Hugging Face's logo

CMU-AIR2
/
math-deepseek-FULL-ArithHard-50k

Text Generation
Transformers
Safetensors
llama
conversational
Inference Endpoints
text-generation-inference
Model card Files Community
slackingfred commited on
Commit
9056c91
•
1 Parent(s): 9404cfe

Save best model

Browse files
Files changed (11) hide show
  1. config.json +31 -0
  2. generation_config.json +6 -0
  3. model.safetensors +3 -0
  4. optimizer.pt +3 -0
  5. rng_state.pth +3 -0
  6. run_config.json +77 -0
  7. scheduler.pt +3 -0
  8. tokenizer.json +0 -0
  9. tokenizer_config.json +35 -0
  10. trainer_state.json +3681 -0
  11. training_args.bin +3 -0
config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "deepseek-ai/deepseek-coder-1.3b-instruct",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 32013,
9
+ "eos_token_id": 32021,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 5504,
14
+ "max_position_embeddings": 16384,
15
+ "model_type": "llama",
16
+ "num_attention_heads": 16,
17
+ "num_hidden_layers": 24,
18
+ "num_key_value_heads": 16,
19
+ "pretraining_tp": 1,
20
+ "rms_norm_eps": 1e-06,
21
+ "rope_scaling": {
22
+ "factor": 4.0,
23
+ "type": "linear"
24
+ },
25
+ "rope_theta": 100000,
26
+ "tie_word_embeddings": false,
27
+ "torch_dtype": "float16",
28
+ "transformers_version": "4.39.0.dev0",
29
+ "use_cache": true,
30
+ "vocab_size": 32256
31
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 32013,
4
+ "eos_token_id": 32021,
5
+ "transformers_version": "4.39.0.dev0"
6
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8758b81eca050aa3f80bec273985f52d367674d347ca321c73d08695b7ab39ca
3
+ size 2692968904
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4ca2eb9f51827959c2286e3cc9e0bd3faffca1e95448b230fbec0ca2292b16c
3
+ size 2699039674
rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4ceca872404f155cf0d9853014eb18c0cc7e23cd52acf7a31c07149383dc6f9e
3
+ size 14244
run_config.json ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "base_model_id": "deepseek-ai/deepseek-coder-1.3b-instruct",
3
+ "quantitize": "fp16",
4
+ "dataset": "Arithmetic_Hard",
5
+ "data_collator": "DataCollatorForLanguageModeling",
6
+ "peft_config": {
7
+ "lora": {
8
+ "r": 32,
9
+ "lora_alpha": 64,
10
+ "target_modules": [
11
+ "q_proj",
12
+ "k_proj",
13
+ "v_proj",
14
+ "o_proj",
15
+ "gate_proj",
16
+ "up_proj",
17
+ "down_proj"
18
+ ],
19
+ "bias": "none",
20
+ "lora_dropout": 0.05,
21
+ "task_type": "CAUSAL_LM"
22
+ },
23
+ "lora_large": {
24
+ "r": 128,
25
+ "lora_alpha": 256,
26
+ "target_modules": [
27
+ "q_proj",
28
+ "k_proj",
29
+ "v_proj",
30
+ "o_proj",
31
+ "gate_proj",
32
+ "up_proj",
33
+ "down_proj"
34
+ ],
35
+ "bias": "none",
36
+ "lora_dropout": 0.05,
37
+ "task_type": "CAUSAL_LM"
38
+ },
39
+ "p_tuning": {
40
+ "num_virtual_tokens": 16,
41
+ "num_transformer_submodules": 1,
42
+ "token_dim": 2048,
43
+ "encoder_hidden_size": 2048,
44
+ "task_type": "CAUSAL_LM"
45
+ }
46
+ },
47
+ "training_args": {
48
+ "warmup_steps": 500,
49
+ "per_device_train_batch_size": 4,
50
+ "per_device_eval_batch_size": 4,
51
+ "gradient_accumulation_steps": 1,
52
+ "max_steps": 100000,
53
+ "learning_rate": 0.0001,
54
+ "optim": "paged_adamw_8bit",
55
+ "logging_steps": 100,
56
+ "save_strategy": "steps",
57
+ "save_total_limit": 5,
58
+ "save_steps": 2500,
59
+ "evaluation_strategy": "steps",
60
+ "eval_steps": 2500,
61
+ "weight_decay": 0.01,
62
+ "report_to": "wandb",
63
+ "dataloader_num_workers": 4,
64
+ "load_best_model_at_end": true,
65
+ "fp16": false,
66
+ "output_dir": "runs/deepseek-full-hard",
67
+ "logging_dir": "runs/deepseek-full-hard/logs"
68
+ },
69
+ "tokenizer": {
70
+ "tokenize_config": {
71
+ "truncation": true,
72
+ "max_length": 512,
73
+ "padding": "max_length"
74
+ },
75
+ "prompt_template": "config/qa_template.txt"
76
+ }
77
+ }
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c33e06479db97dbbf4d8dd72315d27433ca92d8110cff4086f2d8380597fd0f8
3
+ size 1064
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "bos_token": {
5
+ "__type": "AddedToken",
6
+ "content": "<｜begin▁of▁sentence｜>",
7
+ "lstrip": false,
8
+ "normalized": true,
9
+ "rstrip": false,
10
+ "single_word": false
11
+ },
12
+ "clean_up_tokenization_spaces": false,
13
+ "eos_token": {
14
+ "__type": "AddedToken",
15
+ "content": "<|EOT|>",
16
+ "lstrip": false,
17
+ "normalized": true,
18
+ "rstrip": false,
19
+ "single_word": false
20
+ },
21
+ "legacy": true,
22
