Hugging Face's logo

joseagmz
/
TinyLlama-medtext-epochs-1-lr-0002_Train_On_Input

Text Generation
Transformers
PyTorch
Safetensors
llama
Generated from Trainer
text-generation-inference
Model card Files
xet
Community
joseagmz commited on
Commit
9468c30
·
verified ·
1 Parent(s): c5d58d2

Upload folder using huggingface_hub

Browse files
Files changed (15) hide show
  1. README.md +132 -0
  2. checkpoint-620/config.json +28 -0
  3. checkpoint-620/generation_config.json +8 -0
  4. checkpoint-620/model.safetensors +3 -0
  5. checkpoint-620/optimizer.pt +3 -0
  6. checkpoint-620/rng_state.pth +3 -0
  7. checkpoint-620/scheduler.pt +3 -0
  8. checkpoint-620/trainer_state.json +4401 -0
  9. checkpoint-620/training_args.bin +3 -0
  10. config.json +28 -0
  11. generation_config.json +8 -0
  12. pytorch_model.bin +3 -0
  13. special_tokens_map.json +24 -0
  14. tokenizer.model +3 -0
  15. tokenizer_config.json +44 -0
README.md ADDED
@@ -0,0 +1,132 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
4
+ tags:
5
+ - generated_from_trainer
6
+ model-index:
7
+ - name: TinyLlama-medtext-epochs-1-lr-0002_Train_On_Input
8
+ results: []
9
+ ---
10
+
11
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
12
+ should probably proofread and complete it, then remove this comment. -->
13
+
14
+ [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
15
+ <details><summary>See axolotl config</summary>
16
+
17
+ axolotl version: `0.4.0`
18
+ ```yaml
19
+ adapter: null
20
+ base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
21
+ bf16: auto
22
+ dataset_prepared_path: last_run_prepared
23
+ datasets:
24
+ - path: utrgvseniorproject/medtext
25
+ type: completion
26
+ debug: null
27
+ deepspeed: null
28
+ early_stopping_patience: null
29
+ eval_sample_packing: false
30
+ eval_table_size: null
31
+ evals_per_epoch: 4
32
+ flash_attention: true
33
+ flash_attn_cross_entropy: false
34
+ flash_attn_fuse_mlp: true
35
+ flash_attn_fuse_qkv: false
36
+ flash_attn_rms_norm: true
37
+ fp16: null
38
+ fsdp: null
39
+ fsdp_config: null
40
+ gradient_accumulation_steps: 1
41
+ gradient_checkpointing: true
42
+ group_by_length: false
43
+ learning_rate: 0.0002
44
+ load_in_4bit: false
45
+ load_in_8bit: false
46
+ local_rank: null
47
+ logging_steps: 1
48
+ lora_alpha: null
49
+ lora_dropout: null
50
+ lora_fan_in_fan_out: null
51
+ lora_model_dir: null
52
+ lora_r: null
53
+ lora_target_linear: null
54
+ lr_scheduler: cosine
55
+ micro_batch_size: 1
56
+ model_type: LlamaForCausalLM
57
+ num_epochs: 1
58
+ optimizer: adamw_bnb_8bit
59
+ output_dir: ./TinyLlama-medtext-epochs-1-lr-0002_Train_On_Input
60
+ pad_to_sequence_len: true
61
+ resume_from_checkpoint: null
62
+ sample_packing: true
63
+ saves_per_epoch: 1
64
+ sequence_len: 2048
65
+ special_tokens: null
66
+ strict: false
67
+ tf32: false
68
+ tokenizer_type: LlamaTokenizer
69
+ train_on_inputs: true
70
+ val_set_size: 0.05
71
+ wandb_entity: utrgvmedai
72
+ wandb_log_model: null
73
+ wandb_name: tinyLama_medtext_epochs_1_lr_0002
74
+ wandb_project: TinyLlama-Train-On-Input
75
+ wandb_watch: null
76
+ warmup_steps: 100
77
+ weight_decay: 0.1
78
+ xformers_attention: null
79
+
80
+ ```
81
+
82
+ </details><br>
83
+
84
+ # TinyLlama-medtext-epochs-1-lr-0002_Train_On_Input
85
+
86
+ This model is a fine-tuned version of [TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T](https://huggingface.co/TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T) on the None dataset.
87
+ It achieves the following results on the evaluation set:
88
+ - Loss: 3.2155
89
+
90
+ ## Model description
91
+
92
+ More information needed
93
+
94
+ ## Intended uses & limitations
95
+
96
+ More information needed
97
+
98
+ ## Training and evaluation data
99
+
100
+ More information needed
101
+
102
+ ## Training procedure
103
+
104
+ ### Training hyperparameters
105
+
106
+ The following hyperparameters were used during training:
107
+ - learning_rate: 0.0002
108
+ - train_batch_size: 1
109
+ - eval_batch_size: 1
110
+ - seed: 42
111
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
112
+ - lr_scheduler_type: cosine
113
+ - lr_scheduler_warmup_steps: 100
114
+ - num_epochs: 1
115
+
116
+ ### Training results
117
+
118
+ | Training Loss | Epoch | Step | Validation Loss |
119
+ |:-------------:|:-----:|:----:|:---------------:|
120
+ | 1.7582 | 0.0 | 1 | 2.1282 |
121
+ | 2.837 | 0.25 | 155 | 3.0603 |
122
+ | 2.9122 | 0.5 | 310 | 3.6874 |
123
+ | 2.6502 | 0.75 | 465 | 3.2784 |
124
+ | 1.7745 | 1.0 | 620 | 3.2155 |
125
+
126
+
127
+ ### Framework versions
128
+
129
+ - Transformers 4.38.2
130
+ - Pytorch 2.1.2+cu121
131
+ - Datasets 2.18.0
132
+ - Tokenizers 0.15.0
checkpoint-620/config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 5632,
14
+ "max_position_embeddings": 2048,
15
+ "model_type": "llama",
16
+ "num_attention_heads": 32,
17
+ "num_hidden_layers": 22,
18
+ "num_key_value_heads": 4,
19
+ "pretraining_tp": 1,
20
+ "rms_norm_eps": 1e-05,
21
+ "rope_scaling": null,
22
+ "rope_theta": 10000.0,
23
+ "tie_word_embeddings": false,
24
+ "torch_dtype": "bfloat16",
25
+ "transformers_version": "4.38.2",
26
+ "use_cache": false,
27
+ "vocab_size": 32000
28
+ }
checkpoint-620/generation_config.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 1,
3
+ "do_sample": true,
4
+ "eos_token_id": 2,
5
+ "max_length": 2048,
6
+ "pad_token_id": 0,
7
+ "transformers_version": "4.38.2"
8
+ }
checkpoint-620/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a3e43fc3f62ee9f36a9e27c45c18c706b9d949a85cb693fbdbbe564cfdf10b8
3
+ size 2200117448
checkpoint-620/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9585085b0b168c3fea53ac114e002dee5587f9dd8e8f968f214f73e815caaba
3
+ size 2205132218
checkpoint-620/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c062f7f375beded48b5337f5a3f3a5cb38807fa3e85dbf3e294c0ab6b627bfc2
3
+ size 14244
checkpoint-620/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb100b247904563b3d34e79013feb358059333bd862f9413e61e228a3d530d82
3
+ size 1064
checkpoint-620/trainer_state.json ADDED
@@ -0,0 +1,4401 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 1.0,
5
+ "eval_steps": 155,
6
+ "global_step": 620,
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
+ "grad_norm": 5.8125,
14
+ "learning_rate": 2.0000000000000003e-06,
15
+ "loss": 1.7582,
16
+ "step": 1
17
+ },
18
+ {
19
+ "epoch": 0.0,
20
+ "eval_loss": 2.128222703933716,
21
+ "eval_runtime": 1.7364,
22
+ "eval_samples_per_second": 19.58,
23
+ "eval_steps_per_second": 19.58,
24
+ "step": 1
25
+ },
26
+ {
27
+ "epoch": 0.0,
28
+ "grad_norm": 8.0625,
29
+ "learning_rate": 4.000000000000001e-06,
30
+ "loss": 2.0767,
31
+ "step": 2
32
+ },
33
+ {
34
+ "epoch": 0.0,
35
+ "grad_norm": 5.875,
36
+ "learning_rate": 6e-06,
37
+ "loss": 2.0738,
38
+ "step": 3
39
+ },
40
+ {
41
+ "epoch": 0.01,
42
+ "grad_norm": 6.21875,
43
+ "learning_rate": 8.000000000000001e-06,
44
+ "loss": 1.9586,
45
+ "step": 4
46
+ },
47
+ {
48
+ "epoch": 0.01,
49
+ "grad_norm": 7.75,
50
+ "learning_rate": 1e-05,
51
+ "loss": 2.1597,
52
+ "step": 5
53
+ },
54
+ {
55
+ "epoch": 0.01,
56
+ "grad_norm": 7.1875,
57
+ "learning_rate": 1.2e-05,
58
+ "loss": 1.5356,
59
+ "step": 6
60
+ },
61
+ {
62
+ "epoch": 0.01,
63
+ "grad_norm": 6.53125,
64
+ "learning_rate": 1.4000000000000001e-05,
65
+ "loss": 1.796,
66
+ "step": 7
67
+ },
68
+ {
69
+ "epoch": 0.01,
70
+ "grad_norm": 6.84375,
71
+ "learning_rate": 1.6000000000000003e-05,
72
+ "loss": 2.0644,
73
+ "step": 8
74
+ },
75
+ {
76
+ "epoch": 0.01,
77
+ "grad_norm": 5.4375,
78
+ "learning_rate": 1.8e-05,
79
+ "loss": 1.5361,
80
+ "step": 9
81
+ },
82
+ {
83
+ "epoch": 0.02,
84
+ "grad_norm": 7.125,
85
+ "learning_rate": 2e-05,
86
+ "loss": 2.2476,
87
+ "step": 10
88
+ },
89
+ {
90
+ "epoch": 0.02,
91
+ "grad_norm": 5.90625,
92
+ "learning_rate": 2.2000000000000003e-05,
93
+ "loss": 1.8926,
94
+ "step": 11
95
+ },
96
+ {
97
+ "epoch": 0.02,
98
+ "grad_norm": 6.0625,
99
+ "learning_rate": 2.4e-05,
100
+ "loss": 1.5785,
101
+ "step": 12
102
+ },
103
+ {
104
+ "epoch": 0.02,
105
+ "grad_norm": 5.9375,
106
+ "learning_rate": 2.6000000000000002e-05,
107
+ "loss": 1.5668,
108
+ "step": 13
109
+ },
110
+ {
111
+ "epoch": 0.02,
112
+ "grad_norm": 6.28125,
113
+ "learning_rate": 2.8000000000000003e-05,
114
+ "loss": 1.9217,
115
+ "step": 14
116
+ },
117
+ {
118
+ "epoch": 0.02,
119
+ "grad_norm": 7.78125,
120
+ "learning_rate": 3e-05,
121
+ "loss": 2.0163,
122
+ "step": 15
123
+ },
124
+ {
125
+ "epoch": 0.03,
126
+ "grad_norm": 5.53125,
127
+ "learning_rate": 3.2000000000000005e-05,
128
+ "loss": 2.059,
129
+ "step": 16
130
+ },
131
+ {
132
+ "epoch": 0.03,
133
+ "grad_norm": 5.84375,
134
+ "learning_rate": 3.4000000000000007e-05,
135
+ "loss": 1.9966,
136
+ "step": 17
137
+ },
138
+ {
139
+ "epoch": 0.03,
140
+ "grad_norm": 28.25,
141
+ "learning_rate": 3.6e-05,
142
+ "loss": 2.2663,
143
+ "step": 18
144
+ },
145
+ {
146
+ "epoch": 0.03,
147
+ "grad_norm": 7.53125,
148
+ "learning_rate": 3.8e-05,
149
+ "loss": 2.3443,
150
+ "step": 19
151
+ },
152
+ {
153
+ "epoch": 0.03,
154
+ "grad_norm": 8.1875,
155
+ "learning_rate": 4e-05,
156
+ "loss": 2.3054,
157
+ "step": 20
158
+ },
159
+ {
160
+ "epoch": 0.03,
161
+ "grad_norm": 5.71875,
162
+ "learning_rate": 4.2e-05,
163
+ "loss": 2.1128,
164
+ "step": 21
165
+ },
166
+ {
167
+ "epoch": 0.04,
168
+ "grad_norm": 9.1875,
169
+ "learning_rate": 4.4000000000000006e-05,
170
+ "loss": 2.8513,
171
+ "step": 22
172
+ },
173
+ {
174
+ "epoch": 0.04,
175
+ "grad_norm": 6.0,
176
+ "learning_rate": 4.600000000000001e-05,
177
+ "loss": 1.9679,
178
+ "step": 23
179
+ },
180
+ {
181
+ "epoch": 0.04,
182
+ "grad_norm": 7.5625,
183
+ "learning_rate": 4.8e-05,
184
+ "loss": 2.3664,
185
+ "step": 24
186
+ },
187
+ {
188
+ "epoch": 0.04,
189
+ "grad_norm": 5.34375,
190
+ "learning_rate": 5e-05,
191
+ "loss": 1.8031,
192
+ "step": 25
193
+ },
194
+ {
195
+ "epoch": 0.04,
196
+ "grad_norm": 17.125,
197
+ "learning_rate": 5.2000000000000004e-05,
198
+ "loss": 2.0872,
199
+ "step": 26
200
+ },
201
+ {
202
+ "epoch": 0.04,
203
+ "grad_norm": 5.4375,
204
+ "learning_rate": 5.4000000000000005e-05,
205
+ "loss": 1.8351,
206
+ "step": 27
207
+ },
208
+ {
209
+ "epoch": 0.05,
210
+ "grad_norm": 7.40625,
211
+ "learning_rate": 5.6000000000000006e-05,
212
+ "loss": 2.0825,
213
+ "step": 28
214
+ },
215
+ {
216
+ "epoch": 0.05,
217
+ "grad_norm": 8.25,
218
+ "learning_rate": 5.8e-05,
219
+ "loss": 2.3783,
220
+ "step": 29
221
+ },
222
+ {
223
+ "epoch": 0.05,
224
+ "grad_norm": 5.90625,
225
+ "learning_rate": 6e-05,
226
+ "loss": 1.84,
227
+ "step": 30
228
+ },
229
+ {
230
+ "epoch": 0.05,
231
+ "grad_norm": 6.8125,
232
+ "learning_rate": 6.2e-05,
233
+ "loss": 1.7072,
234
+ "step": 31
235
+ },
236
+ {
237
+ "epoch": 0.05,
238
+ "grad_norm": 5.6875,
239
+ "learning_rate": 6.400000000000001e-05,
240
+ "loss": 1.8357,
241
+ "step": 32
242
+ },
243
+ {
244
+ "epoch": 0.05,
245
+ "grad_norm": 7.15625,
246
+ "learning_rate": 6.6e-05,
247
+ "loss": 1.954,
248
+ "step": 33
249
+ },
250
+ {
251
+ "epoch": 0.05,
252
+ "grad_norm": 7.375,
253
+ "learning_rate": 6.800000000000001e-05,
254
+ "loss": 2.3207,
255
+ "step": 34
256
+ },
257
+ {
258
+ "epoch": 0.06,
259
+ "grad_norm": 7.46875,
260
+ "learning_rate": 7e-05,
261
+ "loss": 2.2977,
262
+ "step": 35
263
+ },
264
+ {
265
+ "epoch": 0.06,
266
+ "grad_norm": 6.71875,
267
+ "learning_rate": 7.2e-05,
268
+ "loss": 2.2092,
269
+ "step": 36
270
+ },
271
+ {
272
+ "epoch": 0.06,
273
+ "grad_norm": 5.4375,
274
+ "learning_rate": 7.4e-05,
275
+ "loss": 1.8733,
276
+ "step": 37
277
+ },
278
+ {
279
+ "epoch": 0.06,
280
+ "grad_norm": 9.125,
281
+ "learning_rate": 7.6e-05,
282
+ "loss": 2.8407,
283
+ "step": 38
284
+ },
285
+ {
286
+ "epoch": 0.06,
287
+ "grad_norm": 7.84375,
288
+ "learning_rate": 7.800000000000001e-05,
289
+ "loss": 2.1865,
290
+ "step": 39
291
+ },
292
+ {
293
+ "epoch": 0.06,
294
+ "grad_norm": 5.40625,
295
+ "learning_rate": 8e-05,
296
+ "loss": 1.7405,
297
+ "step": 40
298
+ },
299
+ {
300
+ "epoch": 0.07,
301
+ "grad_norm": 10.4375,
302
+ "learning_rate": 8.2e-05,
303
+ "loss": 2.3453,
304
+ "step": 41
305
+ },
306
+ {
307
+ "epoch": 0.07,
308
+ "grad_norm": 5.90625,
309
+ "learning_rate": 8.4e-05,
310
+ "loss": 1.6975,
311
+ "step": 42
312
+ },
313
+ {
314
+ "epoch": 0.07,
315
+ "grad_norm": 5.78125,
316
+ "learning_rate": 8.6e-05,
317
+ "loss": 1.9287,
318
+ "step": 43
319
+ },
320
+ {
321
+ "epoch": 0.07,
322
+ "grad_norm": 6.3125,
323
+ "learning_rate": 8.800000000000001e-05,
324
+ "loss": 1.8328,
325
+ "step": 44
326
+ },
327
+ {
328
+ "epoch": 0.07,
329
+ "grad_norm": 11.0,
330
+ "learning_rate": 9e-05,
331
+ "loss": 1.4144,
332
+ "step": 45
333
+ },
334
+ {
335
+ "epoch": 0.07,
336
+ "grad_norm": 6.0,
337
+ "learning_rate": 9.200000000000001e-05,
338
+ "loss": 1.6557,
339
+ "step": 46
340
+ },
341
+ {
342
+ "epoch": 0.08,
343
+ "grad_norm": 9.0,
344
+ "learning_rate": 9.4e-05,
345
+ "loss": 2.4638,
346
+ "step": 47
347
+ },
348
+ {
349
+ "epoch": 0.08,
350
+ "grad_norm": 5.8125,
351
+ "learning_rate": 9.6e-05,
352
+ "loss": 1.8715,
353
+ "step": 48
354
+ },
355
+ {
356
+ "epoch": 0.08,
357
+ "grad_norm": 7.5625,
358
+ "learning_rate": 9.8e-05,
359
+ "loss": 2.3932,
360
+ "step": 49
361
+ },
362
+ {
