
+ },
1706
+ {
1707
+ "epoch": 0.49,
1708
+ "learning_rate": 2.5368722756314567e-05,
1709
+ "loss": 3.5455,
1710
+ "step": 132000
1711
+ },
1712
+ {
1713
+ "epoch": 0.49,
1714
+ "learning_rate": 2.5275422463724845e-05,
1715
+ "loss": 3.5381,
1716
+ "step": 132500
1717
+ },
1718
+ {
1719
+ "epoch": 0.5,
1720
+ "learning_rate": 2.518212217113513e-05,
1721
+ "loss": 3.541,
1722
+ "step": 133000
1723
+ },
1724
+ {
1725
+ "epoch": 0.5,
1726
+ "learning_rate": 2.508882187854541e-05,
1727
+ "loss": 3.5375,
1728
+ "step": 133500
1729
+ },
1730
+ {
1731
+ "epoch": 0.5,
1732
+ "learning_rate": 2.4995521585955694e-05,
1733
+ "loss": 3.5389,
1734
+ "step": 134000
1735
+ },
1736
+ {
1737
+ "epoch": 0.5,
1738
+ "learning_rate": 2.4902221293365976e-05,
1739
+ "loss": 3.5375,
1740
+ "step": 134500
1741
+ },
1742
+ {
1743
+ "epoch": 0.5,
1744
+ "learning_rate": 2.4808921000776258e-05,
1745
+ "loss": 3.5344,
1746
+ "step": 135000
1747
+ },
1748
+ {
1749
+ "epoch": 0.51,
1750
+ "learning_rate": 2.4715620708186543e-05,
1751
+ "loss": 3.5342,
1752
+ "step": 135500
1753
+ },
1754
+ {
1755
+ "epoch": 0.51,
1756
+ "learning_rate": 2.4622320415596825e-05,
1757
+ "loss": 3.5444,
1758
+ "step": 136000
1759
+ },
1760
+ {
1761
+ "epoch": 0.51,
1762
+ "learning_rate": 2.4529020123007107e-05,
1763
+ "loss": 3.5318,
1764
+ "step": 136500
1765
+ },
1766
+ {
1767
+ "epoch": 0.51,
1768
+ "learning_rate": 2.443571983041739e-05,
1769
+ "loss": 3.5402,
1770
+ "step": 137000
1771
+ },
1772
+ {
1773
+ "epoch": 0.51,
1774
+ "learning_rate": 2.434241953782767e-05,
1775
+ "loss": 3.5408,
1776
+ "step": 137500
1777
+ },
1778
+ {
1779
+ "epoch": 0.52,
1780
+ "learning_rate": 2.4249119245237953e-05,
1781
+ "loss": 3.5314,
1782
+ "step": 138000
1783
+ },
1784
+ {
1785
+ "epoch": 0.52,
1786
+ "learning_rate": 2.4155818952648238e-05,
1787
+ "loss": 3.5276,
1788
+ "step": 138500
1789
+ },
1790
+ {
1791
+ "epoch": 0.52,
1792
+ "learning_rate": 2.406251866005852e-05,
1793
+ "loss": 3.5309,
1794
+ "step": 139000
1795
+ },
1796
+ {
1797
+ "epoch": 0.52,
1798
+ "learning_rate": 2.39692183674688e-05,
1799
+ "loss": 3.533,
1800
+ "step": 139500
1801
+ },
1802
+ {
1803
+ "epoch": 0.52,
1804
+ "learning_rate": 2.3875918074879084e-05,
1805
+ "loss": 3.5304,
1806
+ "step": 140000
1807
+ },
1808
+ {
1809
+ "epoch": 0.52,
1810
+ "eval_accuracy": 0.3857324143630419,
1811
+ "eval_loss": 3.4535393714904785,
1812
+ "eval_runtime": 6160.8797,
1813
+ "eval_samples_per_second": 73.234,
1814
+ "eval_steps_per_second": 1.144,
1815
+ "step": 140000
1816
+ },
1817
+ {
1818
+ "epoch": 0.52,
1819
+ "learning_rate": 2.3782617782289365e-05,
1820
+ "loss": 3.536,
1821
+ "step": 140500
1822
+ },
1823
+ {
1824
+ "epoch": 0.53,
1825
+ "learning_rate": 2.368931748969965e-05,
1826
+ "loss": 3.5374,
1827
+ "step": 141000
1828
+ },
1829
+ {
1830
+ "epoch": 0.53,
1831
+ "learning_rate": 2.3596017197109933e-05,
1832
+ "loss": 3.5279,
1833
+ "step": 141500
1834
+ },
1835
+ {
1836
+ "epoch": 0.53,
1837
+ "learning_rate": 2.3502716904520214e-05,
1838
+ "loss": 3.5334,
1839
+ "step": 142000
1840
+ },
1841
+ {
1842
+ "epoch": 0.53,
1843
+ "learning_rate": 2.3409416611930496e-05,
1844
+ "loss": 3.531,
1845
+ "step": 142500
1846
+ },
1847
+ {
1848
+ "epoch": 0.53,
1849
+ "learning_rate": 2.3316116319340778e-05,
1850
+ "loss": 3.5285,
1851
+ "step": 143000
1852
+ },
1853
+ {
1854
+ "epoch": 0.54,
1855
+ "learning_rate": 2.322281602675106e-05,
1856
+ "loss": 3.5266,
1857
+ "step": 143500
1858
+ },
1859
+ {
1860
+ "epoch": 0.54,
1861
+ "learning_rate": 2.3129515734161345e-05,
1862
+ "loss": 3.5307,
1863
+ "step": 144000
1864
+ },
1865
+ {
1866
+ "epoch": 0.54,
1867
+ "learning_rate": 2.3036215441571627e-05,
1868
+ "loss": 3.534,
1869
+ "step": 144500
1870
+ },
1871
+ {
1872
+ "epoch": 0.54,
1873
+ "learning_rate": 2.294291514898191e-05,
1874
+ "loss": 3.5278,
1875
+ "step": 145000
1876
+ },
1877
+ {
1878
+ "epoch": 0.54,
1879
+ "learning_rate": 2.284961485639219e-05,
1880
+ "loss": 3.522,
1881
+ "step": 145500
1882
+ },
1883
+ {
1884
+ "epoch": 0.54,
1885
+ "learning_rate": 2.2756314563802473e-05,
1886
+ "loss": 3.5234,
1887
+ "step": 146000
1888
+ },
1889
+ {
1890
+ "epoch": 0.55,
1891
+ "learning_rate": 2.2663014271212758e-05,
1892
+ "loss": 3.5258,
1893
+ "step": 146500
1894
+ },
1895
+ {
1896
+ "epoch": 0.55,
1897
+ "learning_rate": 2.256971397862304e-05,
1898
+ "loss": 3.5226,
1899
+ "step": 147000
1900
+ },
1901
+ {
1902
+ "epoch": 0.55,
1903
+ "learning_rate": 2.2476413686033322e-05,
1904
+ "loss": 3.5204,
1905
+ "step": 147500
1906
+ },
1907
+ {
1908
+ "epoch": 0.55,
1909
+ "learning_rate": 2.2383113393443604e-05,
1910
+ "loss": 3.5237,
1911
+ "step": 148000
1912
+ },
1913
+ {
1914
+ "epoch": 0.55,
1915
+ "learning_rate": 2.2289813100853886e-05,
1916
+ "loss": 3.5225,
1917
+ "step": 148500
1918
+ },
1919
+ {
1920
+ "epoch": 0.56,
1921
+ "learning_rate": 2.2196512808264167e-05,
1922
+ "loss": 3.5218,
1923
+ "step": 149000
1924
+ },
1925
+ {
1926
+ "epoch": 0.56,
1927
+ "learning_rate": 2.2103212515674453e-05,
1928
+ "loss": 3.5269,
1929