+ "model_max_length": 16384,
23
+ "pad_token": {
24
+ "__type": "AddedToken",
25
+ "content": "<｜end▁of▁sentence｜>",
26
+ "lstrip": false,
27
+ "normalized": true,
28
+ "rstrip": false,
29
+ "single_word": false
30
+ },
31
+ "sp_model_kwargs": {},
32
+ "unk_token": null,
33
+ "tokenizer_class": "LlamaTokenizerFast",
34
+ "chat_template": "{% if not add_generation_prompt is defined %}\n{% set add_generation_prompt = false %}\n{% endif %}\n{%- set ns = namespace(found=false) -%}\n{%- for message in messages -%}\n {%- if message['role'] == 'system' -%}\n {%- set ns.found = true -%}\n {%- endif -%}\n{%- endfor -%}\n{{bos_token}}{%- if not ns.found -%}\n{{'You are an AI programming assistant, utilizing the Deepseek Coder model, developed by Deepseek Company, and you only answer questions related to computer science. For politically sensitive questions, security and privacy issues, and other non-computer science questions, you will refuse to answer\\n'}}\n{%- endif %}\n{%- for message in messages %}\n {%- if message['role'] == 'system' %}\n{{ message['content'] }}\n {%- else %}\n {%- if message['role'] == 'user' %}\n{{'### Instruction:\\n' + message['content'] + '\\n'}}\n {%- else %}\n{{'### Response:\\n' + message['content'] + '\\n<|EOT|>\\n'}}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{% if add_generation_prompt %}\n{{'### Response:'}}\n{% endif %}"
35
+ }
trainer_state.json ADDED
@@ -0,0 +1,3681 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 1.704933762550354,
3
+ "best_model_checkpoint": "runs/deepseek-full-hard/checkpoint-50000",
4
+ "epoch": 1.25,
5
+ "eval_steps": 2500,
6
+ "global_step": 50000,
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": 24.3125,
14
+ "learning_rate": 2e-05,
15
+ "loss": 1.9182,
16
+ "step": 100
17
+ },
18
+ {
19
+ "epoch": 0.01,
20
+ "grad_norm": 13.9296875,
21
+ "learning_rate": 4e-05,
22
+ "loss": 1.9313,
23
+ "step": 200
24
+ },
25
+ {
26
+ "epoch": 0.01,
27
+ "grad_norm": 7.0,
28
+ "learning_rate": 6e-05,
29
+ "loss": 1.9142,
30
+ "step": 300
31
+ },
32
+ {
33
+ "epoch": 0.01,
34
+ "grad_norm": 10.0703125,
35
+ "learning_rate": 8e-05,
36
+ "loss": 1.9665,
37
+ "step": 400
38
+ },
39
+ {
40
+ "epoch": 0.01,
41
+ "grad_norm": 8.0703125,
42
+ "learning_rate": 0.0001,
43
+ "loss": 1.9457,
44
+ "step": 500
45
+ },
46
+ {
47
+ "epoch": 0.01,
48
+ "grad_norm": 13.6484375,
49
+ "learning_rate": 9.989949748743719e-05,
50
+ "loss": 1.9216,
51
+ "step": 600
52
+ },
53
+ {
54
+ "epoch": 0.02,
55
+ "grad_norm": 7.98046875,
56
+ "learning_rate": 9.979899497487438e-05,
57
+ "loss": 1.9332,
58
+ "step": 700
59
+ },
60
+ {
61
+ "epoch": 0.02,
62
+ "grad_norm": 5.02734375,
63
+ "learning_rate": 9.969849246231156e-05,
64
+ "loss": 1.9248,
65
+ "step": 800
66
+ },
67
+ {
68
+ "epoch": 0.02,
69
+ "grad_norm": 6.359375,
70
+ "learning_rate": 9.959798994974875e-05,
71
+ "loss": 1.9724,
72
+ "step": 900
73
+ },
74
+ {
75
+ "epoch": 0.03,
76
+ "grad_norm": 8.6328125,
77
+ "learning_rate": 9.949748743718594e-05,
78
+ "loss": 1.9189,
79
+ "step": 1000
80
+ },
81
+ {
82
+ "epoch": 0.03,
83
+ "grad_norm": 9.0390625,
84
+ "learning_rate": 9.939698492462311e-05,
85
+ "loss": 1.9435,
86
+ "step": 1100
87
+ },
88
+ {
89
+ "epoch": 0.03,
90
+ "grad_norm": 7.85546875,
91
+ "learning_rate": 9.929648241206031e-05,
92
+ "loss": 1.9003,
93
+ "step": 1200
94
+ },
95
+ {
96
+ "epoch": 0.03,
97
+ "grad_norm": 11.4296875,
98
+ "learning_rate": 9.91959798994975e-05,
99
+ "loss": 1.8982,
100
+ "step": 1300
101
+ },
102
+ {
103
+ "epoch": 0.04,
104
+ "grad_norm": 4.10546875,
105
+ "learning_rate": 9.909547738693468e-05,
106
+ "loss": 1.8895,
107
+ "step": 1400
108
+ },
109
+ {
110
+ "epoch": 0.04,
111
+ "grad_norm": 8.8671875,
112
+ "learning_rate": 9.899497487437186e-05,
113
+ "loss": 1.9317,
114
+ "step": 1500
115
+ },
116
+ {
117
+ "epoch": 0.04,
118
+ "grad_norm": 8.546875,
119
+ "learning_rate": 9.889447236180906e-05,
120
+ "loss": 1.8948,
121
+ "step": 1600
122
+ },
123
+ {
124
+ "epoch": 0.04,
125
+ "grad_norm": 7.5625,
126
+ "learning_rate": 9.879396984924624e-05,
127
+ "loss": 1.8883,
128
+ "step": 1700
129
+ },
130
+ {
131
+ "epoch": 0.04,
132
+ "grad_norm": 3.732421875,
133
+ "learning_rate": 9.869346733668342e-05,
134
+ "loss": 1.8749,
135
+ "step": 1800
136
+ },
137
+ {
138
+ "epoch": 0.05,
139
+ "grad_norm": 9.140625,
140
+ "learning_rate": 9.85929648241206e-05,
141
+ "loss": 1.8626,
142
+ "step": 1900
143
+ },
144
+ {
145
+ "epoch": 0.05,