363
+ "epoch": 0.08,
364
+ "grad_norm": 8.0625,
365
+ "learning_rate": 0.0001,
366
+ "loss": 2.0249,
367
+ "step": 50
368
+ },
369
+ {
370
+ "epoch": 0.08,
371
+ "grad_norm": 9.0625,
372
+ "learning_rate": 0.00010200000000000001,
373
+ "loss": 2.2521,
374
+ "step": 51
375
+ },
376
+ {
377
+ "epoch": 0.08,
378
+ "grad_norm": 7.4375,
379
+ "learning_rate": 0.00010400000000000001,
380
+ "loss": 2.2294,
381
+ "step": 52
382
+ },
383
+ {
384
+ "epoch": 0.09,
385
+ "grad_norm": 8.625,
386
+ "learning_rate": 0.00010600000000000002,
387
+ "loss": 2.0424,
388
+ "step": 53
389
+ },
390
+ {
391
+ "epoch": 0.09,
392
+ "grad_norm": 8.25,
393
+ "learning_rate": 0.00010800000000000001,
394
+ "loss": 2.4381,
395
+ "step": 54
396
+ },
397
+ {
398
+ "epoch": 0.09,
399
+ "grad_norm": 7.5625,
400
+ "learning_rate": 0.00011000000000000002,
401
+ "loss": 1.982,
402
+ "step": 55
403
+ },
404
+ {
405
+ "epoch": 0.09,
406
+ "grad_norm": 8.6875,
407
+ "learning_rate": 0.00011200000000000001,
408
+ "loss": 1.9878,
409
+ "step": 56
410
+ },
411
+ {
412
+ "epoch": 0.09,
413
+ "grad_norm": 6.21875,
414
+ "learning_rate": 0.00011399999999999999,
415
+ "loss": 2.0957,
416
+ "step": 57
417
+ },
418
+ {
419
+ "epoch": 0.09,
420
+ "grad_norm": 5.59375,
421
+ "learning_rate": 0.000116,
422
+ "loss": 2.0157,
423
+ "step": 58
424
+ },
425
+ {
426
+ "epoch": 0.1,
427
+ "grad_norm": 7.71875,
428
+ "learning_rate": 0.000118,
429
+ "loss": 1.8377,
430
+ "step": 59
431
+ },
432
+ {
433
+ "epoch": 0.1,
434
+ "grad_norm": 5.40625,
435
+ "learning_rate": 0.00012,
436
+ "loss": 2.0343,
437
+ "step": 60
438
+ },
439
+ {
440
+ "epoch": 0.1,
441
+ "grad_norm": 6.65625,
442
+ "learning_rate": 0.000122,
443
+ "loss": 1.8884,
444
+ "step": 61
445
+ },
446
+ {
447
+ "epoch": 0.1,
448
+ "grad_norm": 14.3125,
449
+ "learning_rate": 0.000124,
450
+ "loss": 2.3979,
451
+ "step": 62
452
+ },
453
+ {
454
+ "epoch": 0.1,
455
+ "grad_norm": 10.3125,
456
+ "learning_rate": 0.000126,
457
+ "loss": 2.0596,
458
+ "step": 63
459
+ },
460
+ {
461
+ "epoch": 0.1,
462
+ "grad_norm": 13.375,
463
+ "learning_rate": 0.00012800000000000002,
464
+ "loss": 2.7197,
465
+ "step": 64
466
+ },
467
+ {
468
+ "epoch": 0.1,
469
+ "grad_norm": 6.75,
470
+ "learning_rate": 0.00013000000000000002,
471
+ "loss": 1.6793,
472
+ "step": 65
473
+ },
474
+ {
475
+ "epoch": 0.11,
476
+ "grad_norm": 12.8125,
477
+ "learning_rate": 0.000132,
478
+ "loss": 2.9125,
479
+ "step": 66
480
+ },
481
+ {
482
+ "epoch": 0.11,
483
+ "grad_norm": 6.90625,
484
+ "learning_rate": 0.000134,
485
+ "loss": 2.0604,
486
+ "step": 67
487
+ },
488
+ {
489
+ "epoch": 0.11,
490
+ "grad_norm": 8.4375,
491
+ "learning_rate": 0.00013600000000000003,
492
+ "loss": 2.1453,
493
+ "step": 68
494
+ },
495
+ {
496
+ "epoch": 0.11,
497
+ "grad_norm": 6.9375,
498
+ "learning_rate": 0.000138,
499
+ "loss": 2.4704,
500
+ "step": 69
501
+ },
502
+ {
503
+ "epoch": 0.11,
504
+ "grad_norm": 11.125,
505
+ "learning_rate": 0.00014,
506
+ "loss": 2.575,
507
+ "step": 70
508
+ },
509
+ {
510
+ "epoch": 0.11,
511
+ "grad_norm": 10.0,
512
+ "learning_rate": 0.000142,
513
+ "loss": 2.8627,
514
+ "step": 71
515
+ },
516
+ {
517
+ "epoch": 0.12,
518
+ "grad_norm": 8.625,
519
+ "learning_rate": 0.000144,
520
+ "loss": 2.3511,
521
+ "step": 72
522
+ },
523
+ {
524
+ "epoch": 0.12,
525
+ "grad_norm": 7.84375,
526
+ "learning_rate": 0.000146,
527
+ "loss": 2.2555,
528
+ "step": 73
529
+ },
530
+ {
531
+ "epoch": 0.12,
532
+ "grad_norm": 10.0,
533
+ "learning_rate": 0.000148,
534
+ "loss": 2.2663,
535
+ "step": 74
536
+ },
537
+ {
538
+ "epoch": 0.12,
539
+ "grad_norm": 10.0,
540
+ "learning_rate": 0.00015000000000000001,
541
+ "loss": 2.4242,
542
+ "step": 75
543
+ },
544
+ {
545
+ "epoch": 0.12,
546
+ "grad_norm": 9.1875,
547
+ "learning_rate": 0.000152,
548
+ "loss": 3.314,
549
+ "step": 76
550
+ },
551
+ {
552
+ "epoch": 0.12,
553
+ "grad_norm": 12.75,
554
+ "learning_rate": 0.000154,
555
+ "loss": 2.9263,
556
+ "step": 77
557
+ },
558
+ {
559
+ "epoch": 0.13,
560
+ "grad_norm": 8.8125,
561
+ "learning_rate": 0.00015600000000000002,
562
+ "loss": 2.7491,
563
+ "step": 78
564
+ },
565
+ {
566
+ "epoch": 0.13,
567
+ "grad_norm": 6.34375,
568
+ "learning_rate": 0.00015800000000000002,
569
+ "loss": 2.2093,
570
+ "step": 79
571
+ },
572
+ {
573
+ "epoch": 0.13,
574
+ "grad_norm": 15.0,
575
+ "learning_rate": 0.00016,
576
+ "loss": 2.6103,
577
+ "step": 80
578
+ },
579
+ {
580
+ "epoch": 0.13,
581
+ "grad_norm": 14.625,
582
+ "learning_rate": 0.000162,
583
+ "loss": 3.2124,
584
+ "step": 81
585
+ },
586
+ {
587
+ "epoch": 0.13,
588
+ "grad_norm": 8.0,
589
+ "learning_rate": 0.000164,
590
+ "loss": 2.3414,
591
+ "step": 82
592
+ },
593
+ {
594
+ "epoch": 0.13,
595
+ "grad_norm": 13.3125,
596
+ "learning_rate": 0.000166,
597
+ "loss": 2.1735,
598
+ "step": 83
599
+ },
600
+ {
601
+ "epoch": 0.14,
602
+ "grad_norm": 11.1875,
603
+ "learning_rate": 0.000168,
604
+ "loss": 2.4399,
605
+ "step": 84
606
+ },
607
+ {
608
+ "epoch": 0.14,
609
+ "grad_norm": 10.25,
610
+ "learning_rate": 0.00017,
611
+ "loss": 2.8319,
612
+ "step": 85
613
+ },
614
+ {
615
+ "epoch": 0.14,
616
+ "grad_norm": 6.28125,
617
+ "learning_rate": 0.000172,
618
+ "loss": 2.3573,
619
+ "step": 86
620
+ },
621
+ {
622
+ "epoch": 0.14,
623
+ "grad_norm": 20.875,
624
+ "learning_rate": 0.000174,
625
+ "loss": 3.0671,
626
+ "step": 87
627
+ },
628
+ {
629
+ "epoch": 0.14,
630
+ "grad_norm": 8.75,
631
+ "learning_rate": 0.00017600000000000002,
632
+ "loss": 2.5408,
633
+ "step": 88
634
+ },
635
+ {
636
+ "epoch": 0.14,
637
+ "grad_norm": 6.40625,
638
+ "learning_rate": 0.00017800000000000002,
639
+ "loss": 2.2219,
640
+ "step": 89
641
+ },
642
+ {
643
+ "epoch": 0.15,
644
+ "grad_norm": 9.125,
645
+ "learning_rate": 0.00018,
646
+ "loss": 2.4576,
647
+ "step": 90
648
+ },
649
+ {
650
+ "epoch": 0.15,
651
+ "grad_norm": 19.375,
652
+ "learning_rate": 0.000182,
653
+ "loss": 2.8095,
654
+ "step": 91
655
+ },
656
+ {
657
+ "epoch": 0.15,
658
+ "grad_norm": 9.75,
659
+ "learning_rate": 0.00018400000000000003,
660
+ "loss": 2.7362,
661
+ "step": 92
662
+ },
663
+ {
664
+ "epoch": 0.15,
665
+ "grad_norm": 10.25,
666
+ "learning_rate": 0.00018600000000000002,
667
+ "loss": 2.1255,
668
+ "step": 93
669
+ },
670
+ {
671
+ "epoch": 0.15,
672
+ "grad_norm": 11.375,
673
+ "learning_rate": 0.000188,
674
+ "loss": 2.5562,
675
+ "step": 94
676
+ },
677
+ {
678
+ "epoch": 0.15,
679
+ "grad_norm": 8.8125,
680
+ "learning_rate": 0.00019,
681
+ "loss": 2.2949,
682
+ "step": 95
683
+ },
684
+ {
685
+ "epoch": 0.15,
686
+ "grad_norm": 9.0,
687
+ "learning_rate": 0.000192,
688
+ "loss": 2.5435,
689
+ "step": 96
690
+ },
691
+ {
692
+ "epoch": 0.16,
693
+ "grad_norm": 11.25,
694
+ "learning_rate": 0.000194,
695
+ "loss": 2.6474,
696
+ "step": 97
697
+ },
698
+ {
699
+ "epoch": 0.16,
700
+ "grad_norm": 12.625,
701
+ "learning_rate": 0.000196,
702
+ "loss": 2.4485,
703
+ "step": 98
704
+ },
705
+ {
706
+ "epoch": 0.16,
707
+ "grad_norm": 11.875,
708
+ "learning_rate": 0.00019800000000000002,
709
+ "loss": 2.402,
710
+ "step": 99
711
+ },
712
+ {
713
+ "epoch": 0.16,
714
+ "grad_norm": 8.9375,
715
+ "learning_rate": 0.0002,
716
+ "loss": 2.0389,
717
+ "step": 100
718
+ },
719
+ {
720
+ "epoch": 0.16,
721
+ "grad_norm": 7.40625,
722
+ "learning_rate": 0.00019999817500473724,
723
+ "loss": 2.392,
724
+ "step": 101
725
+ },
726
+ {
727
+ "epoch": 0.16,
728
+ "grad_norm": 12.625,
729
+ "learning_rate": 0.00019999270008556108,
730
+ "loss": 2.6223,
731
+ "step": 102
732
+ },
733
+ {
734
+ "epoch": 0.17,
735
+ "grad_norm": 9.9375,
736
+ "learning_rate": 0.00019998357544230558,
737
+ "loss": 2.7335,
738
+ "step": 103
739
+ },
740
+ {
741
+ "epoch": 0.17,
742
+ "grad_norm": 7.75,
743
+ "learning_rate": 0.00019997080140801932,
744
+ "loss": 2.0729,
745
+ "step": 104
746
+ },
747
+ {
748
+ "epoch": 0.17,
749
+ "grad_norm": 7.46875,
750
+ "learning_rate": 0.00019995437844895334,
751
+ "loss": 2.6476,
752
+ "step": 105
753
+ },
754
+ {
755
+ "epoch": 0.17,
756
+ "grad_norm": 8.625,
757
+ "learning_rate": 0.00019993430716454413,
758
+ "loss": 2.0718,
759
+ "step": 106
760
+ },
761
+ {
762
+ "epoch": 0.17,
763
+ "grad_norm": 8.8125,
764
+ "learning_rate": 0.00019991058828739165,
765
+ "loss": 2.5581,
766
+ "step": 107
767
+ },
768
+ {
769
+ "epoch": 0.17,
770
+ "grad_norm": 8.5625,
771
+ "learning_rate": 0.00019988322268323268,
772
+ "loss": 2.6858,
773
+ "step": 108
774
+ },
775
+ {
776
+ "epoch": 0.18,
777
+ "grad_norm": 7.375,
778
+ "learning_rate": 0.00019985221135090914,
779
+ "loss": 2.3407,
780
+ "step": 109
781
+ },
782
+ {
783
+ "epoch": 0.18,
784
+ "grad_norm": 10.0625,
785
+ "learning_rate": 0.00019981755542233177,
786
+ "loss": 2.302,
787
+ "step": 110
788
+ },
789
+ {
790
+ "epoch": 0.18,
791
+ "grad_norm": 8.75,
792
+ "learning_rate": 0.00019977925616243862,
793
+ "loss": 2.5327,
794
+ "step": 111
795
+ },
796
+ {
797
+ "epoch": 0.18,
798
+ "grad_norm": 12.25,
799
+ "learning_rate": 0.00019973731496914914,
800
+ "loss": 2.9085,
801
+ "step": 112
802
+ },
803
+ {
804
+ "epoch": 0.18,
805
+ "grad_norm": 9.0,
806
+ "learning_rate": 0.0001996917333733128,
807
+ "loss": 2.5231,
808
+ "step": 113
809
+ },
810
+ {
811
+ "epoch": 0.18,
812
+ "grad_norm": 20.875,
813
+ "learning_rate": 0.00019964251303865362,
814
+ "loss": 3.2295,
815
+ "step": 114
816
+ },
817
+ {
818
+ "epoch": 0.19,
819
+ "grad_norm": 556.0,
820
+ "learning_rate": 0.00019958965576170908,
821
+ "loss": 3.058,
822
+ "step": 115
823
+ },
824
+ {
825
+ "epoch": 0.19,
826
+ "grad_norm": 14.75,
827
+ "learning_rate": 0.00019953316347176488,
828
+ "loss": 2.9174,
829
+ "step": 116
830
+ },
831
+ {
832
+ "epoch": 0.19,
833
+ "grad_norm": 9.75,
834
+ "learning_rate": 0.00019947303823078416,
835
+ "loss": 2.8341,
836
+ "step": 117
837
+ },
838
+ {
839
+ "epoch": 0.19,
840
+ "grad_norm": 11.0625,
841
+ "learning_rate": 0.00019940928223333252,
842
+ "loss": 2.374,
843
+ "step": 118
844
+ },
845
+ {
846
+ "epoch": 0.19,
847
+ "grad_norm": 12.0,
848
+ "learning_rate": 0.0001993418978064979,
849
+ "loss": 2.6879,
850
+ "step": 119
851
+ },
852
+ {
853
+ "epoch": 0.19,
854
+ "grad_norm": 138.0,
855
+ "learning_rate": 0.0001992708874098054,
856
+ "loss": 4.0829,
857
+ "step": 120
858
+ },
859
+ {
860
+ "epoch": 0.2,
861
+ "grad_norm": 101.5,
862
+ "learning_rate": 0.00019919625363512786,
863
+ "loss": 7.0349,
864
+ "step": 121
865
+ },
866
+ {
867
+ "epoch": 0.2,
868
+ "grad_norm": 9.5625,
869
+ "learning_rate": 0.00019911799920659093,
870
+ "loss": 2.2249,
871
+ "step": 122
872
+ },
873
+ {
874
+ "epoch": 0.2,
875
+ "grad_norm": 11.625,
876
+ "learning_rate": 0.00019903612698047383,
877
+ "loss": 2.9503,
878
+ "step": 123
879
+ },
880
+ {
881
+ "epoch": 0.2,
882
+ "grad_norm": 24.625,
883
+ "learning_rate": 0.0001989506399451051,
884
+ "loss": 3.3981,
885
+ "step": 124
886
+ },
887
+ {
888
+ "epoch": 0.2,
889
+ "grad_norm": 8.5625,
890
+ "learning_rate": 0.00019886154122075343,
891
+ "loss": 2.082,
892
+ "step": 125
893
+ },
894
+ {
895
+ "epoch": 0.2,
896
+ "grad_norm": 21.125,
897
+ "learning_rate": 0.00019876883405951377,
898
+ "loss": 2.2787,
899
+ "step": 126
900
+ },
901
+ {
902
+ "epoch": 0.2,
903
+ "grad_norm": 10.1875,
904
+ "learning_rate": 0.00019867252184518878,
905
+ "loss": 2.3117,
906
+ "step": 127
907
+ },
908
+ {
909
+ "epoch": 0.21,
910
+ "grad_norm": 13.3125,
911
+ "learning_rate": 0.0001985726080931651,
912
+ "loss": 2.2723,
913
+ "step": 128
914
+ },
915
+ {
916
+ "epoch": 0.21,
917
+ "grad_norm": 19.875,
918
+ "learning_rate": 0.00019846909645028523,
919
+ "loss": 2.2222,
920
+ "step": 129
921
+ },
922
+ {
923
+ "epoch": 0.21,
924
+ "grad_norm": 14.375,
925
+ "learning_rate": 0.00019836199069471437,
926
+ "loss": 2.6681,
927
+ "step": 130
928
+ },
929
+ {
930
+ "epoch": 0.21,
931
+ "grad_norm": 9.6875,
932
+ "learning_rate": 0.0001982512947358024,
933
+ "loss": 2.2032,
934
+ "step": 131
935
+ },
936
+ {
937
+ "epoch": 0.21,
938
+ "grad_norm": 17.5,
939
+ "learning_rate": 0.00019813701261394136,
940
+ "loss": 3.0575,
941
+ "step": 132
942
+ },
943
+ {
944
+ "epoch": 0.21,
945
+ "grad_norm": 31.0,
946
+ "learning_rate": 0.00019801914850041784,
947
+ "loss": 2.4558,
948
+ "step": 133
949
+ },
950
+ {
951
+ "epoch": 0.22,
952
+ "grad_norm": 18.75,
953
+ "learning_rate": 0.00019789770669726087,
954
+ "loss": 2.741,
955
+ "step": 134
956
+ },
957
+ {
958
+ "epoch": 0.22,
959
+ "grad_norm": 8.625,
960
+ "learning_rate": 0.00019777269163708468,
961
+ "loss": 1.9201,
962
+ "step": 135
963
+ },
964
+ {
965
+ "epoch": 0.22,
966
+ "grad_norm": 14.1875,
967
+ "learning_rate": 0.00019764410788292722,
968
+ "loss": 2.8184,
969
+ "step": 136
970
+ },
971
+ {
972
+ "epoch": 0.22,
973
+ "grad_norm": 30.875,
974
+ "learning_rate": 0.00019751196012808325,
975
+ "loss": 2.7528,
976
+ "step": 137
977
+ },
978
+ {
979
+ "epoch": 0.22,
980
+ "grad_norm": 67.0,
981
+ "learning_rate": 0.00019737625319593335,
982
+ "loss": 3.0919,
983
+ "step": 138
984
+ },
985
+ {
986
+ "epoch": 0.22,
987
+ "grad_norm": 9.125,
988
+ "learning_rate": 0.00019723699203976766,
989
+ "loss": 2.6063,
990
+ "step": 139
991
+ },
992
+ {
993
+ "epoch": 0.23,
994
+ "grad_norm": 17.625,
995
+ "learning_rate": 0.0001970941817426052,
996
+ "loss": 2.9237,
997
+ "step": 140
