Hugging Face's logo

jonasknobloch
/
gpt2_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
6c17a3a
•
1 Parent(s): 580bab5

Initial commit

Browse files
Files changed (12) hide show
  1. README.md +94 -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 +3234 -0
  12. training_args.bin +3 -0
README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - generated_from_trainer
4
+ datasets:
5
+ - uonlp/CulturaX
6
+ metrics:
7
+ - accuracy
8
+ model-index:
9
+ - name: gpt2_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: 0.36165373273858764
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_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: 3.5834
32
+ - Accuracy: 0.3617
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
+ | 4.586 | 0.04 | 10000 | 4.4977 | 0.2821 |
64
+ | 4.2706 | 0.08 | 20000 | 4.1928 | 0.3058 |
65
+ | 4.1189 | 0.12 | 30000 | 4.0469 | 0.3179 |
66
+ | 4.0314 | 0.16 | 40000 | 3.9610 | 0.3253 |
67
+ | 3.9704 | 0.2 | 50000 | 3.8977 | 0.3311 |
68
+ | 3.923 | 0.24 | 60000 | 3.8486 | 0.3353 |
69
+ | 3.888 | 0.28 | 70000 | 3.8084 | 0.3390 |
70
+ | 3.8529 | 0.32 | 80000 | 3.7777 | 0.3423 |
71
+ | 3.832 | 0.36 | 90000 | 3.7526 | 0.3446 |
72
+ | 3.8102 | 0.4 | 100000 | 3.7277 | 0.3470 |
73
+ | 3.7876 | 0.44 | 110000 | 3.7073 | 0.3490 |
74
+ | 3.7686 | 0.48 | 120000 | 3.6922 | 0.3506 |
75
+ | 3.7585 | 0.52 | 130000 | 3.6750 | 0.3522 |
76
+ | 3.7459 | 0.56 | 140000 | 3.6620 | 0.3535 |
77
+ | 3.7378 | 0.6 | 150000 | 3.6501 | 0.3545 |
78
+ | 3.7181 | 0.64 | 160000 | 3.6385 | 0.3559 |
79
+ | 3.7139 | 0.68 | 170000 | 3.6293 | 0.3568 |
80
+ | 3.6958 | 0.72 | 180000 | 3.6201 | 0.3578 |
81
+ | 3.6872 | 0.76 | 190000 | 3.6122 | 0.3585 |
82
+ | 3.6888 | 0.8 | 200000 | 3.6060 | 0.3592 |
83
+ | 3.6765 | 0.84 | 210000 | 3.6001 | 0.3599 |
84
+ | 3.6734 | 0.88 | 220000 | 3.5945 | 0.3604 |
85
+ | 3.6669 | 0.92 | 230000 | 3.5891 | 0.3611 |
86
+ | 3.6696 | 0.96 | 240000 | 3.5856 | 0.3614 |
87
+
88
+
89
+ ### Framework versions
90
+
91
+ - Transformers 4.37.1
92
+ - Pytorch 2.1.2+cu121
93
+ - Datasets 2.16.1
94
+ - Tokenizers 0.15.1
all_results.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "eval_accuracy": 0.36165373273858764,
4
+ "eval_loss": 3.583437442779541,
5
+ "eval_runtime": 4627.9586,
6
+ "eval_samples": 419639,
7
+ "eval_samples_per_second": 90.675,
8
+ "eval_steps_per_second": 1.417,
9
+ "perplexity": 35.99706627518808,
10
+ "train_loss": 3.8837062227938306,
11
+ "train_runtime": 230124.717,
12
+ "train_samples": 7975299,
13
+ "train_samples_per_second": 34.656,
14
+ "train_steps_per_second": 1.083
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": 0.36165373273858764,
4
+ "eval_loss": 3.583437442779541,
5
+ "eval_runtime": 4627.9586,
6
+ "eval_samples": 419639,
7
+ "eval_samples_per_second": 90.675,
8
+ "eval_steps_per_second": 1.417,
9
+ "perplexity": 35.99706627518808
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:dff5250c9649d1f253392a94130366d6cd26647af25be271a40c60b5aafa1ac3
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": 3.8837062227938306,
4
+ "train_runtime": 230124.717,
5
+ "train_samples": 7975299,
6
+ "train_samples_per_second": 34.656,
7
+ "train_steps_per_second": 1.083
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,3234 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 10000,
6
+ "global_step": 249229,
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": 4.989969064595212e-05,
14
+ "loss": 7.3065,
15
+ "step": 500
16
+ },
17
+ {
18
+ "epoch": 0.0,
19
+ "learning_rate": 4.9799381291904234e-05,
20
+ "loss": 6.5187,
21
+ "step": 1000
22
+ },
23
+ {
24
+ "epoch": 0.01,
25
+ "learning_rate": 4.9699071937856354e-05,
26
+ "loss": 6.1861,
27
+ "step": 1500
28
+ },
29
+ {
30
+ "epoch": 0.01,
31
+ "learning_rate": 4.9598762583808466e-05,
32
+ "loss": 5.9496,
33
+ "step": 2000
34
+ },
35
+ {
36
+ "epoch": 0.01,
37
+ "learning_rate": 4.9498453229760586e-05,
38
+ "loss": 5.7513,
39
+ "step": 2500
40
+ },
41
+ {
42
+ "epoch": 0.01,
43
+ "learning_rate": 4.93981438757127e-05,
44
+ "loss": 5.5881,
45
+ "step": 3000
46
+ },
47
+ {
48
+ "epoch": 0.01,
49
+ "learning_rate": 4.929783452166482e-05,
50
+ "loss": 5.4459,
51
+ "step": 3500
52
+ },
53
+ {
54
+ "epoch": 0.02,
55
+ "learning_rate": 4.919752516761693e-05,
56
+ "loss": 5.3082,
57
+ "step": 4000
58
+ },
59
+ {
60
+ "epoch": 0.02,
61
+ "learning_rate": 4.909721581356905e-05,
62
+ "loss": 5.1931,
63
+ "step": 4500
64
+ },
65
+ {
66
+ "epoch": 0.02,
67
+ "learning_rate": 4.899690645952116e-05,
68
+ "loss": 5.0948,
69
+ "step": 5000
70
+ },
71
+ {
72
+ "epoch": 0.02,
73
+ "learning_rate": 4.889659710547328e-05,
74
+ "loss": 5.0116,
75
+ "step": 5500
76
+ },
77
+ {
78
+ "epoch": 0.02,
79
+ "learning_rate": 4.8796287751425394e-05,
80
+ "loss": 4.9386,
81
+ "step": 6000
82
+ },
83
+ {
84
+ "epoch": 0.03,
85
+ "learning_rate": 4.8695978397377514e-05,
86
+ "loss": 4.8745,
87
+ "step": 6500
88
+ },
89
+ {
90
+ "epoch": 0.03,
91
+ "learning_rate": 4.859566904332963e-05,
92
+ "loss": 4.8234,
93
+ "step": 7000
94
+ },
95
+ {
96
+ "epoch": 0.03,
97
+ "learning_rate": 4.849535968928175e-05,
98
+ "loss": 4.7714,
99
+ "step": 7500
100
+ },
101
+ {
102
+ "epoch": 0.03,
103
+ "learning_rate": 4.8395050335233865e-05,
104
+ "loss": 4.7328,
105
+ "step": 8000
106
+ },
107
+ {
108
+ "epoch": 0.03,
109
+ "learning_rate": 4.8294740981185984e-05,
110
+ "loss": 4.6888,
111
+ "step": 8500
112
+ },
113
+ {
114
+ "epoch": 0.04,
115
+ "learning_rate": 4.81944316271381e-05,
116
+ "loss": 4.6569,
117
+ "step": 9000
118
+ },
119
+ {
120
+ "epoch": 0.04,
121
+ "learning_rate": 4.8094122273090216e-05,
122
+ "loss": 4.6302,
123
+ "step": 9500
124
+ },
125
+ {
126
+ "epoch": 0.04,
127
+ "learning_rate": 4.799381291904233e-05,
128
+ "loss": 4.586,
129
+ "step": 10000
130
+ },
131
+ {
132
+ "epoch": 0.04,
133
+ "eval_accuracy": 0.2821468712144023,
134
+ "eval_loss": 4.497698783874512,
135
+ "eval_runtime": 4655.4137,
136
+ "eval_samples_per_second": 90.14,
137
+ "eval_steps_per_second": 1.408,
138
+ "step": 10000
139
+ },
140
+ {
141
+ "epoch": 0.04,
142
+ "learning_rate": 4.789350356499445e-05,
143
+ "loss": 4.5693,
144
+ "step": 10500
145
+ },
146
+ {
147
+ "epoch": 0.04,
148
+ "learning_rate": 4.779319421094656e-05,
149
+ "loss": 4.5425,
150
+ "step": 11000
151
+ },
152
+ {
153
+ "epoch": 0.05,
154
+ "learning_rate": 4.769288485689868e-05,
155
+ "loss": 4.5203,
156
+ "step": 11500
157
+ },
158
+ {
159
+ "epoch": 0.05,
160
+ "learning_rate": 4.759257550285079e-05,
161
+ "loss": 4.4994,
162
+ "step": 12000
163
+ },
164
+ {
165
+ "epoch": 0.05,
166
+ "learning_rate": 4.749226614880291e-05,
167
+ "loss": 4.4772,
168
+ "step": 12500
169
+ },
170
+ {
171
+ "epoch": 0.05,
172
+ "learning_rate": 4.7391956794755025e-05,
173
+ "loss": 4.4553,
174
+ "step": 13000
175
+ },
176
+ {
177
+ "epoch": 0.05,
178
+ "learning_rate": 4.7291647440707144e-05,
179
+ "loss": 4.4413,
180
+ "step": 13500
181
+ },
182
+ {
183
+ "epoch": 0.06,
184
+ "learning_rate": 4.719133808665926e-05,
185
+ "loss": 4.4206,
186
+ "step": 14000
187
+ },
188
+ {
189
+ "epoch": 0.06,
190
+ "learning_rate": 4.7091028732611376e-05,
191
+ "loss": 4.4077,
192
+ "step": 14500
193
+ },
194
+ {
195
+ "epoch": 0.06,
196
+ "learning_rate": 4.699071937856349e-05,
197
+ "loss": 4.3925,
198
+ "step": 15000
199
+ },
200
+ {
201
+ "epoch": 0.06,
202
+ "learning_rate": 4.689041002451561e-05,
203
+ "loss": 4.3732,
204
+ "step": 15500
205
+ },
206
+ {
207
+ "epoch": 0.06,
208
+ "learning_rate": 4.679010067046772e-05,
209
+ "loss": 4.3648,
210
+ "step": 16000
211
+ },
212
+ {
213
+ "epoch": 0.07,
214
+ "learning_rate": 4.668979131641984e-05,
215
+ "loss": 4.3451,
216
+ "step": 16500
217
+ },
218
+ {
219
+ "epoch": 0.07,
220
+ "learning_rate": 4.658948196237195e-05,
221
+ "loss": 4.3398,
222
+ "step": 17000
223
+ },
224
+ {
225
+ "epoch": 0.07,
226
+ "learning_rate": 4.648917260832407e-05,
227
+ "loss": 4.3228,
228
+ "step": 17500
229
+ },
230
+ {
231
+ "epoch": 0.07,
232
+ "learning_rate": 4.6388863254276185e-05,
233
+ "loss": 4.317,
234
+ "step": 18000
235
+ },
236
+ {
237
+ "epoch": 0.07,
238
+ "learning_rate": 4.6288553900228304e-05,
239
+ "loss": 4.3033,
240
+ "step": 18500
241
+ },
242
+ {
243
+ "epoch": 0.08,
244
+ "learning_rate": 4.618824454618042e-05,
245
+ "loss": 4.2927,
246
+ "step": 19000
247
+ },
248
+ {
249
+ "epoch": 0.08,
250
+ "learning_rate": 4.6087935192132536e-05,
251
+ "loss": 4.2853,
252
+ "step": 19500
253
+ },
254
+ {
255
+ "epoch": 0.08,
256
+ "learning_rate": 4.5987625838084656e-05,
257
+ "loss": 4.2706,
258
+ "step": 20000
259
+ },
260
+ {
261
+ "epoch": 0.08,
262
+ "eval_accuracy": 0.3058443984869302,
263
+ "eval_loss": 4.192770004272461,
264
+ "eval_runtime": 4658.7505,
265
+ "eval_samples_per_second": 90.075,
266
+ "eval_steps_per_second": 1.407,
267
+ "step": 20000
268
+ },
269
+ {
270
+ "epoch": 0.08,
271
+ "learning_rate": 4.5887316484036775e-05,
272
+ "loss": 4.2613,
273
+ "step": 20500
274
+ },
275
+ {
276
+ "epoch": 0.08,
277
+ "learning_rate": 4.578700712998889e-05,
278
+ "loss": 4.255,
279
+ "step": 21000
280
+ },
281
+ {
282
+ "epoch": 0.09,
283
+ "learning_rate": 4.568669777594101e-05,
284
+ "loss": 4.2422,
285
+ "step": 21500
286
+ },
287
+ {
288
+ "epoch": 0.09,
289
+ "learning_rate": 4.558638842189312e-05,
290
+ "loss": 4.2364,
291
+ "step": 22000
292
+ },
293
+ {
294
+ "epoch": 0.09,
295
+ "learning_rate": 4.548607906784524e-05,
296
+ "loss": 4.2239,
297
+ "step": 22500
298
+ },
299
+ {