146
+ "grad_norm": 8.7265625,
147
+ "learning_rate": 9.84924623115578e-05,
148
+ "loss": 1.8761,
149
+ "step": 2000
150
+ },
151
+ {
152
+ "epoch": 0.05,
153
+ "grad_norm": 5.890625,
154
+ "learning_rate": 9.839195979899497e-05,
155
+ "loss": 1.8842,
156
+ "step": 2100
157
+ },
158
+ {
159
+ "epoch": 0.06,
160
+ "grad_norm": 7.0234375,
161
+ "learning_rate": 9.829145728643216e-05,
162
+ "loss": 1.9056,
163
+ "step": 2200
164
+ },
165
+ {
166
+ "epoch": 0.06,
167
+ "grad_norm": 8.0078125,
168
+ "learning_rate": 9.819095477386935e-05,
169
+ "loss": 1.9,
170
+ "step": 2300
171
+ },
172
+ {
173
+ "epoch": 0.06,
174
+ "grad_norm": 8.734375,
175
+ "learning_rate": 9.809045226130655e-05,
176
+ "loss": 1.9253,
177
+ "step": 2400
178
+ },
179
+ {
180
+ "epoch": 0.06,
181
+ "grad_norm": 9.1640625,
182
+ "learning_rate": 9.798994974874372e-05,
183
+ "loss": 1.9373,
184
+ "step": 2500
185
+ },
186
+ {
187
+ "epoch": 0.06,
188
+ "eval_loss": 1.9465688467025757,
189
+ "eval_runtime": 58.7017,
190
+ "eval_samples_per_second": 17.035,
191
+ "eval_steps_per_second": 4.259,
192
+ "step": 2500
193
+ },
194
+ {
195
+ "epoch": 0.07,
196
+ "grad_norm": 8.359375,
197
+ "learning_rate": 9.788944723618091e-05,
198
+ "loss": 1.9574,
199
+ "step": 2600
200
+ },
201
+ {
202
+ "epoch": 0.07,
203
+ "grad_norm": 6.40234375,
204
+ "learning_rate": 9.77889447236181e-05,
205
+ "loss": 1.8468,
206
+ "step": 2700
207
+ },
208
+ {
209
+ "epoch": 0.07,
210
+ "grad_norm": 7.328125,
211
+ "learning_rate": 9.768844221105528e-05,
212
+ "loss": 1.8394,
213
+ "step": 2800
214
+ },
215
+ {
216
+ "epoch": 0.07,
217
+ "grad_norm": 9.2578125,
218
+ "learning_rate": 9.758793969849247e-05,
219
+ "loss": 1.8358,
220
+ "step": 2900
221
+ },
222
+ {
223
+ "epoch": 0.07,
224
+ "grad_norm": 8.53125,
225
+ "learning_rate": 9.748743718592965e-05,
226
+ "loss": 1.8669,
227
+ "step": 3000
228
+ },
229
+ {
230
+ "epoch": 0.08,
231
+ "grad_norm": 5.25390625,
232
+ "learning_rate": 9.738693467336684e-05,
233
+ "loss": 1.8772,
234
+ "step": 3100
235
+ },
236
+ {
237
+ "epoch": 0.08,
238
+ "grad_norm": 7.84375,
239
+ "learning_rate": 9.728643216080403e-05,
240
+ "loss": 1.8632,
241
+ "step": 3200
242
+ },
243
+ {
244
+ "epoch": 0.08,
245
+ "grad_norm": 10.2578125,
246
+ "learning_rate": 9.718592964824121e-05,
247
+ "loss": 1.8813,
248
+ "step": 3300
249
+ },
250
+ {
251
+ "epoch": 0.09,
252
+ "grad_norm": 5.78515625,
253
+ "learning_rate": 9.70854271356784e-05,
254
+ "loss": 1.8944,
255
+ "step": 3400
256
+ },
257
+ {
258
+ "epoch": 0.09,
259
+ "grad_norm": 5.18359375,
260
+ "learning_rate": 9.698492462311559e-05,
261
+ "loss": 1.8789,
262
+ "step": 3500
263
+ },
264
+ {
265
+ "epoch": 0.09,
266
+ "grad_norm": 8.2578125,
267
+ "learning_rate": 9.688442211055276e-05,
268
+ "loss": 1.8867,
269
+ "step": 3600
270
+ },
271
+ {
272
+ "epoch": 0.09,
273
+ "grad_norm": 12.1171875,
274
+ "learning_rate": 9.678391959798996e-05,
275
+ "loss": 1.8426,
276
+ "step": 3700
277
+ },
278
+ {
279
+ "epoch": 0.1,
280
+ "grad_norm": 8.875,
281
+ "learning_rate": 9.668341708542715e-05,
282
+ "loss": 1.8427,
283
+ "step": 3800
284
+ },
285
+ {
286
+ "epoch": 0.1,
287
+ "grad_norm": 7.8203125,
288
+ "learning_rate": 9.658291457286432e-05,
289
+ "loss": 1.8416,
290
+ "step": 3900
291
+ },
292
+ {
293
+ "epoch": 0.1,
294
+ "grad_norm": 5.94140625,
295
+ "learning_rate": 9.64824120603015e-05,
296
+ "loss": 1.8496,
297
+ "step": 4000
298
+ },
299
+ {
300
+ "epoch": 0.1,
301
+ "grad_norm": 8.1328125,
302
+ "learning_rate": 9.63819095477387e-05,
303
+ "loss": 1.865,
304
+ "step": 4100
305
+ },
306
+ {
307
+ "epoch": 0.1,
308
+ "grad_norm": 8.578125,
309
+ "learning_rate": 9.628140703517589e-05,
310
+ "loss": 1.8689,
311
+ "step": 4200
312
+ },
313
+ {
314
+ "epoch": 0.11,
315
+ "grad_norm": 6.1171875,
316
+ "learning_rate": 9.618090452261306e-05,
317
+ "loss": 1.8711,
318
+ "step": 4300
319
+ },
320
+ {
321
+ "epoch": 0.11,
322
+ "grad_norm": 7.08203125,
323
+ "learning_rate": 9.608040201005025e-05,
324
+ "loss": 1.8723,
325
+ "step": 4400
326
+ },
327
+ {
328
+ "epoch": 0.11,
329
+ "grad_norm": 4.77734375,
330
+ "learning_rate": 9.597989949748745e-05,
331
+ "loss": 1.8686,
332
+ "step": 4500
333
+ },
334
+ {
335
+ "epoch": 0.12,
336
+ "grad_norm": 6.6875,
337
+ "learning_rate": 9.587939698492462e-05,
338
+ "loss": 1.8718,
339
+ "step": 4600
340
+ },
341
+ {
342
+ "epoch": 0.12,
343
+ "grad_norm": 9.5,
344
+ "learning_rate": 9.577889447236181e-05,