998
+ },
999
+ {
1000
+ "epoch": 0.23,
1001
+ "grad_norm": 115.0,
1002
+ "learning_rate": 0.00019694782751700828,
1003
+ "loss": 3.5595,
1004
+ "step": 141
1005
+ },
1006
+ {
1007
+ "epoch": 0.23,
1008
+ "grad_norm": 15.125,
1009
+ "learning_rate": 0.00019679793470489228,
1010
+ "loss": 2.4841,
1011
+ "step": 142
1012
+ },
1013
+ {
1014
+ "epoch": 0.23,
1015
+ "grad_norm": 9.9375,
1016
+ "learning_rate": 0.00019664450877733062,
1017
+ "loss": 2.1646,
1018
+ "step": 143
1019
+ },
1020
+ {
1021
+ "epoch": 0.23,
1022
+ "grad_norm": 58.0,
1023
+ "learning_rate": 0.00019648755533435518,
1024
+ "loss": 2.4607,
1025
+ "step": 144
1026
+ },
1027
+ {
1028
+ "epoch": 0.23,
1029
+ "grad_norm": 10.9375,
1030
+ "learning_rate": 0.00019632708010475165,
1031
+ "loss": 1.9877,
1032
+ "step": 145
1033
+ },
1034
+ {
1035
+ "epoch": 0.24,
1036
+ "grad_norm": 9.1875,
1037
+ "learning_rate": 0.00019616308894585078,
1038
+ "loss": 2.1341,
1039
+ "step": 146
1040
+ },
1041
+ {
1042
+ "epoch": 0.24,
1043
+ "grad_norm": 11.375,
1044
+ "learning_rate": 0.0001959955878433143,
1045
+ "loss": 3.0362,
1046
+ "step": 147
1047
+ },
1048
+ {
1049
+ "epoch": 0.24,
1050
+ "grad_norm": 2736.0,
1051
+ "learning_rate": 0.00019582458291091663,
1052
+ "loss": 2.8518,
1053
+ "step": 148
1054
+ },
1055
+ {
1056
+ "epoch": 0.24,
1057
+ "grad_norm": 14.0,
1058
+ "learning_rate": 0.00019565008039032158,
1059
+ "loss": 2.9418,
1060
+ "step": 149
1061
+ },
1062
+ {
1063
+ "epoch": 0.24,
1064
+ "grad_norm": 16.125,
1065
+ "learning_rate": 0.00019547208665085457,
1066
+ "loss": 2.6276,
1067
+ "step": 150
1068
+ },
1069
+ {
1070
+ "epoch": 0.24,
1071
+ "grad_norm": 24.0,
1072
+ "learning_rate": 0.0001952906081892703,
1073
+ "loss": 2.4923,
1074
+ "step": 151
1075
+ },
1076
+ {
1077
+ "epoch": 0.25,
1078
+ "grad_norm": 12.6875,
1079
+ "learning_rate": 0.00019510565162951537,
1080
+ "loss": 3.0933,
1081
+ "step": 152
1082
+ },
1083
+ {
1084
+ "epoch": 0.25,
1085
+ "grad_norm": 188.0,
1086
+ "learning_rate": 0.0001949172237224867,
1087
+ "loss": 2.8266,
1088
+ "step": 153
1089
+ },
1090
+ {
1091
+ "epoch": 0.25,
1092
+ "grad_norm": 115.5,
1093
+ "learning_rate": 0.00019472533134578507,
1094
+ "loss": 2.5288,
1095
+ "step": 154
1096
+ },
1097
+ {
1098
+ "epoch": 0.25,
1099
+ "grad_norm": 42.25,
1100
+ "learning_rate": 0.00019452998150346401,
1101
+ "loss": 2.837,
1102
+ "step": 155
1103
+ },
1104
+ {
1105
+ "epoch": 0.25,
1106
+ "eval_loss": 3.0602517127990723,
1107
+ "eval_runtime": 1.7317,
1108
+ "eval_samples_per_second": 19.633,
1109
+ "eval_steps_per_second": 19.633,
1110
+ "step": 155
1111
+ },
1112
+ {
1113
+ "epoch": 0.25,
1114
+ "grad_norm": 9.125,
1115
+ "learning_rate": 0.0001943311813257743,
1116
+ "loss": 2.2013,
1117
+ "step": 156
1118
+ },
1119
+ {
1120
+ "epoch": 0.25,
1121
+ "grad_norm": 12.8125,
1122
+ "learning_rate": 0.00019412893806890357,
1123
+ "loss": 2.3497,
1124
+ "step": 157
1125
+ },
1126
+ {
1127
+ "epoch": 0.25,
1128
+ "grad_norm": 12.6875,
1129
+ "learning_rate": 0.00019392325911471155,
1130
+ "loss": 2.4373,
1131
+ "step": 158
1132
+ },
1133
+ {
1134
+ "epoch": 0.26,
1135
+ "grad_norm": 11.875,
1136
+ "learning_rate": 0.00019371415197046052,
1137
+ "loss": 2.4359,
1138
+ "step": 159
1139
+ },
1140
+ {
1141
+ "epoch": 0.26,
1142
+ "grad_norm": 7.3125,
1143
+ "learning_rate": 0.0001935016242685415,
1144
+ "loss": 2.5483,
1145
+ "step": 160
1146
+ },
1147
+ {
1148
+ "epoch": 0.26,
1149
+ "grad_norm": 1256.0,
1150
+ "learning_rate": 0.00019328568376619543,
1151
+ "loss": 3.7604,
1152
+ "step": 161
1153
+ },
1154
+ {
1155
+ "epoch": 0.26,
1156
+ "grad_norm": 10.4375,
1157
+ "learning_rate": 0.00019306633834523024,
1158
+ "loss": 2.0609,
1159
+ "step": 162
1160
+ },
1161
+ {
1162
+ "epoch": 0.26,
1163
+ "grad_norm": 237.0,
1164
+ "learning_rate": 0.00019284359601173294,
1165
+ "loss": 8.9874,
1166
+ "step": 163
1167
+ },
1168
+ {
1169
+ "epoch": 0.26,
1170
+ "grad_norm": 9.75,
1171
+ "learning_rate": 0.00019261746489577765,
1172
+ "loss": 2.4526,
1173
+ "step": 164
1174
+ },
1175
+ {
1176
+ "epoch": 0.27,
1177
+ "grad_norm": 13.1875,
1178
+ "learning_rate": 0.0001923879532511287,
1179
+ "loss": 3.1912,
1180
+ "step": 165
1181
+ },
1182
+ {
1183
+ "epoch": 0.27,
1184
+ "grad_norm": 8.125,
1185
+ "learning_rate": 0.0001921550694549393,
1186
+ "loss": 2.632,
1187
+ "step": 166
1188
+ },
1189
+ {
1190
+ "epoch": 0.27,
1191
+ "grad_norm": 9.375,
1192
+ "learning_rate": 0.000191918822007446,
1193
+ "loss": 2.5058,
1194
+ "step": 167
1195
+ },
1196
+ {
1197
+ "epoch": 0.27,
1198
+ "grad_norm": 78.5,
1199
+ "learning_rate": 0.00019167921953165825,
1200
+ "loss": 3.5444,
1201
+ "step": 168
1202
+ },
1203
+ {
1204
+ "epoch": 0.27,
1205
+ "grad_norm": 22.875,
1206
+ "learning_rate": 0.0001914362707730437,
1207
+ "loss": 2.5383,
1208
+ "step": 169
1209
+ },
1210
+ {
1211
+ "epoch": 0.27,
1212
+ "grad_norm": 12.8125,
1213
+ "learning_rate": 0.00019118998459920902,
1214
+ "loss": 2.8637,
1215
+ "step": 170
1216
+ },
1217
+ {
1218
+ "epoch": 0.28,
1219
+ "grad_norm": 19.5,
1220
+ "learning_rate": 0.00019094036999957624,
1221
+ "loss": 2.8096,
1222
+ "step": 171
1223
+ },
1224
+ {
1225
+ "epoch": 0.28,
1226
+ "grad_norm": 9.0625,
1227
+ "learning_rate": 0.00019068743608505455,
1228
+ "loss": 2.3494,
1229
+ "step": 172
1230
+ },
1231
+ {
1232
+ "epoch": 0.28,
1233
+ "grad_norm": 14.875,
1234
+ "learning_rate": 0.00019043119208770793,
1235
+ "loss": 2.2318,
1236
+ "step": 173
1237
+ },
1238
+ {
1239
+ "epoch": 0.28,
1240
+ "grad_norm": 11.4375,
1241
+ "learning_rate": 0.00019017164736041795,
1242
+ "loss": 2.6877,
1243
+ "step": 174
1244
+ },
1245
+ {
1246
+ "epoch": 0.28,
1247
+ "grad_norm": 7.28125,
1248
+ "learning_rate": 0.00018990881137654258,
1249
+ "loss": 2.918,
1250
+ "step": 175
1251
+ },
1252
+ {
1253
+ "epoch": 0.28,
1254
+ "grad_norm": 8.1875,
1255
+ "learning_rate": 0.00018964269372957038,
1256
+ "loss": 2.6105,
1257
+ "step": 176
1258
+ },
1259
+ {
1260
+ "epoch": 0.29,
1261
+ "grad_norm": 8.6875,
1262
+ "learning_rate": 0.0001893733041327702,
1263
+ "loss": 2.3586,
1264
+ "step": 177
1265
+ },
1266
+ {
1267
+ "epoch": 0.29,
1268
+ "grad_norm": 9.625,
1269
+ "learning_rate": 0.0001891006524188368,
1270
+ "loss": 2.3024,
1271
+ "step": 178
1272
+ },
1273
+ {
1274
+ "epoch": 0.29,
1275
+ "grad_norm": 48.25,
1276
+ "learning_rate": 0.0001888247485395319,
1277
+ "loss": 2.403,
1278
+ "step": 179
1279
+ },
1280
+ {
1281
+ "epoch": 0.29,
1282
+ "grad_norm": 50.25,
1283
+ "learning_rate": 0.000188545602565321,
1284
+ "loss": 3.2945,
1285
+ "step": 180
1286
+ },
1287
+ {
1288
+ "epoch": 0.29,
1289
+ "grad_norm": 21.0,
1290
+ "learning_rate": 0.00018826322468500566,
1291
+ "loss": 2.7344,
1292
+ "step": 181
1293
+ },
1294
+ {
1295
+ "epoch": 0.29,
1296
+ "grad_norm": 10.75,
1297
+ "learning_rate": 0.00018797762520535177,
1298
+ "loss": 2.9573,
1299
+ "step": 182
1300
+ },
1301
+ {
1302
+ "epoch": 0.3,
1303
+ "grad_norm": 10.25,
1304
+ "learning_rate": 0.00018768881455071332,
1305
+ "loss": 2.4458,
1306
+ "step": 183
1307
+ },
1308
+ {
1309
+ "epoch": 0.3,
1310
+ "grad_norm": 9.5,
1311
+ "learning_rate": 0.0001873968032626518,
1312
+ "loss": 3.0015,
1313
+ "step": 184
1314
+ },
1315
+ {
1316
+ "epoch": 0.3,
1317
+ "grad_norm": 7.28125,
1318
+ "learning_rate": 0.00018710160199955156,
1319
+ "loss": 2.7771,
1320
+ "step": 185
1321
+ },
1322
+ {
1323
+ "epoch": 0.3,
1324
+ "grad_norm": 8.875,
1325
+ "learning_rate": 0.00018680322153623075,
1326
+ "loss": 2.7048,
1327
+ "step": 186
1328
+ },
1329
+ {
1330
+ "epoch": 0.3,
1331
+ "grad_norm": 11.375,
1332
+ "learning_rate": 0.000186501672763548,
1333
+ "loss": 2.7448,
1334
+ "step": 187
1335
+ },
1336
+ {
1337
+ "epoch": 0.3,
1338
+ "grad_norm": 11.125,
1339
+ "learning_rate": 0.00018619696668800492,
1340
+ "loss": 2.8452,
1341
+ "step": 188
1342
+ },
1343
+ {
1344
+ "epoch": 0.3,
1345
+ "grad_norm": 49.0,
1346
+ "learning_rate": 0.00018588911443134448,
1347
+ "loss": 3.1947,
1348
+ "step": 189
1349
+ },
1350
+ {
1351
+ "epoch": 0.31,
1352
+ "grad_norm": 28.625,
1353
+ "learning_rate": 0.00018557812723014476,
1354
+ "loss": 2.7165,
1355
+ "step": 190
1356
+ },
1357
+ {
1358
+ "epoch": 0.31,
1359
+ "grad_norm": 230.0,
1360
+ "learning_rate": 0.00018526401643540922,
1361
+ "loss": 10.3406,
1362
+ "step": 191
1363
+ },
1364
+ {
1365
+ "epoch": 0.31,
1366
+ "grad_norm": 21.125,
1367
+ "learning_rate": 0.0001849467935121521,
1368
+ "loss": 3.2117,
1369
+ "step": 192
1370
+ },
1371
+ {
1372
+ "epoch": 0.31,
1373
+ "grad_norm": 26.0,
1374
+ "learning_rate": 0.00018462647003898006,
1375
+ "loss": 3.3466,
1376
+ "step": 193
1377
+ },
1378
+ {
1379
+ "epoch": 0.31,
1380
+ "grad_norm": 816.0,
1381
+ "learning_rate": 0.00018430305770766948,
1382
+ "loss": 11.3285,
1383
+ "step": 194
1384
+ },
1385
+ {
1386
+ "epoch": 0.31,
1387
+ "grad_norm": 15.375,
1388
+ "learning_rate": 0.0001839765683227398,
1389
+ "loss": 3.1025,
1390
+ "step": 195
1391
+ },
1392
+ {
1393
+ "epoch": 0.32,
1394
+ "grad_norm": 18.5,
1395
+ "learning_rate": 0.00018364701380102266,
1396
+ "loss": 2.7839,
1397
+ "step": 196
1398
+ },
1399
+ {
1400
+ "epoch": 0.32,
1401
+ "grad_norm": 45.0,
1402
+ "learning_rate": 0.00018331440617122696,
1403
+ "loss": 9.9088,
1404
+ "step": 197
1405
+ },
1406
+ {
1407
+ "epoch": 0.32,
1408
+ "grad_norm": 10.1875,
1409
+ "learning_rate": 0.00018297875757349952,
1410
+ "loss": 2.9432,
1411
+ "step": 198
1412
+ },
1413
+ {
1414
+ "epoch": 0.32,
1415
+ "grad_norm": 9.4375,
1416
+ "learning_rate": 0.00018264008025898248,
1417
+ "loss": 2.5301,
1418
+ "step": 199
1419
+ },
1420
+ {
1421
+ "epoch": 0.32,
1422
+ "grad_norm": 7.53125,
1423
+ "learning_rate": 0.00018229838658936564,
1424
+ "loss": 2.7123,
1425
+ "step": 200
1426
+ },
1427
+ {
1428
+ "epoch": 0.32,
1429
+ "grad_norm": 24.0,
1430
+ "learning_rate": 0.00018195368903643563,
1431
+ "loss": 2.5865,
1432
+ "step": 201
1433
+ },
1434
+ {
1435
+ "epoch": 0.33,
1436
+ "grad_norm": 9.5625,
1437
+ "learning_rate": 0.0001816060001816205,
1438
+ "loss": 2.0398,
1439
+ "step": 202
1440
+ },
1441
+ {
1442
+ "epoch": 0.33,
1443
+ "grad_norm": 16.625,
1444
+ "learning_rate": 0.00018125533271553043,
1445
+ "loss": 2.7481,
1446
+ "step": 203
1447
+ },
1448
+ {
1449
+ "epoch": 0.33,
1450
+ "grad_norm": 10.9375,
1451
+ "learning_rate": 0.00018090169943749476,
1452
+ "loss": 2.0515,
1453
+ "step": 204
1454
+ },
1455
+ {
1456
+ "epoch": 0.33,
1457
+ "grad_norm": 318.0,
1458
+ "learning_rate": 0.0001805451132550946,
1459
+ "loss": 2.908,
1460
+ "step": 205
1461
+ },
1462
+ {
1463
+ "epoch": 0.33,
1464
+ "grad_norm": 7.8125,
1465
+ "learning_rate": 0.00018018558718369186,
1466
+ "loss": 2.2113,
1467
+ "step": 206
1468
+ },
1469
+ {
1470
+ "epoch": 0.33,
1471
+ "grad_norm": 37.75,
1472
+ "learning_rate": 0.00017982313434595406,
1473
+ "loss": 2.0329,
1474
+ "step": 207
1475
+ },
1476
+ {
1477
+ "epoch": 0.34,
1478
+ "grad_norm": 612.0,
1479
+ "learning_rate": 0.00017945776797137543,
1480
+ "loss": 7.6253,
1481
+ "step": 208
1482
+ },
1483
+ {
1484
+ "epoch": 0.34,
1485
+ "grad_norm": 18.5,
1486
+ "learning_rate": 0.00017908950139579406,
1487
+ "loss": 3.1603,
1488
+ "step": 209
1489
+ },
1490
+ {
1491
+ "epoch": 0.34,
1492
+ "grad_norm": 16.0,
1493
+ "learning_rate": 0.00017871834806090501,
1494
+ "loss": 2.2803,
1495
+ "step": 210
1496
+ },
1497
+ {
1498
+ "epoch": 0.34,
1499
+ "grad_norm": 14.25,
1500
+ "learning_rate": 0.0001783443215137699,
1501
+ "loss": 2.2342,
1502
+ "step": 211
1503
+ },
1504
+ {
1505
+ "epoch": 0.34,
1506
+ "grad_norm": 7.28125,
1507
+ "learning_rate": 0.00017796743540632223,
1508
+ "loss": 2.0045,
1509
+ "step": 212
1510
+ },
1511
+ {
1512
+ "epoch": 0.34,
1513
+ "grad_norm": 14.0,
1514
+ "learning_rate": 0.00017758770349486923,
1515
+ "loss": 2.4731,
1516
+ "step": 213
1517
+ },
1518
+ {
1519
+ "epoch": 0.35,
1520
+ "grad_norm": 11.0,
1521
+ "learning_rate": 0.00017720513963958968,
1522
+ "loss": 2.3612,
1523
+ "step": 214
1524
+ },
1525
+ {
1526
+ "epoch": 0.35,
1527
+ "grad_norm": 16.125,
1528
+ "learning_rate": 0.00017681975780402807,
1529
+ "loss": 3.2419,
1530
+ "step": 215
1531
+ },
1532
+ {
1533
+ "epoch": 0.35,
1534
+ "grad_norm": 245.0,
1535
+ "learning_rate": 0.00017643157205458483,
1536
+ "loss": 8.9317,
1537
+ "step": 216
1538
+ },
1539
+ {
1540
+ "epoch": 0.35,
1541
+ "grad_norm": 19.875,
1542
+ "learning_rate": 0.0001760405965600031,
1543
+ "loss": 2.0673,
1544
+ "step": 217
1545
+ },
1546
+ {
1547
+ "epoch": 0.35,
1548
+ "grad_norm": 8.8125,
1549
+ "learning_rate": 0.00017564684559085136,
1550
+ "loss": 2.2646,
1551
+ "step": 218
1552
+ },
1553
+ {
1554
+ "epoch": 0.35,
1555
+ "grad_norm": 1192.0,
1556
+ "learning_rate": 0.00017525033351900268,
1557
+ "loss": 2.9879,
1558
+ "step": 219
1559
+ },
1560
+ {
1561
+ "epoch": 0.35,
1562
+ "grad_norm": 9.5625,
1563
+ "learning_rate": 0.00017485107481711012,
1564
+ "loss": 2.7486,
1565
+ "step": 220
1566
+ },
1567
+ {
1568
+ "epoch": 0.36,
1569
+ "grad_norm": 284.0,
1570
+ "learning_rate": 0.00017444908405807845,
1571
+ "loss": 7.0221,
1572
+ "step": 221
1573
+ },
1574
+ {
1575
+ "epoch": 0.36,
1576
+ "grad_norm": 9.25,
1577
+ "learning_rate": 0.00017404437591453235,
1578
+ "loss": 2.4782,
1579
+ "step": 222
1580
+ },
1581
+ {
1582
+ "epoch": 0.36,
1583