300
+ "epoch": 0.09,
301
+ "learning_rate": 4.538576971379735e-05,
302
+ "loss": 4.2258,
303
+ "step": 23000
304
+ },
305
+ {
306
+ "epoch": 0.09,
307
+ "learning_rate": 4.528546035974947e-05,
308
+ "loss": 4.2067,
309
+ "step": 23500
310
+ },
311
+ {
312
+ "epoch": 0.1,
313
+ "learning_rate": 4.5185151005701584e-05,
314
+ "loss": 4.2101,
315
+ "step": 24000
316
+ },
317
+ {
318
+ "epoch": 0.1,
319
+ "learning_rate": 4.50848416516537e-05,
320
+ "loss": 4.1927,
321
+ "step": 24500
322
+ },
323
+ {
324
+ "epoch": 0.1,
325
+ "learning_rate": 4.498453229760582e-05,
326
+ "loss": 4.1868,
327
+ "step": 25000
328
+ },
329
+ {
330
+ "epoch": 0.1,
331
+ "learning_rate": 4.4884222943557935e-05,
332
+ "loss": 4.1779,
333
+ "step": 25500
334
+ },
335
+ {
336
+ "epoch": 0.1,
337
+ "learning_rate": 4.4783913589510054e-05,
338
+ "loss": 4.1711,
339
+ "step": 26000
340
+ },
341
+ {
342
+ "epoch": 0.11,
343
+ "learning_rate": 4.468360423546217e-05,
344
+ "loss": 4.166,
345
+ "step": 26500
346
+ },
347
+ {
348
+ "epoch": 0.11,
349
+ "learning_rate": 4.4583294881414286e-05,
350
+ "loss": 4.1639,
351
+ "step": 27000
352
+ },
353
+ {
354
+ "epoch": 0.11,
355
+ "learning_rate": 4.44829855273664e-05,
356
+ "loss": 4.1586,
357
+ "step": 27500
358
+ },
359
+ {
360
+ "epoch": 0.11,
361
+ "learning_rate": 4.438267617331852e-05,
362
+ "loss": 4.1508,
363
+ "step": 28000
364
+ },
365
+ {
366
+ "epoch": 0.11,
367
+ "learning_rate": 4.428236681927063e-05,
368
+ "loss": 4.1471,
369
+ "step": 28500
370
+ },
371
+ {
372
+ "epoch": 0.12,
373
+ "learning_rate": 4.418205746522275e-05,
374
+ "loss": 4.1396,
375
+ "step": 29000
376
+ },
377
+ {
378
+ "epoch": 0.12,
379
+ "learning_rate": 4.408174811117486e-05,
380
+ "loss": 4.1354,
381
+ "step": 29500
382
+ },
383
+ {
384
+ "epoch": 0.12,
385
+ "learning_rate": 4.398143875712698e-05,
386
+ "loss": 4.1189,
387
+ "step": 30000
388
+ },
389
+ {
390
+ "epoch": 0.12,
391
+ "eval_accuracy": 0.31791789096235645,
392
+ "eval_loss": 4.0469441413879395,
393
+ "eval_runtime": 4672.6798,
394
+ "eval_samples_per_second": 89.807,
395
+ "eval_steps_per_second": 1.403,
396
+ "step": 30000
397
+ },
398
+ {
399
+ "epoch": 0.12,
400
+ "learning_rate": 4.3881129403079095e-05,
401
+ "loss": 4.1212,
402
+ "step": 30500
403
+ },
404
+ {
405
+ "epoch": 0.12,
406
+ "learning_rate": 4.3780820049031214e-05,
407
+ "loss": 4.1141,
408
+ "step": 31000
409
+ },
410
+ {
411
+ "epoch": 0.13,
412
+ "learning_rate": 4.368051069498333e-05,
413
+ "loss": 4.1145,
414
+ "step": 31500
415
+ },
416
+ {
417
+ "epoch": 0.13,
418
+ "learning_rate": 4.3580201340935446e-05,
419
+ "loss": 4.0995,
420
+ "step": 32000
421
+ },
422
+ {
423
+ "epoch": 0.13,
424
+ "learning_rate": 4.347989198688756e-05,
425
+ "loss": 4.1028,
426
+ "step": 32500
427
+ },
428
+ {
429
+ "epoch": 0.13,
430
+ "learning_rate": 4.337958263283968e-05,
431
+ "loss": 4.0968,
432
+ "step": 33000
433
+ },
434
+ {
435
+ "epoch": 0.13,
436
+ "learning_rate": 4.32792732787918e-05,
437
+ "loss": 4.0937,
438
+ "step": 33500
439
+ },
440
+ {
441
+ "epoch": 0.14,
442
+ "learning_rate": 4.317896392474392e-05,
443
+ "loss": 4.0851,
444
+ "step": 34000
445
+ },
446
+ {
447
+ "epoch": 0.14,
448
+ "learning_rate": 4.307865457069603e-05,
449
+ "loss": 4.0801,
450
+ "step": 34500
451
+ },
452
+ {
453
+ "epoch": 0.14,
454
+ "learning_rate": 4.297834521664815e-05,
455
+ "loss": 4.0803,
456
+ "step": 35000
457
+ },
458
+ {
459
+ "epoch": 0.14,
460
+ "learning_rate": 4.287803586260026e-05,
461
+ "loss": 4.0716,
462
+ "step": 35500
463
+ },
464
+ {
465
+ "epoch": 0.14,
466
+ "learning_rate": 4.277772650855238e-05,
467
+ "loss": 4.0719,
468
+ "step": 36000
469
+ },
470
+ {
471
+ "epoch": 0.15,
472
+ "learning_rate": 4.2677417154504494e-05,
473
+ "loss": 4.0665,
474
+ "step": 36500
475
+ },
476
+ {
477
+ "epoch": 0.15,
478
+ "learning_rate": 4.257710780045661e-05,
479
+ "loss": 4.0619,
480
+ "step": 37000
481
+ },
482
+ {
483
+ "epoch": 0.15,
484
+ "learning_rate": 4.2476798446408726e-05,
485
+ "loss": 4.0561,
486
+ "step": 37500
487
+ },
488
+ {
489
+ "epoch": 0.15,
490
+ "learning_rate": 4.2376489092360845e-05,
491
+ "loss": 4.0577,
492
+ "step": 38000
493
+ },
494
+ {
495
+ "epoch": 0.15,
496
+ "learning_rate": 4.227617973831296e-05,
497
+ "loss": 4.0542,
498
+ "step": 38500
499
+ },
500
+ {
501
+ "epoch": 0.16,
502
+ "learning_rate": 4.217587038426508e-05,
503
+ "loss": 4.0338,
504
+ "step": 39000
505
+ },
506
+ {
507
+ "epoch": 0.16,
508
+ "learning_rate": 4.207556103021719e-05,
509
+ "loss": 4.0415,
510
+ "step": 39500
511
+ },
512
+ {
513
+ "epoch": 0.16,
514
+ "learning_rate": 4.197525167616931e-05,
515
+ "loss": 4.0314,
516
+ "step": 40000
517
+ },
518
+ {
519
+ "epoch": 0.16,
520
+ "eval_accuracy": 0.3253092461027638,
521
+ "eval_loss": 3.96097469329834,
522
+ "eval_runtime": 4672.6679,
523
+ "eval_samples_per_second": 89.807,
524
+ "eval_steps_per_second": 1.403,
525
+ "step": 40000
526
+ },
527
+ {
528
+ "epoch": 0.16,
529
+ "learning_rate": 4.187494232212142e-05,
530
+ "loss": 4.0337,
531
+ "step": 40500
532
+ },
533
+ {
534
+ "epoch": 0.16,
535
+ "learning_rate": 4.177463296807354e-05,
536
+ "loss": 4.023,
537
+ "step": 41000
538
+ },
539
+ {
540
+ "epoch": 0.17,
541
+ "learning_rate": 4.1674323614025654e-05,
542
+ "loss": 4.0296,
543
+ "step": 41500
544
+ },
545
+ {
546
+ "epoch": 0.17,
547
+ "learning_rate": 4.157401425997777e-05,
548
+ "loss": 4.0172,
549
+ "step": 42000
550
+ },
551
+ {
552
+ "epoch": 0.17,
553
+ "learning_rate": 4.1473704905929886e-05,
554
+ "loss": 4.0191,
555
+ "step": 42500
556
+ },
557
+ {
558
+ "epoch": 0.17,
559
+ "learning_rate": 4.1373395551882005e-05,
560
+ "loss": 4.0218,
561
+ "step": 43000
562
+ },
563
+ {
564
+ "epoch": 0.17,
565
+ "learning_rate": 4.127308619783412e-05,
566
+ "loss": 4.0075,
567
+ "step": 43500
568
+ },
569
+ {
570
+ "epoch": 0.18,
571
+ "learning_rate": 4.117277684378624e-05,
572
+ "loss": 4.0105,
573
+ "step": 44000
574
+ },
575
+ {
576
+ "epoch": 0.18,
577
+ "learning_rate": 4.1072467489738356e-05,
578
+ "loss": 4.0073,
579
+ "step": 44500
580
+ },
581
+ {
582
+ "epoch": 0.18,
583
+ "learning_rate": 4.097215813569047e-05,
584
+ "loss": 3.9932,
585
+ "step": 45000
586
+ },
587
+ {
588
+ "epoch": 0.18,
589
+ "learning_rate": 4.087184878164259e-05,
590
+ "loss": 3.9959,
591
+ "step": 45500
592
+ },
593
+ {
594
+ "epoch": 0.18,
595
+ "learning_rate": 4.07715394275947e-05,
596
+ "loss": 3.9903,
597
+ "step": 46000
598
+ },
599
+ {
600
+ "epoch": 0.19,
601
+ "learning_rate": 4.067123007354682e-05,
602
+ "loss": 3.9957,
603
+ "step": 46500
604
+ },
605
+ {
606
+ "epoch": 0.19,
607
+ "learning_rate": 4.057092071949894e-05,
608
+ "loss": 3.9921,
609
+ "step": 47000
610
+ },
611
+ {
612
+ "epoch": 0.19,
613
+ "learning_rate": 4.047061136545105e-05,
614
+ "loss": 3.9841,
615
+ "step": 47500
616
+ },
617
+ {
618
+ "epoch": 0.19,
619
+ "learning_rate": 4.037030201140317e-05,
620
+ "loss": 3.987,
621
+ "step": 48000
622
+ },
623
+ {
624
+ "epoch": 0.19,
625
+ "learning_rate": 4.0269992657355284e-05,
626
+ "loss": 3.9793,
627
+ "step": 48500
628
+ },
629
+ {
630
+ "epoch": 0.2,
631
+ "learning_rate": 4.0169683303307404e-05,
632
+ "loss": 3.9824,
633
+ "step": 49000
634
+ },
635
+ {
636
+ "epoch": 0.2,
637
+ "learning_rate": 4.006937394925952e-05,
638
+ "loss": 3.9745,
639
+ "step": 49500
640
+ },
641
+ {
642
+ "epoch": 0.2,
643
+ "learning_rate": 3.9969064595211636e-05,
644
+ "loss": 3.9704,
645
+ "step": 50000
646
+ },
647
+ {
648
+ "epoch": 0.2,
649
+ "eval_accuracy": 0.33107534589784043,
650
+ "eval_loss": 3.897735834121704,
651
+ "eval_runtime": 4799.5493,
652
+ "eval_samples_per_second": 87.433,
653
+ "eval_steps_per_second": 1.366,
654
+ "step": 50000
655
+ },
656
+ {
657
+ "epoch": 0.2,
658
+ "learning_rate": 3.9868755241163755e-05,
659
+ "loss": 3.9677,
660
+ "step": 50500
661
+ },
662
+ {
663
+ "epoch": 0.2,
664
+ "learning_rate": 3.976844588711587e-05,
665
+ "loss": 3.9727,
666
+ "step": 51000
667
+ },
668
+ {
669
+ "epoch": 0.21,
670
+ "learning_rate": 3.966813653306799e-05,
671
+ "loss": 3.9714,
672
+ "step": 51500
673
+ },
674
+ {
675
+ "epoch": 0.21,
676
+ "learning_rate": 3.95678271790201e-05,
677
+ "loss": 3.9605,
678
+ "step": 52000
679
+ },
680
+ {
681
+ "epoch": 0.21,
682
+ "learning_rate": 3.946751782497222e-05,
683
+ "loss": 3.9584,
684
+ "step": 52500
685
+ },
686
+ {
687
+ "epoch": 0.21,
688
+ "learning_rate": 3.936720847092433e-05,
689
+ "loss": 3.9582,
690
+ "step": 53000
691
+ },
692
+ {
693
+ "epoch": 0.21,
694
+ "learning_rate": 3.926689911687645e-05,
695
+ "loss": 3.9571,
696
+ "step": 53500
697
+ },
698
+ {
699
+ "epoch": 0.22,
700
+ "learning_rate": 3.9166589762828564e-05,
701
+ "loss": 3.9517,
702
+ "step": 54000
703
+ },
704
+ {
705
+ "epoch": 0.22,
706
+ "learning_rate": 3.906628040878068e-05,
707
+ "loss": 3.9453,
708
+ "step": 54500
709
+ },
710
+ {
711
+ "epoch": 0.22,
712
+ "learning_rate": 3.8965971054732796e-05,
713
+ "loss": 3.952,
714
+ "step": 55000
715
+ },
716
+ {
717
+ "epoch": 0.22,
718
+ "learning_rate": 3.8865661700684915e-05,
719
+ "loss": 3.9473,
720
+ "step": 55500
721
+ },
722
+ {
723
+ "epoch": 0.22,
724
+ "learning_rate": 3.876535234663703e-05,
725
+ "loss": 3.9465,
726
+ "step": 56000
727
+ },
728
+ {
729
+ "epoch": 0.23,
730
+ "learning_rate": 3.866504299258915e-05,
731
+ "loss": 3.9409,
732
+ "step": 56500
733
+ },
734
+ {
735
+ "epoch": 0.23,
736
+ "learning_rate": 3.856473363854126e-05,
737
+ "loss": 3.9323,
738