+ "step": 149500
1930
+ },
1931
+ {
1932
+ "epoch": 0.56,
1933
+ "learning_rate": 2.2009912223084735e-05,
1934
+ "loss": 3.5245,
1935
+ "step": 150000
1936
+ },
1937
+ {
1938
+ "epoch": 0.56,
1939
+ "eval_accuracy": 0.3867262152967859,
1940
+ "eval_loss": 3.442957639694214,
1941
+ "eval_runtime": 6174.4314,
1942
+ "eval_samples_per_second": 73.073,
1943
+ "eval_steps_per_second": 1.142,
1944
+ "step": 150000
1945
+ },
1946
+ {
1947
+ "epoch": 0.56,
1948
+ "learning_rate": 2.1916611930495016e-05,
1949
+ "loss": 3.5227,
1950
+ "step": 150500
1951
+ },
1952
+ {
1953
+ "epoch": 0.56,
1954
+ "learning_rate": 2.18233116379053e-05,
1955
+ "loss": 3.5117,
1956
+ "step": 151000
1957
+ },
1958
+ {
1959
+ "epoch": 0.57,
1960
+ "learning_rate": 2.173001134531558e-05,
1961
+ "loss": 3.5203,
1962
+ "step": 151500
1963
+ },
1964
+ {
1965
+ "epoch": 0.57,
1966
+ "learning_rate": 2.1636711052725862e-05,
1967
+ "loss": 3.5177,
1968
+ "step": 152000
1969
+ },
1970
+ {
1971
+ "epoch": 0.57,
1972
+ "learning_rate": 2.1543410760136147e-05,
1973
+ "loss": 3.5175,
1974
+ "step": 152500
1975
+ },
1976
+ {
1977
+ "epoch": 0.57,
1978
+ "learning_rate": 2.145011046754643e-05,
1979
+ "loss": 3.5186,
1980
+ "step": 153000
1981
+ },
1982
+ {
1983
+ "epoch": 0.57,
1984
+ "learning_rate": 2.135681017495671e-05,
1985
+ "loss": 3.5208,
1986
+ "step": 153500
1987
+ },
1988
+ {
1989
+ "epoch": 0.57,
1990
+ "learning_rate": 2.1263509882366993e-05,
1991
+ "loss": 3.5199,
1992
+ "step": 154000
1993
+ },
1994
+ {
1995
+ "epoch": 0.58,
1996
+ "learning_rate": 2.1170209589777275e-05,
1997
+ "loss": 3.5188,
1998
+ "step": 154500
1999
+ },
2000
+ {
2001
+ "epoch": 0.58,
2002
+ "learning_rate": 2.1076909297187557e-05,
2003
+ "loss": 3.5235,
2004
+ "step": 155000
2005
+ },
2006
+ {
2007
+ "epoch": 0.58,
2008
+ "learning_rate": 2.098360900459784e-05,
2009
+ "loss": 3.516,
2010
+ "step": 155500
2011
+ },
2012
+ {
2013
+ "epoch": 0.58,
2014
+ "learning_rate": 2.089030871200812e-05,
2015
+ "loss": 3.5124,
2016
+ "step": 156000
2017
+ },
2018
+ {
2019
+ "epoch": 0.58,
2020
+ "learning_rate": 2.0797008419418402e-05,
2021
+ "loss": 3.5145,
2022
+ "step": 156500
2023
+ },
2024
+ {
2025
+ "epoch": 0.59,
2026
+ "learning_rate": 2.0703708126828684e-05,
2027
+ "loss": 3.5092,
2028
+ "step": 157000
2029
+ },
2030
+ {
2031
+ "epoch": 0.59,
2032
+ "learning_rate": 2.061040783423897e-05,
2033
+ "loss": 3.5142,
2034
+ "step": 157500
2035
+ },
2036
+ {
2037
+ "epoch": 0.59,
2038
+ "learning_rate": 2.051710754164925e-05,
2039
+ "loss": 3.5084,
2040
+ "step": 158000
2041
+ },
2042
+ {
2043
+ "epoch": 0.59,
2044
+ "learning_rate": 2.0423807249059533e-05,
2045
+ "loss": 3.5147,
2046
+ "step": 158500
2047
+ },
2048
+ {
2049
+ "epoch": 0.59,
2050
+ "learning_rate": 2.0330506956469815e-05,
2051
+ "loss": 3.5112,
2052
+ "step": 159000
2053
+ },
2054
+ {
2055
+ "epoch": 0.6,
2056
+ "learning_rate": 2.0237206663880097e-05,
2057
+ "loss": 3.5137,
2058
+ "step": 159500
2059
+ },
2060
+ {
2061
+ "epoch": 0.6,
2062
+ "learning_rate": 2.014390637129038e-05,
2063
+ "loss": 3.5124,
2064
+ "step": 160000
2065
+ },
2066
+ {
2067
+ "epoch": 0.6,
2068
+ "eval_accuracy": 0.3877192016058592,
2069
+ "eval_loss": 3.432858943939209,
2070
+ "eval_runtime": 6161.9947,
2071
+ "eval_samples_per_second": 73.221,
2072
+ "eval_steps_per_second": 1.144,
2073
+ "step": 160000
2074
+ },
2075
+ {
2076
+ "epoch": 0.6,
2077
+ "learning_rate": 2.0050606078700664e-05,
2078
+ "loss": 3.5108,
2079
+ "step": 160500
2080
+ },
2081
+ {
2082
+ "epoch": 0.6,
2083
+ "learning_rate": 1.9957305786110946e-05,
2084
+ "loss": 3.5035,
2085
+ "step": 161000
2086
+ },
2087
+ {
2088
+ "epoch": 0.6,
2089
+ "learning_rate": 1.9864005493521228e-05,
2090
+ "loss": 3.5044,
2091
+ "step": 161500
2092
+ },
2093
+ {
2094
+ "epoch": 0.6,
2095
+ "learning_rate": 1.977070520093151e-05,
2096
+ "loss": 3.5069,
2097
+ "step": 162000
2098
+ },
2099
+ {
2100
+ "epoch": 0.61,
2101
+ "learning_rate": 1.967740490834179e-05,
2102
+ "loss": 3.5078,
2103
+ "step": 162500
2104
+ },
2105
+ {
2106
+ "epoch": 0.61,
2107
+ "learning_rate": 1.9584104615752074e-05,
2108
+ "loss": 3.5003,
2109
+ "step": 163000
2110
+ },
2111
+ {
2112
+ "epoch": 0.61,
2113
+ "learning_rate": 1.949080432316236e-05,
2114
+ "loss": 3.5056,
2115
+ "step": 163500
2116
+ },
2117
+ {
2118
+ "epoch": 0.61,
2119
+ "learning_rate": 1.939750403057264e-05,
2120
+ "loss": 3.5082,
2121
+ "step": 164000
2122
+ },
2123
+ {
2124
+ "epoch": 0.61,
2125
+ "learning_rate": 1.9304203737982923e-05,
2126
+ "loss": 3.5086,
2127
+ "step": 164500
2128
+ },
2129
+ {
2130
+ "epoch": 0.62,
2131
+ "learning_rate": 1.9210903445393204e-05,
2132
+ "loss": 3.4997,
2133
+ "step": 165000
2134
+ },
2135
+ {
2136
+ "epoch": 0.62,
2137
+ "learning_rate": 1.9117603152803486e-05,
2138
+ "loss": 3.5013,
2139
+ "step": 165500
2140
+ },
2141
+ {
2142
+ "epoch": 0.62,
2143
+ "learning_rate": 1.902430286021377e-05,
2144
+ "loss": 3.5048,
2145
+ "step": 166000
2146
+ },
2147
+ {
2148
+ "epoch": 0.62,
2149
+ "learning_rate": 1.8931002567624053e-05,
2150