+ "grad_norm": 8.0625,
1584
+ "learning_rate": 0.00017363696515828062,
1585
+ "loss": 2.2747,
1586
+ "step": 223
1587
+ },
1588
+ {
1589
+ "epoch": 0.36,
1590
+ "grad_norm": 6.90625,
1591
+ "learning_rate": 0.00017322686665977737,
1592
+ "loss": 2.3441,
1593
+ "step": 224
1594
+ },
1595
+ {
1596
+ "epoch": 0.36,
1597
+ "grad_norm": 16.25,
1598
+ "learning_rate": 0.00017281409538757883,
1599
+ "loss": 2.7646,
1600
+ "step": 225
1601
+ },
1602
+ {
1603
+ "epoch": 0.36,
1604
+ "grad_norm": 26.125,
1605
+ "learning_rate": 0.00017239866640779745,
1606
+ "loss": 2.8155,
1607
+ "step": 226
1608
+ },
1609
+ {
1610
+ "epoch": 0.37,
1611
+ "grad_norm": 72.5,
1612
+ "learning_rate": 0.0001719805948835515,
1613
+ "loss": 3.3018,
1614
+ "step": 227
1615
+ },
1616
+ {
1617
+ "epoch": 0.37,
1618
+ "grad_norm": 10.25,
1619
+ "learning_rate": 0.00017155989607441213,
1620
+ "loss": 2.4764,
1621
+ "step": 228
1622
+ },
1623
+ {
1624
+ "epoch": 0.37,
1625
+ "grad_norm": 16.75,
1626
+ "learning_rate": 0.00017113658533584594,
1627
+ "loss": 2.4821,
1628
+ "step": 229
1629
+ },
1630
+ {
1631
+ "epoch": 0.37,
1632
+ "grad_norm": 7.90625,
1633
+ "learning_rate": 0.00017071067811865476,
1634
+ "loss": 2.1399,
1635
+ "step": 230
1636
+ },
1637
+ {
1638
+ "epoch": 0.37,
1639
+ "grad_norm": 15.0,
1640
+ "learning_rate": 0.00017028218996841172,
1641
+ "loss": 2.36,
1642
+ "step": 231
1643
+ },
1644
+ {
1645
+ "epoch": 0.37,
1646
+ "grad_norm": 11.875,
1647
+ "learning_rate": 0.00016985113652489374,
1648
+ "loss": 2.307,
1649
+ "step": 232
1650
+ },
1651
+ {
1652
+ "epoch": 0.38,
1653
+ "grad_norm": 8.875,
1654
+ "learning_rate": 0.00016941753352151055,
1655
+ "loss": 2.7008,
1656
+ "step": 233
1657
+ },
1658
+ {
1659
+ "epoch": 0.38,
1660
+ "grad_norm": 9.5625,
1661
+ "learning_rate": 0.00016898139678473076,
1662
+ "loss": 2.6393,
1663
+ "step": 234
1664
+ },
1665
+ {
1666
+ "epoch": 0.38,
1667
+ "grad_norm": 6.6875,
1668
+ "learning_rate": 0.00016854274223350397,
1669
+ "loss": 2.5139,
1670
+ "step": 235
1671
+ },
1672
+ {
1673
+ "epoch": 0.38,
1674
+ "grad_norm": 11.1875,
1675
+ "learning_rate": 0.00016810158587867973,
1676
+ "loss": 2.6936,
1677
+ "step": 236
1678
+ },
1679
+ {
1680
+ "epoch": 0.38,
1681
+ "grad_norm": 7.4375,
1682
+ "learning_rate": 0.00016765794382242314,
1683
+ "loss": 2.5852,
1684
+ "step": 237
1685
+ },
1686
+ {
1687
+ "epoch": 0.38,
1688
+ "grad_norm": 5.4375,
1689
+ "learning_rate": 0.00016721183225762727,
1690
+ "loss": 1.7845,
1691
+ "step": 238
1692
+ },
1693
+ {
1694
+ "epoch": 0.39,
1695
+ "grad_norm": 12.9375,
1696
+ "learning_rate": 0.00016676326746732195,
1697
+ "loss": 1.889,
1698
+ "step": 239
1699
+ },
1700
+ {
1701
+ "epoch": 0.39,
1702
+ "grad_norm": 5.8125,
1703
+ "learning_rate": 0.00016631226582407952,
1704
+ "loss": 2.2385,
1705
+ "step": 240
1706
+ },
1707
+ {
1708
+ "epoch": 0.39,
1709
+ "grad_norm": 8.625,
1710
+ "learning_rate": 0.00016585884378941725,
1711
+ "loss": 2.6846,
1712
+ "step": 241
1713
+ },
1714
+ {
1715
+ "epoch": 0.39,
1716
+ "grad_norm": 7.125,
1717
+ "learning_rate": 0.00016540301791319645,
1718
+ "loss": 2.2453,
1719
+ "step": 242
1720
+ },
1721
+ {
1722
+ "epoch": 0.39,
1723
+ "grad_norm": 9.75,
1724
+ "learning_rate": 0.00016494480483301836,
1725
+ "loss": 2.7644,
1726
+ "step": 243
1727
+ },
1728
+ {
1729
+ "epoch": 0.39,
1730
+ "grad_norm": 8.5625,
1731
+ "learning_rate": 0.00016448422127361706,
1732
+ "loss": 1.9651,
1733
+ "step": 244
1734
+ },
1735
+ {
1736
+ "epoch": 0.4,
1737
+ "grad_norm": 13.125,
1738
+ "learning_rate": 0.00016402128404624882,
1739
+ "loss": 2.6637,
1740
+ "step": 245
1741
+ },
1742
+ {
1743
+ "epoch": 0.4,
1744
+ "grad_norm": 8.125,
1745
+ "learning_rate": 0.00016355601004807856,
1746
+ "loss": 2.2727,
1747
+ "step": 246
1748
+ },
1749
+ {
1750
+ "epoch": 0.4,
1751
+ "grad_norm": 7.34375,
1752
+ "learning_rate": 0.00016308841626156307,
1753
+ "loss": 2.198,
1754
+ "step": 247
1755
+ },
1756
+ {
1757
+ "epoch": 0.4,
1758
+ "grad_norm": 9.0,
1759
+ "learning_rate": 0.00016261851975383137,
1760
+ "loss": 2.9655,
1761
+ "step": 248
1762
+ },
1763
+ {
1764
+ "epoch": 0.4,
1765
+ "grad_norm": 12.5625,
1766
+ "learning_rate": 0.00016214633767606143,
1767
+ "loss": 3.3041,
1768
+ "step": 249
1769
+ },
1770
+ {
1771
+ "epoch": 0.4,
1772
+ "grad_norm": 72.5,
1773
+ "learning_rate": 0.00016167188726285434,
1774
+ "loss": 2.7481,
1775
+ "step": 250
1776
+ },
1777
+ {
1778
+ "epoch": 0.4,
1779
+ "grad_norm": 6.875,
1780
+ "learning_rate": 0.0001611951858316052,
1781
+ "loss": 2.8002,
1782
+ "step": 251
1783
+ },
1784
+ {
1785
+ "epoch": 0.41,
1786
+ "grad_norm": 24.0,
1787
+ "learning_rate": 0.00016071625078187114,
1788
+ "loss": 3.5713,
1789
+ "step": 252
1790
+ },
1791
+ {
1792
+ "epoch": 0.41,
1793
+ "grad_norm": 6.90625,
1794
+ "learning_rate": 0.00016023509959473605,
1795
+ "loss": 2.7173,
1796
+ "step": 253
1797
+ },
1798
+ {
1799
+ "epoch": 0.41,
1800
+ "grad_norm": 8.1875,
1801
+ "learning_rate": 0.00015975174983217275,
1802
+ "loss": 2.3719,
1803
+ "step": 254
1804
+ },
1805
+ {
1806
+ "epoch": 0.41,
1807
+ "grad_norm": 51.0,
1808
+ "learning_rate": 0.0001592662191364017,
1809
+ "loss": 2.5691,
1810
+ "step": 255
1811
+ },
1812
+ {
1813
+ "epoch": 0.41,
1814
+ "grad_norm": 15.5,
1815
+ "learning_rate": 0.00015877852522924732,
1816
+ "loss": 8.3287,
1817
+ "step": 256
1818
+ },
1819
+ {
1820
+ "epoch": 0.41,
1821
+ "grad_norm": 19.25,
1822
+ "learning_rate": 0.00015828868591149104,
1823
+ "loss": 2.8875,
1824
+ "step": 257
1825
+ },
1826
+ {
1827
+ "epoch": 0.42,
1828
+ "grad_norm": 12.875,
1829
+ "learning_rate": 0.0001577967190622215,
1830
+ "loss": 7.894,
1831
+ "step": 258
1832
+ },
1833
+ {
1834
+ "epoch": 0.42,
1835
+ "grad_norm": 18.875,
1836
+ "learning_rate": 0.00015730264263818212,
1837
+ "loss": 3.3559,
1838
+ "step": 259
1839
+ },
1840
+ {
1841
+ "epoch": 0.42,
1842
+ "grad_norm": 9.875,
1843
+ "learning_rate": 0.00015680647467311557,
1844
+ "loss": 2.7433,
1845
+ "step": 260
1846
+ },
1847
+ {
1848
+ "epoch": 0.42,
1849
+ "grad_norm": 11.8125,
1850
+ "learning_rate": 0.00015630823327710558,
1851
+ "loss": 2.8969,
1852
+ "step": 261
1853
+ },
1854
+ {
1855
+ "epoch": 0.42,
1856
+ "grad_norm": 14.0,
1857
+ "learning_rate": 0.00015580793663591585,
1858
+ "loss": 2.2271,
1859
+ "step": 262
1860
+ },
1861
+ {
1862
+ "epoch": 0.42,
1863
+ "grad_norm": 10.625,
1864
+ "learning_rate": 0.0001553056030103264,
1865
+ "loss": 2.9629,
1866
+ "step": 263
1867
+ },
1868
+ {
1869
+ "epoch": 0.43,
1870
+ "grad_norm": 8.3125,
1871
+ "learning_rate": 0.00015480125073546704,
1872
+ "loss": 2.7916,
1873
+ "step": 264
1874
+ },
1875
+ {
1876
+ "epoch": 0.43,
1877
+ "grad_norm": 10.4375,
1878
+ "learning_rate": 0.0001542948982201479,
1879
+ "loss": 2.8387,
1880
+ "step": 265
1881
+ },
1882
+ {
1883
+ "epoch": 0.43,
1884
+ "grad_norm": 5.46875,
1885
+ "learning_rate": 0.00015378656394618787,
1886
+ "loss": 2.7407,
1887
+ "step": 266
1888
+ },
1889
+ {
1890
+ "epoch": 0.43,
1891
+ "grad_norm": 6.09375,
1892
+ "learning_rate": 0.00015327626646773976,
1893
+ "loss": 2.2058,
1894
+ "step": 267
1895
+ },
1896
+ {
1897
+ "epoch": 0.43,
1898
+ "grad_norm": 10.8125,
1899
+ "learning_rate": 0.0001527640244106133,
1900
+ "loss": 2.963,
1901
+ "step": 268
1902
+ },
1903
+ {
1904
+ "epoch": 0.43,
1905
+ "grad_norm": 8.25,
1906
+ "learning_rate": 0.0001522498564715949,
1907
+ "loss": 2.8504,
1908
+ "step": 269
1909
+ },
1910
+ {
1911
+ "epoch": 0.44,
1912
+ "grad_norm": 6.5,
1913
+ "learning_rate": 0.00015173378141776568,
1914
+ "loss": 2.6087,
1915
+ "step": 270
1916
+ },
1917
+ {
1918
+ "epoch": 0.44,
1919
+ "grad_norm": 12.5,
1920
+ "learning_rate": 0.00015121581808581622,
1921
+ "loss": 2.5967,
1922
+ "step": 271
1923
+ },
1924
+ {
1925
+ "epoch": 0.44,
1926
+ "grad_norm": 13.125,
1927
+ "learning_rate": 0.00015069598538135906,
1928
+ "loss": 2.6404,
1929
+ "step": 272
1930
+ },
1931
+ {
1932
+ "epoch": 0.44,
1933
+ "grad_norm": 6.9375,
1934
+ "learning_rate": 0.00015017430227823864,
1935
+ "loss": 2.8854,
1936
+ "step": 273
1937
+ },
1938
+ {
1939
+ "epoch": 0.44,
1940
+ "grad_norm": 6.53125,
1941
+ "learning_rate": 0.0001496507878178388,
1942
+ "loss": 2.5134,
1943
+ "step": 274
1944
+ },
1945
+ {
1946
+ "epoch": 0.44,
1947
+ "grad_norm": 6.96875,
1948
+ "learning_rate": 0.00014912546110838775,
1949
+ "loss": 3.1168,
1950
+ "step": 275
1951
+ },
1952
+ {
1953
+ "epoch": 0.45,
1954
+ "grad_norm": 10.0625,
1955
+ "learning_rate": 0.0001485983413242606,
1956
+ "loss": 2.0887,
1957
+ "step": 276
1958
+ },
1959
+ {
1960
+ "epoch": 0.45,
1961
+ "grad_norm": 6.09375,
1962
+ "learning_rate": 0.00014806944770527958,
1963
+ "loss": 2.7354,
1964
+ "step": 277
1965
+ },
1966
+ {
1967
+ "epoch": 0.45,
1968
+ "grad_norm": 11.8125,
1969
+ "learning_rate": 0.00014753879955601163,
1970
+ "loss": 3.7037,
1971
+ "step": 278
1972
+ },
1973
+ {
1974
+ "epoch": 0.45,
1975
+ "grad_norm": 5.0625,
1976
+ "learning_rate": 0.00014700641624506392,
1977
+ "loss": 2.0548,
1978
+ "step": 279
1979
+ },
1980
+ {
1981
+ "epoch": 0.45,
1982
+ "grad_norm": 5.65625,
1983
+ "learning_rate": 0.00014647231720437686,
1984
+ "loss": 2.5235,
1985
+ "step": 280
1986
+ },
1987
+ {
1988
+ "epoch": 0.45,
1989
+ "grad_norm": 10.8125,
1990
+ "learning_rate": 0.00014593652192851486,
1991
+ "loss": 2.9558,
1992
+ "step": 281
1993
+ },
1994
+ {
1995
+ "epoch": 0.45,
1996
+ "grad_norm": 5.34375,
1997
+ "learning_rate": 0.00014539904997395468,
1998
+ "loss": 2.012,
1999
+ "step": 282
2000
+ },
2001
+ {
2002
+ "epoch": 0.46,
2003
+ "grad_norm": 5.21875,
2004
+ "learning_rate": 0.00014485992095837177,
2005
+ "loss": 1.7521,
2006
+ "step": 283
2007
+ },
2008
+ {
2009
+ "epoch": 0.46,
2010
+ "grad_norm": 8.0625,
2011
+ "learning_rate": 0.00014431915455992414,
2012
+ "loss": 2.2712,
2013
+ "step": 284
2014
+ },
2015
+ {
2016
+ "epoch": 0.46,
2017
+ "grad_norm": 6.78125,
2018
+ "learning_rate": 0.00014377677051653404,
2019
+ "loss": 2.4476,
2020
+ "step": 285
2021
+ },
2022
+ {
2023
+ "epoch": 0.46,
2024
+ "grad_norm": 9.375,
2025
+ "learning_rate": 0.00014323278862516775,
2026
+ "loss": 2.6188,
2027
+ "step": 286
2028
+ },
2029
+ {
2030
+ "epoch": 0.46,
2031
+ "grad_norm": 7.125,
2032
+ "learning_rate": 0.00014268722874111265,
2033
+ "loss": 2.4349,
2034
+ "step": 287
2035
+ },
2036
+ {
2037
+ "epoch": 0.46,
2038
+ "grad_norm": 20.75,
2039
+ "learning_rate": 0.00014214011077725292,
2040
+ "loss": 2.5096,
2041
+ "step": 288
2042
+ },
2043
+ {
2044
+ "epoch": 0.47,
2045
+ "grad_norm": 11.3125,
2046
+ "learning_rate": 0.00014159145470334235,
2047
+ "loss": 2.8644,
2048
+ "step": 289
2049
+ },
2050
+ {
2051
+ "epoch": 0.47,
2052
+ "grad_norm": 9.1875,
2053
+ "learning_rate": 0.0001410412805452757,
2054
+ "loss": 2.23,
2055
+ "step": 290
2056
+ },
2057
+ {
2058
+ "epoch": 0.47,
2059
+ "grad_norm": 119.5,
2060
+ "learning_rate": 0.00014048960838435753,
2061
+ "loss": 2.3854,
2062
+ "step": 291
2063
+ },
2064
+ {
2065
+ "epoch": 0.47,
2066
+ "grad_norm": 6.09375,
2067
+ "learning_rate": 0.00013993645835656953,
2068
+ "loss": 2.8378,
2069
+ "step": 292
2070
+ },
2071
+ {
2072
+ "epoch": 0.47,
2073
+ "grad_norm": 5.3125,
2074
+ "learning_rate": 0.00013938185065183532,
2075
+ "loss": 2.2624,
2076
+ "step": 293
2077
+ },
2078
+ {
2079
+ "epoch": 0.47,
2080
+ "grad_norm": 8.125,
2081
+ "learning_rate": 0.0001388258055132835,
2082
+ "loss": 2.9981,
2083
+ "step": 294
2084
+ },
2085
+ {
2086
+ "epoch": 0.48,
2087
+ "grad_norm": 288.0,
2088
+ "learning_rate": 0.000138268343236509,
2089
+ "loss": 9.5386,
2090
+ "step": 295
2091
+ },
2092
+ {
2093
+ "epoch": 0.48,
2094
+ "grad_norm": 8.625,
2095
+ "learning_rate": 0.00013770948416883205,
2096
+ "loss": 3.1489,
2097
+ "step": 296
2098
+ },
2099
+ {
2100
+ "epoch": 0.48,
2101
+ "grad_norm": 6.46875,
2102
+ "learning_rate": 0.00013714924870855571,
2103
+ "loss": 2.6677,
2104
+ "step": 297
2105
+ },
2106
+ {
2107
+ "epoch": 0.48,
2108
+ "grad_norm": 7.8125,
2109
+ "learning_rate": 0.00013658765730422125,
2110
+ "loss": 2.3183,
2111
+ "step": 298
2112
+ },
2113
+ {
2114
+ "epoch": 0.48,
2115
+ "grad_norm": 8.125,
2116
+ "learning_rate": 0.00013602473045386165,
2117
+ "loss": 2.4839,
2118
+ "step": 299
2119
+ },
2120
+ {
2121
+ "epoch": 0.48,
2122
+ "grad_norm": 6.59375,
2123
+ "learning_rate": 0.00013546048870425356,
2124
+ "loss": 2.882,
2125
+ "step": 300
2126
+ },
2127
+ {
2128
+ "epoch": 0.49,
2129
+ "grad_norm": 8.25,
2130
+ "learning_rate": 0.0001348949526501675,
2131
+ "loss": 3.1237,
2132
+ "step": 301
2133
+ },
2134
+ {
2135
+ "epoch": 0.49,
2136
+ "grad_norm": 48.5,
2137
+ "learning_rate": 0.00013432814293361584,
2138
+ "loss": 2.3554,
2139
+ "step": 302
2140
+ },
2141
+ {
2142
+ "epoch": 0.49,
2143
+ "grad_norm": 11.0,
2144
+ "learning_rate": 0.00013376008024309948,
2145
+ "loss": 2.4418,
2146
+ "step": 303
2147
+ },
2148
+ {
2149
+ "epoch": 0.49,
2150
+ "grad_norm": 133.0,
2151
+ "learning_rate": 0.00013319078531285285,
2152
+ "loss": 3.1814,
2153
+ "step": 304
2154
+ },
2155
+ {
2156
+ "epoch": 0.49,
2157
+ "grad_norm": 6.0,
2158
+ "learning_rate": 0.00013262027892208694,
2159
+ "loss": 2.5243,
2160
+ "step": 305
2161
+ },
2162
+ {
2163
+ "epoch": 0.49,
2164
+ "grad_norm": 9.125,
2165
+ "learning_rate": 0.00013204858189423097,
2166
+ "loss": 2.9641,