+ "step": 57000
739
+ },
740
+ {
741
+ "epoch": 0.23,
742
+ "learning_rate": 3.846442428449338e-05,
743
+ "loss": 3.9435,
744
+ "step": 57500
745
+ },
746
+ {
747
+ "epoch": 0.23,
748
+ "learning_rate": 3.836411493044549e-05,
749
+ "loss": 3.9361,
750
+ "step": 58000
751
+ },
752
+ {
753
+ "epoch": 0.23,
754
+ "learning_rate": 3.826380557639761e-05,
755
+ "loss": 3.9332,
756
+ "step": 58500
757
+ },
758
+ {
759
+ "epoch": 0.24,
760
+ "learning_rate": 3.8163496222349724e-05,
761
+ "loss": 3.9293,
762
+ "step": 59000
763
+ },
764
+ {
765
+ "epoch": 0.24,
766
+ "learning_rate": 3.806318686830184e-05,
767
+ "loss": 3.9302,
768
+ "step": 59500
769
+ },
770
+ {
771
+ "epoch": 0.24,
772
+ "learning_rate": 3.7962877514253956e-05,
773
+ "loss": 3.923,
774
+ "step": 60000
775
+ },
776
+ {
777
+ "epoch": 0.24,
778
+ "eval_accuracy": 0.33527885650874006,
779
+ "eval_loss": 3.848573923110962,
780
+ "eval_runtime": 4791.4499,
781
+ "eval_samples_per_second": 87.581,
782
+ "eval_steps_per_second": 1.368,
783
+ "step": 60000
784
+ },
785
+ {
786
+ "epoch": 0.24,
787
+ "learning_rate": 3.786256816020608e-05,
788
+ "loss": 3.9301,
789
+ "step": 60500
790
+ },
791
+ {
792
+ "epoch": 0.24,
793
+ "learning_rate": 3.7762258806158194e-05,
794
+ "loss": 3.9248,
795
+ "step": 61000
796
+ },
797
+ {
798
+ "epoch": 0.25,
799
+ "learning_rate": 3.7661949452110314e-05,
800
+ "loss": 3.915,
801
+ "step": 61500
802
+ },
803
+ {
804
+ "epoch": 0.25,
805
+ "learning_rate": 3.7561640098062426e-05,
806
+ "loss": 3.9202,
807
+ "step": 62000
808
+ },
809
+ {
810
+ "epoch": 0.25,
811
+ "learning_rate": 3.7461330744014546e-05,
812
+ "loss": 3.9194,
813
+ "step": 62500
814
+ },
815
+ {
816
+ "epoch": 0.25,
817
+ "learning_rate": 3.736102138996666e-05,
818
+ "loss": 3.9193,
819
+ "step": 63000
820
+ },
821
+ {
822
+ "epoch": 0.25,
823
+ "learning_rate": 3.726071203591878e-05,
824
+ "loss": 3.9076,
825
+ "step": 63500
826
+ },
827
+ {
828
+ "epoch": 0.26,
829
+ "learning_rate": 3.716040268187089e-05,
830
+ "loss": 3.9139,
831
+ "step": 64000
832
+ },
833
+ {
834
+ "epoch": 0.26,
835
+ "learning_rate": 3.706009332782301e-05,
836
+ "loss": 3.9101,
837
+ "step": 64500
838
+ },
839
+ {
840
+ "epoch": 0.26,
841
+ "learning_rate": 3.695978397377512e-05,
842
+ "loss": 3.9046,
843
+ "step": 65000
844
+ },
845
+ {
846
+ "epoch": 0.26,
847
+ "learning_rate": 3.685947461972724e-05,
848
+ "loss": 3.9086,
849
+ "step": 65500
850
+ },
851
+ {
852
+ "epoch": 0.26,
853
+ "learning_rate": 3.6759165265679354e-05,
854
+ "loss": 3.9011,
855
+ "step": 66000
856
+ },
857
+ {
858
+ "epoch": 0.27,
859
+ "learning_rate": 3.6658855911631474e-05,
860
+ "loss": 3.901,
861
+ "step": 66500
862
+ },
863
+ {
864
+ "epoch": 0.27,
865
+ "learning_rate": 3.6558546557583586e-05,
866
+ "loss": 3.9051,
867
+ "step": 67000
868
+ },
869
+ {
870
+ "epoch": 0.27,
871
+ "learning_rate": 3.6458237203535706e-05,
872
+ "loss": 3.8936,
873
+ "step": 67500
874
+ },
875
+ {
876
+ "epoch": 0.27,
877
+ "learning_rate": 3.6357927849487825e-05,
878
+ "loss": 3.8905,
879
+ "step": 68000
880
+ },
881
+ {
882
+ "epoch": 0.27,
883
+ "learning_rate": 3.625761849543994e-05,
884
+ "loss": 3.8844,
885
+ "step": 68500
886
+ },
887
+ {
888
+ "epoch": 0.28,
889
+ "learning_rate": 3.615730914139206e-05,
890
+ "loss": 3.8948,
891
+ "step": 69000
892
+ },
893
+ {
894
+ "epoch": 0.28,
895
+ "learning_rate": 3.605699978734417e-05,
896
+ "loss": 3.8902,
897
+ "step": 69500
898
+ },
899
+ {
900
+ "epoch": 0.28,
901
+ "learning_rate": 3.595669043329629e-05,
902
+ "loss": 3.888,
903
+ "step": 70000
904
+ },
905
+ {
906
+ "epoch": 0.28,
907
+ "eval_accuracy": 0.33901861842582626,
908
+ "eval_loss": 3.808432102203369,
909
+ "eval_runtime": 4676.4302,
910
+ "eval_samples_per_second": 89.735,
911
+ "eval_steps_per_second": 1.402,
912
+ "step": 70000
913
+ },
914
+ {
915
+ "epoch": 0.28,
916
+ "learning_rate": 3.58563810792484e-05,
917
+ "loss": 3.8876,
918
+ "step": 70500
919
+ },
920
+ {
921
+ "epoch": 0.28,
922
+ "learning_rate": 3.575607172520052e-05,
923
+ "loss": 3.8866,
924
+ "step": 71000
925
+ },
926
+ {
927
+ "epoch": 0.29,
928
+ "learning_rate": 3.5655762371152634e-05,
929
+ "loss": 3.8802,
930
+ "step": 71500
931
+ },
932
+ {
933
+ "epoch": 0.29,
934
+ "learning_rate": 3.555545301710475e-05,
935
+ "loss": 3.8877,
936
+ "step": 72000
937
+ },
938
+ {
939
+ "epoch": 0.29,
940
+ "learning_rate": 3.5455143663056866e-05,
941
+ "loss": 3.881,
942
+ "step": 72500
943
+ },
944
+ {
945
+ "epoch": 0.29,
946
+ "learning_rate": 3.5354834309008985e-05,
947
+ "loss": 3.8813,
948
+ "step": 73000
949
+ },
950
+ {
951
+ "epoch": 0.29,
952
+ "learning_rate": 3.52545249549611e-05,
953
+ "loss": 3.8704,
954
+ "step": 73500
955
+ },
956
+ {
957
+ "epoch": 0.3,
958
+ "learning_rate": 3.5154215600913224e-05,
959
+ "loss": 3.8745,
960
+ "step": 74000
961
+ },
962
+ {
963
+ "epoch": 0.3,
964
+ "learning_rate": 3.5053906246865336e-05,
965
+ "loss": 3.8739,
966
+ "step": 74500
967
+ },
968
+ {
969
+ "epoch": 0.3,
970
+ "learning_rate": 3.4953596892817456e-05,
971
+ "loss": 3.8779,
972
+ "step": 75000
973
+ },
974
+ {
975
+ "epoch": 0.3,
976
+ "learning_rate": 3.485328753876957e-05,
977
+ "loss": 3.8753,
978
+ "step": 75500
979
+ },
980
+ {
981
+ "epoch": 0.3,
982
+ "learning_rate": 3.475297818472169e-05,
983
+ "loss": 3.8621,
984
+ "step": 76000
985
+ },
986
+ {
987
+ "epoch": 0.31,
988
+ "learning_rate": 3.46526688306738e-05,
989
+ "loss": 3.8634,
990
+ "step": 76500
991
+ },
992
+ {
993
+ "epoch": 0.31,
994
+ "learning_rate": 3.455235947662592e-05,
995
+ "loss": 3.8675,
996
+ "step": 77000
997
+ },
998
+ {
999
+ "epoch": 0.31,
1000
+ "learning_rate": 3.445205012257803e-05,
1001
+ "loss": 3.8672,
1002
+ "step": 77500
1003
+ },
1004
+ {
1005
+ "epoch": 0.31,
1006
+ "learning_rate": 3.435174076853015e-05,
1007
+ "loss": 3.8639,
1008
+ "step": 78000
1009
+ },
1010
+ {
1011
+ "epoch": 0.31,
1012
+ "learning_rate": 3.4251431414482264e-05,
1013
+ "loss": 3.8547,
1014
+ "step": 78500
1015
+ },
1016
+ {
1017
+ "epoch": 0.32,
1018
+ "learning_rate": 3.4151122060434384e-05,
1019
+ "loss": 3.8573,
1020
+ "step": 79000
1021
+ },
1022
+ {
1023
+ "epoch": 0.32,
1024
+ "learning_rate": 3.4050812706386496e-05,
1025
+ "loss": 3.8516,
1026
+ "step": 79500
1027
+ },
1028
+ {
1029
+ "epoch": 0.32,
1030
+ "learning_rate": 3.3950503352338616e-05,
1031
+ "loss": 3.8529,
1032
+ "step": 80000
1033
+ },
1034
+ {
1035
+ "epoch": 0.32,
1036
+ "eval_accuracy": 0.3423139332553484,
1037
+ "eval_loss": 3.7777137756347656,
1038
+ "eval_runtime": 4683.7605,
1039
+ "eval_samples_per_second": 89.594,
1040
+ "eval_steps_per_second": 1.4,
1041
+ "step": 80000
1042
+ },
1043
+ {
1044
+ "epoch": 0.32,
1045
+ "learning_rate": 3.385019399829073e-05,
1046
+ "loss": 3.8512,
1047
+ "step": 80500
1048
+ },
1049
+ {
1050
+ "epoch": 0.33,
1051
+ "learning_rate": 3.374988464424285e-05,
1052
+ "loss": 3.8508,
1053
+ "step": 81000
1054
+ },
1055
+ {
1056
+ "epoch": 0.33,
1057
+ "learning_rate": 3.364957529019496e-05,
1058
+ "loss": 3.8484,
1059
+ "step": 81500
1060
+ },
1061
+ {
1062
+ "epoch": 0.33,
1063
+ "learning_rate": 3.354926593614708e-05,
1064
+ "loss": 3.8489,
1065
+ "step": 82000
1066
+ },
1067
+ {
1068
+ "epoch": 0.33,
1069
+ "learning_rate": 3.344895658209919e-05,
1070
+ "loss": 3.8471,
1071
+ "step": 82500
1072
+ },
1073
+ {
1074
+ "epoch": 0.33,
1075
+ "learning_rate": 3.334864722805131e-05,
1076
+ "loss": 3.8505,
1077
+ "step": 83000
1078
+ },
1079
+ {
1080
+ "epoch": 0.34,
1081
+ "learning_rate": 3.3248337874003424e-05,
1082
+ "loss": 3.839,
1083
+ "step": 83500
1084
+ },
1085
+ {
1086
+ "epoch": 0.34,
1087
+ "learning_rate": 3.3148028519955544e-05,
1088
+ "loss": 3.8493,
1089
+ "step": 84000
1090
+ },
1091
+ {
1092
+ "epoch": 0.34,
1093
+ "learning_rate": 3.3047719165907656e-05,
1094
+ "loss": 3.839,
1095
+ "step": 84500
1096
+ },
1097
+ {
1098
+ "epoch": 0.34,
1099
+ "learning_rate": 3.2947409811859776e-05,
1100
+ "loss": 3.8447,
1101
+ "step": 85000
1102
+ },
1103
+ {
1104
+ "epoch": 0.34,
1105
+ "learning_rate": 3.284710045781189e-05,
1106
+ "loss": 3.8401,
1107
+ "step": 85500
1108
+ },
1109
+ {
1110
+ "epoch": 0.35,
1111
+ "learning_rate": 3.274679110376401e-05,
1112
+ "loss": 3.8343,
1113
+ "step": 86000
1114
+ },
1115
+ {
1116
+ "epoch": 0.35,
1117
+ "learning_rate": 3.264648174971612e-05,
1118
+ "loss": 3.8379,
1119
+ "step": 86500
1120
+ },
1121
+ {
1122
+ "epoch": 0.35,
1123
+ "learning_rate": 3.2546172395668246e-05,
1124
+ "loss": 3.8378,
1125
+ "step": 87000
1126
+ },
1127
+ {
1128
+ "epoch": 0.35,
1129
+ "learning_rate": 3.244586304162036e-05,
1130
+ "loss": 3.8355,
1131
+ "step": 87500
1132
+ },
1133
+ {
1134
+ "epoch": 0.35,
1135
+ "learning_rate": 3.234555368757248e-05,
1136
+ "loss": 3.8363,
1137
+ "step": 88000
1138
+ },
1139
+ {
1140
+ "epoch": 0.36,
1141
+ "learning_rate": 3.224524433352459e-05,
1142
+ "loss": 3.8349,
1143
+ "step": 88500
1144
+ },
1145
+ {
1146
+ "epoch": 0.36,
1147
+ "learning_rate": 3.214493497947671e-05,
1148
+ "loss": 3.8364,
1149
+ "step": 89000
1150
+ },
1151
+ {
1152
+ "epoch": 0.36,
1153
+ "learning_rate": 3.204462562542882e-05,
1154
+ "loss": 3.8271,
1155
+ "step": 89500
1156
+ },
1157
+ {
1158
+ "epoch": 0.36,
1159
+ "learning_rate": 3.194431627138094e-05,
1160
+ "loss": 3.832,
1161
+ "step": 90000
1162
+ },
1163
+ {
1164
+ "epoch": 0.36,
1165
+ "eval_accuracy": 0.3446369162758726,
1166
+ "eval_loss": 3.752626419067383,
1167
+ "eval_runtime": 4676.2979,