+ "loss": 3.5067,
2151
+ "step": 166500
2152
+ },
2153
+ {
2154
+ "epoch": 0.62,
2155
+ "learning_rate": 1.8837702275034335e-05,
2156
+ "loss": 3.5029,
2157
+ "step": 167000
2158
+ },
2159
+ {
2160
+ "epoch": 0.63,
2161
+ "learning_rate": 1.8744401982444617e-05,
2162
+ "loss": 3.503,
2163
+ "step": 167500
2164
+ },
2165
+ {
2166
+ "epoch": 0.63,
2167
+ "learning_rate": 1.86511016898549e-05,
2168
+ "loss": 3.5041,
2169
+ "step": 168000
2170
+ },
2171
+ {
2172
+ "epoch": 0.63,
2173
+ "learning_rate": 1.855780139726518e-05,
2174
+ "loss": 3.5038,
2175
+ "step": 168500
2176
+ },
2177
+ {
2178
+ "epoch": 0.63,
2179
+ "learning_rate": 1.8464501104675466e-05,
2180
+ "loss": 3.5015,
2181
+ "step": 169000
2182
+ },
2183
+ {
2184
+ "epoch": 0.63,
2185
+ "learning_rate": 1.8371200812085748e-05,
2186
+ "loss": 3.4985,
2187
+ "step": 169500
2188
+ },
2189
+ {
2190
+ "epoch": 0.63,
2191
+ "learning_rate": 1.827790051949603e-05,
2192
+ "loss": 3.501,
2193
+ "step": 170000
2194
+ },
2195
+ {
2196
+ "epoch": 0.63,
2197
+ "eval_accuracy": 0.3889908155509813,
2198
+ "eval_loss": 3.422307252883911,
2199
+ "eval_runtime": 6181.3169,
2200
+ "eval_samples_per_second": 72.992,
2201
+ "eval_steps_per_second": 1.141,
2202
+ "step": 170000
2203
+ },
2204
+ {
2205
+ "epoch": 0.64,
2206
+ "learning_rate": 1.8184600226906312e-05,
2207
+ "loss": 3.4997,
2208
+ "step": 170500
2209
+ },
2210
+ {
2211
+ "epoch": 0.64,
2212
+ "learning_rate": 1.8091299934316594e-05,
2213
+ "loss": 3.5017,
2214
+ "step": 171000
2215
+ },
2216
+ {
2217
+ "epoch": 0.64,
2218
+ "learning_rate": 1.799799964172688e-05,
2219
+ "loss": 3.4982,
2220
+ "step": 171500
2221
+ },
2222
+ {
2223
+ "epoch": 0.64,
2224
+ "learning_rate": 1.790469934913716e-05,
2225
+ "loss": 3.4967,
2226
+ "step": 172000
2227
+ },
2228
+ {
2229
+ "epoch": 0.64,
2230
+ "learning_rate": 1.7811399056547443e-05,
2231
+ "loss": 3.4937,
2232
+ "step": 172500
2233
+ },
2234
+ {
2235
+ "epoch": 0.65,
2236
+ "learning_rate": 1.7718098763957725e-05,
2237
+ "loss": 3.5034,
2238
+ "step": 173000
2239
+ },
2240
+ {
2241
+ "epoch": 0.65,
2242
+ "learning_rate": 1.7624798471368006e-05,
2243
+ "loss": 3.494,
2244
+ "step": 173500
2245
+ },
2246
+ {
2247
+ "epoch": 0.65,
2248
+ "learning_rate": 1.753149817877829e-05,
2249
+ "loss": 3.5028,
2250
+ "step": 174000
2251
+ },
2252
+ {
2253
+ "epoch": 0.65,
2254
+ "learning_rate": 1.7438197886188574e-05,
2255
+ "loss": 3.4946,
2256
+ "step": 174500
2257
+ },
2258
+ {
2259
+ "epoch": 0.65,
2260
+ "learning_rate": 1.7344897593598855e-05,
2261
+ "loss": 3.4916,
2262
+ "step": 175000
2263
+ },
2264
+ {
2265
+ "epoch": 0.65,
2266
+ "learning_rate": 1.7251597301009137e-05,
2267
+ "loss": 3.4945,
2268
+ "step": 175500
2269
+ },
2270
+ {
2271
+ "epoch": 0.66,
2272
+ "learning_rate": 1.715829700841942e-05,
2273
+ "loss": 3.4958,
2274
+ "step": 176000
2275
+ },
2276
+ {
2277
+ "epoch": 0.66,
2278
+ "learning_rate": 1.70649967158297e-05,
2279
+ "loss": 3.4939,
2280
+ "step": 176500
2281
+ },
2282
+ {
2283
+ "epoch": 0.66,
2284
+ "learning_rate": 1.6971696423239983e-05,
2285
+ "loss": 3.4976,
2286
+ "step": 177000
2287
+ },
2288
+ {
2289
+ "epoch": 0.66,
2290
+ "learning_rate": 1.6878396130650268e-05,
2291
+ "loss": 3.4994,
2292
+ "step": 177500
2293
+ },
2294
+ {
2295
+ "epoch": 0.66,
2296
+ "learning_rate": 1.678509583806055e-05,
2297
+ "loss": 3.4868,
2298
+ "step": 178000
2299
+ },
2300
+ {
2301
+ "epoch": 0.67,
2302
+ "learning_rate": 1.6691795545470832e-05,
2303
+ "loss": 3.4933,
2304
+ "step": 178500
2305
+ },
2306
+ {
2307
+ "epoch": 0.67,
2308
+ "learning_rate": 1.6598495252881114e-05,
2309
+ "loss": 3.4922,
2310
+ "step": 179000
2311
+ },
2312
+ {
2313
+ "epoch": 0.67,
2314
+ "learning_rate": 1.6505194960291396e-05,
2315
+ "loss": 3.4896,
2316
+ "step": 179500
2317
+ },
2318
+ {
2319
+ "epoch": 0.67,
2320
+ "learning_rate": 1.641189466770168e-05,
2321
+ "loss": 3.4934,
2322
+ "step": 180000
2323
+ },
2324
+ {
2325
+ "epoch": 0.67,
2326
+ "eval_accuracy": 0.3900964280538928,
2327
+ "eval_loss": 3.4130406379699707,
2328
+ "eval_runtime": 6165.5394,
2329
+ "eval_samples_per_second": 73.179,
2330
+ "eval_steps_per_second": 1.143,
2331
+ "step": 180000
2332
+ },
2333
+ {
2334
+ "epoch": 0.67,
2335
+ "learning_rate": 1.6318594375111963e-05,
2336
+ "loss": 3.4938,
2337
+ "step": 180500
2338
+ },
2339
+ {
2340
+ "epoch": 0.68,
2341
+ "learning_rate": 1.6225294082522245e-05,
2342
+ "loss": 3.4904,
2343
+ "step": 181000
2344
+ },
2345
+ {
2346
+ "epoch": 0.68,
2347
+ "learning_rate": 1.6131993789932527e-05,
2348
+ "loss": 3.4894,
2349
+ "step": 181500
2350
+ },
2351
+ {
2352
+ "epoch": 0.68,
2353
+ "learning_rate": 1.603869349734281e-05,
2354
+ "loss": 3.49,
2355
+ "step": 182000
2356
+ },
2357
+ {
2358
+ "epoch": 0.68,
2359
+ "learning_rate": 1.594539320475309e-05,
2360
+ "loss": 3.4879,
2361
+ "step": 182500
2362
+ },
2363
+ {
2364
+ "epoch": 0.68,
2365
+ "learning_rate": 1.5852092912163376e-05,
2366
+ "loss": 3.4914,
2367
+ "step": 183000
2368
+ },
2369
+ {
2370
+ "epoch": 0.68,
2371
+ "learning_rate": 1.5758792619573658e-05,