2167
+ "step": 306
2168
+ },
2169
+ {
2170
+ "epoch": 0.5,
2171
+ "grad_norm": 7.71875,
2172
+ "learning_rate": 0.00013147571509617228,
2173
+ "loss": 2.9313,
2174
+ "step": 307
2175
+ },
2176
+ {
2177
+ "epoch": 0.5,
2178
+ "grad_norm": 7.75,
2179
+ "learning_rate": 0.00013090169943749476,
2180
+ "loss": 2.1694,
2181
+ "step": 308
2182
+ },
2183
+ {
2184
+ "epoch": 0.5,
2185
+ "grad_norm": 5.375,
2186
+ "learning_rate": 0.00013032655586971552,
2187
+ "loss": 2.6369,
2188
+ "step": 309
2189
+ },
2190
+ {
2191
+ "epoch": 0.5,
2192
+ "grad_norm": 7.28125,
2193
+ "learning_rate": 0.00012975030538552032,
2194
+ "loss": 2.9122,
2195
+ "step": 310
2196
+ },
2197
+ {
2198
+ "epoch": 0.5,
2199
+ "eval_loss": 3.687380075454712,
2200
+ "eval_runtime": 1.7338,
2201
+ "eval_samples_per_second": 19.61,
2202
+ "eval_steps_per_second": 19.61,
2203
+ "step": 310
2204
+ },
2205
+ {
2206
+ "epoch": 0.5,
2207
+ "grad_norm": 7.59375,
2208
+ "learning_rate": 0.0001291729690179972,
2209
+ "loss": 2.6619,
2210
+ "step": 311
2211
+ },
2212
+ {
2213
+ "epoch": 0.5,
2214
+ "grad_norm": 163.0,
2215
+ "learning_rate": 0.00012859456783986893,
2216
+ "loss": 2.7945,
2217
+ "step": 312
2218
+ },
2219
+ {
2220
+ "epoch": 0.5,
2221
+ "grad_norm": 9.5,
2222
+ "learning_rate": 0.00012801512296272368,
2223
+ "loss": 3.0543,
2224
+ "step": 313
2225
+ },
2226
+ {
2227
+ "epoch": 0.51,
2228
+ "grad_norm": 6.78125,
2229
+ "learning_rate": 0.0001274346555362446,
2230
+ "loss": 2.6224,
2231
+ "step": 314
2232
+ },
2233
+ {
2234
+ "epoch": 0.51,
2235
+ "grad_norm": 7.75,
2236
+ "learning_rate": 0.0001268531867474377,
2237
+ "loss": 2.4459,
2238
+ "step": 315
2239
+ },
2240
+ {
2241
+ "epoch": 0.51,
2242
+ "grad_norm": 5.46875,
2243
+ "learning_rate": 0.0001262707378198587,
2244
+ "loss": 2.1947,
2245
+ "step": 316
2246
+ },
2247
+ {
2248
+ "epoch": 0.51,
2249
+ "grad_norm": 6.28125,
2250
+ "learning_rate": 0.00012568733001283827,
2251
+ "loss": 2.2675,
2252
+ "step": 317
2253
+ },
2254
+ {
2255
+ "epoch": 0.51,
2256
+ "grad_norm": 7.5,
2257
+ "learning_rate": 0.00012510298462070619,
2258
+ "loss": 2.6113,
2259
+ "step": 318
2260
+ },
2261
+ {
2262
+ "epoch": 0.51,
2263
+ "grad_norm": 6.28125,
2264
+ "learning_rate": 0.00012451772297201376,
2265
+ "loss": 2.2417,
2266
+ "step": 319
2267
+ },
2268
+ {
2269
+ "epoch": 0.52,
2270
+ "grad_norm": 4.875,
2271
+ "learning_rate": 0.0001239315664287558,
2272
+ "loss": 2.4538,
2273
+ "step": 320
2274
+ },
2275
+ {
2276
+ "epoch": 0.52,
2277
+ "grad_norm": 18.0,
2278
+ "learning_rate": 0.00012334453638559057,
2279
+ "loss": 2.2888,
2280
+ "step": 321
2281
+ },
2282
+ {
2283
+ "epoch": 0.52,
2284
+ "grad_norm": 6.75,
2285
+ "learning_rate": 0.000122756654269059,
2286
+ "loss": 2.743,
2287
+ "step": 322
2288
+ },
2289
+ {
2290
+ "epoch": 0.52,
2291
+ "grad_norm": 9.9375,
2292
+ "learning_rate": 0.00012216794153680274,
2293
+ "loss": 2.8164,
2294
+ "step": 323
2295
+ },
2296
+ {
2297
+ "epoch": 0.52,
2298
+ "grad_norm": 7.65625,
2299
+ "learning_rate": 0.00012157841967678063,
2300
+ "loss": 2.4418,
2301
+ "step": 324
2302
+ },
2303
+ {
2304
+ "epoch": 0.52,
2305
+ "grad_norm": 48.0,
2306
+ "learning_rate": 0.00012098811020648475,
2307
+ "loss": 9.2518,
2308
+ "step": 325
2309
+ },
2310
+ {
2311
+ "epoch": 0.53,
2312
+ "grad_norm": 6.03125,
2313
+ "learning_rate": 0.00012039703467215488,
2314
+ "loss": 2.7601,
2315
+ "step": 326
2316
+ },
2317
+ {
2318
+ "epoch": 0.53,
2319
+ "grad_norm": 16.625,
2320
+ "learning_rate": 0.00011980521464799198,
2321
+ "loss": 2.604,
2322
+ "step": 327
2323
+ },
2324
+ {
2325
+ "epoch": 0.53,
2326
+ "grad_norm": 19.25,
2327
+ "learning_rate": 0.00011921267173537086,
2328
+ "loss": 2.4513,
2329
+ "step": 328
2330
+ },
2331
+ {
2332
+ "epoch": 0.53,
2333
+ "grad_norm": 7.96875,
2334
+ "learning_rate": 0.00011861942756205169,
2335
+ "loss": 2.4684,
2336
+ "step": 329
2337
+ },
2338
+ {
2339
+ "epoch": 0.53,
2340
+ "grad_norm": 5.03125,
2341
+ "learning_rate": 0.0001180255037813906,
2342
+ "loss": 2.2713,
2343
+ "step": 330
2344
+ },
2345
+ {
2346
+ "epoch": 0.53,
2347
+ "grad_norm": 11.625,
2348
+ "learning_rate": 0.00011743092207154929,
2349
+ "loss": 2.673,
2350
+ "step": 331
2351
+ },
2352
+ {
2353
+ "epoch": 0.54,
2354
+ "grad_norm": 2928.0,
2355
+ "learning_rate": 0.00011683570413470383,
2356
+ "loss": 2.6462,
2357
+ "step": 332
2358
+ },
2359
+ {
2360
+ "epoch": 0.54,
2361
+ "grad_norm": 6.4375,
2362
+ "learning_rate": 0.00011623987169625261,
2363
+ "loss": 2.5405,
2364
+ "step": 333
2365
+ },
2366
+ {
2367
+ "epoch": 0.54,
2368
+ "grad_norm": 6.3125,
2369
+ "learning_rate": 0.0001156434465040231,
2370
+ "loss": 2.9086,
2371
+ "step": 334
2372
+ },
2373
+ {
2374
+ "epoch": 0.54,
2375
+ "grad_norm": 5.0625,
2376
+ "learning_rate": 0.00011504645032747832,
2377
+ "loss": 2.2515,
2378
+ "step": 335
2379
+ },
2380
+ {
2381
+ "epoch": 0.54,
2382
+ "grad_norm": 10.0625,
2383
+ "learning_rate": 0.00011444890495692213,
2384
+ "loss": 2.4309,
2385
+ "step": 336
2386
+ },
2387
+ {
2388
+ "epoch": 0.54,
2389
+ "grad_norm": 5.46875,
2390
+ "learning_rate": 0.00011385083220270401,
2391
+ "loss": 2.7969,
2392
+ "step": 337
2393
+ },
2394
+ {
2395
+ "epoch": 0.55,
2396
+ "grad_norm": 8.875,
2397
+ "learning_rate": 0.00011325225389442277,
2398
+ "loss": 3.1781,
2399
+ "step": 338
2400
+ },
2401
+ {
2402
+ "epoch": 0.55,
2403
+ "grad_norm": 5.65625,
2404
+ "learning_rate": 0.00011265319188012994,
2405
+ "loss": 2.5028,
2406
+ "step": 339
2407
+ },
2408
+ {
2409
+ "epoch": 0.55,
2410
+ "grad_norm": 7.75,
2411
+ "learning_rate": 0.0001120536680255323,
2412
+ "loss": 2.906,
2413
+ "step": 340
2414
+ },
2415
+ {
2416
+ "epoch": 0.55,
2417
+ "grad_norm": 19.5,
2418
+ "learning_rate": 0.00011145370421319377,
2419
+ "loss": 2.1992,
2420
+ "step": 341
2421
+ },
2422
+ {
2423
+ "epoch": 0.55,
2424
+ "grad_norm": 7.5,
2425
+ "learning_rate": 0.00011085332234173664,
2426
+ "loss": 2.9335,
2427
+ "step": 342
2428
+ },
2429
+ {
2430
+ "epoch": 0.55,
2431
+ "grad_norm": 9.25,
2432
+ "learning_rate": 0.00011025254432504233,
2433
+ "loss": 2.6833,
2434
+ "step": 343
2435
+ },
2436
+ {
2437
+ "epoch": 0.55,
2438
+ "grad_norm": 9.625,
2439
+ "learning_rate": 0.00010965139209145152,
2440
+ "loss": 2.8628,
2441
+ "step": 344
2442
+ },
2443
+ {
2444
+ "epoch": 0.56,
2445
+ "grad_norm": 5.65625,
2446
+ "learning_rate": 0.0001090498875829638,
2447
+ "loss": 2.3975,
2448
+ "step": 345
2449
+ },
2450
+ {
2451
+ "epoch": 0.56,
2452
+ "grad_norm": 5.125,
2453
+ "learning_rate": 0.00010844805275443673,
2454
+ "loss": 2.3744,
2455
+ "step": 346
2456
+ },
2457
+ {
2458
+ "epoch": 0.56,
2459
+ "grad_norm": 6.3125,
2460
+ "learning_rate": 0.0001078459095727845,
2461
+ "loss": 2.2086,
2462
+ "step": 347
2463
+ },
2464
+ {
2465
+ "epoch": 0.56,
2466
+ "grad_norm": 55.5,
2467
+ "learning_rate": 0.00010724348001617625,
2468
+ "loss": 3.0402,
2469
+ "step": 348
2470
+ },
2471
+ {
2472
+ "epoch": 0.56,
2473
+ "grad_norm": 31.0,
2474
+ "learning_rate": 0.00010664078607323367,
2475
+ "loss": 2.2517,
2476
+ "step": 349
2477
+ },
2478
+ {
2479
+ "epoch": 0.56,
2480
+ "grad_norm": 6.65625,
2481
+ "learning_rate": 0.00010603784974222861,
2482
+ "loss": 2.2291,
2483
+ "step": 350
2484
+ },
2485
+ {
2486
+ "epoch": 0.57,
2487
+ "grad_norm": 5.5,
2488
+ "learning_rate": 0.00010543469303028002,
2489
+ "loss": 2.6128,
2490
+ "step": 351
2491
+ },
2492
+ {
2493
+ "epoch": 0.57,
2494
+ "grad_norm": 6.15625,
2495
+ "learning_rate": 0.00010483133795255071,
2496
+ "loss": 2.7162,
2497
+ "step": 352
2498
+ },
2499
+ {
2500
+ "epoch": 0.57,
2501
+ "grad_norm": 113.0,
2502
+ "learning_rate": 0.0001042278065314439,
2503
+ "loss": 2.3415,
2504
+ "step": 353
2505
+ },
2506
+ {
2507
+ "epoch": 0.57,
2508
+ "grad_norm": 5.6875,
2509
+ "learning_rate": 0.00010362412079579924,
2510
+ "loss": 2.2554,
2511
+ "step": 354
2512
+ },
2513
+ {
2514
+ "epoch": 0.57,
2515
+ "grad_norm": 25.5,
2516
+ "learning_rate": 0.0001030203027800889,
2517
+ "loss": 2.708,
2518
+ "step": 355
2519
+ },
2520
+ {
2521
+ "epoch": 0.57,
2522
+ "grad_norm": 6.34375,
2523
+ "learning_rate": 0.00010241637452361323,
2524
+ "loss": 2.8018,
2525
+ "step": 356
2526
+ },
2527
+ {
2528
+ "epoch": 0.58,
2529
+ "grad_norm": 760.0,
2530
+ "learning_rate": 0.0001018123580696964,
2531
+ "loss": 2.6352,
2532
+ "step": 357
2533
+ },
2534
+ {
2535
+ "epoch": 0.58,
2536
+ "grad_norm": 10.5625,
2537
+ "learning_rate": 0.00010120827546488174,
2538
+ "loss": 2.4275,
2539
+ "step": 358
2540
+ },
2541
+ {
2542
+ "epoch": 0.58,
2543
+ "grad_norm": 5.84375,
2544
+ "learning_rate": 0.00010060414875812709,
2545
+ "loss": 2.0376,
2546
+ "step": 359
2547
+ },
2548
+ {
2549
+ "epoch": 0.58,
2550
+ "grad_norm": 10.3125,
2551
+ "learning_rate": 0.0001,
2552
+ "loss": 2.3766,
2553
+ "step": 360
2554
+ },
2555
+ {
2556
+ "epoch": 0.58,
2557
+ "grad_norm": 26.25,
2558
+ "learning_rate": 9.939585124187292e-05,
2559
+ "loss": 2.6808,
2560
+ "step": 361
2561
+ },
2562
+ {
2563
+ "epoch": 0.58,
2564
+ "grad_norm": 7.15625,
2565
+ "learning_rate": 9.879172453511827e-05,
2566
+ "loss": 2.2098,
2567
+ "step": 362
2568
+ },
2569
+ {
2570
+ "epoch": 0.59,
2571
+ "grad_norm": 7.84375,
2572
+ "learning_rate": 9.818764193030363e-05,
2573
+ "loss": 2.2265,
2574
+ "step": 363
2575
+ },
2576
+ {
2577
+ "epoch": 0.59,
2578
+ "grad_norm": 6.65625,
2579
+ "learning_rate": 9.75836254763868e-05,
2580
+ "loss": 2.4252,
2581
+ "step": 364
2582
+ },
2583
+ {
2584
+ "epoch": 0.59,
2585
+ "grad_norm": 5.09375,
2586
+ "learning_rate": 9.697969721991114e-05,
2587
+ "loss": 2.1414,
2588
+ "step": 365
2589
+ },
2590
+ {
2591
+ "epoch": 0.59,
2592
+ "grad_norm": 8.8125,
2593
+ "learning_rate": 9.63758792042008e-05,
2594
+ "loss": 2.8087,
2595
+ "step": 366
2596
+ },
2597
+ {
2598
+ "epoch": 0.59,
2599
+ "grad_norm": 5.84375,
2600
+ "learning_rate": 9.577219346855613e-05,
2601
+ "loss": 2.3886,
2602
+ "step": 367
2603
+ },
2604
+ {
2605
+ "epoch": 0.59,
2606
+ "grad_norm": 6.71875,
2607
+ "learning_rate": 9.516866204744931e-05,
2608
+ "loss": 2.4171,
2609
+ "step": 368
2610
+ },
2611
+ {
2612
+ "epoch": 0.6,
2613
+ "grad_norm": 4.90625,
2614
+ "learning_rate": 9.456530696971999e-05,
2615
+ "loss": 2.2413,
2616
+ "step": 369
2617
+ },
2618
+ {
2619
+ "epoch": 0.6,
2620
+ "grad_norm": 6.28125,
2621
+ "learning_rate": 9.396215025777139e-05,
2622
+ "loss": 2.8211,
2623
+ "step": 370
2624
+ },
2625
+ {
2626
+ "epoch": 0.6,
2627
+ "grad_norm": 4.90625,
2628
+ "learning_rate": 9.335921392676631e-05,
2629
+ "loss": 2.354,
2630
+ "step": 371
2631
+ },
2632
+ {
2633
+ "epoch": 0.6,
2634
+ "grad_norm": 4160.0,
2635
+ "learning_rate": 9.275651998382377e-05,
2636
+ "loss": 2.0496,
2637
+ "step": 372
2638
+ },
2639
+ {
2640
+ "epoch": 0.6,
2641
+ "grad_norm": 7.0,
2642
+ "learning_rate": 9.215409042721552e-05,
2643
+ "loss": 1.8566,
2644
+ "step": 373
2645
+ },
2646
+ {
2647
+ "epoch": 0.6,
2648
+ "grad_norm": 5.15625,
2649
+ "learning_rate": 9.155194724556331e-05,
2650
+ "loss": 2.2705,
2651
+ "step": 374
2652
+ },
2653
+ {
2654
+ "epoch": 0.6,
2655
+ "grad_norm": 5.53125,
2656
+ "learning_rate": 9.095011241703623e-05,
2657
+ "loss": 2.1012,
2658
+ "step": 375
2659
+ },
2660
+ {
2661
+ "epoch": 0.61,
2662
+ "grad_norm": 42.5,
2663
+ "learning_rate": 9.034860790854849e-05,
2664
+ "loss": 2.342,
2665
+ "step": 376
2666
+ },
2667
+ {
2668
+ "epoch": 0.61,
2669
+ "grad_norm": 5.25,
2670
+ "learning_rate": 8.974745567495768e-05,
2671
+ "loss": 2.1868,
2672
+ "step": 377
2673
+ },
2674
+ {
2675
+ "epoch": 0.61,
2676
+ "grad_norm": 5.78125,
2677
+ "learning_rate": 8.914667765826338e-05,
2678
+ "loss": 2.6162,
2679
+ "step": 378
2680
+ },
2681
+ {
2682
+ "epoch": 0.61,
2683
+ "grad_norm": 8.375,
2684
+ "learning_rate": 8.854629578680624e-05,
2685
+ "loss": 2.507,
2686
+ "step": 379
2687
+ },
2688
+ {
2689
+ "epoch": 0.61,
2690
+ "grad_norm": 6.28125,
2691
+ "learning_rate": 8.79463319744677e-05,
2692
+ "loss": 2.4593,
2693
+ "step": 380
2694
+ },
2695
+ {
2696
+ "epoch": 0.61,
2697
+ "grad_norm": 5.59375,
2698
+ "learning_rate": 8.73468081198701e-05,
2699
+ "loss": 2.4898,
2700
+ "step": 381
2701
+ },
2702
+ {
2703
+ "epoch": 0.62,
2704
+ "grad_norm": 4.71875,
2705
+ "learning_rate": 8.674774610557728e-05,
2706
+ "loss": 2.2433,
2707
+ "step": 382
2708
+ },
2709
+ {
2710
+ "epoch": 0.62,
2711
+ "grad_norm": 6.34375,
2712
+ "learning_rate": 8.614916779729603e-05,
2713
+ "loss": 2.1837,
2714
+ "step": 383
2715
+ },
2716
+ {
2717
+ "epoch": 0.62,
2718
+ "grad_norm": 6.34375,
2719
+ "learning_rate": 8.55510950430779e-05,
2720
+ "loss": 2.5492,
2721
+ "step": 384
2722
+ },
2723
+ {
2724
+ "epoch": 0.62,
2725
+ "grad_norm": 5.4375,
2726
+ "learning_rate": 8.495354967252169e-05,
2727
+ "loss": 1.4873,
2728
+ "step": 385
2729
+ },
2730
+ {
2731
+ "epoch": 0.62,
2732
+ "grad_norm": 5.5,
2733
+ "learning_rate": 8.435655349597689e-05,
2734
+ "loss": 2.4969,
2735
+ "step": 386
2736
+ },
2737
+ {
2738
+ "epoch": 0.62,
2739
+ "grad_norm": 7.71875,
2740
+ "learning_rate": 8.37601283037474e-05,
2741
+ "loss": 3.0314,
2742
+ "step": 387
2743
+ },
2744
+ {
2745
+ "epoch": 0.63,
2746
+ "grad_norm": 5.9375,
2747
+ "learning_rate": 8.316429586529615e-05,
2748
+ "loss": 2.4772,
2749
+ "step": 388
2750
+ },
2751
+ {
2752
+ "epoch": 0.63,
2753
+ "grad_norm": 4.9375,