345
+ "loss": 1.8171,
346
+ "step": 4700
347
+ },
348
+ {
349
+ "epoch": 0.12,
350
+ "grad_norm": 11.125,
351
+ "learning_rate": 9.5678391959799e-05,
352
+ "loss": 1.9202,
353
+ "step": 4800
354
+ },
355
+ {
356
+ "epoch": 0.12,
357
+ "grad_norm": 8.09375,
358
+ "learning_rate": 9.55778894472362e-05,
359
+ "loss": 1.8685,
360
+ "step": 4900
361
+ },
362
+ {
363
+ "epoch": 0.12,
364
+ "grad_norm": 7.26171875,
365
+ "learning_rate": 9.547738693467337e-05,
366
+ "loss": 1.8742,
367
+ "step": 5000
368
+ },
369
+ {
370
+ "epoch": 0.12,
371
+ "eval_loss": 1.8054884672164917,
372
+ "eval_runtime": 58.5502,
373
+ "eval_samples_per_second": 17.079,
374
+ "eval_steps_per_second": 4.27,
375
+ "step": 5000
376
+ },
377
+ {
378
+ "epoch": 0.13,
379
+ "grad_norm": 7.7734375,
380
+ "learning_rate": 9.537688442211056e-05,
381
+ "loss": 1.8839,
382
+ "step": 5100
383
+ },
384
+ {
385
+ "epoch": 0.13,
386
+ "grad_norm": 9.90625,
387
+ "learning_rate": 9.527638190954774e-05,
388
+ "loss": 1.8702,
389
+ "step": 5200
390
+ },
391
+ {
392
+ "epoch": 0.13,
393
+ "grad_norm": 6.0078125,
394
+ "learning_rate": 9.517587939698493e-05,
395
+ "loss": 1.8273,
396
+ "step": 5300
397
+ },
398
+ {
399
+ "epoch": 0.14,
400
+ "grad_norm": 10.109375,
401
+ "learning_rate": 9.507537688442212e-05,
402
+ "loss": 1.8637,
403
+ "step": 5400
404
+ },
405
+ {
406
+ "epoch": 0.14,
407
+ "grad_norm": 6.84765625,
408
+ "learning_rate": 9.49748743718593e-05,
409
+ "loss": 1.8654,
410
+ "step": 5500
411
+ },
412
+ {
413
+ "epoch": 0.14,
414
+ "grad_norm": 6.609375,
415
+ "learning_rate": 9.487437185929649e-05,
416
+ "loss": 1.8233,
417
+ "step": 5600
418
+ },
419
+ {
420
+ "epoch": 0.14,
421
+ "grad_norm": 6.83984375,
422
+ "learning_rate": 9.477386934673366e-05,
423
+ "loss": 1.8636,
424
+ "step": 5700
425
+ },
426
+ {
427
+ "epoch": 0.14,
428
+ "grad_norm": 7.73046875,
429
+ "learning_rate": 9.467336683417086e-05,
430
+ "loss": 1.8542,
431
+ "step": 5800
432
+ },
433
+ {
434
+ "epoch": 0.15,
435
+ "grad_norm": 9.46875,
436
+ "learning_rate": 9.457286432160805e-05,
437
+ "loss": 1.8727,
438
+ "step": 5900
439
+ },
440
+ {
441
+ "epoch": 0.15,
442
+ "grad_norm": 8.828125,
443
+ "learning_rate": 9.447236180904523e-05,
444
+ "loss": 1.8413,
445
+ "step": 6000
446
+ },
447
+ {
448
+ "epoch": 0.15,
449
+ "grad_norm": 9.125,
450
+ "learning_rate": 9.437185929648241e-05,
451
+ "loss": 1.8322,
452
+ "step": 6100
453
+ },
454
+ {
455
+ "epoch": 0.15,
456
+ "grad_norm": 4.12890625,
457
+ "learning_rate": 9.427135678391961e-05,
458
+ "loss": 1.862,
459
+ "step": 6200
460
+ },
461
+ {
462
+ "epoch": 0.16,
463
+ "grad_norm": 8.6640625,
464
+ "learning_rate": 9.41708542713568e-05,
465
+ "loss": 1.8537,
466
+ "step": 6300
467
+ },
468
+ {
469
+ "epoch": 0.16,
470
+ "grad_norm": 5.04296875,
471
+ "learning_rate": 9.407035175879397e-05,
472
+ "loss": 1.847,
473
+ "step": 6400
474
+ },
475
+ {
476
+ "epoch": 0.16,
477
+ "grad_norm": 7.30078125,
478
+ "learning_rate": 9.396984924623115e-05,
479
+ "loss": 1.8379,
480
+ "step": 6500
481
+ },
482
+ {
483
+ "epoch": 0.17,
484
+ "grad_norm": 6.3046875,
485
+ "learning_rate": 9.386934673366835e-05,
486
+ "loss": 1.8492,
487
+ "step": 6600
488
+ },
489
+ {
490
+ "epoch": 0.17,
491
+ "grad_norm": 9.671875,
492
+ "learning_rate": 9.376884422110554e-05,
493
+ "loss": 1.845,
494
+ "step": 6700
495
+ },
496
+ {
497
+ "epoch": 0.17,
498
+ "grad_norm": 5.0546875,
499
+ "learning_rate": 9.366834170854271e-05,
500
+ "loss": 1.8557,
501
+ "step": 6800
502
+ },
503
+ {
504
+ "epoch": 0.17,
505
+ "grad_norm": 5.66015625,
506
+ "learning_rate": 9.35678391959799e-05,
507
+ "loss": 1.8256,
508
+ "step": 6900
509
+ },
510
+ {
511
+ "epoch": 0.17,
512
+ "grad_norm": 7.19921875,
513
+ "learning_rate": 9.34673366834171e-05,
514
+ "loss": 1.8081,
515
+ "step": 7000
516
+ },
517
+ {
518
+ "epoch": 0.18,
519
+ "grad_norm": 7.00390625,
520
+ "learning_rate": 9.336683417085427e-05,
521
+ "loss": 1.8326,
522
+ "step": 7100
523
+ },
524
+ {
525
+ "epoch": 0.18,
526
+ "grad_norm": 7.5390625,
527
+ "learning_rate": 9.326633165829146e-05,
528
+ "loss": 1.8197,
529
+ "step": 7200
530
+ },
531
+ {
532
+ "epoch": 0.18,
533
+ "grad_norm": 6.75,
534
+ "learning_rate": 9.316582914572864e-05,
535
+ "loss": 1.81,
536
+ "step": 7300
537
+ },
538
+ {
539
+ "epoch": 0.18,
540
+ "grad_norm": 9.28125,
541
+ "learning_rate": 9.306532663316585e-05,
542
+ "loss": 1.8062,
543
+ "step": 7400
544