2372
+ "loss": 3.4971,
2373
+ "step": 183500
2374
+ },
2375
+ {
2376
+ "epoch": 0.69,
2377
+ "learning_rate": 1.566549232698394e-05,
2378
+ "loss": 3.4892,
2379
+ "step": 184000
2380
+ },
2381
+ {
2382
+ "epoch": 0.69,
2383
+ "learning_rate": 1.557219203439422e-05,
2384
+ "loss": 3.4822,
2385
+ "step": 184500
2386
+ },
2387
+ {
2388
+ "epoch": 0.69,
2389
+ "learning_rate": 1.5478891741804503e-05,
2390
+ "loss": 3.4924,
2391
+ "step": 185000
2392
+ },
2393
+ {
2394
+ "epoch": 0.69,
2395
+ "learning_rate": 1.538559144921479e-05,
2396
+ "loss": 3.4869,
2397
+ "step": 185500
2398
+ },
2399
+ {
2400
+ "epoch": 0.69,
2401
+ "learning_rate": 1.529229115662507e-05,
2402
+ "loss": 3.4916,
2403
+ "step": 186000
2404
+ },
2405
+ {
2406
+ "epoch": 0.7,
2407
+ "learning_rate": 1.519899086403535e-05,
2408
+ "loss": 3.4782,
2409
+ "step": 186500
2410
+ },
2411
+ {
2412
+ "epoch": 0.7,
2413
+ "learning_rate": 1.5105690571445632e-05,
2414
+ "loss": 3.4797,
2415
+ "step": 187000
2416
+ },
2417
+ {
2418
+ "epoch": 0.7,
2419
+ "learning_rate": 1.5012390278855914e-05,
2420
+ "loss": 3.4762,
2421
+ "step": 187500
2422
+ },
2423
+ {
2424
+ "epoch": 0.7,
2425
+ "learning_rate": 1.4919089986266196e-05,
2426
+ "loss": 3.4915,
2427
+ "step": 188000
2428
+ },
2429
+ {
2430
+ "epoch": 0.7,
2431
+ "learning_rate": 1.4825789693676481e-05,
2432
+ "loss": 3.4816,
2433
+ "step": 188500
2434
+ },
2435
+ {
2436
+ "epoch": 0.71,
2437
+ "learning_rate": 1.4732489401086763e-05,
2438
+ "loss": 3.4886,
2439
+ "step": 189000
2440
+ },
2441
+ {
2442
+ "epoch": 0.71,
2443
+ "learning_rate": 1.4639189108497045e-05,
2444
+ "loss": 3.4776,
2445
+ "step": 189500
2446
+ },
2447
+ {
2448
+ "epoch": 0.71,
2449
+ "learning_rate": 1.4545888815907327e-05,
2450
+ "loss": 3.4863,
2451
+ "step": 190000
2452
+ },
2453
+ {
2454
+ "epoch": 0.71,
2455
+ "eval_accuracy": 0.390879254197248,
2456
+ "eval_loss": 3.4041781425476074,
2457
+ "eval_runtime": 6176.7947,
2458
+ "eval_samples_per_second": 73.045,
2459
+ "eval_steps_per_second": 1.141,
2460
+ "step": 190000
2461
+ },
2462
+ {
2463
+ "epoch": 0.71,
2464
+ "learning_rate": 1.4452588523317609e-05,
2465
+ "loss": 3.4775,
2466
+ "step": 190500
2467
+ },
2468
+ {
2469
+ "epoch": 0.71,
2470
+ "learning_rate": 1.435928823072789e-05,
2471
+ "loss": 3.4816,
2472
+ "step": 191000
2473
+ },
2474
+ {
2475
+ "epoch": 0.71,
2476
+ "learning_rate": 1.4265987938138176e-05,
2477
+ "loss": 3.4877,
2478
+ "step": 191500
2479
+ },
2480
+ {
2481
+ "epoch": 0.72,
2482
+ "learning_rate": 1.4172687645548458e-05,
2483
+ "loss": 3.479,
2484
+ "step": 192000
2485
+ },
2486
+ {
2487
+ "epoch": 0.72,
2488
+ "learning_rate": 1.407938735295874e-05,
2489
+ "loss": 3.4829,
2490
+ "step": 192500
2491
+ },
2492
+ {
2493
+ "epoch": 0.72,
2494
+ "learning_rate": 1.3986087060369022e-05,
2495
+ "loss": 3.4774,
2496
+ "step": 193000
2497
+ },
2498
+ {
2499
+ "epoch": 0.72,
2500
+ "learning_rate": 1.3892786767779304e-05,
2501
+ "loss": 3.4793,
2502
+ "step": 193500
2503
+ },
2504
+ {
2505
+ "epoch": 0.72,
2506
+ "learning_rate": 1.3799486475189589e-05,
2507
+ "loss": 3.4811,
2508
+ "step": 194000
2509
+ },
2510
+ {
2511
+ "epoch": 0.73,
2512
+ "learning_rate": 1.370618618259987e-05,
2513
+ "loss": 3.4737,
2514
+ "step": 194500
2515
+ },
2516
+ {
2517
+ "epoch": 0.73,
2518
+ "learning_rate": 1.3612885890010153e-05,
2519
+ "loss": 3.4784,
2520
+ "step": 195000
2521
+ },
2522
+ {
2523
+ "epoch": 0.73,
2524
+ "learning_rate": 1.3519585597420434e-05,
2525
+ "loss": 3.4769,
2526
+ "step": 195500
2527
+ },
2528
+ {
2529
+ "epoch": 0.73,
2530
+ "learning_rate": 1.3426285304830716e-05,
2531
+ "loss": 3.4766,
2532
+ "step": 196000
2533
+ },
2534
+ {
2535
+ "epoch": 0.73,
2536
+ "learning_rate": 1.3332985012240998e-05,
2537
+ "loss": 3.48,
2538
+ "step": 196500
2539
+ },
2540
+ {
2541
+ "epoch": 0.74,
2542
+ "learning_rate": 1.3239684719651282e-05,
2543
+ "loss": 3.4781,
2544
+ "step": 197000
2545
+ },
2546
+ {
2547
+ "epoch": 0.74,
2548
+ "learning_rate": 1.3146384427061564e-05,
2549
+ "loss": 3.4774,
2550
+ "step": 197500
2551
+ },
2552
+ {
2553
+ "epoch": 0.74,
2554
+ "learning_rate": 1.3053084134471845e-05,
2555
+ "loss": 3.4706,
2556
+ "step": 198000
2557
+ },
2558
+ {
2559
+ "epoch": 0.74,
2560
+ "learning_rate": 1.2959783841882129e-05,
2561
+ "loss": 3.474,
2562
+ "step": 198500
2563
+ },
2564
+ {
2565
+ "epoch": 0.74,
2566
+ "learning_rate": 1.2866483549292411e-05,
2567
+ "loss": 3.4722,
2568
+ "step": 199000
2569
+ },
2570
+ {
2571
+ "epoch": 0.74,
2572
+ "learning_rate": 1.2773183256702694e-05,
2573
+ "loss": 3.4734,
2574
+ "step": 199500
2575
+ },
2576
+ {
2577
+ "epoch": 0.75,
2578
+ "learning_rate": 1.2679882964112976e-05,
2579
+ "loss": 3.4799,
2580
+ "step": 200000
2581
+ },
2582
+ {
2583
+ "epoch": 0.75,
2584
+ "eval_accuracy": 0.3914298624786812,
2585
+ "eval_loss": 3.39914870262146,
2586
+ "eval_runtime": 6175.7078,
2587
+ "eval_samples_per_second": 73.058,
2588
+ "eval_steps_per_second": 1.142,
2589
+ "step": 200000
2590
+ },
2591
+ {
2592
+ "epoch": 0.75,
2593