1168
+ "eval_samples_per_second": 89.737,
1169
+ "eval_steps_per_second": 1.402,
1170
+ "step": 90000
1171
+ },
1172
+ {
1173
+ "epoch": 0.36,
1174
+ "learning_rate": 3.1844006917333055e-05,
1175
+ "loss": 3.8362,
1176
+ "step": 90500
1177
+ },
1178
+ {
1179
+ "epoch": 0.37,
1180
+ "learning_rate": 3.1743697563285174e-05,
1181
+ "loss": 3.8261,
1182
+ "step": 91000
1183
+ },
1184
+ {
1185
+ "epoch": 0.37,
1186
+ "learning_rate": 3.164338820923729e-05,
1187
+ "loss": 3.8246,
1188
+ "step": 91500
1189
+ },
1190
+ {
1191
+ "epoch": 0.37,
1192
+ "learning_rate": 3.1543078855189406e-05,
1193
+ "loss": 3.8274,
1194
+ "step": 92000
1195
+ },
1196
+ {
1197
+ "epoch": 0.37,
1198
+ "learning_rate": 3.1442769501141526e-05,
1199
+ "loss": 3.8221,
1200
+ "step": 92500
1201
+ },
1202
+ {
1203
+ "epoch": 0.37,
1204
+ "learning_rate": 3.134246014709364e-05,
1205
+ "loss": 3.823,
1206
+ "step": 93000
1207
+ },
1208
+ {
1209
+ "epoch": 0.38,
1210
+ "learning_rate": 3.124215079304576e-05,
1211
+ "loss": 3.8257,
1212
+ "step": 93500
1213
+ },
1214
+ {
1215
+ "epoch": 0.38,
1216
+ "learning_rate": 3.114184143899787e-05,
1217
+ "loss": 3.8206,
1218
+ "step": 94000
1219
+ },
1220
+ {
1221
+ "epoch": 0.38,
1222
+ "learning_rate": 3.104153208494999e-05,
1223
+ "loss": 3.8241,
1224
+ "step": 94500
1225
+ },
1226
+ {
1227
+ "epoch": 0.38,
1228
+ "learning_rate": 3.09412227309021e-05,
1229
+ "loss": 3.8211,
1230
+ "step": 95000
1231
+ },
1232
+ {
1233
+ "epoch": 0.38,
1234
+ "learning_rate": 3.084091337685422e-05,
1235
+ "loss": 3.8144,
1236
+ "step": 95500
1237
+ },
1238
+ {
1239
+ "epoch": 0.39,
1240
+ "learning_rate": 3.0740604022806334e-05,
1241
+ "loss": 3.8205,
1242
+ "step": 96000
1243
+ },
1244
+ {
1245
+ "epoch": 0.39,
1246
+ "learning_rate": 3.0640294668758454e-05,
1247
+ "loss": 3.8178,
1248
+ "step": 96500
1249
+ },
1250
+ {
1251
+ "epoch": 0.39,
1252
+ "learning_rate": 3.0539985314710566e-05,
1253
+ "loss": 3.8113,
1254
+ "step": 97000
1255
+ },
1256
+ {
1257
+ "epoch": 0.39,
1258
+ "learning_rate": 3.0439675960662682e-05,
1259
+ "loss": 3.8081,
1260
+ "step": 97500
1261
+ },
1262
+ {
1263
+ "epoch": 0.39,
1264
+ "learning_rate": 3.03393666066148e-05,
1265
+ "loss": 3.8163,
1266
+ "step": 98000
1267
+ },
1268
+ {
1269
+ "epoch": 0.4,
1270
+ "learning_rate": 3.0239057252566914e-05,
1271
+ "loss": 3.8163,
1272
+ "step": 98500
1273
+ },
1274
+ {
1275
+ "epoch": 0.4,
1276
+ "learning_rate": 3.013874789851903e-05,
1277
+ "loss": 3.808,
1278
+ "step": 99000
1279
+ },
1280
+ {
1281
+ "epoch": 0.4,
1282
+ "learning_rate": 3.003843854447115e-05,
1283
+ "loss": 3.8116,
1284
+ "step": 99500
1285
+ },
1286
+ {
1287
+ "epoch": 0.4,
1288
+ "learning_rate": 2.9938129190423266e-05,
1289
+ "loss": 3.8102,
1290
+ "step": 100000
1291
+ },
1292
+ {
1293
+ "epoch": 0.4,
1294
+ "eval_accuracy": 0.3469910693173023,
1295
+ "eval_loss": 3.7277164459228516,
1296
+ "eval_runtime": 4664.5454,
1297
+ "eval_samples_per_second": 89.964,
1298
+ "eval_steps_per_second": 1.406,
1299
+ "step": 100000
1300
+ },
1301
+ {
1302
+ "epoch": 0.4,
1303
+ "learning_rate": 2.9837819836375385e-05,
1304
+ "loss": 3.8106,
1305
+ "step": 100500
1306
+ },
1307
+ {
1308
+ "epoch": 0.41,
1309
+ "learning_rate": 2.97375104823275e-05,
1310
+ "loss": 3.8051,
1311
+ "step": 101000
1312
+ },
1313
+ {
1314
+ "epoch": 0.41,
1315
+ "learning_rate": 2.9637201128279617e-05,
1316
+ "loss": 3.8013,
1317
+ "step": 101500
1318
+ },
1319
+ {
1320
+ "epoch": 0.41,
1321
+ "learning_rate": 2.9536891774231733e-05,
1322
+ "loss": 3.8023,
1323
+ "step": 102000
1324
+ },
1325
+ {
1326
+ "epoch": 0.41,
1327
+ "learning_rate": 2.943658242018385e-05,
1328
+ "loss": 3.8013,
1329
+ "step": 102500
1330
+ },
1331
+ {
1332
+ "epoch": 0.41,
1333
+ "learning_rate": 2.9336273066135965e-05,
1334
+ "loss": 3.7951,
1335
+ "step": 103000
1336
+ },
1337
+ {
1338
+ "epoch": 0.42,
1339
+ "learning_rate": 2.923596371208808e-05,
1340
+ "loss": 3.806,
1341
+ "step": 103500
1342
+ },
1343
+ {
1344
+ "epoch": 0.42,
1345
+ "learning_rate": 2.9135654358040197e-05,
1346
+ "loss": 3.8028,
1347
+ "step": 104000
1348
+ },
1349
+ {
1350
+ "epoch": 0.42,
1351
+ "learning_rate": 2.9035345003992316e-05,
1352
+ "loss": 3.8002,
1353
+ "step": 104500
1354
+ },
1355
+ {
1356
+ "epoch": 0.42,
1357
+ "learning_rate": 2.8935035649944432e-05,
1358
+ "loss": 3.7978,
1359
+ "step": 105000
1360
+ },
1361
+ {
1362
+ "epoch": 0.42,
1363
+ "learning_rate": 2.883472629589655e-05,
1364
+ "loss": 3.7911,
1365
+ "step": 105500
1366
+ },
1367
+ {
1368
+ "epoch": 0.43,
1369
+ "learning_rate": 2.8734416941848664e-05,
1370
+ "loss": 3.792,
1371
+ "step": 106000
1372
+ },
1373
+ {
1374
+ "epoch": 0.43,
1375
+ "learning_rate": 2.863410758780078e-05,
1376
+ "loss": 3.7943,
1377
+ "step": 106500
1378
+ },
1379
+ {
1380
+ "epoch": 0.43,
1381
+ "learning_rate": 2.8533798233752896e-05,
1382
+ "loss": 3.788,
1383
+ "step": 107000
1384
+ },
1385
+ {
1386
+ "epoch": 0.43,
1387
+ "learning_rate": 2.8433488879705012e-05,
1388
+ "loss": 3.7955,
1389
+ "step": 107500
1390
+ },
1391
+ {
1392
+ "epoch": 0.43,
1393
+ "learning_rate": 2.833317952565713e-05,
1394
+ "loss": 3.7807,
1395
+ "step": 108000
1396
+ },
1397
+ {
1398
+ "epoch": 0.44,
1399
+ "learning_rate": 2.8232870171609244e-05,
1400
+ "loss": 3.7892,
1401
+ "step": 108500
1402
+ },
1403
+ {
1404
+ "epoch": 0.44,
1405
+ "learning_rate": 2.813256081756136e-05,
1406
+ "loss": 3.7892,
1407
+ "step": 109000
1408
+ },
1409
+ {
1410
+ "epoch": 0.44,
1411
+ "learning_rate": 2.8032251463513476e-05,
1412
+ "loss": 3.7916,
1413
+ "step": 109500
1414
+ },
1415
+ {
1416
+ "epoch": 0.44,
1417
+ "learning_rate": 2.7931942109465592e-05,
1418
+ "loss": 3.7876,
1419
+ "step": 110000
1420
+ },
1421
+ {
1422
+ "epoch": 0.44,
1423
+ "eval_accuracy": 0.34899459281783596,
1424
+ "eval_loss": 3.7072978019714355,
1425
+ "eval_runtime": 4653.1421,
1426
+ "eval_samples_per_second": 90.184,
1427
+ "eval_steps_per_second": 1.409,
1428
+ "step": 110000
1429
+ },
1430
+ {
1431
+ "epoch": 0.44,
1432
+ "learning_rate": 2.783163275541771e-05,
1433
+ "loss": 3.7845,
1434
+ "step": 110500
1435
+ },
1436
+ {
1437
+ "epoch": 0.45,
1438
+ "learning_rate": 2.7731323401369824e-05,
1439
+ "loss": 3.7859,
1440
+ "step": 111000
1441
+ },
1442
+ {
1443
+ "epoch": 0.45,
1444
+ "learning_rate": 2.763101404732194e-05,
1445
+ "loss": 3.7828,
1446
+ "step": 111500
1447
+ },
1448
+ {
1449
+ "epoch": 0.45,
1450
+ "learning_rate": 2.7530704693274056e-05,
1451
+ "loss": 3.7866,
1452
+ "step": 112000
1453
+ },
1454
+ {
1455
+ "epoch": 0.45,
1456
+ "learning_rate": 2.7430395339226172e-05,
1457
+ "loss": 3.7789,
1458
+ "step": 112500
1459
+ },
1460
+ {
1461
+ "epoch": 0.45,
1462
+ "learning_rate": 2.733008598517829e-05,
1463
+ "loss": 3.7793,
1464
+ "step": 113000
1465
+ },
1466
+ {
1467
+ "epoch": 0.46,
1468
+ "learning_rate": 2.7229776631130404e-05,
1469
+ "loss": 3.7869,
1470
+ "step": 113500
1471
+ },
1472
+ {
1473
+ "epoch": 0.46,
1474
+ "learning_rate": 2.7129467277082527e-05,
1475
+ "loss": 3.7817,
1476
+ "step": 114000
1477
+ },
1478
+ {
1479
+ "epoch": 0.46,
1480
+ "learning_rate": 2.7029157923034643e-05,
1481
+ "loss": 3.7724,
1482
+ "step": 114500
1483
+ },
1484
+ {
1485
+ "epoch": 0.46,
1486
+ "learning_rate": 2.692884856898676e-05,
1487
+ "loss": 3.7779,
1488
+ "step": 115000
1489
+ },
1490
+ {
1491
+ "epoch": 0.46,
1492
+ "learning_rate": 2.6828539214938875e-05,
1493
+ "loss": 3.7778,
1494
+ "step": 115500
1495
+ },
1496
+ {
1497
+ "epoch": 0.47,
1498
+ "learning_rate": 2.672822986089099e-05,
1499
+ "loss": 3.784,
1500
+ "step": 116000
1501
+ },
1502
+ {
1503
+ "epoch": 0.47,
1504
+ "learning_rate": 2.6627920506843107e-05,
1505
+ "loss": 3.7777,
1506
+ "step": 116500
1507
+ },
1508
+ {
1509
+ "epoch": 0.47,
1510
+ "learning_rate": 2.6527611152795223e-05,
1511
+ "loss": 3.7763,
1512
+ "step": 117000
1513
+ },
1514
+ {
1515
+ "epoch": 0.47,
1516
+ "learning_rate": 2.642730179874734e-05,
1517
+ "loss": 3.7693,
1518
+ "step": 117500
1519
+ },
1520
+ {
1521
+ "epoch": 0.47,
1522
+ "learning_rate": 2.6326992444699455e-05,
1523
+ "loss": 3.7743,
1524
+ "step": 118000
1525
+ },
1526
+ {
1527
+ "epoch": 0.48,
1528
+ "learning_rate": 2.622668309065157e-05,
1529
+ "loss": 3.771,
1530
+ "step": 118500
1531
+ },
1532
+ {
1533
+ "epoch": 0.48,
1534
+ "learning_rate": 2.6126373736603687e-05,
1535
+ "loss": 3.7796,
1536
+ "step": 119000
1537
+ },
1538
+ {
1539
+ "epoch": 0.48,
1540
+ "learning_rate": 2.6026064382555803e-05,
1541
+ "loss": 3.765,
1542
+ "step": 119500
1543
+ },
1544
+ {
1545
+ "epoch": 0.48,
1546
+ "learning_rate": 2.592575502850792e-05,
1547
+ "loss": 3.7686,
1548
+ "step": 120000
1549
+ },
1550
+ {
1551
+ "epoch": 0.48,
1552
+ "eval_accuracy": 0.35060705962607897,
1553
+ "eval_loss": 3.6921615600585938,
1554
+ "eval_runtime": 4753.6745,
1555
+ "eval_samples_per_second": 88.277,
1556
+ "eval_steps_per_second": 1.379,
1557
+ "step": 120000
1558
+ },
1559
+ {
1560
+ "epoch": 0.48,
1561
+ "learning_rate": 2.5825445674460035e-05,
1562
+ "loss": 3.7761,
1563
+ "step": 120500
1564
+ },
1565
+ {
1566
+ "epoch": 0.49,
1567
+ "learning_rate": 2.572513632041215e-05,
1568
+ "loss": 3.772,
1569
+ "step": 121000
1570
+ },
1571
+ {
1572
+ "epoch": 0.49,
1573
+ "learning_rate": 2.5624826966364267e-05,
1574
+ "loss": 3.7714,
1575
+ "step": 121500
1576
+ },
1577
+ {
1578
+ "epoch": 0.49,
1579
+ "learning_rate": 2.5524517612316383e-05,
1580
+ "loss": 3.7655,
1581
+ "step": 122000
1582
+ },
1583