+ },
545
+ {
546
+ "epoch": 0.19,
547
+ "grad_norm": 4.9765625,
548
+ "learning_rate": 9.296482412060302e-05,
549
+ "loss": 1.8546,
550
+ "step": 7500
551
+ },
552
+ {
553
+ "epoch": 0.19,
554
+ "eval_loss": 1.7942283153533936,
555
+ "eval_runtime": 58.5803,
556
+ "eval_samples_per_second": 17.071,
557
+ "eval_steps_per_second": 4.268,
558
+ "step": 7500
559
+ },
560
+ {
561
+ "epoch": 0.19,
562
+ "grad_norm": 6.27734375,
563
+ "learning_rate": 9.28643216080402e-05,
564
+ "loss": 1.8519,
565
+ "step": 7600
566
+ },
567
+ {
568
+ "epoch": 0.19,
569
+ "grad_norm": 9.65625,
570
+ "learning_rate": 9.276381909547739e-05,
571
+ "loss": 1.8272,
572
+ "step": 7700
573
+ },
574
+ {
575
+ "epoch": 0.2,
576
+ "grad_norm": 6.54296875,
577
+ "learning_rate": 9.266331658291458e-05,
578
+ "loss": 1.8251,
579
+ "step": 7800
580
+ },
581
+ {
582
+ "epoch": 0.2,
583
+ "grad_norm": 5.80859375,
584
+ "learning_rate": 9.256281407035176e-05,
585
+ "loss": 1.7912,
586
+ "step": 7900
587
+ },
588
+ {
589
+ "epoch": 0.2,
590
+ "grad_norm": 12.078125,
591
+ "learning_rate": 9.246231155778895e-05,
592
+ "loss": 1.8036,
593
+ "step": 8000
594
+ },
595
+ {
596
+ "epoch": 0.2,
597
+ "grad_norm": 8.5546875,
598
+ "learning_rate": 9.236180904522614e-05,
599
+ "loss": 1.8208,
600
+ "step": 8100
601
+ },
602
+ {
603
+ "epoch": 0.2,
604
+ "grad_norm": 8.59375,
605
+ "learning_rate": 9.226130653266331e-05,
606
+ "loss": 1.8396,
607
+ "step": 8200
608
+ },
609
+ {
610
+ "epoch": 0.21,
611
+ "grad_norm": 9.0078125,
612
+ "learning_rate": 9.216080402010051e-05,
613
+ "loss": 1.8225,
614
+ "step": 8300
615
+ },
616
+ {
617
+ "epoch": 0.21,
618
+ "grad_norm": 4.7421875,
619
+ "learning_rate": 9.20603015075377e-05,
620
+ "loss": 1.7912,
621
+ "step": 8400
622
+ },
623
+ {
624
+ "epoch": 0.21,
625
+ "grad_norm": 9.8828125,
626
+ "learning_rate": 9.195979899497488e-05,
627
+ "loss": 1.8061,
628
+ "step": 8500
629
+ },
630
+ {
631
+ "epoch": 0.21,
632
+ "grad_norm": 4.69140625,
633
+ "learning_rate": 9.185929648241206e-05,
634
+ "loss": 1.7964,
635
+ "step": 8600
636
+ },
637
+ {
638
+ "epoch": 0.22,
639
+ "grad_norm": 7.5625,
640
+ "learning_rate": 9.175879396984926e-05,
641
+ "loss": 1.8634,
642
+ "step": 8700
643
+ },
644
+ {
645
+ "epoch": 0.22,
646
+ "grad_norm": 5.3046875,
647
+ "learning_rate": 9.165829145728644e-05,
648
+ "loss": 1.8434,
649
+ "step": 8800
650
+ },
651
+ {
652
+ "epoch": 0.22,
653
+ "grad_norm": 6.30859375,
654
+ "learning_rate": 9.155778894472362e-05,
655
+ "loss": 1.8358,
656
+ "step": 8900
657
+ },
658
+ {
659
+ "epoch": 0.23,
660
+ "grad_norm": 6.71484375,
661
+ "learning_rate": 9.14572864321608e-05,
662
+ "loss": 1.8229,
663
+ "step": 9000
664
+ },
665
+ {
666
+ "epoch": 0.23,
667
+ "grad_norm": 9.875,
668
+ "learning_rate": 9.1356783919598e-05,
669
+ "loss": 1.8039,
670
+ "step": 9100
671
+ },
672
+ {
673
+ "epoch": 0.23,
674
+ "grad_norm": 6.8828125,
675
+ "learning_rate": 9.125628140703519e-05,
676
+ "loss": 1.8105,
677
+ "step": 9200
678
+ },
679
+ {
680
+ "epoch": 0.23,
681
+ "grad_norm": 5.0859375,
682
+ "learning_rate": 9.115577889447236e-05,
683
+ "loss": 1.8134,
684
+ "step": 9300
685
+ },
686
+ {
687
+ "epoch": 0.23,
688
+ "grad_norm": 4.5078125,
689
+ "learning_rate": 9.105527638190955e-05,
690
+ "loss": 1.8096,
691
+ "step": 9400
692
+ },
693
+ {
694
+ "epoch": 0.24,
695
+ "grad_norm": 6.5,
696
+ "learning_rate": 9.095477386934675e-05,
697
+ "loss": 1.7956,
698
+ "step": 9500
699
+ },
700
+ {
701
+ "epoch": 0.24,
702
+ "grad_norm": 4.234375,
703
+ "learning_rate": 9.085427135678392e-05,
704
+ "loss": 1.8053,
705
+ "step": 9600
706
+ },
707
+ {
708
+ "epoch": 0.24,
709
+ "grad_norm": 9.6171875,
710
+ "learning_rate": 9.075376884422111e-05,
711
+ "loss": 1.8205,
712
+ "step": 9700
713
+ },
714
+ {
715
+ "epoch": 0.24,
716
+ "grad_norm": 5.046875,
717
+ "learning_rate": 9.06532663316583e-05,
718
+ "loss": 1.8414,
719
+ "step": 9800
720
+ },
721
+ {
722
+ "epoch": 0.25,
723
+ "grad_norm": 6.66796875,
724
+ "learning_rate": 9.055276381909548e-05,
725
+ "loss": 1.8026,
726
+ "step": 9900
727
+ },
728
+ {
729
+ "epoch": 0.25,
730
+ "grad_norm": 6.7421875,
731
+ "learning_rate": 9.045226130653267e-05,
732
+ "loss": 1.8133,
733
+ "step": 10000
734
+ },
735
+ {
736
+ "epoch": 0.25,
737
+ "eval_loss": 1.7820364236831665,
738
+ "eval_runtime": 58.5615,
739
+ "eval_samples_per_second": 17.076,
740
+ "eval_steps_per_second": 4.269,
741
+ "step": 10000
742
+ },