+ "learning_rate": 1.2586582671523258e-05,
2594
+ "loss": 3.4762,
2595
+ "step": 200500
2596
+ },
2597
+ {
2598
+ "epoch": 0.75,
2599
+ "learning_rate": 1.249328237893354e-05,
2600
+ "loss": 3.4779,
2601
+ "step": 201000
2602
+ },
2603
+ {
2604
+ "epoch": 0.75,
2605
+ "learning_rate": 1.2399982086343824e-05,
2606
+ "loss": 3.476,
2607
+ "step": 201500
2608
+ },
2609
+ {
2610
+ "epoch": 0.75,
2611
+ "learning_rate": 1.2306681793754106e-05,
2612
+ "loss": 3.473,
2613
+ "step": 202000
2614
+ },
2615
+ {
2616
+ "epoch": 0.76,
2617
+ "learning_rate": 1.2213381501164387e-05,
2618
+ "loss": 3.4828,
2619
+ "step": 202500
2620
+ },
2621
+ {
2622
+ "epoch": 0.76,
2623
+ "learning_rate": 1.2120081208574671e-05,
2624
+ "loss": 3.4759,
2625
+ "step": 203000
2626
+ },
2627
+ {
2628
+ "epoch": 0.76,
2629
+ "learning_rate": 1.2026780915984953e-05,
2630
+ "loss": 3.4801,
2631
+ "step": 203500
2632
+ },
2633
+ {
2634
+ "epoch": 0.76,
2635
+ "learning_rate": 1.1933480623395235e-05,
2636
+ "loss": 3.4724,
2637
+ "step": 204000
2638
+ },
2639
+ {
2640
+ "epoch": 0.76,
2641
+ "learning_rate": 1.1840180330805518e-05,
2642
+ "loss": 3.4795,
2643
+ "step": 204500
2644
+ },
2645
+ {
2646
+ "epoch": 0.77,
2647
+ "learning_rate": 1.17468800382158e-05,
2648
+ "loss": 3.4742,
2649
+ "step": 205000
2650
+ },
2651
+ {
2652
+ "epoch": 0.77,
2653
+ "learning_rate": 1.1653579745626082e-05,
2654
+ "loss": 3.4712,
2655
+ "step": 205500
2656
+ },
2657
+ {
2658
+ "epoch": 0.77,
2659
+ "learning_rate": 1.1560279453036366e-05,
2660
+ "loss": 3.4743,
2661
+ "step": 206000
2662
+ },
2663
+ {
2664
+ "epoch": 0.77,
2665
+ "learning_rate": 1.1466979160446648e-05,
2666
+ "loss": 3.471,
2667
+ "step": 206500
2668
+ },
2669
+ {
2670
+ "epoch": 0.77,
2671
+ "learning_rate": 1.137367886785693e-05,
2672
+ "loss": 3.4698,
2673
+ "step": 207000
2674
+ },
2675
+ {
2676
+ "epoch": 0.77,
2677
+ "learning_rate": 1.1280378575267213e-05,
2678
+ "loss": 3.4809,
2679
+ "step": 207500
2680
+ },
2681
+ {
2682
+ "epoch": 0.78,
2683
+ "learning_rate": 1.1187078282677495e-05,
2684
+ "loss": 3.472,
2685
+ "step": 208000
2686
+ },
2687
+ {
2688
+ "epoch": 0.78,
2689
+ "learning_rate": 1.1093777990087778e-05,
2690
+ "loss": 3.4675,
2691
+ "step": 208500
2692
+ },
2693
+ {
2694
+ "epoch": 0.78,
2695
+ "learning_rate": 1.100047769749806e-05,
2696
+ "loss": 3.4722,
2697
+ "step": 209000
2698
+ },
2699
+ {
2700
+ "epoch": 0.78,
2701
+ "learning_rate": 1.0907177404908342e-05,
2702
+ "loss": 3.4684,
2703
+ "step": 209500
2704
+ },
2705
+ {
2706
+ "epoch": 0.78,
2707
+ "learning_rate": 1.0813877112318626e-05,
2708
+ "loss": 3.4682,
2709
+ "step": 210000
2710
+ },
2711
+ {
2712
+ "epoch": 0.78,
2713
+ "eval_accuracy": 0.3923717267565564,
2714
+ "eval_loss": 3.3909456729888916,
2715
+ "eval_runtime": 6173.3331,
2716
+ "eval_samples_per_second": 73.086,
2717
+ "eval_steps_per_second": 1.142,
2718
+ "step": 210000
2719
+ },
2720
+ {
2721
+ "epoch": 0.79,
2722
+ "learning_rate": 1.0720576819728908e-05,
2723
+ "loss": 3.465,
2724
+ "step": 210500
2725
+ },
2726
+ {
2727
+ "epoch": 0.79,
2728
+ "learning_rate": 1.062727652713919e-05,
2729
+ "loss": 3.4655,
2730
+ "step": 211000
2731
+ },
2732
+ {
2733
+ "epoch": 0.79,
2734
+ "learning_rate": 1.0533976234549473e-05,
2735
+ "loss": 3.4601,
2736
+ "step": 211500
2737
+ },
2738
+ {
2739
+ "epoch": 0.79,
2740
+ "learning_rate": 1.0440675941959755e-05,
2741
+ "loss": 3.4688,
2742
+ "step": 212000
2743
+ },
2744
+ {
2745
+ "epoch": 0.79,
2746
+ "learning_rate": 1.0347375649370037e-05,
2747
+ "loss": 3.4694,
2748
+ "step": 212500
2749
+ },
2750
+ {
2751
+ "epoch": 0.79,
2752
+ "learning_rate": 1.025407535678032e-05,
2753
+ "loss": 3.4675,
2754
+ "step": 213000
2755
+ },
2756
+ {
2757
+ "epoch": 0.8,
2758
+ "learning_rate": 1.0160775064190602e-05,
2759
+ "loss": 3.4701,
2760
+ "step": 213500
2761
+ },
2762
+ {
2763
+ "epoch": 0.8,
2764
+ "learning_rate": 1.0067474771600884e-05,
2765
+ "loss": 3.4659,
2766
+ "step": 214000
2767
+ },
2768
+ {
2769
+ "epoch": 0.8,
2770
+ "learning_rate": 9.974174479011166e-06,
2771
+ "loss": 3.4678,
2772
+ "step": 214500
2773
+ },
2774
+ {
2775
+ "epoch": 0.8,
2776
+ "learning_rate": 9.88087418642145e-06,
2777
+ "loss": 3.4694,
2778
+ "step": 215000
2779
+ },
2780
+ {
2781
+ "epoch": 0.8,
2782
+ "learning_rate": 9.787573893831731e-06,
2783
+ "loss": 3.4714,
2784
+ "step": 215500
2785
+ },
2786
+ {
2787
+ "epoch": 0.81,
2788
+ "learning_rate": 9.694273601242013e-06,
2789
+ "loss": 3.4613,
2790
+ "step": 216000
2791
+ },
2792
+ {
2793
+ "epoch": 0.81,
2794
+ "learning_rate": 9.600973308652295e-06,
2795
+ "loss": 3.4647,
2796
+ "step": 216500
2797
+ },
2798
+ {
2799
+ "epoch": 0.81,
2800
+ "learning_rate": 9.507673016062579e-06,
2801
+ "loss": 3.4664,
2802
+ "step": 217000
2803
+ },
2804
+ {
2805
+ "epoch": 0.81,
2806
+ "learning_rate": 9.41437272347286e-06,
2807
+ "loss": 3.4631,
2808
+ "step": 217500
2809
+ },
2810
+ {
2811
+ "epoch": 0.81,
2812
+ "learning_rate": 9.321072430883143e-06,
2813
+ "loss": 3.4666,
2814
+ "step": 218000
2815
+ },
2816
+ {
2817
+ "epoch": 0.82,
2818