+ {
1584
+ "epoch": 0.49,
1585
+ "learning_rate": 2.54242082582685e-05,
1586
+ "loss": 3.7657,
1587
+ "step": 122500
1588
+ },
1589
+ {
1590
+ "epoch": 0.49,
1591
+ "learning_rate": 2.5323898904220615e-05,
1592
+ "loss": 3.7732,
1593
+ "step": 123000
1594
+ },
1595
+ {
1596
+ "epoch": 0.5,
1597
+ "learning_rate": 2.522358955017273e-05,
1598
+ "loss": 3.7634,
1599
+ "step": 123500
1600
+ },
1601
+ {
1602
+ "epoch": 0.5,
1603
+ "learning_rate": 2.512328019612485e-05,
1604
+ "loss": 3.7609,
1605
+ "step": 124000
1606
+ },
1607
+ {
1608
+ "epoch": 0.5,
1609
+ "learning_rate": 2.5022970842076966e-05,
1610
+ "loss": 3.7604,
1611
+ "step": 124500
1612
+ },
1613
+ {
1614
+ "epoch": 0.5,
1615
+ "learning_rate": 2.4922661488029082e-05,
1616
+ "loss": 3.7602,
1617
+ "step": 125000
1618
+ },
1619
+ {
1620
+ "epoch": 0.5,
1621
+ "learning_rate": 2.48223521339812e-05,
1622
+ "loss": 3.7664,
1623
+ "step": 125500
1624
+ },
1625
+ {
1626
+ "epoch": 0.51,
1627
+ "learning_rate": 2.4722042779933314e-05,
1628
+ "loss": 3.7629,
1629
+ "step": 126000
1630
+ },
1631
+ {
1632
+ "epoch": 0.51,
1633
+ "learning_rate": 2.4621733425885434e-05,
1634
+ "loss": 3.7592,
1635
+ "step": 126500
1636
+ },
1637
+ {
1638
+ "epoch": 0.51,
1639
+ "learning_rate": 2.452142407183755e-05,
1640
+ "loss": 3.7627,
1641
+ "step": 127000
1642
+ },
1643
+ {
1644
+ "epoch": 0.51,
1645
+ "learning_rate": 2.4421114717789666e-05,
1646
+ "loss": 3.7622,
1647
+ "step": 127500
1648
+ },
1649
+ {
1650
+ "epoch": 0.51,
1651
+ "learning_rate": 2.4320805363741782e-05,
1652
+ "loss": 3.7569,
1653
+ "step": 128000
1654
+ },
1655
+ {
1656
+ "epoch": 0.52,
1657
+ "learning_rate": 2.4220496009693898e-05,
1658
+ "loss": 3.7576,
1659
+ "step": 128500
1660
+ },
1661
+ {
1662
+ "epoch": 0.52,
1663
+ "learning_rate": 2.4120186655646014e-05,
1664
+ "loss": 3.7599,
1665
+ "step": 129000
1666
+ },
1667
+ {
1668
+ "epoch": 0.52,
1669
+ "learning_rate": 2.401987730159813e-05,
1670
+ "loss": 3.764,
1671
+ "step": 129500
1672
+ },
1673
+ {
1674
+ "epoch": 0.52,
1675
+ "learning_rate": 2.3919567947550246e-05,
1676
+ "loss": 3.7585,
1677
+ "step": 130000
1678
+ },
1679
+ {
1680
+ "epoch": 0.52,
1681
+ "eval_accuracy": 0.3521617310984962,
1682
+ "eval_loss": 3.6750001907348633,
1683
+ "eval_runtime": 4751.082,
1684
+ "eval_samples_per_second": 88.325,
1685
+ "eval_steps_per_second": 1.38,
1686
+ "step": 130000
1687
+ },
1688
+ {
1689
+ "epoch": 0.52,
1690
+ "learning_rate": 2.3819258593502365e-05,
1691
+ "loss": 3.754,
1692
+ "step": 130500
1693
+ },
1694
+ {
1695
+ "epoch": 0.53,
1696
+ "learning_rate": 2.371894923945448e-05,
1697
+ "loss": 3.751,
1698
+ "step": 131000
1699
+ },
1700
+ {
1701
+ "epoch": 0.53,
1702
+ "learning_rate": 2.3618639885406597e-05,
1703
+ "loss": 3.7583,
1704
+ "step": 131500
1705
+ },
1706
+ {
1707
+ "epoch": 0.53,
1708
+ "learning_rate": 2.3518330531358713e-05,
1709
+ "loss": 3.7475,
1710
+ "step": 132000
1711
+ },
1712
+ {
1713
+ "epoch": 0.53,
1714
+ "learning_rate": 2.341802117731083e-05,
1715
+ "loss": 3.7542,
1716
+ "step": 132500
1717
+ },
1718
+ {
1719
+ "epoch": 0.53,
1720
+ "learning_rate": 2.3317711823262945e-05,
1721
+ "loss": 3.7468,
1722
+ "step": 133000
1723
+ },
1724
+ {
1725
+ "epoch": 0.54,
1726
+ "learning_rate": 2.321740246921506e-05,
1727
+ "loss": 3.746,
1728
+ "step": 133500
1729
+ },
1730
+ {
1731
+ "epoch": 0.54,
1732
+ "learning_rate": 2.3117093115167177e-05,
1733
+ "loss": 3.7527,
1734
+ "step": 134000
1735
+ },
1736
+ {
1737
+ "epoch": 0.54,
1738
+ "learning_rate": 2.3016783761119293e-05,
1739
+ "loss": 3.748,
1740
+ "step": 134500
1741
+ },
1742
+ {
1743
+ "epoch": 0.54,
1744
+ "learning_rate": 2.291647440707141e-05,
1745
+ "loss": 3.7517,
1746
+ "step": 135000
1747
+ },
1748
+ {
1749
+ "epoch": 0.54,
1750
+ "learning_rate": 2.2816165053023525e-05,
1751
+ "loss": 3.7453,
1752
+ "step": 135500
1753
+ },
1754
+ {
1755
+ "epoch": 0.55,
1756
+ "learning_rate": 2.271585569897564e-05,
1757
+ "loss": 3.751,
1758
+ "step": 136000
1759
+ },
1760
+ {
1761
+ "epoch": 0.55,
1762
+ "learning_rate": 2.2615546344927757e-05,
1763
+ "loss": 3.7542,
1764
+ "step": 136500
1765
+ },
1766
+ {
1767
+ "epoch": 0.55,
1768
+ "learning_rate": 2.2515236990879873e-05,
1769
+ "loss": 3.7535,
1770
+ "step": 137000
1771
+ },
1772
+ {
1773
+ "epoch": 0.55,
1774
+ "learning_rate": 2.2414927636831992e-05,
1775
+ "loss": 3.7499,
1776
+ "step": 137500
1777
+ },
1778
+ {
1779
+ "epoch": 0.55,
1780
+ "learning_rate": 2.231461828278411e-05,
1781
+ "loss": 3.7421,
1782
+ "step": 138000
1783
+ },
1784
+ {
1785
+ "epoch": 0.56,
1786
+ "learning_rate": 2.2214308928736224e-05,
1787
+ "loss": 3.7428,
1788
+ "step": 138500
1789
+ },
1790
+ {
1791
+ "epoch": 0.56,
1792
+ "learning_rate": 2.211399957468834e-05,
1793
+ "loss": 3.7429,
1794
+ "step": 139000
1795
+ },
1796
+ {
1797
+ "epoch": 0.56,
1798
+ "learning_rate": 2.2013690220640456e-05,
1799
+ "loss": 3.7469,
1800
+ "step": 139500
1801
+ },
1802
+ {
1803
+ "epoch": 0.56,
1804
+ "learning_rate": 2.1913380866592572e-05,
1805
+ "loss": 3.7459,
1806
+ "step": 140000
1807
+ },
1808
+ {
1809
+ "epoch": 0.56,
1810
+ "eval_accuracy": 0.35345456367995787,
1811
+ "eval_loss": 3.6619808673858643,
1812
+ "eval_runtime": 4644.9079,
1813
+ "eval_samples_per_second": 90.344,
1814
+ "eval_steps_per_second": 1.412,
1815
+ "step": 140000
1816
+ },
1817
+ {
1818
+ "epoch": 0.56,
1819
+ "learning_rate": 2.181307151254469e-05,
1820
+ "loss": 3.7399,
1821
+ "step": 140500
1822
+ },
1823
+ {
1824
+ "epoch": 0.57,
1825
+ "learning_rate": 2.1712762158496804e-05,
1826
+ "loss": 3.7392,
1827
+ "step": 141000
1828
+ },
1829
+ {
1830
+ "epoch": 0.57,
1831
+ "learning_rate": 2.161245280444892e-05,
1832
+ "loss": 3.7433,
1833
+ "step": 141500
1834
+ },
1835
+ {
1836
+ "epoch": 0.57,
1837
+ "learning_rate": 2.1512143450401036e-05,
1838
+ "loss": 3.7379,
1839
+ "step": 142000
1840
+ },
1841
+ {
1842
+ "epoch": 0.57,
1843
+ "learning_rate": 2.1411834096353152e-05,
1844
+ "loss": 3.7382,
1845
+ "step": 142500
1846
+ },
1847
+ {
1848
+ "epoch": 0.57,
1849
+ "learning_rate": 2.131152474230527e-05,
1850
+ "loss": 3.7347,
1851
+ "step": 143000
1852
+ },
1853
+ {
1854
+ "epoch": 0.58,
1855
+ "learning_rate": 2.1211215388257384e-05,
1856
+ "loss": 3.7424,
1857
+ "step": 143500
1858
+ },
1859
+ {
1860
+ "epoch": 0.58,
1861
+ "learning_rate": 2.1110906034209504e-05,
1862
+ "loss": 3.7371,
1863
+ "step": 144000
1864
+ },
1865
+ {
1866
+ "epoch": 0.58,
1867
+ "learning_rate": 2.101059668016162e-05,
1868
+ "loss": 3.7472,
1869
+ "step": 144500
1870
+ },
1871
+ {
1872
+ "epoch": 0.58,
1873
+ "learning_rate": 2.0910287326113736e-05,
1874
+ "loss": 3.7358,
1875
+ "step": 145000
1876
+ },
1877
+ {
1878
+ "epoch": 0.58,
1879
+ "learning_rate": 2.080997797206585e-05,
1880
+ "loss": 3.7351,
1881
+ "step": 145500
1882
+ },
1883
+ {
1884
+ "epoch": 0.59,
1885
+ "learning_rate": 2.0709668618017968e-05,
1886
+ "loss": 3.7308,
1887
+ "step": 146000
1888
+ },
1889
+ {
1890
+ "epoch": 0.59,
1891
+ "learning_rate": 2.0609359263970084e-05,
1892
+ "loss": 3.7387,
1893
+ "step": 146500
1894
+ },
1895
+ {
1896
+ "epoch": 0.59,
1897
+ "learning_rate": 2.05090499099222e-05,
1898
+ "loss": 3.7255,
1899
+ "step": 147000
1900
+ },
1901
+ {
1902
+ "epoch": 0.59,
1903
+ "learning_rate": 2.0408740555874316e-05,
1904
+ "loss": 3.7346,
1905
+ "step": 147500
1906
+ },
1907
+ {
1908
+ "epoch": 0.59,
1909
+ "learning_rate": 2.030843120182643e-05,
1910
+ "loss": 3.7399,
1911
+ "step": 148000
1912
+ },
1913
+ {
1914
+ "epoch": 0.6,
1915
+ "learning_rate": 2.020812184777855e-05,
1916
+ "loss": 3.7363,
1917
+ "step": 148500
1918
+ },
1919
+ {
1920
+ "epoch": 0.6,
1921
+ "learning_rate": 2.0107812493730667e-05,
1922
+ "loss": 3.7349,
1923
+ "step": 149000
1924
+ },
1925
+ {
1926
+ "epoch": 0.6,
1927
+ "learning_rate": 2.0007503139682783e-05,
1928
+ "loss": 3.7266,
1929
+ "step": 149500
1930
+ },
1931
+ {
1932
+ "epoch": 0.6,
1933
+ "learning_rate": 1.99071937856349e-05,
1934
+ "loss": 3.7378,
1935
+ "step": 150000
1936
+ },
1937
+ {
1938
+ "epoch": 0.6,
1939
+ "eval_accuracy": 0.3545389268009225,
1940
+ "eval_loss": 3.650148630142212,
1941
+ "eval_runtime": 4647.2835,
1942
+ "eval_samples_per_second": 90.298,
1943
+ "eval_steps_per_second": 1.411,
1944
+ "step": 150000
1945
+ },
1946
+ {
1947
+ "epoch": 0.6,
1948
+ "learning_rate": 1.9806884431587015e-05,
1949
+ "loss": 3.7285,
1950
+ "step": 150500
1951
+ },
1952
+ {
1953
+ "epoch": 0.61,
1954
+ "learning_rate": 1.9706575077539134e-05,
1955
+ "loss": 3.7343,
1956
+ "step": 151000
1957
+ },
1958
+ {
1959
+ "epoch": 0.61,
1960
+ "learning_rate": 1.960626572349125e-05,
1961
+ "loss": 3.7287,
1962
+ "step": 151500
1963
+ },
1964
+ {
1965
+ "epoch": 0.61,
1966
+ "learning_rate": 1.9505956369443366e-05,
1967
+ "loss": 3.728,
1968
+ "step": 152000
1969
+ },
1970
+ {
1971
+ "epoch": 0.61,
1972
+ "learning_rate": 1.9405647015395482e-05,
1973
+ "loss": 3.7368,
1974
+ "step": 152500
1975
+ },
1976
+ {
1977
+ "epoch": 0.61,
1978
+ "learning_rate": 1.93053376613476e-05,
1979
+ "loss": 3.7284,
1980
+ "step": 153000
1981
+ },
1982
+ {
1983
+ "epoch": 0.62,
1984
+ "learning_rate": 1.9205028307299714e-05,
1985
+ "loss": 3.7245,
1986
+ "step": 153500
1987
+ },
1988
+ {
1989
+ "epoch": 0.62,
1990
+ "learning_rate": 1.910471895325183e-05,
1991
+ "loss": 3.7256,
1992
+ "step": 154000
1993
+ },
1994
+ {
1995
+ "epoch": 0.62,
1996
+ "learning_rate": 1.9004409599203946e-05,
1997
+ "loss": 3.7238,
1998