743
+ {
744
+ "epoch": 0.25,
745
+ "grad_norm": 6.05078125,
746
+ "learning_rate": 9.035175879396985e-05,
747
+ "loss": 1.8395,
748
+ "step": 10100
749
+ },
750
+ {
751
+ "epoch": 0.26,
752
+ "grad_norm": 6.93359375,
753
+ "learning_rate": 9.025125628140704e-05,
754
+ "loss": 1.7843,
755
+ "step": 10200
756
+ },
757
+ {
758
+ "epoch": 0.26,
759
+ "grad_norm": 9.09375,
760
+ "learning_rate": 9.015075376884423e-05,
761
+ "loss": 1.8097,
762
+ "step": 10300
763
+ },
764
+ {
765
+ "epoch": 0.26,
766
+ "grad_norm": 5.5078125,
767
+ "learning_rate": 9.005025125628141e-05,
768
+ "loss": 1.824,
769
+ "step": 10400
770
+ },
771
+ {
772
+ "epoch": 0.26,
773
+ "grad_norm": 5.0703125,
774
+ "learning_rate": 8.99497487437186e-05,
775
+ "loss": 1.7863,
776
+ "step": 10500
777
+ },
778
+ {
779
+ "epoch": 0.27,
780
+ "grad_norm": 5.8203125,
781
+ "learning_rate": 8.984924623115579e-05,
782
+ "loss": 1.863,
783
+ "step": 10600
784
+ },
785
+ {
786
+ "epoch": 0.27,
787
+ "grad_norm": 5.4375,
788
+ "learning_rate": 8.974874371859296e-05,
789
+ "loss": 1.8308,
790
+ "step": 10700
791
+ },
792
+ {
793
+ "epoch": 0.27,
794
+ "grad_norm": 5.1171875,
795
+ "learning_rate": 8.964824120603016e-05,
796
+ "loss": 1.7928,
797
+ "step": 10800
798
+ },
799
+ {
800
+ "epoch": 0.27,
801
+ "grad_norm": 4.72265625,
802
+ "learning_rate": 8.954773869346734e-05,
803
+ "loss": 1.836,
804
+ "step": 10900
805
+ },
806
+ {
807
+ "epoch": 0.28,
808
+ "grad_norm": 5.35546875,
809
+ "learning_rate": 8.944723618090453e-05,
810
+ "loss": 1.7759,
811
+ "step": 11000
812
+ },
813
+ {
814
+ "epoch": 0.28,
815
+ "grad_norm": 6.04296875,
816
+ "learning_rate": 8.93467336683417e-05,
817
+ "loss": 1.8561,
818
+ "step": 11100
819
+ },
820
+ {
821
+ "epoch": 0.28,
822
+ "grad_norm": 6.49609375,
823
+ "learning_rate": 8.92462311557789e-05,
824
+ "loss": 1.7855,
825
+ "step": 11200
826
+ },
827
+ {
828
+ "epoch": 0.28,
829
+ "grad_norm": 4.94140625,
830
+ "learning_rate": 8.914572864321609e-05,
831
+ "loss": 1.7924,
832
+ "step": 11300
833
+ },
834
+ {
835
+ "epoch": 0.28,
836
+ "grad_norm": 7.515625,
837
+ "learning_rate": 8.904522613065326e-05,
838
+ "loss": 1.7844,
839
+ "step": 11400
840
+ },
841
+ {
842
+ "epoch": 0.29,
843
+ "grad_norm": 10.3515625,
844
+ "learning_rate": 8.894472361809045e-05,
845
+ "loss": 1.7828,
846
+ "step": 11500
847
+ },
848
+ {
849
+ "epoch": 0.29,
850
+ "grad_norm": 5.5078125,
851
+ "learning_rate": 8.884422110552765e-05,
852
+ "loss": 1.8021,
853
+ "step": 11600
854
+ },
855
+ {
856
+ "epoch": 0.29,
857
+ "grad_norm": 5.91015625,
858
+ "learning_rate": 8.874371859296484e-05,
859
+ "loss": 1.7883,
860
+ "step": 11700
861
+ },
862
+ {
863
+ "epoch": 0.29,
864
+ "grad_norm": 6.11328125,
865
+ "learning_rate": 8.864321608040201e-05,
866
+ "loss": 1.8127,
867
+ "step": 11800
868
+ },
869
+ {
870
+ "epoch": 0.3,
871
+ "grad_norm": 7.7734375,
872
+ "learning_rate": 8.85427135678392e-05,
873
+ "loss": 1.7899,
874
+ "step": 11900
875
+ },
876
+ {
877
+ "epoch": 0.3,
878
+ "grad_norm": 5.88671875,
879
+ "learning_rate": 8.84422110552764e-05,
880
+ "loss": 1.8298,
881
+ "step": 12000
882
+ },
883
+ {
884
+ "epoch": 0.3,
885
+ "grad_norm": 7.35546875,
886
+ "learning_rate": 8.834170854271357e-05,
887
+ "loss": 1.8107,
888
+ "step": 12100
889
+ },
890
+ {
891
+ "epoch": 0.3,
892
+ "grad_norm": 8.984375,
893
+ "learning_rate": 8.824120603015076e-05,
894
+ "loss": 1.7587,
895
+ "step": 12200
896
+ },
897
+ {
898
+ "epoch": 0.31,
899
+ "grad_norm": 7.0078125,
900
+ "learning_rate": 8.814070351758794e-05,
901
+ "loss": 1.8383,
902
+ "step": 12300
903
+ },
904
+ {
905
+ "epoch": 0.31,
906
+ "grad_norm": 4.87890625,
907
+ "learning_rate": 8.804020100502513e-05,
908
+ "loss": 1.8341,
909
+ "step": 12400
910
+ },
911
+ {
912
+ "epoch": 0.31,
913
+ "grad_norm": 7.375,
914
+ "learning_rate": 8.793969849246232e-05,
915
+ "loss": 1.7689,
916
+ "step": 12500
917
+ },
918
+ {
919
+ "epoch": 0.31,
920
+ "eval_loss": 1.7876064777374268,
921
+ "eval_runtime": 58.5853,
922
+ "eval_samples_per_second": 17.069,
923
+ "eval_steps_per_second": 4.267,
924
+ "step": 12500
925
+ },
926
+ {
927
+ "epoch": 0.32,
928
+ "grad_norm": 6.21484375,
929
+ "learning_rate": 8.78391959798995e-05,
930
+ "loss": 1.7756,
931
+ "step": 12600
932
+ },
933
+ {
934
+ "epoch": 0.32,
935
+ "grad_norm": 4.9765625,
936
+ "learning_rate": 8.773869346733669e-05,
937
+ "loss": 1.8252,
938
+ "step": 12700
939
+ },
940
+ {
941