+ "learning_rate": 9.227772138293426e-06,
2819
+ "loss": 3.467,
2820
+ "step": 218500
2821
+ },
2822
+ {
2823
+ "epoch": 0.82,
2824
+ "learning_rate": 9.134471845703708e-06,
2825
+ "loss": 3.4636,
2826
+ "step": 219000
2827
+ },
2828
+ {
2829
+ "epoch": 0.82,
2830
+ "learning_rate": 9.04117155311399e-06,
2831
+ "loss": 3.4637,
2832
+ "step": 219500
2833
+ },
2834
+ {
2835
+ "epoch": 0.82,
2836
+ "learning_rate": 8.947871260524273e-06,
2837
+ "loss": 3.4667,
2838
+ "step": 220000
2839
+ },
2840
+ {
2841
+ "epoch": 0.82,
2842
+ "eval_accuracy": 0.3929634584179766,
2843
+ "eval_loss": 3.385192394256592,
2844
+ "eval_runtime": 6176.1319,
2845
+ "eval_samples_per_second": 73.053,
2846
+ "eval_steps_per_second": 1.141,
2847
+ "step": 220000
2848
+ },
2849
+ {
2850
+ "epoch": 0.82,
2851
+ "learning_rate": 8.854570967934555e-06,
2852
+ "loss": 3.4613,
2853
+ "step": 220500
2854
+ },
2855
+ {
2856
+ "epoch": 0.82,
2857
+ "learning_rate": 8.761270675344839e-06,
2858
+ "loss": 3.4618,
2859
+ "step": 221000
2860
+ },
2861
+ {
2862
+ "epoch": 0.83,
2863
+ "learning_rate": 8.66797038275512e-06,
2864
+ "loss": 3.4675,
2865
+ "step": 221500
2866
+ },
2867
+ {
2868
+ "epoch": 0.83,
2869
+ "learning_rate": 8.574670090165403e-06,
2870
+ "loss": 3.4575,
2871
+ "step": 222000
2872
+ },
2873
+ {
2874
+ "epoch": 0.83,
2875
+ "learning_rate": 8.481369797575686e-06,
2876
+ "loss": 3.4609,
2877
+ "step": 222500
2878
+ },
2879
+ {
2880
+ "epoch": 0.83,
2881
+ "learning_rate": 8.388069504985968e-06,
2882
+ "loss": 3.4639,
2883
+ "step": 223000
2884
+ },
2885
+ {
2886
+ "epoch": 0.83,
2887
+ "learning_rate": 8.29476921239625e-06,
2888
+ "loss": 3.4629,
2889
+ "step": 223500
2890
+ },
2891
+ {
2892
+ "epoch": 0.84,
2893
+ "learning_rate": 8.201468919806534e-06,
2894
+ "loss": 3.4579,
2895
+ "step": 224000
2896
+ },
2897
+ {
2898
+ "epoch": 0.84,
2899
+ "learning_rate": 8.108168627216815e-06,
2900
+ "loss": 3.461,
2901
+ "step": 224500
2902
+ },
2903
+ {
2904
+ "epoch": 0.84,
2905
+ "learning_rate": 8.014868334627097e-06,
2906
+ "loss": 3.4586,
2907
+ "step": 225000
2908
+ },
2909
+ {
2910
+ "epoch": 0.84,
2911
+ "learning_rate": 7.92156804203738e-06,
2912
+ "loss": 3.4619,
2913
+ "step": 225500
2914
+ },
2915
+ {
2916
+ "epoch": 0.84,
2917
+ "learning_rate": 7.828267749447663e-06,
2918
+ "loss": 3.4657,
2919
+ "step": 226000
2920
+ },
2921
+ {
2922
+ "epoch": 0.85,
2923
+ "learning_rate": 7.734967456857945e-06,
2924
+ "loss": 3.4628,
2925
+ "step": 226500
2926
+ },
2927
+ {
2928
+ "epoch": 0.85,
2929
+ "learning_rate": 7.641667164268228e-06,
2930
+ "loss": 3.461,
2931
+ "step": 227000
2932
+ },
2933
+ {
2934
+ "epoch": 0.85,
2935
+ "learning_rate": 7.54836687167851e-06,
2936
+ "loss": 3.4642,
2937
+ "step": 227500
2938
+ },
2939
+ {
2940
+ "epoch": 0.85,
2941
+ "learning_rate": 7.455066579088793e-06,
2942
+ "loss": 3.459,
2943
+ "step": 228000
2944
+ },
2945
+ {
2946
+ "epoch": 0.85,
2947
+ "learning_rate": 7.361766286499075e-06,
2948
+ "loss": 3.4579,
2949
+ "step": 228500
2950
+ },
2951
+ {
2952
+ "epoch": 0.85,
2953
+ "learning_rate": 7.2684659939093565e-06,
2954
+ "loss": 3.4543,
2955
+ "step": 229000
2956
+ },
2957
+ {
2958
+ "epoch": 0.86,
2959
+ "learning_rate": 7.17516570131964e-06,
2960
+ "loss": 3.4529,
2961
+ "step": 229500
2962
+ },
2963
+ {
2964
+ "epoch": 0.86,
2965
+ "learning_rate": 7.081865408729922e-06,
2966
+ "loss": 3.4564,
2967
+ "step": 230000
2968
+ },
2969
+ {
2970
+ "epoch": 0.86,
2971
+ "eval_accuracy": 0.39360805835390505,
2972
+ "eval_loss": 3.3790297508239746,
2973
+ "eval_runtime": 6181.6044,
2974
+ "eval_samples_per_second": 72.988,
2975
+ "eval_steps_per_second": 1.14,
2976
+ "step": 230000
2977
+ },
2978
+ {
2979
+ "epoch": 0.86,
2980
+ "learning_rate": 6.988565116140204e-06,
2981
+ "loss": 3.4576,
2982
+ "step": 230500
2983
+ },
2984
+ {
2985
+ "epoch": 0.86,
2986
+ "learning_rate": 6.895264823550487e-06,
2987
+ "loss": 3.4563,
2988
+ "step": 231000
2989
+ },
2990
+ {
2991
+ "epoch": 0.86,
2992
+ "learning_rate": 6.801964530960769e-06,
2993
+ "loss": 3.4598,
2994
+ "step": 231500
2995
+ },
2996
+ {
2997
+ "epoch": 0.87,
2998
+ "learning_rate": 6.708664238371051e-06,
2999
+ "loss": 3.4568,
3000
+ "step": 232000
3001
+ },
3002
+ {
3003
+ "epoch": 0.87,
3004
+ "learning_rate": 6.615363945781335e-06,
3005
+ "loss": 3.4592,
3006
+ "step": 232500
3007
+ },
3008
+ {
3009
+ "epoch": 0.87,
3010
+ "learning_rate": 6.522063653191617e-06,
3011
+ "loss": 3.4551,
3012
+ "step": 233000
3013
+ },
3014
+ {
3015
+ "epoch": 0.87,
3016
+ "learning_rate": 6.4287633606018985e-06,
3017
+ "loss": 3.4606,
3018
+ "step": 233500
3019
+ },
3020
+ {
3021
+ "epoch": 0.87,
3022
+ "learning_rate": 6.335463068012182e-06,
3023
+ "loss": 3.4538,
3024
+ "step": 234000
3025
+ },
3026
+ {
3027
+ "epoch": 0.88,
3028
+ "learning_rate": 6.242162775422464e-06,
3029
+ "loss": 3.4554,
3030
+ "step": 234500
3031
+ },
3032
+ {
3033
+ "epoch": 0.88,
3034
+ "learning_rate": 6.148862482832747e-06,
3035
+ "loss": 3.4551,
3036
+ "step": 235000
3037
+ },
3038
+ {
3039
+ "epoch": 0.88,
3040