+ "step": 154500
1999
+ },
2000
+ {
2001
+ "epoch": 0.62,
2002
+ "learning_rate": 1.8904100245156062e-05,
2003
+ "loss": 3.7256,
2004
+ "step": 155000
2005
+ },
2006
+ {
2007
+ "epoch": 0.62,
2008
+ "learning_rate": 1.880379089110818e-05,
2009
+ "loss": 3.7265,
2010
+ "step": 155500
2011
+ },
2012
+ {
2013
+ "epoch": 0.63,
2014
+ "learning_rate": 1.8703481537060294e-05,
2015
+ "loss": 3.725,
2016
+ "step": 156000
2017
+ },
2018
+ {
2019
+ "epoch": 0.63,
2020
+ "learning_rate": 1.860317218301241e-05,
2021
+ "loss": 3.7152,
2022
+ "step": 156500
2023
+ },
2024
+ {
2025
+ "epoch": 0.63,
2026
+ "learning_rate": 1.8502862828964526e-05,
2027
+ "loss": 3.7176,
2028
+ "step": 157000
2029
+ },
2030
+ {
2031
+ "epoch": 0.63,
2032
+ "learning_rate": 1.8402553474916646e-05,
2033
+ "loss": 3.7206,
2034
+ "step": 157500
2035
+ },
2036
+ {
2037
+ "epoch": 0.63,
2038
+ "learning_rate": 1.8302244120868762e-05,
2039
+ "loss": 3.7272,
2040
+ "step": 158000
2041
+ },
2042
+ {
2043
+ "epoch": 0.64,
2044
+ "learning_rate": 1.8201934766820878e-05,
2045
+ "loss": 3.7181,
2046
+ "step": 158500
2047
+ },
2048
+ {
2049
+ "epoch": 0.64,
2050
+ "learning_rate": 1.8101625412772994e-05,
2051
+ "loss": 3.7257,
2052
+ "step": 159000
2053
+ },
2054
+ {
2055
+ "epoch": 0.64,
2056
+ "learning_rate": 1.800131605872511e-05,
2057
+ "loss": 3.7215,
2058
+ "step": 159500
2059
+ },
2060
+ {
2061
+ "epoch": 0.64,
2062
+ "learning_rate": 1.7901006704677226e-05,
2063
+ "loss": 3.7181,
2064
+ "step": 160000
2065
+ },
2066
+ {
2067
+ "epoch": 0.64,
2068
+ "eval_accuracy": 0.3558949659698775,
2069
+ "eval_loss": 3.638543128967285,
2070
+ "eval_runtime": 4644.9417,
2071
+ "eval_samples_per_second": 90.343,
2072
+ "eval_steps_per_second": 1.412,
2073
+ "step": 160000
2074
+ },
2075
+ {
2076
+ "epoch": 0.64,
2077
+ "learning_rate": 1.780069735062934e-05,
2078
+ "loss": 3.719,
2079
+ "step": 160500
2080
+ },
2081
+ {
2082
+ "epoch": 0.65,
2083
+ "learning_rate": 1.7700387996581458e-05,
2084
+ "loss": 3.7194,
2085
+ "step": 161000
2086
+ },
2087
+ {
2088
+ "epoch": 0.65,
2089
+ "learning_rate": 1.7600078642533574e-05,
2090
+ "loss": 3.7197,
2091
+ "step": 161500
2092
+ },
2093
+ {
2094
+ "epoch": 0.65,
2095
+ "learning_rate": 1.749976928848569e-05,
2096
+ "loss": 3.7198,
2097
+ "step": 162000
2098
+ },
2099
+ {
2100
+ "epoch": 0.65,
2101
+ "learning_rate": 1.7399459934437806e-05,
2102
+ "loss": 3.7223,
2103
+ "step": 162500
2104
+ },
2105
+ {
2106
+ "epoch": 0.65,
2107
+ "learning_rate": 1.729915058038992e-05,
2108
+ "loss": 3.7201,
2109
+ "step": 163000
2110
+ },
2111
+ {
2112
+ "epoch": 0.66,
2113
+ "learning_rate": 1.7198841226342038e-05,
2114
+ "loss": 3.7158,
2115
+ "step": 163500
2116
+ },
2117
+ {
2118
+ "epoch": 0.66,
2119
+ "learning_rate": 1.7098531872294157e-05,
2120
+ "loss": 3.7124,
2121
+ "step": 164000
2122
+ },
2123
+ {
2124
+ "epoch": 0.66,
2125
+ "learning_rate": 1.6998222518246273e-05,
2126
+ "loss": 3.7162,
2127
+ "step": 164500
2128
+ },
2129
+ {
2130
+ "epoch": 0.66,
2131
+ "learning_rate": 1.689791316419839e-05,
2132
+ "loss": 3.7146,
2133
+ "step": 165000
2134
+ },
2135
+ {
2136
+ "epoch": 0.66,
2137
+ "learning_rate": 1.6797603810150505e-05,
2138
+ "loss": 3.7111,
2139
+ "step": 165500
2140
+ },
2141
+ {
2142
+ "epoch": 0.67,
2143
+ "learning_rate": 1.669729445610262e-05,
2144
+ "loss": 3.7226,
2145
+ "step": 166000
2146
+ },
2147
+ {
2148
+ "epoch": 0.67,
2149
+ "learning_rate": 1.6596985102054737e-05,
2150
+ "loss": 3.7134,
2151
+ "step": 166500
2152
+ },
2153
+ {
2154
+ "epoch": 0.67,
2155
+ "learning_rate": 1.6496675748006853e-05,
2156
+ "loss": 3.7138,
2157
+ "step": 167000
2158
+ },
2159
+ {
2160
+ "epoch": 0.67,
2161
+ "learning_rate": 1.639636639395897e-05,
2162
+ "loss": 3.7155,
2163
+ "step": 167500
2164
+ },
2165
+ {
2166
+ "epoch": 0.67,
2167
+ "learning_rate": 1.6296057039911085e-05,
2168
+ "loss": 3.7161,
2169
+ "step": 168000
2170
+ },
2171
+ {
2172
+ "epoch": 0.68,
2173
+ "learning_rate": 1.61957476858632e-05,
2174
+ "loss": 3.7155,
2175
+ "step": 168500
2176
+ },
2177
+ {
2178
+ "epoch": 0.68,
2179
+ "learning_rate": 1.6095438331815317e-05,
2180
+ "loss": 3.7118,
2181
+ "step": 169000
2182
+ },
2183
+ {
2184
+ "epoch": 0.68,
2185
+ "learning_rate": 1.5995128977767433e-05,
2186
+ "loss": 3.714,
2187
+ "step": 169500
2188
+ },
2189
+ {
2190
+ "epoch": 0.68,
2191
+ "learning_rate": 1.5894819623719552e-05,
2192
+ "loss": 3.7139,
2193
+ "step": 170000
2194
+ },
2195
+ {
2196
+ "epoch": 0.68,
2197
+ "eval_accuracy": 0.35682303406635435,
2198
+ "eval_loss": 3.629270315170288,
2199
+ "eval_runtime": 4647.4689,
2200
+ "eval_samples_per_second": 90.294,
2201
+ "eval_steps_per_second": 1.411,
2202
+ "step": 170000
2203
+ },
2204
+ {
2205
+ "epoch": 0.68,
2206
+ "learning_rate": 1.579451026967167e-05,
2207
+ "loss": 3.7157,
2208
+ "step": 170500
2209
+ },
2210
+ {
2211
+ "epoch": 0.69,
2212
+ "learning_rate": 1.5694200915623784e-05,
2213
+ "loss": 3.7122,
2214
+ "step": 171000
2215
+ },
2216
+ {
2217
+ "epoch": 0.69,
2218
+ "learning_rate": 1.55938915615759e-05,
2219
+ "loss": 3.7123,
2220
+ "step": 171500
2221
+ },
2222
+ {
2223
+ "epoch": 0.69,
2224
+ "learning_rate": 1.5493582207528016e-05,
2225
+ "loss": 3.7171,
2226
+ "step": 172000
2227
+ },
2228
+ {
2229
+ "epoch": 0.69,
2230
+ "learning_rate": 1.5393272853480136e-05,
2231
+ "loss": 3.7049,
2232
+ "step": 172500
2233
+ },
2234
+ {
2235
+ "epoch": 0.69,
2236
+ "learning_rate": 1.5292963499432252e-05,
2237
+ "loss": 3.7011,
2238
+ "step": 173000
2239
+ },
2240
+ {
2241
+ "epoch": 0.7,
2242
+ "learning_rate": 1.5192654145384366e-05,
2243
+ "loss": 3.71,
2244
+ "step": 173500
2245
+ },
2246
+ {
2247
+ "epoch": 0.7,
2248
+ "learning_rate": 1.5092344791336482e-05,
2249
+ "loss": 3.7042,
2250
+ "step": 174000
2251
+ },
2252
+ {
2253
+ "epoch": 0.7,
2254
+ "learning_rate": 1.4992035437288598e-05,
2255
+ "loss": 3.7035,
2256
+ "step": 174500
2257
+ },
2258
+ {
2259
+ "epoch": 0.7,
2260
+ "learning_rate": 1.4891726083240714e-05,
2261
+ "loss": 3.7043,
2262
+ "step": 175000
2263
+ },
2264
+ {
2265
+ "epoch": 0.7,
2266
+ "learning_rate": 1.479141672919283e-05,
2267
+ "loss": 3.7054,
2268
+ "step": 175500
2269
+ },
2270
+ {
2271
+ "epoch": 0.71,
2272
+ "learning_rate": 1.4691107375144948e-05,
2273
+ "loss": 3.7039,
2274
+ "step": 176000
2275
+ },
2276
+ {
2277
+ "epoch": 0.71,
2278
+ "learning_rate": 1.4590798021097064e-05,
2279
+ "loss": 3.7069,
2280
+ "step": 176500
2281
+ },
2282
+ {
2283
+ "epoch": 0.71,
2284
+ "learning_rate": 1.449048866704918e-05,
2285
+ "loss": 3.7039,
2286
+ "step": 177000
2287
+ },
2288
+ {
2289
+ "epoch": 0.71,
2290
+ "learning_rate": 1.4390179313001297e-05,
2291
+ "loss": 3.7046,
2292
+ "step": 177500
2293
+ },
2294
+ {
2295
+ "epoch": 0.71,
2296
+ "learning_rate": 1.4289869958953413e-05,
2297
+ "loss": 3.7101,
2298
+ "step": 178000
2299
+ },
2300
+ {
2301
+ "epoch": 0.72,
2302
+ "learning_rate": 1.418956060490553e-05,
2303
+ "loss": 3.7113,
2304
+ "step": 178500
2305
+ },
2306
+ {
2307
+ "epoch": 0.72,
2308
+ "learning_rate": 1.4089251250857647e-05,
2309
+ "loss": 3.7057,
2310
+ "step": 179000
2311
+ },
2312
+ {
2313
+ "epoch": 0.72,
2314
+ "learning_rate": 1.3988941896809763e-05,
2315
+ "loss": 3.7077,
2316
+ "step": 179500
2317
+ },
2318
+ {
2319
+ "epoch": 0.72,
2320
+ "learning_rate": 1.3888632542761879e-05,
2321
+ "loss": 3.6958,
2322
+ "step": 180000
2323
+ },
2324
+ {
2325
+ "epoch": 0.72,
2326
+ "eval_accuracy": 0.35775385321243053,
2327
+ "eval_loss": 3.6201136112213135,
2328
+ "eval_runtime": 4642.689,
2329
+ "eval_samples_per_second": 90.387,
2330
+ "eval_steps_per_second": 1.412,
2331
+ "step": 180000
2332
+ },
2333
+ {
2334
+ "epoch": 0.72,
2335
+ "learning_rate": 1.3788323188713995e-05,
2336
+ "loss": 3.7004,
2337
+ "step": 180500
2338
+ },
2339
+ {
2340
+ "epoch": 0.73,
2341
+ "learning_rate": 1.3688013834666111e-05,
2342
+ "loss": 3.7042,
2343
+ "step": 181000
2344
+ },
2345
+ {
2346
+ "epoch": 0.73,
2347
+ "learning_rate": 1.3587704480618227e-05,
2348
+ "loss": 3.6989,
2349
+ "step": 181500
2350
+ },
2351
+ {
2352
+ "epoch": 0.73,
2353
+ "learning_rate": 1.3487395126570343e-05,
2354
+ "loss": 3.7118,
2355
+ "step": 182000
2356
+ },
2357
+ {
2358
+ "epoch": 0.73,
2359
+ "learning_rate": 1.3387085772522459e-05,
2360
+ "loss": 3.6947,
2361
+ "step": 182500
2362
+ },
2363
+ {
2364
+ "epoch": 0.73,
2365
+ "learning_rate": 1.3286776418474575e-05,
2366
+ "loss": 3.6928,
2367
+ "step": 183000
2368
+ },
2369
+ {
2370
+ "epoch": 0.74,
2371
+ "learning_rate": 1.3186467064426691e-05,
2372
+ "loss": 3.6959,
2373
+ "step": 183500
2374
+ },
2375
+ {
2376
+ "epoch": 0.74,
2377
+ "learning_rate": 1.308615771037881e-05,
2378
+ "loss": 3.7036,
2379
+ "step": 184000
2380
+ },
2381
+ {
2382
+ "epoch": 0.74,
2383
+ "learning_rate": 1.2985848356330926e-05,
2384
+ "loss": 3.7079,
2385
+ "step": 184500
2386
+ },
2387
+ {
2388
+ "epoch": 0.74,
2389
+ "learning_rate": 1.2885539002283042e-05,
2390
+ "loss": 3.7025,
2391
+ "step": 185000
2392
+ },
2393
+ {
2394
+ "epoch": 0.74,
2395
+ "learning_rate": 1.2785229648235158e-05,
2396
+ "loss": 3.7001,
2397
+ "step": 185500
2398
+ },
2399
+ {
2400
+ "epoch": 0.75,
2401
+ "learning_rate": 1.2684920294187274e-05,
2402
+ "loss": 3.6922,
2403
+ "step": 186000
2404
+ },
2405
+ {
2406
+ "epoch": 0.75,
2407
+ "learning_rate": 1.258461094013939e-05,
2408
+ "loss": 3.6997,
2409
+ "step": 186500
2410
+ },
2411
+ {
2412
+ "epoch": 0.75,
2413
+ "learning_rate": 1.2484301586091506e-05,