+ "epoch": 0.32,
942
+ "grad_norm": 8.1484375,
943
+ "learning_rate": 8.763819095477387e-05,
944
+ "loss": 1.8265,
945
+ "step": 12800
946
+ },
947
+ {
948
+ "epoch": 0.32,
949
+ "grad_norm": 6.59375,
950
+ "learning_rate": 8.753768844221106e-05,
951
+ "loss": 1.7877,
952
+ "step": 12900
953
+ },
954
+ {
955
+ "epoch": 0.33,
956
+ "grad_norm": 5.40234375,
957
+ "learning_rate": 8.743718592964825e-05,
958
+ "loss": 1.7729,
959
+ "step": 13000
960
+ },
961
+ {
962
+ "epoch": 0.33,
963
+ "grad_norm": 5.43359375,
964
+ "learning_rate": 8.733668341708543e-05,
965
+ "loss": 1.8078,
966
+ "step": 13100
967
+ },
968
+ {
969
+ "epoch": 0.33,
970
+ "grad_norm": 5.33984375,
971
+ "learning_rate": 8.723618090452261e-05,
972
+ "loss": 1.8294,
973
+ "step": 13200
974
+ },
975
+ {
976
+ "epoch": 0.33,
977
+ "grad_norm": 7.26171875,
978
+ "learning_rate": 8.713567839195981e-05,
979
+ "loss": 1.7883,
980
+ "step": 13300
981
+ },
982
+ {
983
+ "epoch": 0.34,
984
+ "grad_norm": 10.5859375,
985
+ "learning_rate": 8.7035175879397e-05,
986
+ "loss": 1.8201,
987
+ "step": 13400
988
+ },
989
+ {
990
+ "epoch": 0.34,
991
+ "grad_norm": 6.0546875,
992
+ "learning_rate": 8.693467336683418e-05,
993
+ "loss": 1.747,
994
+ "step": 13500
995
+ },
996
+ {
997
+ "epoch": 0.34,
998
+ "grad_norm": 4.40234375,
999
+ "learning_rate": 8.683417085427135e-05,
1000
+ "loss": 1.8044,
1001
+ "step": 13600
1002
+ },
1003
+ {
1004
+ "epoch": 0.34,
1005
+ "grad_norm": 5.55078125,
1006
+ "learning_rate": 8.673366834170855e-05,
1007
+ "loss": 1.7687,
1008
+ "step": 13700
1009
+ },
1010
+ {
1011
+ "epoch": 0.34,
1012
+ "grad_norm": 4.03125,
1013
+ "learning_rate": 8.663316582914574e-05,
1014
+ "loss": 1.7941,
1015
+ "step": 13800
1016
+ },
1017
+ {
1018
+ "epoch": 0.35,
1019
+ "grad_norm": 5.046875,
1020
+ "learning_rate": 8.653266331658291e-05,
1021
+ "loss": 1.7781,
1022
+ "step": 13900
1023
+ },
1024
+ {
1025
+ "epoch": 0.35,
1026
+ "grad_norm": 8.3828125,
1027
+ "learning_rate": 8.64321608040201e-05,
1028
+ "loss": 1.7927,
1029
+ "step": 14000
1030
+ },
1031
+ {
1032
+ "epoch": 0.35,
1033
+ "grad_norm": 5.33984375,
1034
+ "learning_rate": 8.63316582914573e-05,
1035
+ "loss": 1.7793,
1036
+ "step": 14100
1037
+ },
1038
+ {
1039
+ "epoch": 0.35,
1040
+ "grad_norm": 6.73828125,
1041
+ "learning_rate": 8.623115577889449e-05,
1042
+ "loss": 1.7785,
1043
+ "step": 14200
1044
+ },
1045
+ {
1046
+ "epoch": 0.36,
1047
+ "grad_norm": 4.859375,
1048
+ "learning_rate": 8.613065326633166e-05,
1049
+ "loss": 1.7857,
1050
+ "step": 14300
1051
+ },
1052
+ {
1053
+ "epoch": 0.36,
1054
+ "grad_norm": 5.09765625,
1055
+ "learning_rate": 8.603015075376884e-05,
1056
+ "loss": 1.836,
1057
+ "step": 14400
1058
+ },
1059
+ {
1060
+ "epoch": 0.36,
1061
+ "grad_norm": 8.71875,
1062
+ "learning_rate": 8.592964824120603e-05,
1063
+ "loss": 1.8173,
1064
+ "step": 14500
1065
+ },
1066
+ {
1067
+ "epoch": 0.36,
1068
+ "grad_norm": 4.703125,
1069
+ "learning_rate": 8.582914572864322e-05,
1070
+ "loss": 1.8225,
1071
+ "step": 14600
1072
+ },
1073
+ {
1074
+ "epoch": 0.37,
1075
+ "grad_norm": 5.14453125,
1076
+ "learning_rate": 8.57286432160804e-05,
1077
+ "loss": 1.7832,
1078
+ "step": 14700
1079
+ },
1080
+ {
1081
+ "epoch": 0.37,
1082
+ "grad_norm": 5.51953125,
1083
+ "learning_rate": 8.562814070351759e-05,
1084
+ "loss": 1.7766,
1085
+ "step": 14800
1086
+ },
1087
+ {
1088
+ "epoch": 0.37,
1089
+ "grad_norm": 5.1484375,
1090
+ "learning_rate": 8.552763819095478e-05,
1091
+ "loss": 1.7659,
1092
+ "step": 14900
1093
+ },
1094
+ {
1095
+ "epoch": 0.38,
1096
+ "grad_norm": 3.9375,
1097
+ "learning_rate": 8.542713567839196e-05,
1098
+ "loss": 1.8042,
1099
+ "step": 15000
1100
+ },
1101
+ {
1102
+ "epoch": 0.38,
1103
+ "eval_loss": 1.7957104444503784,
1104
+ "eval_runtime": 58.7068,
1105
+ "eval_samples_per_second": 17.034,
1106
+ "eval_steps_per_second": 4.258,
1107
+ "step": 15000
1108
+ },
1109
+ {
1110
+ "epoch": 0.38,
1111
+ "grad_norm": 5.37109375,
1112
+ "learning_rate": 8.532663316582915e-05,
1113
+ "loss": 1.7498,
1114
+ "step": 15100
1115
+ },
1116
+ {
1117
+ "epoch": 0.38,
1118
+ "grad_norm": 6.34765625,
1119
+ "learning_rate": 8.522613065326634e-05,
1120
+ "loss": 1.8001,
1121
+ "step": 15200
1122
+ },
1123
+ {
1124
+ "epoch": 0.38,
1125
+ "grad_norm": 7.6953125,
1126
+ "learning_rate": 8.512562814070352e-05,
1127
+ "loss": 1.8107,
1128
+ "step": 15300
1129
+ },
1130
+ {
1131
+ "epoch": 0.39,
1132
+ "grad_norm": 4.55859375,