+ "learning_rate": 6.055562190243029e-06,
3041
+ "loss": 3.4592,
3042
+ "step": 235500
3043
+ },
3044
+ {
3045
+ "epoch": 0.88,
3046
+ "learning_rate": 5.962261897653311e-06,
3047
+ "loss": 3.4535,
3048
+ "step": 236000
3049
+ },
3050
+ {
3051
+ "epoch": 0.88,
3052
+ "learning_rate": 5.868961605063594e-06,
3053
+ "loss": 3.4534,
3054
+ "step": 236500
3055
+ },
3056
+ {
3057
+ "epoch": 0.88,
3058
+ "learning_rate": 5.775661312473876e-06,
3059
+ "loss": 3.458,
3060
+ "step": 237000
3061
+ },
3062
+ {
3063
+ "epoch": 0.89,
3064
+ "learning_rate": 5.6823610198841586e-06,
3065
+ "loss": 3.4605,
3066
+ "step": 237500
3067
+ },
3068
+ {
3069
+ "epoch": 0.89,
3070
+ "learning_rate": 5.5890607272944404e-06,
3071
+ "loss": 3.4597,
3072
+ "step": 238000
3073
+ },
3074
+ {
3075
+ "epoch": 0.89,
3076
+ "learning_rate": 5.495760434704723e-06,
3077
+ "loss": 3.4539,
3078
+ "step": 238500
3079
+ },
3080
+ {
3081
+ "epoch": 0.89,
3082
+ "learning_rate": 5.402460142115006e-06,
3083
+ "loss": 3.4585,
3084
+ "step": 239000
3085
+ },
3086
+ {
3087
+ "epoch": 0.89,
3088
+ "learning_rate": 5.309159849525289e-06,
3089
+ "loss": 3.4474,
3090
+ "step": 239500
3091
+ },
3092
+ {
3093
+ "epoch": 0.9,
3094
+ "learning_rate": 5.2158595569355705e-06,
3095
+ "loss": 3.4581,
3096
+ "step": 240000
3097
+ },
3098
+ {
3099
+ "epoch": 0.9,
3100
+ "eval_accuracy": 0.3941211310703905,
3101
+ "eval_loss": 3.3752598762512207,
3102
+ "eval_runtime": 6171.8723,
3103
+ "eval_samples_per_second": 73.103,
3104
+ "eval_steps_per_second": 1.142,
3105
+ "step": 240000
3106
+ },
3107
+ {
3108
+ "epoch": 0.9,
3109
+ "learning_rate": 5.122559264345853e-06,
3110
+ "loss": 3.4558,
3111
+ "step": 240500
3112
+ },
3113
+ {
3114
+ "epoch": 0.9,
3115
+ "learning_rate": 5.029258971756136e-06,
3116
+ "loss": 3.4538,
3117
+ "step": 241000
3118
+ },
3119
+ {
3120
+ "epoch": 0.9,
3121
+ "learning_rate": 4.935958679166418e-06,
3122
+ "loss": 3.4561,
3123
+ "step": 241500
3124
+ },
3125
+ {
3126
+ "epoch": 0.9,
3127
+ "learning_rate": 4.8426583865767005e-06,
3128
+ "loss": 3.4564,
3129
+ "step": 242000
3130
+ },
3131
+ {
3132
+ "epoch": 0.91,
3133
+ "learning_rate": 4.749358093986983e-06,
3134
+ "loss": 3.4544,
3135
+ "step": 242500
3136
+ },
3137
+ {
3138
+ "epoch": 0.91,
3139
+ "learning_rate": 4.656057801397266e-06,
3140
+ "loss": 3.4481,
3141
+ "step": 243000
3142
+ },
3143
+ {
3144
+ "epoch": 0.91,
3145
+ "learning_rate": 4.562757508807548e-06,
3146
+ "loss": 3.4535,
3147
+ "step": 243500
3148
+ },
3149
+ {
3150
+ "epoch": 0.91,
3151
+ "learning_rate": 4.4694572162178306e-06,
3152
+ "loss": 3.4538,
3153
+ "step": 244000
3154
+ },
3155
+ {
3156
+ "epoch": 0.91,
3157
+ "learning_rate": 4.376156923628113e-06,
3158
+ "loss": 3.4467,
3159
+ "step": 244500
3160
+ },
3161
+ {
3162
+ "epoch": 0.91,
3163
+ "learning_rate": 4.282856631038395e-06,
3164
+ "loss": 3.4495,
3165
+ "step": 245000
3166
+ },
3167
+ {
3168
+ "epoch": 0.92,
3169
+ "learning_rate": 4.189556338448677e-06,
3170
+ "loss": 3.4485,
3171
+ "step": 245500
3172
+ },
3173
+ {
3174
+ "epoch": 0.92,
3175
+ "learning_rate": 4.09625604585896e-06,
3176
+ "loss": 3.4535,
3177
+ "step": 246000
3178
+ },
3179
+ {
3180
+ "epoch": 0.92,
3181
+ "learning_rate": 4.0029557532692425e-06,
3182
+ "loss": 3.45,
3183
+ "step": 246500
3184
+ },
3185
+ {
3186
+ "epoch": 0.92,
3187
+ "learning_rate": 3.909655460679524e-06,
3188
+ "loss": 3.4558,
3189
+ "step": 247000
3190
+ },
3191
+ {
3192
+ "epoch": 0.92,
3193
+ "learning_rate": 3.816355168089807e-06,
3194
+ "loss": 3.4502,
3195
+ "step": 247500
3196
+ },
3197
+ {
3198
+ "epoch": 0.93,
3199
+ "learning_rate": 3.72305487550009e-06,
3200
+ "loss": 3.4509,
3201
+ "step": 248000
3202
+ },
3203
+ {
3204
+ "epoch": 0.93,
3205
+ "learning_rate": 3.6297545829103717e-06,
3206
+ "loss": 3.4575,
3207
+ "step": 248500
3208
+ },
3209
+ {
3210
+ "epoch": 0.93,
3211
+ "learning_rate": 3.5364542903206544e-06,
3212
+ "loss": 3.451,
3213
+ "step": 249000
3214
+ },
3215
+ {
3216
+ "epoch": 0.93,
3217
+ "learning_rate": 3.443153997730937e-06,
3218
+ "loss": 3.4456,
3219
+ "step": 249500
3220
+ },
3221
+ {
3222
+ "epoch": 0.93,
3223
+ "learning_rate": 3.34985370514122e-06,
3224
+ "loss": 3.4553,
3225
+ "step": 250000
3226
+ },
3227
+ {
3228
+ "epoch": 0.93,
3229
+ "eval_accuracy": 0.3945191770501251,
3230
+ "eval_loss": 3.370955467224121,
3231
+ "eval_runtime": 6175.512,
3232
+ "eval_samples_per_second": 73.06,
3233
+ "eval_steps_per_second": 1.142,
3234
+ "step": 250000
3235
+ },
3236
+ {
3237
+ "epoch": 0.93,
3238
+ "learning_rate": 3.2565534125515017e-06,
3239
+ "loss": 3.4459,
3240
+ "step": 250500
3241
+ },
3242
+ {
3243
+ "epoch": 0.94,
3244
+ "learning_rate": 3.1632531199617845e-06,
3245
+ "loss": 3.4544,
3246
+ "step": 251000
3247
+ },
3248
+ {
3249
+ "epoch": 0.94,
3250
+ "learning_rate": 3.0699528273720668e-06,
3251
+ "loss": 3.4487,
3252
+ "step": 251500
3253
+ },
3254
+ {
3255
+ "epoch": 0.94,
3256
+ "learning_rate": 2.976652534782349e-06,
3257
+ "loss": 3.4497,
3258
+ "step": 252000
3259
+ },
3260
+ {
3261
+ "epoch": 0.94,
3262