2754
+ "learning_rate": 8.256907792845072e-05,
2755
+ "loss": 2.4245,
2756
+ "step": 389
2757
+ },
2758
+ {
2759
+ "epoch": 0.63,
2760
+ "grad_norm": 4.90625,
2761
+ "learning_rate": 8.197449621860943e-05,
2762
+ "loss": 1.8508,
2763
+ "step": 390
2764
+ },
2765
+ {
2766
+ "epoch": 0.63,
2767
+ "grad_norm": 6.65625,
2768
+ "learning_rate": 8.138057243794833e-05,
2769
+ "loss": 2.1701,
2770
+ "step": 391
2771
+ },
2772
+ {
2773
+ "epoch": 0.63,
2774
+ "grad_norm": 7.09375,
2775
+ "learning_rate": 8.078732826462915e-05,
2776
+ "loss": 3.0743,
2777
+ "step": 392
2778
+ },
2779
+ {
2780
+ "epoch": 0.63,
2781
+ "grad_norm": 4.75,
2782
+ "learning_rate": 8.019478535200806e-05,
2783
+ "loss": 2.0487,
2784
+ "step": 393
2785
+ },
2786
+ {
2787
+ "epoch": 0.64,
2788
+ "grad_norm": 4.40625,
2789
+ "learning_rate": 7.960296532784515e-05,
2790
+ "loss": 1.8642,
2791
+ "step": 394
2792
+ },
2793
+ {
2794
+ "epoch": 0.64,
2795
+ "grad_norm": 4.96875,
2796
+ "learning_rate": 7.901188979351526e-05,
2797
+ "loss": 2.5672,
2798
+ "step": 395
2799
+ },
2800
+ {
2801
+ "epoch": 0.64,
2802
+ "grad_norm": 5.6875,
2803
+ "learning_rate": 7.84215803232194e-05,
2804
+ "loss": 2.6458,
2805
+ "step": 396
2806
+ },
2807
+ {
2808
+ "epoch": 0.64,
2809
+ "grad_norm": 4.15625,
2810
+ "learning_rate": 7.78320584631973e-05,
2811
+ "loss": 1.894,
2812
+ "step": 397
2813
+ },
2814
+ {
2815
+ "epoch": 0.64,
2816
+ "grad_norm": 5.125,
2817
+ "learning_rate": 7.7243345730941e-05,
2818
+ "loss": 2.2436,
2819
+ "step": 398
2820
+ },
2821
+ {
2822
+ "epoch": 0.64,
2823
+ "grad_norm": 8.125,
2824
+ "learning_rate": 7.66554636144095e-05,
2825
+ "loss": 1.9708,
2826
+ "step": 399
2827
+ },
2828
+ {
2829
+ "epoch": 0.65,
2830
+ "grad_norm": 5.78125,
2831
+ "learning_rate": 7.606843357124426e-05,
2832
+ "loss": 2.5617,
2833
+ "step": 400
2834
+ },
2835
+ {
2836
+ "epoch": 0.65,
2837
+ "grad_norm": 5.3125,
2838
+ "learning_rate": 7.548227702798624e-05,
2839
+ "loss": 2.3473,
2840
+ "step": 401
2841
+ },
2842
+ {
2843
+ "epoch": 0.65,
2844
+ "grad_norm": 9.4375,
2845
+ "learning_rate": 7.489701537929384e-05,
2846
+ "loss": 2.9168,
2847
+ "step": 402
2848
+ },
2849
+ {
2850
+ "epoch": 0.65,
2851
+ "grad_norm": 5.28125,
2852
+ "learning_rate": 7.431266998716171e-05,
2853
+ "loss": 2.5894,
2854
+ "step": 403
2855
+ },
2856
+ {
2857
+ "epoch": 0.65,
2858
+ "grad_norm": 4.53125,
2859
+ "learning_rate": 7.372926218014131e-05,
2860
+ "loss": 2.2084,
2861
+ "step": 404
2862
+ },
2863
+ {
2864
+ "epoch": 0.65,
2865
+ "grad_norm": 5.71875,
2866
+ "learning_rate": 7.314681325256232e-05,
2867
+ "loss": 2.2265,
2868
+ "step": 405
2869
+ },
2870
+ {
2871
+ "epoch": 0.65,
2872
+ "grad_norm": 5.28125,
2873
+ "learning_rate": 7.256534446375542e-05,
2874
+ "loss": 2.1905,
2875
+ "step": 406
2876
+ },
2877
+ {
2878
+ "epoch": 0.66,
2879
+ "grad_norm": 7.46875,
2880
+ "learning_rate": 7.198487703727632e-05,
2881
+ "loss": 2.0264,
2882
+ "step": 407
2883
+ },
2884
+ {
2885
+ "epoch": 0.66,
2886
+ "grad_norm": 5.90625,
2887
+ "learning_rate": 7.14054321601311e-05,
2888
+ "loss": 2.732,
2889
+ "step": 408
2890
+ },
2891
+ {
2892
+ "epoch": 0.66,
2893
+ "grad_norm": 4.3125,
2894
+ "learning_rate": 7.082703098200282e-05,
2895
+ "loss": 1.7073,
2896
+ "step": 409
2897
+ },
2898
+ {
2899
+ "epoch": 0.66,
2900
+ "grad_norm": 6.875,
2901
+ "learning_rate": 7.024969461447972e-05,
2902
+ "loss": 3.0878,
2903
+ "step": 410
2904
+ },
2905
+ {
2906
+ "epoch": 0.66,
2907
+ "grad_norm": 5.21875,
2908
+ "learning_rate": 6.967344413028452e-05,
2909
+ "loss": 2.2497,
2910
+ "step": 411
2911
+ },
2912
+ {
2913
+ "epoch": 0.66,
2914
+ "grad_norm": 6.15625,
2915
+ "learning_rate": 6.909830056250527e-05,
2916
+ "loss": 2.6836,
2917
+ "step": 412
2918
+ },
2919
+ {
2920
+ "epoch": 0.67,
2921
+ "grad_norm": 4.875,
2922
+ "learning_rate": 6.852428490382773e-05,
2923
+ "loss": 2.2731,
2924
+ "step": 413
2925
+ },
2926
+ {
2927
+ "epoch": 0.67,
2928
+ "grad_norm": 3.9375,
2929
+ "learning_rate": 6.795141810576906e-05,
2930
+ "loss": 2.0797,
2931
+ "step": 414
2932
+ },
2933
+ {
2934
+ "epoch": 0.67,
2935
+ "grad_norm": 6.40625,
2936
+ "learning_rate": 6.73797210779131e-05,
2937
+ "loss": 1.9783,
2938
+ "step": 415
2939
+ },
2940
+ {
2941
+ "epoch": 0.67,
2942
+ "grad_norm": 4.40625,
2943
+ "learning_rate": 6.680921468714719e-05,
2944
+ "loss": 1.856,
2945
+ "step": 416
2946
+ },
2947
+ {
2948
+ "epoch": 0.67,
2949
+ "grad_norm": 4.8125,
2950
+ "learning_rate": 6.623991975690051e-05,
2951
+ "loss": 2.4563,
2952
+ "step": 417
2953
+ },
2954
+ {
2955
+ "epoch": 0.67,
2956
+ "grad_norm": 5.53125,
2957
+ "learning_rate": 6.567185706638417e-05,
2958
+ "loss": 2.475,
2959
+ "step": 418
2960
+ },
2961
+ {
2962
+ "epoch": 0.68,
2963
+ "grad_norm": 4.4375,
2964
+ "learning_rate": 6.510504734983249e-05,
2965
+ "loss": 1.9184,
2966
+ "step": 419
2967
+ },
2968
+ {
2969
+ "epoch": 0.68,
2970
+ "grad_norm": 8.125,
2971
+ "learning_rate": 6.453951129574644e-05,
2972
+ "loss": 1.6104,
2973
+ "step": 420
2974
+ },
2975
+ {
2976
+ "epoch": 0.68,
2977
+ "grad_norm": 29.375,
2978
+ "learning_rate": 6.397526954613839e-05,
2979
+ "loss": 2.0885,
2980
+ "step": 421
2981
+ },
2982
+ {
2983
+ "epoch": 0.68,
2984
+ "grad_norm": 4.3125,
2985
+ "learning_rate": 6.341234269577879e-05,
2986
+ "loss": 2.3858,
2987
+ "step": 422
2988
+ },
2989
+ {
2990
+ "epoch": 0.68,
2991
+ "grad_norm": 5.3125,
2992
+ "learning_rate": 6.285075129144428e-05,
2993
+ "loss": 2.2931,
2994
+ "step": 423
2995
+ },
2996
+ {
2997
+ "epoch": 0.68,
2998
+ "grad_norm": 4.8125,
2999
+ "learning_rate": 6.229051583116796e-05,
3000
+ "loss": 2.0697,
3001
+ "step": 424
3002
+ },
3003
+ {
3004
+ "epoch": 0.69,
3005
+ "grad_norm": 4.78125,
3006
+ "learning_rate": 6.173165676349103e-05,
3007
+ "loss": 2.2237,
3008
+ "step": 425
3009
+ },
3010
+ {
3011
+ "epoch": 0.69,
3012
+ "grad_norm": 5.25,
3013
+ "learning_rate": 6.117419448671651e-05,
3014
+ "loss": 1.7762,
3015
+ "step": 426
3016
+ },
3017
+ {
3018
+ "epoch": 0.69,
3019
+ "grad_norm": 8.9375,
3020
+ "learning_rate": 6.0618149348164696e-05,
3021
+ "loss": 2.2077,
3022
+ "step": 427
3023
+ },
3024
+ {
3025
+ "epoch": 0.69,
3026
+ "grad_norm": 14.8125,
3027
+ "learning_rate": 6.006354164343046e-05,
3028
+ "loss": 7.755,
3029
+ "step": 428
3030
+ },
3031
+ {
3032
+ "epoch": 0.69,
3033
+ "grad_norm": 9.5,
3034
+ "learning_rate": 5.9510391615642466e-05,
3035
+ "loss": 2.1157,
3036
+ "step": 429
3037
+ },
3038
+ {
3039
+ "epoch": 0.69,
3040
+ "grad_norm": 5.4375,
3041
+ "learning_rate": 5.8958719454724346e-05,
3042
+ "loss": 2.1801,
3043
+ "step": 430
3044
+ },
3045
+ {
3046
+ "epoch": 0.7,
3047
+ "grad_norm": 4.96875,
3048
+ "learning_rate": 5.840854529665767e-05,
3049
+ "loss": 1.9679,
3050
+ "step": 431
3051
+ },
3052
+ {
3053
+ "epoch": 0.7,
3054
+ "grad_norm": 5.75,
3055
+ "learning_rate": 5.785988922274711e-05,
3056
+ "loss": 2.4913,
3057
+ "step": 432
3058
+ },
3059
+ {
3060
+ "epoch": 0.7,
3061
+ "grad_norm": 5.1875,
3062
+ "learning_rate": 5.7312771258887386e-05,
3063
+ "loss": 2.231,
3064
+ "step": 433
3065
+ },
3066
+ {
3067
+ "epoch": 0.7,
3068
+ "grad_norm": 45.25,
3069
+ "learning_rate": 5.676721137483225e-05,
3070
+ "loss": 2.1291,
3071
+ "step": 434
3072
+ },
3073
+ {
3074
+ "epoch": 0.7,
3075
+ "grad_norm": 4.78125,
3076
+ "learning_rate": 5.622322948346594e-05,
3077
+ "loss": 2.4463,
3078
+ "step": 435
3079
+ },
3080
+ {
3081
+ "epoch": 0.7,
3082
+ "grad_norm": 52.75,
3083
+ "learning_rate": 5.568084544007588e-05,
3084
+ "loss": 2.8942,
3085
+ "step": 436
3086
+ },
3087
+ {
3088
+ "epoch": 0.7,
3089
+ "grad_norm": 5.625,
3090
+ "learning_rate": 5.5140079041628214e-05,
3091
+ "loss": 2.6779,
3092
+ "step": 437
3093
+ },
3094
+ {
3095
+ "epoch": 0.71,
3096
+ "grad_norm": 4.53125,
3097
+ "learning_rate": 5.4600950026045326e-05,
3098
+ "loss": 1.909,
3099
+ "step": 438
3100
+ },
3101
+ {
3102
+ "epoch": 0.71,
3103
+ "grad_norm": 5.59375,
3104
+ "learning_rate": 5.406347807148515e-05,
3105
+ "loss": 2.54,
3106
+ "step": 439
3107
+ },
3108
+ {
3109
+ "epoch": 0.71,
3110
+ "grad_norm": 4.40625,
3111
+ "learning_rate": 5.3527682795623146e-05,
3112
+ "loss": 2.1972,
3113
+ "step": 440
3114
+ },
3115
+ {
3116
+ "epoch": 0.71,
3117
+ "grad_norm": 4.375,
3118
+ "learning_rate": 5.2993583754936126e-05,
3119
+ "loss": 1.9662,
3120
+ "step": 441
3121
+ },
3122
+ {
3123
+ "epoch": 0.71,
3124
+ "grad_norm": 4.03125,
3125
+ "learning_rate": 5.246120044398839e-05,
3126
+ "loss": 2.0379,
3127
+ "step": 442
3128
+ },
3129
+ {
3130
+ "epoch": 0.71,
3131
+ "grad_norm": 28.875,
3132
+ "learning_rate": 5.193055229472045e-05,
3133
+ "loss": 2.3016,
3134
+ "step": 443
3135
+ },
3136
+ {
3137
+ "epoch": 0.72,
3138
+ "grad_norm": 5.21875,
3139
+ "learning_rate": 5.14016586757394e-05,
3140
+ "loss": 2.1646,
3141
+ "step": 444
3142
+ },
3143
+ {
3144
+ "epoch": 0.72,
3145
+ "grad_norm": 6.09375,
3146
+ "learning_rate": 5.087453889161229e-05,
3147
+ "loss": 7.6577,
3148
+ "step": 445
3149
+ },
3150
+ {
3151
+ "epoch": 0.72,
3152
+ "grad_norm": 4.875,
3153
+ "learning_rate": 5.0349212182161254e-05,
3154
+ "loss": 2.4258,
3155
+ "step": 446
3156
+ },
3157
+ {
3158
+ "epoch": 0.72,
3159
+ "grad_norm": 5.03125,
3160
+ "learning_rate": 4.98256977217614e-05,
3161
+ "loss": 2.1568,
3162
+ "step": 447
3163
+ },
3164
+ {
3165
+ "epoch": 0.72,
3166
+ "grad_norm": 4.5625,
3167
+ "learning_rate": 4.9304014618640995e-05,
3168
+ "loss": 1.8792,
3169
+ "step": 448
3170
+ },
3171
+ {
3172
+ "epoch": 0.72,
3173
+ "grad_norm": 3.53125,
3174
+ "learning_rate": 4.87841819141838e-05,
3175
+ "loss": 1.7279,
3176
+ "step": 449
3177
+ },
3178
+ {
3179
+ "epoch": 0.73,
3180
+ "grad_norm": 3.90625,
3181
+ "learning_rate": 4.826621858223431e-05,
3182
+ "loss": 1.686,
3183
+ "step": 450
3184
+ },
3185
+ {
3186
+ "epoch": 0.73,
3187
+ "grad_norm": 5.90625,
3188
+ "learning_rate": 4.7750143528405126e-05,
3189
+ "loss": 2.5515,
3190
+ "step": 451
3191
+ },
3192
+ {
3193
+ "epoch": 0.73,
3194
+ "grad_norm": 4.3125,
3195
+ "learning_rate": 4.723597558938672e-05,
3196
+ "loss": 2.1101,
3197
+ "step": 452
3198
+ },
3199
+ {
3200
+ "epoch": 0.73,
3201
+ "grad_norm": 9.0625,
3202
+ "learning_rate": 4.672373353226023e-05,
3203
+ "loss": 2.1532,
3204
+ "step": 453
3205
+ },
3206
+ {
3207
+ "epoch": 0.73,
3208
+ "grad_norm": 6.9375,
3209
+ "learning_rate": 4.6213436053812144e-05,
3210
+ "loss": 7.4545,
3211
+ "step": 454
3212
+ },
3213
+ {
3214
+ "epoch": 0.73,
3215
+ "grad_norm": 6.0625,
3216
+ "learning_rate": 4.5705101779852135e-05,
3217
+ "loss": 3.1788,
3218
+ "step": 455
3219
+ },
3220
+ {
3221
+ "epoch": 0.74,
3222
+ "grad_norm": 3.546875,
3223
+ "learning_rate": 4.519874926453302e-05,
3224
+ "loss": 1.7956,
3225
+ "step": 456
3226
+ },
3227
+ {
3228
+ "epoch": 0.74,
3229
+ "grad_norm": 5.5625,
3230
+ "learning_rate": 4.469439698967359e-05,
3231
+ "loss": 1.7923,
3232
+ "step": 457
3233
+ },
3234
+ {
3235
+ "epoch": 0.74,
3236
+ "grad_norm": 4.59375,
3237
+ "learning_rate": 4.419206336408418e-05,
3238
+ "loss": 2.1708,
3239
+ "step": 458
3240
+ },
3241
+ {
3242
+ "epoch": 0.74,
3243
+ "grad_norm": 5.40625,
3244
+ "learning_rate": 4.3691766722894435e-05,
3245
+ "loss": 2.3953,
3246
+ "step": 459
3247
+ },
3248
+ {
3249
+ "epoch": 0.74,
3250
+ "grad_norm": 5.28125,
3251
+ "learning_rate": 4.3193525326884435e-05,
3252
+ "loss": 2.6895,
3253
+ "step": 460
3254
+ },
3255
+ {
3256
+ "epoch": 0.74,
3257
+ "grad_norm": 5.25,
3258
+ "learning_rate": 4.26973573618179e-05,
3259
+ "loss": 2.3356,
3260
+ "step": 461
3261
+ },
3262
+ {
3263
+ "epoch": 0.75,
3264
+ "grad_norm": 5.59375,
3265
+ "learning_rate": 4.220328093777851e-05,
3266
+ "loss": 2.6686,
3267
+ "step": 462
3268
+ },
3269
+ {
3270
+ "epoch": 0.75,
3271
+ "grad_norm": 3.8125,
3272
+ "learning_rate": 4.1711314088509e-05,
3273
+ "loss": 2.2183,
3274
+ "step": 463
3275
+ },
3276
+ {
3277
+ "epoch": 0.75,
3278
+ "grad_norm": 85.5,
3279
+ "learning_rate": 4.12214747707527e-05,
3280
+ "loss": 2.4175,
3281
+ "step": 464
3282
+ },
3283
+ {
3284
+ "epoch": 0.75,
3285
+ "grad_norm": 4.875,
3286
+ "learning_rate": 4.0733780863598335e-05,
3287
+ "loss": 2.6502,
3288
+ "step": 465
3289
+ },
3290
+ {
3291
+ "epoch": 0.75,
3292
+ "eval_loss": 3.27838397026062,
3293
+ "eval_runtime": 1.744,
3294
+ "eval_samples_per_second": 19.496,
3295
+ "eval_steps_per_second": 19.496,
3296
+ "step": 465
3297
+ },
3298
+ {
3299
+ "epoch": 0.75,
3300
+ "grad_norm": 4.21875,
3301
+ "learning_rate": 4.0248250167827275e-05,
3302
+ "loss": 2.2222,
3303
+ "step": 466
3304
+ },
3305
+ {
3306
+ "epoch": 0.75,
3307
+ "grad_norm": 6.0625,
3308
+ "learning_rate": 3.976490040526394e-05,
3309
+ "loss": 2.2477,
3310
+ "step": 467
3311
+ },
3312
+ {
3313
+ "epoch": 0.75,
3314
+ "grad_norm": 4.46875,
3315
+ "learning_rate": 3.9283749218128885e-05,
3316
+ "loss": 2.5111,
3317
+ "step": 468
3318
+ },
3319
+ {
3320
+ "epoch": 0.76,
3321
+ "grad_norm": 4.71875,
3322
+ "learning_rate": 3.88048141683948e-05,
3323
+ "loss": 2.7461,
3324
+ "step": 469
3325
+ },
3326
+ {
3327
+ "epoch": 0.76,
3328
+ "grad_norm": 5.0,
3329
+ "learning_rate": 3.832811273714569e-05,
3330
+ "loss": 2.3025,
3331
+ "step": 470
3332
+ },
3333
+ {
3334
+ "epoch": 0.76,
3335
+ "grad_norm": 35.25,
3336
+ "learning_rate": 3.785366232393861e-05,