2414
+ "loss": 3.6995,
2415
+ "step": 187000
2416
+ },
2417
+ {
2418
+ "epoch": 0.75,
2419
+ "learning_rate": 1.2383992232043622e-05,
2420
+ "loss": 3.6927,
2421
+ "step": 187500
2422
+ },
2423
+ {
2424
+ "epoch": 0.75,
2425
+ "learning_rate": 1.2283682877995738e-05,
2426
+ "loss": 3.6924,
2427
+ "step": 188000
2428
+ },
2429
+ {
2430
+ "epoch": 0.76,
2431
+ "learning_rate": 1.2183373523947856e-05,
2432
+ "loss": 3.6931,
2433
+ "step": 188500
2434
+ },
2435
+ {
2436
+ "epoch": 0.76,
2437
+ "learning_rate": 1.2083064169899972e-05,
2438
+ "loss": 3.6964,
2439
+ "step": 189000
2440
+ },
2441
+ {
2442
+ "epoch": 0.76,
2443
+ "learning_rate": 1.198275481585209e-05,
2444
+ "loss": 3.695,
2445
+ "step": 189500
2446
+ },
2447
+ {
2448
+ "epoch": 0.76,
2449
+ "learning_rate": 1.1882445461804206e-05,
2450
+ "loss": 3.6872,
2451
+ "step": 190000
2452
+ },
2453
+ {
2454
+ "epoch": 0.76,
2455
+ "eval_accuracy": 0.3585257078142553,
2456
+ "eval_loss": 3.612175226211548,
2457
+ "eval_runtime": 4645.5279,
2458
+ "eval_samples_per_second": 90.332,
2459
+ "eval_steps_per_second": 1.411,
2460
+ "step": 190000
2461
+ },
2462
+ {
2463
+ "epoch": 0.76,
2464
+ "learning_rate": 1.1782136107756322e-05,
2465
+ "loss": 3.688,
2466
+ "step": 190500
2467
+ },
2468
+ {
2469
+ "epoch": 0.77,
2470
+ "learning_rate": 1.1681826753708438e-05,
2471
+ "loss": 3.7005,
2472
+ "step": 191000
2473
+ },
2474
+ {
2475
+ "epoch": 0.77,
2476
+ "learning_rate": 1.1581517399660554e-05,
2477
+ "loss": 3.6877,
2478
+ "step": 191500
2479
+ },
2480
+ {
2481
+ "epoch": 0.77,
2482
+ "learning_rate": 1.148120804561267e-05,
2483
+ "loss": 3.7002,
2484
+ "step": 192000
2485
+ },
2486
+ {
2487
+ "epoch": 0.77,
2488
+ "learning_rate": 1.1380898691564787e-05,
2489
+ "loss": 3.6913,
2490
+ "step": 192500
2491
+ },
2492
+ {
2493
+ "epoch": 0.77,
2494
+ "learning_rate": 1.1280589337516903e-05,
2495
+ "loss": 3.6948,
2496
+ "step": 193000
2497
+ },
2498
+ {
2499
+ "epoch": 0.78,
2500
+ "learning_rate": 1.118027998346902e-05,
2501
+ "loss": 3.6953,
2502
+ "step": 193500
2503
+ },
2504
+ {
2505
+ "epoch": 0.78,
2506
+ "learning_rate": 1.1079970629421135e-05,
2507
+ "loss": 3.693,
2508
+ "step": 194000
2509
+ },
2510
+ {
2511
+ "epoch": 0.78,
2512
+ "learning_rate": 1.0979661275373251e-05,
2513
+ "loss": 3.6806,
2514
+ "step": 194500
2515
+ },
2516
+ {
2517
+ "epoch": 0.78,
2518
+ "learning_rate": 1.0879351921325367e-05,
2519
+ "loss": 3.6927,
2520
+ "step": 195000
2521
+ },
2522
+ {
2523
+ "epoch": 0.78,
2524
+ "learning_rate": 1.0779042567277483e-05,
2525
+ "loss": 3.6979,
2526
+ "step": 195500
2527
+ },
2528
+ {
2529
+ "epoch": 0.79,
2530
+ "learning_rate": 1.0678733213229601e-05,
2531
+ "loss": 3.6916,
2532
+ "step": 196000
2533
+ },
2534
+ {
2535
+ "epoch": 0.79,
2536
+ "learning_rate": 1.0578423859181717e-05,
2537
+ "loss": 3.6958,
2538
+ "step": 196500
2539
+ },
2540
+ {
2541
+ "epoch": 0.79,
2542
+ "learning_rate": 1.0478114505133833e-05,
2543
+ "loss": 3.6912,
2544
+ "step": 197000
2545
+ },
2546
+ {
2547
+ "epoch": 0.79,
2548
+ "learning_rate": 1.0377805151085949e-05,
2549
+ "loss": 3.6865,
2550
+ "step": 197500
2551
+ },
2552
+ {
2553
+ "epoch": 0.79,
2554
+ "learning_rate": 1.0277495797038065e-05,
2555
+ "loss": 3.687,
2556
+ "step": 198000
2557
+ },
2558
+ {
2559
+ "epoch": 0.8,
2560
+ "learning_rate": 1.0177186442990181e-05,
2561
+ "loss": 3.6877,
2562
+ "step": 198500
2563
+ },
2564
+ {
2565
+ "epoch": 0.8,
2566
+ "learning_rate": 1.0076877088942299e-05,
2567
+ "loss": 3.6895,
2568
+ "step": 199000
2569
+ },
2570
+ {
2571
+ "epoch": 0.8,
2572
+ "learning_rate": 9.976567734894415e-06,
2573
+ "loss": 3.6805,
2574
+ "step": 199500
2575
+ },
2576
+ {
2577
+ "epoch": 0.8,
2578
+ "learning_rate": 9.87625838084653e-06,
2579
+ "loss": 3.6888,
2580
+ "step": 200000
2581
+ },
2582
+ {
2583
+ "epoch": 0.8,
2584
+ "eval_accuracy": 0.3592407500971306,
2585
+ "eval_loss": 3.6060221195220947,
2586
+ "eval_runtime": 4651.0982,
2587
+ "eval_samples_per_second": 90.224,
2588
+ "eval_steps_per_second": 1.41,
2589
+ "step": 200000
2590
+ },
2591
+ {
2592
+ "epoch": 0.8,
2593
+ "learning_rate": 9.775949026798648e-06,
2594
+ "loss": 3.6844,
2595
+ "step": 200500
2596
+ },
2597
+ {
2598
+ "epoch": 0.81,
2599
+ "learning_rate": 9.675639672750764e-06,
2600
+ "loss": 3.6824,
2601
+ "step": 201000
2602
+ },
2603
+ {
2604
+ "epoch": 0.81,
2605
+ "learning_rate": 9.57533031870288e-06,
2606
+ "loss": 3.6902,
2607
+ "step": 201500
2608
+ },
2609
+ {
2610
+ "epoch": 0.81,
2611
+ "learning_rate": 9.475020964654996e-06,
2612
+ "loss": 3.6883,
2613
+ "step": 202000
2614
+ },
2615
+ {
2616
+ "epoch": 0.81,
2617
+ "learning_rate": 9.374711610607114e-06,
2618
+ "loss": 3.6904,
2619
+ "step": 202500
2620
+ },
2621
+ {
2622
+ "epoch": 0.81,
2623
+ "learning_rate": 9.27440225655923e-06,
2624
+ "loss": 3.6861,
2625
+ "step": 203000
2626
+ },
2627
+ {
2628
+ "epoch": 0.82,
2629
+ "learning_rate": 9.174092902511346e-06,
2630
+ "loss": 3.6838,
2631
+ "step": 203500
2632
+ },
2633
+ {
2634
+ "epoch": 0.82,
2635
+ "learning_rate": 9.073783548463462e-06,
2636
+ "loss": 3.6837,
2637
+ "step": 204000
2638
+ },
2639
+ {
2640
+ "epoch": 0.82,
2641
+ "learning_rate": 8.973474194415578e-06,
2642
+ "loss": 3.6876,
2643
+ "step": 204500
2644
+ },
2645
+ {
2646
+ "epoch": 0.82,
2647
+ "learning_rate": 8.873164840367694e-06,
2648
+ "loss": 3.6857,
2649
+ "step": 205000
2650
+ },
2651
+ {
2652
+ "epoch": 0.82,
2653
+ "learning_rate": 8.77285548631981e-06,
2654
+ "loss": 3.6821,
2655
+ "step": 205500
2656
+ },
2657
+ {
2658
+ "epoch": 0.83,
2659
+ "learning_rate": 8.672546132271928e-06,
2660
+ "loss": 3.6787,
2661
+ "step": 206000
2662
+ },
2663
+ {
2664
+ "epoch": 0.83,
2665
+ "learning_rate": 8.572236778224044e-06,
2666
+ "loss": 3.6878,
2667
+ "step": 206500
2668
+ },
2669
+ {
2670
+ "epoch": 0.83,
2671
+ "learning_rate": 8.47192742417616e-06,
2672
+ "loss": 3.6759,
2673
+ "step": 207000
2674
+ },
2675
+ {
2676
+ "epoch": 0.83,
2677
+ "learning_rate": 8.371618070128276e-06,
2678
+ "loss": 3.6827,
2679
+ "step": 207500
2680
+ },
2681
+ {
2682
+ "epoch": 0.83,
2683
+ "learning_rate": 8.271308716080392e-06,
2684
+ "loss": 3.679,
2685
+ "step": 208000
2686
+ },
2687
+ {
2688
+ "epoch": 0.84,
2689
+ "learning_rate": 8.170999362032508e-06,
2690
+ "loss": 3.6827,
2691
+ "step": 208500
2692
+ },
2693
+ {
2694
+ "epoch": 0.84,
2695
+ "learning_rate": 8.070690007984625e-06,
2696
+ "loss": 3.6825,
2697
+ "step": 209000
2698
+ },
2699
+ {
2700
+ "epoch": 0.84,
2701
+ "learning_rate": 7.970380653936741e-06,
2702
+ "loss": 3.6823,
2703
+ "step": 209500
2704
+ },
2705
+ {
2706
+ "epoch": 0.84,
2707
+ "learning_rate": 7.870071299888857e-06,
2708
+ "loss": 3.6765,
2709
+ "step": 210000
2710
+ },
2711
+ {
2712
+ "epoch": 0.84,
2713
+ "eval_accuracy": 0.35991400717449046,
2714
+ "eval_loss": 3.6000776290893555,
2715
+ "eval_runtime": 4655.0214,
2716
+ "eval_samples_per_second": 90.148,
2717
+ "eval_steps_per_second": 1.409,
2718
+ "step": 210000
2719
+ },
2720
+ {
2721
+ "epoch": 0.84,
2722
+ "learning_rate": 7.769761945840973e-06,
2723
+ "loss": 3.6822,
2724
+ "step": 210500
2725
+ },
2726
+ {
2727
+ "epoch": 0.85,
2728
+ "learning_rate": 7.66945259179309e-06,
2729
+ "loss": 3.6778,
2730
+ "step": 211000
2731
+ },
2732
+ {
2733
+ "epoch": 0.85,
2734
+ "learning_rate": 7.569143237745206e-06,
2735
+ "loss": 3.6786,
2736
+ "step": 211500
2737
+ },
2738
+ {
2739
+ "epoch": 0.85,
2740
+ "learning_rate": 7.468833883697322e-06,
2741
+ "loss": 3.6745,
2742
+ "step": 212000
2743
+ },
2744
+ {
2745
+ "epoch": 0.85,
2746
+ "learning_rate": 7.36852452964944e-06,
2747
+ "loss": 3.6767,
2748
+ "step": 212500
2749
+ },
2750
+ {
2751
+ "epoch": 0.85,
2752
+ "learning_rate": 7.268215175601556e-06,
2753
+ "loss": 3.6785,
2754
+ "step": 213000
2755
+ },
2756
+ {
2757
+ "epoch": 0.86,
2758
+ "learning_rate": 7.167905821553672e-06,
2759
+ "loss": 3.685,
2760
+ "step": 213500
2761
+ },
2762
+ {
2763
+ "epoch": 0.86,
2764
+ "learning_rate": 7.067596467505788e-06,
2765
+ "loss": 3.6694,
2766
+ "step": 214000
2767
+ },
2768
+ {
2769
+ "epoch": 0.86,
2770
+ "learning_rate": 6.967287113457905e-06,
2771
+ "loss": 3.6816,
2772
+ "step": 214500
2773
+ },
2774
+ {
2775
+ "epoch": 0.86,
2776
+ "learning_rate": 6.866977759410021e-06,
2777
+ "loss": 3.6779,
2778
+ "step": 215000
2779
+ },
2780
+ {
2781
+ "epoch": 0.86,
2782
+ "learning_rate": 6.766668405362137e-06,
2783
+ "loss": 3.6772,
2784
+ "step": 215500
2785
+ },
2786
+ {
2787
+ "epoch": 0.87,
2788
+ "learning_rate": 6.666359051314254e-06,
2789
+ "loss": 3.673,
2790
+ "step": 216000
2791
+ },
2792
+ {
2793
+ "epoch": 0.87,
2794
+ "learning_rate": 6.56604969726637e-06,
2795
+ "loss": 3.6848,
2796
+ "step": 216500
2797
+ },
2798
+ {
2799
+ "epoch": 0.87,
2800
+ "learning_rate": 6.465740343218486e-06,
2801
+ "loss": 3.6748,
2802
+ "step": 217000
2803
+ },
2804
+ {
2805
+ "epoch": 0.87,
2806
+ "learning_rate": 6.365430989170602e-06,
2807
+ "loss": 3.6738,
2808
+ "step": 217500
2809
+ },
2810
+ {
2811
+ "epoch": 0.87,
2812
+ "learning_rate": 6.265121635122718e-06,
2813
+ "loss": 3.6802,
2814
+ "step": 218000
2815
+ },
2816
+ {
2817
+ "epoch": 0.88,
2818
+ "learning_rate": 6.164812281074835e-06,
2819
+ "loss": 3.6818,
2820
+ "step": 218500
2821
+ },
2822
+ {
2823
+ "epoch": 0.88,
2824
+ "learning_rate": 6.064502927026951e-06,
2825
+ "loss": 3.6722,
2826
+ "step": 219000
2827
+ },
2828
+ {
2829
+ "epoch": 0.88,
2830