+ "learning_rate": 2.8833522421926314e-06,
3263
+ "loss": 3.4464,
3264
+ "step": 252500
3265
+ },
3266
+ {
3267
+ "epoch": 0.94,
3268
+ "learning_rate": 2.790051949602914e-06,
3269
+ "loss": 3.4464,
3270
+ "step": 253000
3271
+ },
3272
+ {
3273
+ "epoch": 0.95,
3274
+ "learning_rate": 2.6967516570131964e-06,
3275
+ "loss": 3.4446,
3276
+ "step": 253500
3277
+ },
3278
+ {
3279
+ "epoch": 0.95,
3280
+ "learning_rate": 2.603451364423479e-06,
3281
+ "loss": 3.4536,
3282
+ "step": 254000
3283
+ },
3284
+ {
3285
+ "epoch": 0.95,
3286
+ "learning_rate": 2.5101510718337614e-06,
3287
+ "loss": 3.4477,
3288
+ "step": 254500
3289
+ },
3290
+ {
3291
+ "epoch": 0.95,
3292
+ "learning_rate": 2.416850779244044e-06,
3293
+ "loss": 3.4474,
3294
+ "step": 255000
3295
+ },
3296
+ {
3297
+ "epoch": 0.95,
3298
+ "learning_rate": 2.3235504866543264e-06,
3299
+ "loss": 3.4475,
3300
+ "step": 255500
3301
+ },
3302
+ {
3303
+ "epoch": 0.96,
3304
+ "learning_rate": 2.2302501940646087e-06,
3305
+ "loss": 3.449,
3306
+ "step": 256000
3307
+ },
3308
+ {
3309
+ "epoch": 0.96,
3310
+ "learning_rate": 2.136949901474891e-06,
3311
+ "loss": 3.4525,
3312
+ "step": 256500
3313
+ },
3314
+ {
3315
+ "epoch": 0.96,
3316
+ "learning_rate": 2.0436496088851733e-06,
3317
+ "loss": 3.4448,
3318
+ "step": 257000
3319
+ },
3320
+ {
3321
+ "epoch": 0.96,
3322
+ "learning_rate": 1.950349316295456e-06,
3323
+ "loss": 3.4487,
3324
+ "step": 257500
3325
+ },
3326
+ {
3327
+ "epoch": 0.96,
3328
+ "learning_rate": 1.8570490237057383e-06,
3329
+ "loss": 3.4372,
3330
+ "step": 258000
3331
+ },
3332
+ {
3333
+ "epoch": 0.96,
3334
+ "learning_rate": 1.763748731116021e-06,
3335
+ "loss": 3.4422,
3336
+ "step": 258500
3337
+ },
3338
+ {
3339
+ "epoch": 0.97,
3340
+ "learning_rate": 1.6704484385263034e-06,
3341
+ "loss": 3.4483,
3342
+ "step": 259000
3343
+ },
3344
+ {
3345
+ "epoch": 0.97,
3346
+ "learning_rate": 1.5771481459365859e-06,
3347
+ "loss": 3.4515,
3348
+ "step": 259500
3349
+ },
3350
+ {
3351
+ "epoch": 0.97,
3352
+ "learning_rate": 1.4838478533468682e-06,
3353
+ "loss": 3.4508,
3354
+ "step": 260000
3355
+ },
3356
+ {
3357
+ "epoch": 0.97,
3358
+ "eval_accuracy": 0.39486900634888356,
3359
+ "eval_loss": 3.367981433868408,
3360
+ "eval_runtime": 6161.9743,
3361
+ "eval_samples_per_second": 73.221,
3362
+ "eval_steps_per_second": 1.144,
3363
+ "step": 260000
3364
+ },
3365
+ {
3366
+ "epoch": 0.97,
3367
+ "learning_rate": 1.3905475607571507e-06,
3368
+ "loss": 3.4452,
3369
+ "step": 260500
3370
+ },
3371
+ {
3372
+ "epoch": 0.97,
3373
+ "learning_rate": 1.297247268167433e-06,
3374
+ "loss": 3.445,
3375
+ "step": 261000
3376
+ },
3377
+ {
3378
+ "epoch": 0.98,
3379
+ "learning_rate": 1.2039469755777155e-06,
3380
+ "loss": 3.454,
3381
+ "step": 261500
3382
+ },
3383
+ {
3384
+ "epoch": 0.98,
3385
+ "learning_rate": 1.1106466829879978e-06,
3386
+ "loss": 3.4487,
3387
+ "step": 262000
3388
+ },
3389
+ {
3390
+ "epoch": 0.98,
3391
+ "learning_rate": 1.0173463903982803e-06,
3392
+ "loss": 3.4454,
3393
+ "step": 262500
3394
+ },
3395
+ {
3396
+ "epoch": 0.98,
3397
+ "learning_rate": 9.240460978085627e-07,
3398
+ "loss": 3.4432,
3399
+ "step": 263000
3400
+ },
3401
+ {
3402
+ "epoch": 0.98,
3403
+ "learning_rate": 8.307458052188452e-07,
3404
+ "loss": 3.447,
3405
+ "step": 263500
3406
+ },
3407
+ {
3408
+ "epoch": 0.99,
3409
+ "learning_rate": 7.374455126291276e-07,
3410
+ "loss": 3.4449,
3411
+ "step": 264000
3412
+ },
3413
+ {
3414
+ "epoch": 0.99,
3415
+ "learning_rate": 6.4414522003941e-07,
3416
+ "loss": 3.4486,
3417
+ "step": 264500
3418
+ },
3419
+ {
3420
+ "epoch": 0.99,
3421
+ "learning_rate": 5.508449274496924e-07,
3422
+ "loss": 3.4444,
3423
+ "step": 265000
3424
+ },
3425
+ {
3426
+ "epoch": 0.99,
3427
+ "learning_rate": 4.5754463485997496e-07,
3428
+ "loss": 3.4486,
3429
+ "step": 265500
3430
+ },
3431
+ {
3432
+ "epoch": 0.99,
3433
+ "learning_rate": 3.6424434227025736e-07,
3434
+ "loss": 3.4416,
3435
+ "step": 266000
3436
+ },
3437
+ {
3438
+ "epoch": 0.99,
3439
+ "learning_rate": 2.709440496805398e-07,
3440
+ "loss": 3.4419,
3441
+ "step": 266500
3442
+ },
3443
+ {
3444
+ "epoch": 1.0,
3445
+ "learning_rate": 1.7764375709082225e-07,
3446
+ "loss": 3.4406,
3447
+ "step": 267000
3448
+ },
3449
+ {
3450
+ "epoch": 1.0,
3451
+ "learning_rate": 8.434346450110468e-08,
3452
+ "loss": 3.4406,
3453
+ "step": 267500
3454
+ },
3455
+ {
3456
+ "epoch": 1.0,
3457
+ "step": 267952,
3458
+ "total_flos": 4.480871972732928e+18,
3459
+ "train_loss": 3.648317173601172,
3460
+ "train_runtime": 313365.2222,
3461
+ "train_samples_per_second": 27.362,
3462
+ "train_steps_per_second": 0.855
3463
+ }
3464
+ ],
3465
+ "logging_steps": 500,
3466
+ "max_steps": 267952,
3467
+ "num_input_tokens_seen": 0,
3468
+ "num_train_epochs": 1,
3469
+ "save_steps": 10000,
3470
+ "total_flos": 4.480871972732928e+18,
3471
+ "train_batch_size": 32,
3472
+ "trial_name": null,
3473
+ "trial_params": null
3474
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3b4bfd01d070f54bd77067d53791fface2d0ea02e710fa7be9a4534de87b1fb6
3
+ size 4728