3337
+ "loss": 2.2489,
3338
+ "step": 471
3339
+ },
3340
+ {
3341
+ "epoch": 0.76,
3342
+ "grad_norm": 4.53125,
3343
+ "learning_rate": 3.738148024616863e-05,
3344
+ "loss": 2.3456,
3345
+ "step": 472
3346
+ },
3347
+ {
3348
+ "epoch": 0.76,
3349
+ "grad_norm": 4.5,
3350
+ "learning_rate": 3.691158373843694e-05,
3351
+ "loss": 2.0729,
3352
+ "step": 473
3353
+ },
3354
+ {
3355
+ "epoch": 0.76,
3356
+ "grad_norm": 920.0,
3357
+ "learning_rate": 3.644398995192147e-05,
3358
+ "loss": 7.747,
3359
+ "step": 474
3360
+ },
3361
+ {
3362
+ "epoch": 0.77,
3363
+ "grad_norm": 3.875,
3364
+ "learning_rate": 3.597871595375121e-05,
3365
+ "loss": 2.342,
3366
+ "step": 475
3367
+ },
3368
+ {
3369
+ "epoch": 0.77,
3370
+ "grad_norm": 4.1875,
3371
+ "learning_rate": 3.5515778726382966e-05,
3372
+ "loss": 2.0214,
3373
+ "step": 476
3374
+ },
3375
+ {
3376
+ "epoch": 0.77,
3377
+ "grad_norm": 5.4375,
3378
+ "learning_rate": 3.5055195166981645e-05,
3379
+ "loss": 2.6212,
3380
+ "step": 477
3381
+ },
3382
+ {
3383
+ "epoch": 0.77,
3384
+ "grad_norm": 6.21875,
3385
+ "learning_rate": 3.459698208680359e-05,
3386
+ "loss": 7.6408,
3387
+ "step": 478
3388
+ },
3389
+ {
3390
+ "epoch": 0.77,
3391
+ "grad_norm": 9.0,
3392
+ "learning_rate": 3.4141156210582756e-05,
3393
+ "loss": 4.4479,
3394
+ "step": 479
3395
+ },
3396
+ {
3397
+ "epoch": 0.77,
3398
+ "grad_norm": 5.75,
3399
+ "learning_rate": 3.36877341759205e-05,
3400
+ "loss": 2.6872,
3401
+ "step": 480
3402
+ },
3403
+ {
3404
+ "epoch": 0.78,
3405
+ "grad_norm": 4.40625,
3406
+ "learning_rate": 3.3236732532678096e-05,
3407
+ "loss": 2.3633,
3408
+ "step": 481
3409
+ },
3410
+ {
3411
+ "epoch": 0.78,
3412
+ "grad_norm": 4.125,
3413
+ "learning_rate": 3.2788167742372725e-05,
3414
+ "loss": 2.3661,
3415
+ "step": 482
3416
+ },
3417
+ {
3418
+ "epoch": 0.78,
3419
+ "grad_norm": 4.78125,
3420
+ "learning_rate": 3.234205617757686e-05,
3421
+ "loss": 1.8753,
3422
+ "step": 483
3423
+ },
3424
+ {
3425
+ "epoch": 0.78,
3426
+ "grad_norm": 10.8125,
3427
+ "learning_rate": 3.1898414121320276e-05,
3428
+ "loss": 7.2716,
3429
+ "step": 484
3430
+ },
3431
+ {
3432
+ "epoch": 0.78,
3433
+ "grad_norm": 4.8125,
3434
+ "learning_rate": 3.1457257766496015e-05,
3435
+ "loss": 2.298,
3436
+ "step": 485
3437
+ },
3438
+ {
3439
+ "epoch": 0.78,
3440
+ "grad_norm": 5.34375,
3441
+ "learning_rate": 3.101860321526924e-05,
3442
+ "loss": 2.6883,
3443
+ "step": 486
3444
+ },
3445
+ {
3446
+ "epoch": 0.79,
3447
+ "grad_norm": 23.25,
3448
+ "learning_rate": 3.0582466478489455e-05,
3449
+ "loss": 2.2113,
3450
+ "step": 487
3451
+ },
3452
+ {
3453
+ "epoch": 0.79,
3454
+ "grad_norm": 6.375,
3455
+ "learning_rate": 3.0148863475106314e-05,
3456
+ "loss": 2.2339,
3457
+ "step": 488
3458
+ },
3459
+ {
3460
+ "epoch": 0.79,
3461
+ "grad_norm": 4.5625,
3462
+ "learning_rate": 2.9717810031588277e-05,
3463
+ "loss": 2.4111,
3464
+ "step": 489
3465
+ },
3466
+ {
3467
+ "epoch": 0.79,
3468
+ "grad_norm": 4.6875,
3469
+ "learning_rate": 2.9289321881345254e-05,
3470
+ "loss": 2.2198,
3471
+ "step": 490
3472
+ },
3473
+ {
3474
+ "epoch": 0.79,
3475
+ "grad_norm": 4.3125,
3476
+ "learning_rate": 2.886341466415412e-05,
3477
+ "loss": 2.3096,
3478
+ "step": 491
3479
+ },
3480
+ {
3481
+ "epoch": 0.79,
3482
+ "grad_norm": 4.5625,
3483
+ "learning_rate": 2.84401039255879e-05,
3484
+ "loss": 1.8528,
3485
+ "step": 492
3486
+ },
3487
+ {
3488
+ "epoch": 0.8,
3489
+ "grad_norm": 4.46875,
3490
+ "learning_rate": 2.8019405116448516e-05,
3491
+ "loss": 2.6505,
3492
+ "step": 493
3493
+ },
3494
+ {
3495
+ "epoch": 0.8,
3496
+ "grad_norm": 13.25,
3497
+ "learning_rate": 2.7601333592202583e-05,
3498
+ "loss": 1.7653,
3499
+ "step": 494
3500
+ },
3501
+ {
3502
+ "epoch": 0.8,
3503
+ "grad_norm": 3.4375,
3504
+ "learning_rate": 2.7185904612421176e-05,
3505
+ "loss": 1.5667,
3506
+ "step": 495
3507
+ },
3508
+ {
3509
+ "epoch": 0.8,
3510
+ "grad_norm": 5.34375,
3511
+ "learning_rate": 2.677313334022268e-05,
3512
+ "loss": 2.4402,
3513
+ "step": 496
3514
+ },
3515
+ {
3516
+ "epoch": 0.8,
3517
+ "grad_norm": 5.53125,
3518
+ "learning_rate": 2.6363034841719392e-05,
3519
+ "loss": 2.3852,
3520
+ "step": 497
3521
+ },
3522
+ {
3523
+ "epoch": 0.8,
3524
+ "grad_norm": 4.6875,
3525
+ "learning_rate": 2.59556240854677e-05,
3526
+ "loss": 2.2586,
3527
+ "step": 498
3528
+ },
3529
+ {
3530
+ "epoch": 0.8,
3531
+ "grad_norm": 4.25,
3532
+ "learning_rate": 2.5550915941921526e-05,
3533
+ "loss": 2.429,
3534
+ "step": 499
3535
+ },
3536
+ {
3537
+ "epoch": 0.81,
3538
+ "grad_norm": 3.90625,
3539
+ "learning_rate": 2.514892518288988e-05,
3540
+ "loss": 1.9505,
3541
+ "step": 500
3542
+ },
3543
+ {
3544
+ "epoch": 0.81,
3545
+ "grad_norm": 3.875,
3546
+ "learning_rate": 2.4749666480997337e-05,
3547
+ "loss": 1.6284,
3548
+ "step": 501
3549
+ },
3550
+ {
3551
+ "epoch": 0.81,
3552
+ "grad_norm": 4.25,
3553
+ "learning_rate": 2.4353154409148637e-05,
3554
+ "loss": 2.4616,
3555
+ "step": 502
3556
+ },
3557
+ {
3558
+ "epoch": 0.81,
3559
+ "grad_norm": 4.1875,
3560
+ "learning_rate": 2.3959403439996907e-05,
3561
+ "loss": 2.2731,
3562
+ "step": 503
3563
+ },
3564
+ {
3565
+ "epoch": 0.81,
3566
+ "grad_norm": 3.65625,
3567
+ "learning_rate": 2.356842794541516e-05,
3568
+ "loss": 1.829,
3569
+ "step": 504
3570
+ },
3571
+ {
3572
+ "epoch": 0.81,
3573
+ "grad_norm": 4.34375,
3574
+ "learning_rate": 2.318024219597196e-05,
3575
+ "loss": 2.248,
3576
+ "step": 505
3577
+ },
3578
+ {
3579
+ "epoch": 0.82,
3580
+ "grad_norm": 3.640625,
3581
+ "learning_rate": 2.2794860360410342e-05,
3582
+ "loss": 1.652,
3583
+ "step": 506
3584
+ },
3585
+ {
3586
+ "epoch": 0.82,
3587
+ "grad_norm": 4.625,
3588
+ "learning_rate": 2.241229650513077e-05,
3589
+ "loss": 1.8933,
3590
+ "step": 507
3591
+ },
3592
+ {
3593
+ "epoch": 0.82,
3594
+ "grad_norm": 4.03125,
3595
+ "learning_rate": 2.2032564593677774e-05,
3596
+ "loss": 2.0758,
3597
+ "step": 508
3598
+ },
3599
+ {
3600
+ "epoch": 0.82,
3601
+ "grad_norm": 25.125,
3602
+ "learning_rate": 2.165567848623009e-05,
3603
+ "loss": 2.2063,
3604
+ "step": 509
3605
+ },
3606
+ {
3607
+ "epoch": 0.82,
3608
+ "grad_norm": 4.65625,
3609
+ "learning_rate": 2.1281651939094992e-05,
3610
+ "loss": 2.294,
3611
+ "step": 510
3612
+ },
3613
+ {
3614
+ "epoch": 0.82,
3615
+ "grad_norm": 4.34375,
3616
+ "learning_rate": 2.0910498604205986e-05,
3617
+ "loss": 2.1321,
3618
+ "step": 511
3619
+ },
3620
+ {
3621
+ "epoch": 0.83,
3622
+ "grad_norm": 3.828125,
3623
+ "learning_rate": 2.0542232028624586e-05,
3624
+ "loss": 1.805,
3625
+ "step": 512
3626
+ },
3627
+ {
3628
+ "epoch": 0.83,
3629
+ "grad_norm": 6.15625,
3630
+ "learning_rate": 2.0176865654045974e-05,
3631
+ "loss": 2.7504,
3632
+ "step": 513
3633
+ },
3634
+ {
3635
+ "epoch": 0.83,
3636
+ "grad_norm": 4.1875,
3637
+ "learning_rate": 1.981441281630816e-05,
3638
+ "loss": 2.51,
3639
+ "step": 514
3640
+ },
3641
+ {
3642
+ "epoch": 0.83,
3643
+ "grad_norm": 4.875,
3644
+ "learning_rate": 1.94548867449054e-05,
3645
+ "loss": 2.2938,
3646
+ "step": 515
3647
+ },
3648
+ {
3649
+ "epoch": 0.83,
3650
+ "grad_norm": 3.84375,
3651
+ "learning_rate": 1.9098300562505266e-05,
3652
+ "loss": 1.5338,
3653
+ "step": 516
3654
+ },
3655
+ {
3656
+ "epoch": 0.83,
3657
+ "grad_norm": 4.1875,
3658
+ "learning_rate": 1.8744667284469575e-05,
3659
+ "loss": 2.0832,
3660
+ "step": 517
3661
+ },
3662
+ {
3663
+ "epoch": 0.84,
3664
+ "grad_norm": 4.65625,
3665
+ "learning_rate": 1.8393999818379525e-05,
3666
+ "loss": 2.4751,
3667
+ "step": 518
3668
+ },
3669
+ {
3670
+ "epoch": 0.84,
3671
+ "grad_norm": 8.625,
3672
+ "learning_rate": 1.804631096356435e-05,
3673
+ "loss": 1.9146,
3674
+ "step": 519
3675
+ },
3676
+ {
3677
+ "epoch": 0.84,
3678
+ "grad_norm": 3.703125,
3679
+ "learning_rate": 1.7701613410634365e-05,
3680
+ "loss": 1.7048,
3681
+ "step": 520
3682
+ },
3683
+ {
3684
+ "epoch": 0.84,
3685
+ "grad_norm": 8.875,
3686
+ "learning_rate": 1.735991974101756e-05,
3687
+ "loss": 3.7879,
3688
+ "step": 521
3689
+ },
3690
+ {
3691
+ "epoch": 0.84,
3692
+ "grad_norm": 4.125,
3693
+ "learning_rate": 1.7021242426500493e-05,
3694
+ "loss": 1.8485,
3695
+ "step": 522
3696
+ },
3697
+ {
3698
+ "epoch": 0.84,
3699
+ "grad_norm": 9.125,
3700
+ "learning_rate": 1.6685593828773095e-05,
3701
+ "loss": 7.5836,
3702
+ "step": 523
3703
+ },
3704
+ {
3705
+ "epoch": 0.85,
3706
+ "grad_norm": 5.1875,
3707
+ "learning_rate": 1.6352986198977325e-05,
3708
+ "loss": 2.4005,
3709
+ "step": 524
3710
+ },
3711
+ {
3712
+ "epoch": 0.85,
3713
+ "grad_norm": 3.578125,
3714
+ "learning_rate": 1.6023431677260214e-05,
3715
+ "loss": 1.6218,
3716
+ "step": 525
3717
+ },
3718
+ {
3719
+ "epoch": 0.85,
3720
+ "grad_norm": 4.71875,
3721
+ "learning_rate": 1.5696942292330576e-05,
3722
+ "loss": 2.416,
3723
+ "step": 526
3724
+ },
3725
+ {
3726
+ "epoch": 0.85,
3727
+ "grad_norm": 5.5625,
3728
+ "learning_rate": 1.5373529961019974e-05,
3729
+ "loss": 2.3503,
3730
+ "step": 527
3731
+ },
3732
+ {
3733
+ "epoch": 0.85,
3734
+ "grad_norm": 4.75,
3735
+ "learning_rate": 1.5053206487847914e-05,
3736
+ "loss": 2.7853,
3737
+ "step": 528
3738
+ },
3739
+ {
3740
+ "epoch": 0.85,
3741
+ "grad_norm": 5.1875,
3742
+ "learning_rate": 1.4735983564590783e-05,
3743
+ "loss": 2.2424,
3744
+ "step": 529
3745
+ },
3746
+ {
3747
+ "epoch": 0.85,
3748
+ "grad_norm": 6.375,
3749
+ "learning_rate": 1.442187276985526e-05,
3750
+ "loss": 2.0547,
3751
+ "step": 530
3752
+ },
3753
+ {
3754
+ "epoch": 0.86,
3755
+ "grad_norm": 6.59375,
3756
+ "learning_rate": 1.4110885568655564e-05,
3757
+ "loss": 2.848,
3758
+ "step": 531
3759
+ },
3760
+ {
3761
+ "epoch": 0.86,
3762
+ "grad_norm": 3.9375,
3763
+ "learning_rate": 1.3803033311995072e-05,
3764
+ "loss": 1.7906,
3765
+ "step": 532
3766
+ },
3767
+ {
3768
+ "epoch": 0.86,
3769
+ "grad_norm": 5.71875,
3770
+ "learning_rate": 1.3498327236452013e-05,
3771
+ "loss": 2.9018,
3772
+ "step": 533
3773
+ },
3774
+ {
3775
+ "epoch": 0.86,
3776
+ "grad_norm": 4.75,
3777
+ "learning_rate": 1.3196778463769255e-05,
3778
+ "loss": 2.6102,
3779
+ "step": 534
3780
+ },
3781
+ {
3782
+ "epoch": 0.86,
3783
+ "grad_norm": 4.15625,
3784
+ "learning_rate": 1.2898398000448443e-05,
3785
+ "loss": 2.3094,
3786
+ "step": 535
3787
+ },
3788
+ {
3789
+ "epoch": 0.86,
3790
+ "grad_norm": 4.125,
3791
+ "learning_rate": 1.260319673734821e-05,
3792
+ "loss": 1.9745,
3793
+ "step": 536
3794
+ },
3795
+ {
3796
+ "epoch": 0.87,
3797
+ "grad_norm": 4.3125,
3798
+ "learning_rate": 1.2311185449286677e-05,
3799
+ "loss": 1.8915,
3800
+ "step": 537
3801
+ },
3802
+ {
3803
+ "epoch": 0.87,
3804
+ "grad_norm": 5.15625,
3805
+ "learning_rate": 1.2022374794648228e-05,
3806
+ "loss": 2.6586,
3807
+ "step": 538
3808
+ },
3809
+ {
3810
+ "epoch": 0.87,
3811
+ "grad_norm": 3.40625,
3812
+ "learning_rate": 1.1736775314994342e-05,
3813
+ "loss": 1.7468,
3814
+ "step": 539
3815
+ },
3816
+ {
3817
+ "epoch": 0.87,
3818
+ "grad_norm": 4.375,
3819
+ "learning_rate": 1.1454397434679021e-05,
3820
+ "loss": 2.3997,
3821
+ "step": 540
3822
+ },
3823
+ {
3824
+ "epoch": 0.87,
3825
+ "grad_norm": 3.765625,
3826
+ "learning_rate": 1.1175251460468117e-05,
3827
+ "loss": 1.6562,
3828
+ "step": 541
3829
+ },
3830
+ {
3831
+ "epoch": 0.87,
3832
+ "grad_norm": 4.71875,
3833
+ "learning_rate": 1.0899347581163221e-05,
3834
+ "loss": 2.4249,
3835
+ "step": 542
3836
+ },
3837
+ {
3838
+ "epoch": 0.88,
3839
+ "grad_norm": 4.5625,
3840
+ "learning_rate": 1.062669586722983e-05,
3841
+ "loss": 2.2151,
3842
+ "step": 543
3843
+ },
3844
+ {
3845
+ "epoch": 0.88,
3846
+ "grad_norm": 6.25,
3847
+ "learning_rate": 1.0357306270429624e-05,
3848
+ "loss": 7.3873,
3849
+ "step": 544
3850
+ },
3851
+ {
3852
+ "epoch": 0.88,
3853
+ "grad_norm": 3.8125,
3854
+ "learning_rate": 1.0091188623457415e-05,
3855
+ "loss": 2.1178,
3856
+ "step": 545
3857
+ },
3858
+ {
3859
+ "epoch": 0.88,
3860
+ "grad_norm": 4.09375,
3861
+ "learning_rate": 9.828352639582072e-06,
3862
+ "loss": 1.846,
3863
+ "step": 546
3864
+ },
3865
+ {
3866
+ "epoch": 0.88,
3867
+ "grad_norm": 6.03125,
3868
+ "learning_rate": 9.568807912292077e-06,
3869
+ "loss": 2.2441,
3870
+ "step": 547
3871
+ },
3872
+ {
3873
+ "epoch": 0.88,
3874
+ "grad_norm": 4.59375,
3875
+ "learning_rate": 9.31256391494546e-06,
3876
+ "loss": 2.2284,
3877
+ "step": 548
3878
+ },
3879
+ {
3880
+ "epoch": 0.89,
3881
+ "grad_norm": 4.875,
3882
+ "learning_rate": 9.05963000042378e-06,
3883
+ "loss": 2.7284,
3884
+ "step": 549
3885
+ },
3886
+ {
3887
+ "epoch": 0.89,
3888
+ "grad_norm": 4.3125,
3889
+ "learning_rate": 8.810015400790994e-06,
3890
+ "loss": 2.1872,
3891
+ "step": 550
3892
+ },
3893
+ {
3894
+ "epoch": 0.89,
3895
+ "grad_norm": 4.90625,
3896
+ "learning_rate": 8.563729226956319e-06,
3897
+ "loss": 2.4118,
3898
+ "step": 551
3899
+ },
3900
+ {
3901
+ "epoch": 0.89,
3902
+ "grad_norm": 10.0625,
3903
+ "learning_rate": 8.32078046834176e-06,
3904
+ "loss": 7.5981,
3905
+ "step": 552
3906
+ },
3907
+ {
3908
+ "epoch": 0.89,
3909
+ "grad_norm": 46.5,
3910
+ "learning_rate": 8.081177992554013e-06,
3911
+ "loss": 2.3079,
3912
+ "step": 553
3913
+ },
3914
+ {
3915
+ "epoch": 0.89,
3916
+ "grad_norm": 3.984375,
3917
+ "learning_rate": 7.844930545060703e-06,
3918
+ "loss": 2.1643,
3919
+ "step": 554
3920
+ },
3921
+ {
3922
+ "epoch": 0.9,