+ "learning_rate": 5.964193572979067e-06,
2831
+ "loss": 3.6728,
2832
+ "step": 219500
2833
+ },
2834
+ {
2835
+ "epoch": 0.88,
2836
+ "learning_rate": 5.863884218931184e-06,
2837
+ "loss": 3.6734,
2838
+ "step": 220000
2839
+ },
2840
+ {
2841
+ "epoch": 0.88,
2842
+ "eval_accuracy": 0.36040756550566483,
2843
+ "eval_loss": 3.5944714546203613,
2844
+ "eval_runtime": 4642.1498,
2845
+ "eval_samples_per_second": 90.398,
2846
+ "eval_steps_per_second": 1.412,
2847
+ "step": 220000
2848
+ },
2849
+ {
2850
+ "epoch": 0.88,
2851
+ "learning_rate": 5.763574864883301e-06,
2852
+ "loss": 3.6763,
2853
+ "step": 220500
2854
+ },
2855
+ {
2856
+ "epoch": 0.89,
2857
+ "learning_rate": 5.663265510835417e-06,
2858
+ "loss": 3.6743,
2859
+ "step": 221000
2860
+ },
2861
+ {
2862
+ "epoch": 0.89,
2863
+ "learning_rate": 5.562956156787533e-06,
2864
+ "loss": 3.6718,
2865
+ "step": 221500
2866
+ },
2867
+ {
2868
+ "epoch": 0.89,
2869
+ "learning_rate": 5.46264680273965e-06,
2870
+ "loss": 3.6695,
2871
+ "step": 222000
2872
+ },
2873
+ {
2874
+ "epoch": 0.89,
2875
+ "learning_rate": 5.362337448691766e-06,
2876
+ "loss": 3.6704,
2877
+ "step": 222500
2878
+ },
2879
+ {
2880
+ "epoch": 0.89,
2881
+ "learning_rate": 5.262028094643882e-06,
2882
+ "loss": 3.6728,
2883
+ "step": 223000
2884
+ },
2885
+ {
2886
+ "epoch": 0.9,
2887
+ "learning_rate": 5.1617187405959985e-06,
2888
+ "loss": 3.6713,
2889
+ "step": 223500
2890
+ },
2891
+ {
2892
+ "epoch": 0.9,
2893
+ "learning_rate": 5.0614093865481145e-06,
2894
+ "loss": 3.664,
2895
+ "step": 224000
2896
+ },
2897
+ {
2898
+ "epoch": 0.9,
2899
+ "learning_rate": 4.9611000325002305e-06,
2900
+ "loss": 3.6659,
2901
+ "step": 224500
2902
+ },
2903
+ {
2904
+ "epoch": 0.9,
2905
+ "learning_rate": 4.860790678452347e-06,
2906
+ "loss": 3.6739,
2907
+ "step": 225000
2908
+ },
2909
+ {
2910
+ "epoch": 0.9,
2911
+ "learning_rate": 4.760481324404463e-06,
2912
+ "loss": 3.6794,
2913
+ "step": 225500
2914
+ },
2915
+ {
2916
+ "epoch": 0.91,
2917
+ "learning_rate": 4.66017197035658e-06,
2918
+ "loss": 3.6702,
2919
+ "step": 226000
2920
+ },
2921
+ {
2922
+ "epoch": 0.91,
2923
+ "learning_rate": 4.559862616308696e-06,
2924
+ "loss": 3.6667,
2925
+ "step": 226500
2926
+ },
2927
+ {
2928
+ "epoch": 0.91,
2929
+ "learning_rate": 4.459553262260813e-06,
2930
+ "loss": 3.6651,
2931
+ "step": 227000
2932
+ },
2933
+ {
2934
+ "epoch": 0.91,
2935
+ "learning_rate": 4.359243908212929e-06,
2936
+ "loss": 3.6701,
2937
+ "step": 227500
2938
+ },
2939
+ {
2940
+ "epoch": 0.91,
2941
+ "learning_rate": 4.258934554165045e-06,
2942
+ "loss": 3.6626,
2943
+ "step": 228000
2944
+ },
2945
+ {
2946
+ "epoch": 0.92,
2947
+ "learning_rate": 4.158625200117162e-06,
2948
+ "loss": 3.6664,
2949
+ "step": 228500
2950
+ },
2951
+ {
2952
+ "epoch": 0.92,
2953
+ "learning_rate": 4.058315846069278e-06,
2954
+ "loss": 3.6732,
2955
+ "step": 229000
2956
+ },
2957
+ {
2958
+ "epoch": 0.92,
2959
+ "learning_rate": 3.958006492021394e-06,
2960
+ "loss": 3.6714,
2961
+ "step": 229500
2962
+ },
2963
+ {
2964
+ "epoch": 0.92,
2965
+ "learning_rate": 3.857697137973511e-06,
2966
+ "loss": 3.6669,
2967
+ "step": 230000
2968
+ },
2969
+ {
2970
+ "epoch": 0.92,
2971
+ "eval_accuracy": 0.36110471781316056,
2972
+ "eval_loss": 3.5890560150146484,
2973
+ "eval_runtime": 4652.0644,
2974
+ "eval_samples_per_second": 90.205,
2975
+ "eval_steps_per_second": 1.409,
2976
+ "step": 230000
2977
+ },
2978
+ {
2979
+ "epoch": 0.92,
2980
+ "learning_rate": 3.757387783925627e-06,
2981
+ "loss": 3.668,
2982
+ "step": 230500
2983
+ },
2984
+ {
2985
+ "epoch": 0.93,
2986
+ "learning_rate": 3.657078429877743e-06,
2987
+ "loss": 3.6712,
2988
+ "step": 231000
2989
+ },
2990
+ {
2991
+ "epoch": 0.93,
2992
+ "learning_rate": 3.556769075829859e-06,
2993
+ "loss": 3.6695,
2994
+ "step": 231500
2995
+ },
2996
+ {
2997
+ "epoch": 0.93,
2998
+ "learning_rate": 3.456459721781976e-06,
2999
+ "loss": 3.6673,
3000
+ "step": 232000
3001
+ },
3002
+ {
3003
+ "epoch": 0.93,
3004
+ "learning_rate": 3.356150367734092e-06,
3005
+ "loss": 3.6736,
3006
+ "step": 232500
3007
+ },
3008
+ {
3009
+ "epoch": 0.93,
3010
+ "learning_rate": 3.2558410136862083e-06,
3011
+ "loss": 3.6695,
3012
+ "step": 233000
3013
+ },
3014
+ {
3015
+ "epoch": 0.94,
3016
+ "learning_rate": 3.155531659638325e-06,
3017
+ "loss": 3.6642,
3018
+ "step": 233500
3019
+ },
3020
+ {
3021
+ "epoch": 0.94,
3022
+ "learning_rate": 3.055222305590441e-06,
3023
+ "loss": 3.6652,
3024
+ "step": 234000
3025
+ },
3026
+ {
3027
+ "epoch": 0.94,
3028
+ "learning_rate": 2.9549129515425576e-06,
3029
+ "loss": 3.6664,
3030
+ "step": 234500
3031
+ },
3032
+ {
3033
+ "epoch": 0.94,
3034
+ "learning_rate": 2.8546035974946736e-06,
3035
+ "loss": 3.67,
3036
+ "step": 235000
3037
+ },
3038
+ {
3039
+ "epoch": 0.94,
3040
+ "learning_rate": 2.75429424344679e-06,
3041
+ "loss": 3.6692,
3042
+ "step": 235500
3043
+ },
3044
+ {
3045
+ "epoch": 0.95,
3046
+ "learning_rate": 2.6539848893989064e-06,
3047
+ "loss": 3.6647,
3048
+ "step": 236000
3049
+ },
3050
+ {
3051
+ "epoch": 0.95,
3052
+ "learning_rate": 2.553675535351023e-06,
3053
+ "loss": 3.6749,
3054
+ "step": 236500
3055
+ },
3056
+ {
3057
+ "epoch": 0.95,
3058
+ "learning_rate": 2.4533661813031392e-06,
3059
+ "loss": 3.672,
3060
+ "step": 237000
3061
+ },
3062
+ {
3063
+ "epoch": 0.95,
3064
+ "learning_rate": 2.3530568272552552e-06,
3065
+ "loss": 3.6705,
3066
+ "step": 237500
3067
+ },
3068
+ {
3069
+ "epoch": 0.95,
3070
+ "learning_rate": 2.2527474732073717e-06,
3071
+ "loss": 3.6687,
3072
+ "step": 238000
3073
+ },
3074
+ {
3075
+ "epoch": 0.96,
3076
+ "learning_rate": 2.1524381191594877e-06,
3077
+ "loss": 3.666,
3078
+ "step": 238500
3079
+ },
3080
+ {
3081
+ "epoch": 0.96,
3082
+ "learning_rate": 2.0521287651116045e-06,
3083
+ "loss": 3.6686,
3084
+ "step": 239000
3085
+ },
3086
+ {
3087
+ "epoch": 0.96,
3088
+ "learning_rate": 1.951819411063721e-06,
3089
+ "loss": 3.6631,
3090
+ "step": 239500
3091
+ },
3092
+ {
3093
+ "epoch": 0.96,
3094
+ "learning_rate": 1.851510057015837e-06,
3095
+ "loss": 3.6696,
3096
+ "step": 240000
3097
+ },
3098
+ {
3099
+ "epoch": 0.96,
3100
+ "eval_accuracy": 0.3614285682971602,
3101
+ "eval_loss": 3.58561635017395,
3102
+ "eval_runtime": 4649.4492,
3103
+ "eval_samples_per_second": 90.256,
3104
+ "eval_steps_per_second": 1.41,
3105
+ "step": 240000
3106
+ },
3107
+ {
3108
+ "epoch": 0.96,
3109
+ "learning_rate": 1.7512007029679533e-06,
3110
+ "loss": 3.6624,
3111
+ "step": 240500
3112
+ },
3113
+ {
3114
+ "epoch": 0.97,
3115
+ "learning_rate": 1.6508913489200695e-06,
3116
+ "loss": 3.6652,
3117
+ "step": 241000
3118
+ },
3119
+ {
3120
+ "epoch": 0.97,
3121
+ "learning_rate": 1.550581994872186e-06,
3122
+ "loss": 3.6659,
3123
+ "step": 241500
3124
+ },
3125
+ {
3126
+ "epoch": 0.97,
3127
+ "learning_rate": 1.4502726408243022e-06,
3128
+ "loss": 3.6623,
3129
+ "step": 242000
3130
+ },
3131
+ {
3132
+ "epoch": 0.97,
3133
+ "learning_rate": 1.3499632867764186e-06,
3134
+ "loss": 3.6672,
3135
+ "step": 242500
3136
+ },
3137
+ {
3138
+ "epoch": 0.98,
3139
+ "learning_rate": 1.2496539327285348e-06,
3140
+ "loss": 3.6639,
3141
+ "step": 243000
3142
+ },
3143
+ {
3144
+ "epoch": 0.98,
3145
+ "learning_rate": 1.1493445786806512e-06,
3146
+ "loss": 3.6624,
3147
+ "step": 243500
3148
+ },
3149
+ {
3150
+ "epoch": 0.98,
3151
+ "learning_rate": 1.0490352246327676e-06,
3152
+ "loss": 3.6595,
3153
+ "step": 244000
3154
+ },
3155
+ {
3156
+ "epoch": 0.98,
3157
+ "learning_rate": 9.487258705848838e-07,
3158
+ "loss": 3.6607,
3159
+ "step": 244500
3160
+ },
3161
+ {
3162
+ "epoch": 0.98,
3163
+ "learning_rate": 8.484165165370001e-07,
3164
+ "loss": 3.6529,
3165
+ "step": 245000
3166
+ },
3167
+ {
3168
+ "epoch": 0.99,
3169
+ "learning_rate": 7.481071624891164e-07,
3170
+ "loss": 3.6692,
3171
+ "step": 245500
3172
+ },
3173
+ {
3174
+ "epoch": 0.99,
3175
+ "learning_rate": 6.477978084412328e-07,
3176
+ "loss": 3.6662,
3177
+ "step": 246000
3178
+ },
3179
+ {
3180
+ "epoch": 0.99,
3181
+ "learning_rate": 5.474884543933492e-07,
3182
+ "loss": 3.6656,
3183
+ "step": 246500
3184
+ },
3185
+ {
3186
+ "epoch": 0.99,
3187
+ "learning_rate": 4.4717910034546543e-07,
3188
+ "loss": 3.6597,
3189
+ "step": 247000
3190
+ },
3191
+ {
3192
+ "epoch": 0.99,
3193
+ "learning_rate": 3.468697462975818e-07,
3194
+ "loss": 3.6637,
3195
+ "step": 247500
3196
+ },
3197
+ {
3198
+ "epoch": 1.0,
3199
+ "learning_rate": 2.4656039224969806e-07,
3200
+ "loss": 3.6581,
3201
+ "step": 248000
3202
+ },
3203
+ {
3204
+ "epoch": 1.0,
3205
+ "learning_rate": 1.462510382018144e-07,
3206
+ "loss": 3.6684,
3207
+ "step": 248500
3208
+ },
3209
+ {
3210
+ "epoch": 1.0,
3211
+ "learning_rate": 4.594168415393073e-08,
3212
+ "loss": 3.6663,
3213
+ "step": 249000
3214
+ },
3215
+ {
3216
+ "epoch": 1.0,
3217
+ "step": 249229,
3218
+ "total_flos": 4.167764163035136e+18,
3219
+ "train_loss": 3.8837062227938306,
3220
+ "train_runtime": 230124.717,
3221
+ "train_samples_per_second": 34.656,
3222
+ "train_steps_per_second": 1.083
3223
+ }
3224
+ ],
3225
+ "logging_steps": 500,
3226
+ "max_steps": 249229,
3227
+ "num_input_tokens_seen": 0,
3228
+ "num_train_epochs": 1,
3229
+ "save_steps": 10000,
3230
+ "total_flos": 4.167764163035136e+18,
3231
+ "train_batch_size": 32,
3232
+ "trial_name": null,
3233
+ "trial_params": null
3234
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:896e1cb4a3f167b8ae3eb3d8da0e496b788510221febf7004c2602ed5e87cfe6
3
+ size 4728