3923
+ "grad_norm": 4.34375,
3924
+ "learning_rate": 7.612046748871327e-06,
3925
+ "loss": 1.9431,
3926
+ "step": 555
3927
+ },
3928
+ {
3929
+ "epoch": 0.9,
3930
+ "grad_norm": 6.0,
3931
+ "learning_rate": 7.382535104222366e-06,
3932
+ "loss": 2.8449,
3933
+ "step": 556
3934
+ },
3935
+ {
3936
+ "epoch": 0.9,
3937
+ "grad_norm": 5.28125,
3938
+ "learning_rate": 7.156403988267069e-06,
3939
+ "loss": 2.7214,
3940
+ "step": 557
3941
+ },
3942
+ {
3943
+ "epoch": 0.9,
3944
+ "grad_norm": 4.625,
3945
+ "learning_rate": 6.9336616547697965e-06,
3946
+ "loss": 2.7913,
3947
+ "step": 558
3948
+ },
3949
+ {
3950
+ "epoch": 0.9,
3951
+ "grad_norm": 4.25,
3952
+ "learning_rate": 6.714316233804574e-06,
3953
+ "loss": 2.2604,
3954
+ "step": 559
3955
+ },
3956
+ {
3957
+ "epoch": 0.9,
3958
+ "grad_norm": 4.78125,
3959
+ "learning_rate": 6.498375731458528e-06,
3960
+ "loss": 2.154,
3961
+ "step": 560
3962
+ },
3963
+ {
3964
+ "epoch": 0.9,
3965
+ "grad_norm": 5.6875,
3966
+ "learning_rate": 6.28584802953951e-06,
3967
+ "loss": 2.0626,
3968
+ "step": 561
3969
+ },
3970
+ {
3971
+ "epoch": 0.91,
3972
+ "grad_norm": 33.0,
3973
+ "learning_rate": 6.076740885288479e-06,
3974
+ "loss": 2.6221,
3975
+ "step": 562
3976
+ },
3977
+ {
3978
+ "epoch": 0.91,
3979
+ "grad_norm": 4.71875,
3980
+ "learning_rate": 5.8710619310964445e-06,
3981
+ "loss": 2.5808,
3982
+ "step": 563
3983
+ },
3984
+ {
3985
+ "epoch": 0.91,
3986
+ "grad_norm": 4.28125,
3987
+ "learning_rate": 5.668818674225685e-06,
3988
+ "loss": 2.2286,
3989
+ "step": 564
3990
+ },
3991
+ {
3992
+ "epoch": 0.91,
3993
+ "grad_norm": 3.859375,
3994
+ "learning_rate": 5.470018496535967e-06,
3995
+ "loss": 2.0533,
3996
+ "step": 565
3997
+ },
3998
+ {
3999
+ "epoch": 0.91,
4000
+ "grad_norm": 3.796875,
4001
+ "learning_rate": 5.274668654214932e-06,
4002
+ "loss": 2.123,
4003
+ "step": 566
4004
+ },
4005
+ {
4006
+ "epoch": 0.91,
4007
+ "grad_norm": 5.4375,
4008
+ "learning_rate": 5.08277627751329e-06,
4009
+ "loss": 2.3784,
4010
+ "step": 567
4011
+ },
4012
+ {
4013
+ "epoch": 0.92,
4014
+ "grad_norm": 3.875,
4015
+ "learning_rate": 4.8943483704846475e-06,
4016
+ "loss": 1.9758,
4017
+ "step": 568
4018
+ },
4019
+ {
4020
+ "epoch": 0.92,
4021
+ "grad_norm": 3.703125,
4022
+ "learning_rate": 4.709391810729713e-06,
4023
+ "loss": 1.7575,
4024
+ "step": 569
4025
+ },
4026
+ {
4027
+ "epoch": 0.92,
4028
+ "grad_norm": 3.71875,
4029
+ "learning_rate": 4.527913349145441e-06,
4030
+ "loss": 2.0692,
4031
+ "step": 570
4032
+ },
4033
+ {
4034
+ "epoch": 0.92,
4035
+ "grad_norm": 4.75,
4036
+ "learning_rate": 4.349919609678455e-06,
4037
+ "loss": 2.6031,
4038
+ "step": 571
4039
+ },
4040
+ {
4041
+ "epoch": 0.92,
4042
+ "grad_norm": 4.125,
4043
+ "learning_rate": 4.175417089083378e-06,
4044
+ "loss": 2.3837,
4045
+ "step": 572
4046
+ },
4047
+ {
4048
+ "epoch": 0.92,
4049
+ "grad_norm": 5.65625,
4050
+ "learning_rate": 4.004412156685711e-06,
4051
+ "loss": 2.6099,
4052
+ "step": 573
4053
+ },
4054
+ {
4055
+ "epoch": 0.93,
4056
+ "grad_norm": 3.359375,
4057
+ "learning_rate": 3.836911054149239e-06,
4058
+ "loss": 1.7974,
4059
+ "step": 574
4060
+ },
4061
+ {
4062
+ "epoch": 0.93,
4063
+ "grad_norm": 5.03125,
4064
+ "learning_rate": 3.6729198952483724e-06,
4065
+ "loss": 1.8151,
4066
+ "step": 575
4067
+ },
4068
+ {
4069
+ "epoch": 0.93,
4070
+ "grad_norm": 4.8125,
4071
+ "learning_rate": 3.512444665644865e-06,
4072
+ "loss": 2.2943,
4073
+ "step": 576
4074
+ },
4075
+ {
4076
+ "epoch": 0.93,
4077
+ "grad_norm": 4.09375,
4078
+ "learning_rate": 3.355491222669371e-06,
4079
+ "loss": 1.9795,
4080
+ "step": 577
4081
+ },
4082
+ {
4083
+ "epoch": 0.93,
4084
+ "grad_norm": 5.09375,
4085
+ "learning_rate": 3.202065295107726e-06,
4086
+ "loss": 2.1805,
4087
+ "step": 578
4088
+ },
4089
+ {
4090
+ "epoch": 0.93,
4091
+ "grad_norm": 4.21875,
4092
+ "learning_rate": 3.052172482991711e-06,
4093
+ "loss": 1.7644,
4094
+ "step": 579
4095
+ },
4096
+ {
4097
+ "epoch": 0.94,
4098
+ "grad_norm": 27.125,
4099
+ "learning_rate": 2.905818257394799e-06,
4100
+ "loss": 2.1623,
4101
+ "step": 580
4102
+ },
4103
+ {
4104
+ "epoch": 0.94,
4105
+ "grad_norm": 4.34375,
4106
+ "learning_rate": 2.7630079602323442e-06,
4107
+ "loss": 1.8096,
4108
+ "step": 581
4109
+ },
4110
+ {
4111
+ "epoch": 0.94,
4112
+ "grad_norm": 4.375,
4113
+ "learning_rate": 2.6237468040666512e-06,
4114
+ "loss": 2.0634,
4115
+ "step": 582
4116
+ },
4117
+ {
4118
+ "epoch": 0.94,
4119
+ "grad_norm": 4.25,
4120
+ "learning_rate": 2.4880398719167586e-06,
4121
+ "loss": 2.0585,
4122
+ "step": 583
4123
+ },
4124
+ {
4125
+ "epoch": 0.94,
4126
+ "grad_norm": 4.3125,
4127
+ "learning_rate": 2.3558921170727888e-06,
4128
+ "loss": 2.2075,
4129
+ "step": 584
4130
+ },
4131
+ {
4132
+ "epoch": 0.94,
4133
+ "grad_norm": 3.765625,
4134
+ "learning_rate": 2.2273083629153147e-06,
4135
+ "loss": 1.8263,
4136
+ "step": 585
4137
+ },
4138
+ {
4139
+ "epoch": 0.95,
4140
+ "grad_norm": 5.3125,
4141
+ "learning_rate": 2.1022933027391555e-06,
4142
+ "loss": 2.5711,
4143
+ "step": 586
4144
+ },
4145
+ {
4146
+ "epoch": 0.95,
4147
+ "grad_norm": 4.0625,
4148
+ "learning_rate": 1.9808514995821593e-06,
4149
+ "loss": 1.977,
4150
+ "step": 587
4151
+ },
4152
+ {
4153
+ "epoch": 0.95,
4154
+ "grad_norm": 4.625,
4155
+ "learning_rate": 1.8629873860586566e-06,
4156
+ "loss": 2.3325,
4157
+ "step": 588
4158
+ },
4159
+ {
4160
+ "epoch": 0.95,
4161
+ "grad_norm": 4.0625,
4162
+ "learning_rate": 1.7487052641976032e-06,
4163
+ "loss": 1.6915,
4164
+ "step": 589
4165
+ },
4166
+ {
4167
+ "epoch": 0.95,
4168
+ "grad_norm": 5.40625,
4169
+ "learning_rate": 1.6380093052856483e-06,
4170
+ "loss": 7.4379,
4171
+ "step": 590
4172
+ },
4173
+ {
4174
+ "epoch": 0.95,
4175
+ "grad_norm": 5.46875,
4176
+ "learning_rate": 1.5309035497147684e-06,
4177
+ "loss": 2.7398,
4178
+ "step": 591
4179
+ },
4180
+ {
4181
+ "epoch": 0.95,
4182
+ "grad_norm": 7.125,
4183
+ "learning_rate": 1.4273919068349184e-06,
4184
+ "loss": 7.5892,
4185
+ "step": 592
4186
+ },
4187
+ {
4188
+ "epoch": 0.96,
4189
+ "grad_norm": 4.78125,
4190
+ "learning_rate": 1.3274781548112458e-06,
4191
+ "loss": 2.5225,
4192
+ "step": 593
4193
+ },
4194
+ {
4195
+ "epoch": 0.96,
4196
+ "grad_norm": 5.78125,
4197
+ "learning_rate": 1.231165940486234e-06,
4198
+ "loss": 2.7149,
4199
+ "step": 594
4200
+ },
4201
+ {
4202
+ "epoch": 0.96,
4203
+ "grad_norm": 4.1875,
4204
+ "learning_rate": 1.1384587792465872e-06,
4205
+ "loss": 2.4038,
4206
+ "step": 595
4207
+ },
4208
+ {
4209
+ "epoch": 0.96,
4210
+ "grad_norm": 4.84375,
4211
+ "learning_rate": 1.0493600548948878e-06,
4212
+ "loss": 2.4474,
4213
+ "step": 596
4214
+ },
4215
+ {
4216
+ "epoch": 0.96,
4217
+ "grad_norm": 3.9375,
4218
+ "learning_rate": 9.638730195261625e-07,
4219
+ "loss": 1.8508,
4220
+ "step": 597
4221
+ },
4222
+ {
4223
+ "epoch": 0.96,
4224
+ "grad_norm": 4.84375,
4225
+ "learning_rate": 8.820007934090879e-07,
4226
+ "loss": 2.5266,
4227
+ "step": 598
4228
+ },
4229
+ {
4230
+ "epoch": 0.97,
4231
+ "grad_norm": 7.1875,
4232
+ "learning_rate": 8.037463648721488e-07,
4233
+ "loss": 2.5697,
4234
+ "step": 599
4235
+ },
4236
+ {
4237
+ "epoch": 0.97,
4238
+ "grad_norm": 5.6875,
4239
+ "learning_rate": 7.291125901946027e-07,
4240
+ "loss": 2.1599,
4241
+ "step": 600
4242
+ },
4243
+ {
4244
+ "epoch": 0.97,
4245
+ "grad_norm": 24.875,
4246
+ "learning_rate": 6.581021935021304e-07,
4247
+ "loss": 1.7763,
4248
+ "step": 601
4249
+ },
4250
+ {
4251
+ "epoch": 0.97,
4252
+ "grad_norm": 4.46875,
4253
+ "learning_rate": 5.907177666674812e-07,
4254
+ "loss": 2.4318,
4255
+ "step": 602
4256
+ },
4257
+ {
4258
+ "epoch": 0.97,
4259
+ "grad_norm": 4.03125,
4260
+ "learning_rate": 5.269617692158613e-07,
4261
+ "loss": 1.7822,
4262
+ "step": 603
4263
+ },
4264
+ {
4265
+ "epoch": 0.97,
4266
+ "grad_norm": 5.21875,
4267
+ "learning_rate": 4.668365282351372e-07,
4268
+ "loss": 2.5982,
4269
+ "step": 604
4270
+ },
4271
+ {
4272
+ "epoch": 0.98,
4273
+ "grad_norm": 4.0625,
4274
+ "learning_rate": 4.103442382909051e-07,
4275
+ "loss": 2.2378,
4276
+ "step": 605
4277
+ },
4278
+ {
4279
+ "epoch": 0.98,
4280
+ "grad_norm": 3.796875,
4281
+ "learning_rate": 3.5748696134639825e-07,
4282
+ "loss": 2.1997,
4283
+ "step": 606
4284
+ },
4285
+ {
4286
+ "epoch": 0.98,
4287
+ "grad_norm": 4.46875,
4288
+ "learning_rate": 3.0826662668720364e-07,
4289
+ "loss": 2.1681,
4290
+ "step": 607
4291
+ },
4292
+ {
4293
+ "epoch": 0.98,
4294
+ "grad_norm": 5.21875,
4295
+ "learning_rate": 2.6268503085089547e-07,
4296
+ "loss": 2.4847,
4297
+ "step": 608
4298
+ },
4299
+ {
4300
+ "epoch": 0.98,
4301
+ "grad_norm": 4.71875,
4302
+ "learning_rate": 2.2074383756137686e-07,
4303
+ "loss": 2.4499,
4304
+ "step": 609
4305
+ },
4306
+ {
4307
+ "epoch": 0.98,
4308
+ "grad_norm": 5.5,
4309
+ "learning_rate": 1.824445776682504e-07,
4310
+ "loss": 2.7326,
4311
+ "step": 610
4312
+ },
4313
+ {
4314
+ "epoch": 0.99,
4315
+ "grad_norm": 4.15625,
4316
+ "learning_rate": 1.477886490908742e-07,
4317
+ "loss": 1.8444,
4318
+ "step": 611
4319
+ },
4320
+ {
4321
+ "epoch": 0.99,
4322
+ "grad_norm": 5.9375,
4323
+ "learning_rate": 1.1677731676733584e-07,
4324
+ "loss": 7.546,
4325
+ "step": 612
4326
+ },
4327
+ {
4328
+ "epoch": 0.99,
4329
+ "grad_norm": 4.25,
4330
+ "learning_rate": 8.941171260835601e-08,
4331
+ "loss": 1.8015,
4332
+ "step": 613
4333
+ },
4334
+ {
4335
+ "epoch": 0.99,
4336
+ "grad_norm": 4.59375,
4337
+ "learning_rate": 6.569283545587724e-08,
4338
+ "loss": 2.0968,
4339
+ "step": 614
4340
+ },
4341
+ {
4342
+ "epoch": 0.99,
4343
+ "grad_norm": 18.125,
4344
+ "learning_rate": 4.562155104665955e-08,
4345
+ "loss": 2.2729,
4346
+ "step": 615
4347
+ },
4348
+ {
4349
+ "epoch": 0.99,
4350
+ "grad_norm": 4.03125,
4351
+ "learning_rate": 2.9198591980705848e-08,
4352
+ "loss": 2.1077,
4353
+ "step": 616
4354
+ },
4355
+ {
4356
+ "epoch": 1.0,
4357
+ "grad_norm": 4.90625,
4358
+ "learning_rate": 1.642455769444995e-08,
4359
+ "loss": 2.2426,
4360
+ "step": 617
4361
+ },
4362
+ {
4363
+ "epoch": 1.0,
4364
+ "grad_norm": 3.84375,
4365
+ "learning_rate": 7.2999144389296335e-09,
4366
+ "loss": 1.9934,
4367
+ "step": 618
4368
+ },
4369
+ {
4370
+ "epoch": 1.0,
4371
+ "grad_norm": 4.0625,
4372
+ "learning_rate": 1.8249952627669154e-09,
4373
+ "loss": 2.5062,
4374
+ "step": 619
4375
+ },
4376
+ {
4377
+ "epoch": 1.0,
4378
+ "grad_norm": 8.8125,
4379
+ "learning_rate": 0.0,
4380
+ "loss": 1.7745,
4381
+ "step": 620
4382
+ },
4383
+ {
4384
+ "epoch": 1.0,
4385
+ "eval_loss": 3.2154979705810547,
4386
+ "eval_runtime": 1.7336,
4387
+ "eval_samples_per_second": 19.612,
4388
+ "eval_steps_per_second": 19.612,
4389
+ "step": 620
4390
+ }
4391
+ ],
4392
+ "logging_steps": 1,
4393
+ "max_steps": 620,
4394
+ "num_input_tokens_seen": 0,
4395
+ "num_train_epochs": 1,
4396
+ "save_steps": 500,
4397
+ "total_flos": 7881494668247040.0,
4398
+ "train_batch_size": 1,
4399
+ "trial_name": null,
4400
+ "trial_params": null
4401
+ }
checkpoint-620/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e01e08cc22a95eaf6e2958d2dfcf1d9950b1b17a600fc94764a791c627fb5fa9
3
+ size 5752
config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 5632,
14
+ "max_position_embeddings": 2048,
15
+ "model_type": "llama",
16
+ "num_attention_heads": 32,
17
+ "num_hidden_layers": 22,
18
+ "num_key_value_heads": 4,
19
+ "pretraining_tp": 1,
20
+ "rms_norm_eps": 1e-05,
21
+ "rope_scaling": null,
22
+ "rope_theta": 10000.0,
23
+ "tie_word_embeddings": false,
24
+ "torch_dtype": "bfloat16",
25
+ "transformers_version": "4.38.2",
26
+ "use_cache": false,
27
+ "vocab_size": 32000
28
+ }
generation_config.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 1,
3
+ "do_sample": true,
4
+ "eos_token_id": 2,
5
+ "max_length": 2048,
6
+ "pad_token_id": 0,
7
+ "transformers_version": "4.38.2"
8
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c55937a0bf1e53e3ecf5517a13d6c7180b1f2caaf97ca2dbf44fa67acbe3726
3
+ size 2200160278
special_tokens_map.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": "</s>",
17
+ "unk_token": {
18
+ "content": "<unk>",
19
+ "lstrip": false,
20
+ "normalized": false,
21
+ "rstrip": false,
22
+ "single_word": false
23
+ }
24
+ }
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
3
+ size 499723
tokenizer_config.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "add_prefix_space": true,
5
+ "added_tokens_decoder": {
6
+ "0": {
7
+ "content": "<unk>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false,
12
+ "special": true
13
+ },
14
+ "1": {
15
+ "content": "<s>",
16
+ "lstrip": false,
17
+ "normalized": false,
18
+ "rstrip": false,
19
+ "single_word": false,
20
+ "special": true
21
+ },
22
+ "2": {
23
+ "content": "</s>",
24
+ "lstrip": false,
25
+ "normalized": false,
26
+ "rstrip": false,
27
+ "single_word": false,
28
+ "special": true
29
+ }
30
+ },
31
+ "bos_token": "<s>",
32
+ "clean_up_tokenization_spaces": false,
33
+ "eos_token": "</s>",
34
+ "legacy": false,
35
+ "model_max_length": 1000000000000000019884624838656,
36
+ "pad_token": "</s>",
37
+ "padding_side": "right",
38
+ "sp_model_kwargs": {},
39
+ "spaces_between_special_tokens": false,
40
+ "tokenizer_class": "LlamaTokenizer",
41
+ "unk_token": "<unk>",
42
+ "use_default_system_prompt": false,
43
+ "use_fast": true
44
+ }