1133
+ "learning_rate": 8.502512562814071e-05,
1134
+ "loss": 1.8115,
1135
+ "step": 15400
1136
+ },
1137
+ {
1138
+ "epoch": 0.39,
1139
+ "grad_norm": 5.4296875,
1140
+ "learning_rate": 8.49246231155779e-05,
1141
+ "loss": 1.7869,
1142
+ "step": 15500
1143
+ },
1144
+ {
1145
+ "epoch": 0.39,
1146
+ "grad_norm": 5.62890625,
1147
+ "learning_rate": 8.482412060301508e-05,
1148
+ "loss": 1.7731,
1149
+ "step": 15600
1150
+ },
1151
+ {
1152
+ "epoch": 0.39,
1153
+ "grad_norm": 8.6171875,
1154
+ "learning_rate": 8.472361809045226e-05,
1155
+ "loss": 1.8287,
1156
+ "step": 15700
1157
+ },
1158
+ {
1159
+ "epoch": 0.4,
1160
+ "grad_norm": 6.0703125,
1161
+ "learning_rate": 8.462311557788946e-05,
1162
+ "loss": 1.7866,
1163
+ "step": 15800
1164
+ },
1165
+ {
1166
+ "epoch": 0.4,
1167
+ "grad_norm": 5.0625,
1168
+ "learning_rate": 8.452261306532664e-05,
1169
+ "loss": 1.7941,
1170
+ "step": 15900
1171
+ },
1172
+ {
1173
+ "epoch": 0.4,
1174
+ "grad_norm": 6.69140625,
1175
+ "learning_rate": 8.442211055276383e-05,
1176
+ "loss": 1.807,
1177
+ "step": 16000
1178
+ },
1179
+ {
1180
+ "epoch": 0.4,
1181
+ "grad_norm": 5.3125,
1182
+ "learning_rate": 8.4321608040201e-05,
1183
+ "loss": 1.7723,
1184
+ "step": 16100
1185
+ },
1186
+ {
1187
+ "epoch": 0.41,
1188
+ "grad_norm": 4.01171875,
1189
+ "learning_rate": 8.42211055276382e-05,
1190
+ "loss": 1.7878,
1191
+ "step": 16200
1192
+ },
1193
+ {
1194
+ "epoch": 0.41,
1195
+ "grad_norm": 8.21875,
1196
+ "learning_rate": 8.412060301507539e-05,
1197
+ "loss": 1.7906,
1198
+ "step": 16300
1199
+ },
1200
+ {
1201
+ "epoch": 0.41,
1202
+ "grad_norm": 5.55859375,
1203
+ "learning_rate": 8.402010050251256e-05,
1204
+ "loss": 1.8018,
1205
+ "step": 16400
1206
+ },
1207
+ {
1208
+ "epoch": 0.41,
1209
+ "grad_norm": 7.1796875,
1210
+ "learning_rate": 8.391959798994975e-05,
1211
+ "loss": 1.8287,
1212
+ "step": 16500
1213
+ },
1214
+ {
1215
+ "epoch": 0.41,
1216
+ "grad_norm": 5.76953125,
1217
+ "learning_rate": 8.381909547738695e-05,
1218
+ "loss": 1.8516,
1219
+ "step": 16600
1220
+ },
1221
+ {
1222
+ "epoch": 0.42,
1223
+ "grad_norm": 6.8671875,
1224
+ "learning_rate": 8.371859296482413e-05,
1225
+ "loss": 1.7676,
1226
+ "step": 16700
1227
+ },
1228
+ {
1229
+ "epoch": 0.42,
1230
+ "grad_norm": 5.66015625,
1231
+ "learning_rate": 8.36180904522613e-05,
1232
+ "loss": 1.7963,
1233
+ "step": 16800
1234
+ },
1235
+ {
1236
+ "epoch": 0.42,
1237
+ "grad_norm": 6.27734375,
1238
+ "learning_rate": 8.351758793969849e-05,
1239
+ "loss": 1.8459,
1240
+ "step": 16900
1241
+ },
1242
+ {
1243
+ "epoch": 0.42,
1244
+ "grad_norm": 7.44140625,
1245
+ "learning_rate": 8.341708542713568e-05,
1246
+ "loss": 1.7887,
1247
+ "step": 17000
1248
+ },
1249
+ {
1250
+ "epoch": 0.43,
1251
+ "grad_norm": 7.2109375,
1252
+ "learning_rate": 8.331658291457287e-05,
1253
+ "loss": 1.7455,
1254
+ "step": 17100
1255
+ },
1256
+ {
1257
+ "epoch": 0.43,
1258
+ "grad_norm": 6.5625,
1259
+ "learning_rate": 8.321608040201005e-05,
1260
+ "loss": 1.7657,
1261
+ "step": 17200
1262
+ },
1263
+ {
1264
+ "epoch": 0.43,
1265
+ "grad_norm": 8.2578125,
1266
+ "learning_rate": 8.311557788944724e-05,
1267
+ "loss": 1.7667,
1268
+ "step": 17300
1269
+ },
1270
+ {
1271
+ "epoch": 0.43,
1272
+ "grad_norm": 5.5625,
1273
+ "learning_rate": 8.301507537688443e-05,
1274
+ "loss": 1.8137,
1275
+ "step": 17400
1276
+ },
1277
+ {
1278
+ "epoch": 0.44,
1279
+ "grad_norm": 4.91796875,
1280
+ "learning_rate": 8.291457286432161e-05,
1281
+ "loss": 1.7911,
1282
+ "step": 17500
1283
+ },
1284
+ {
1285
+ "epoch": 0.44,
1286
+ "eval_loss": 1.7952567338943481,
1287
+ "eval_runtime": 58.854,
1288
+ "eval_samples_per_second": 16.991,
1289
+ "eval_steps_per_second": 4.248,
1290
+ "step": 17500
1291
+ },
1292
+ {
1293
+ "epoch": 0.44,
1294
+ "grad_norm": 8.5859375,
1295
+ "learning_rate": 8.28140703517588e-05,
1296
+ "loss": 1.7892,
1297
+ "step": 17600
1298
+ },
1299
+ {
1300
+ "epoch": 0.44,
1301
+ "grad_norm": 4.16796875,
1302
+ "learning_rate": 8.271356783919599e-05,
1303
+ "loss": 1.7493,
1304
+ "step": 17700
1305
+ },
1306
+ {
1307
+ "epoch": 0.45,
1308
+ "grad_norm": 6.68359375,
1309
+ "learning_rate": 8.261306532663317e-05,
1310
+ "loss": 1.785,
1311
+ "step": 17800
1312
+ },
1313
+ {
1314
+ "epoch": 0.45,
1315
+ "grad_norm": 4.93359375,
1316
+ "learning_rate": 8.251256281407036e-05,
1317
+ "loss": 1.807,
1318
+ "step": 17900
1319
+ },
1320
+ {
1321
+ "epoch": 0.45,
1322
+ "grad_norm": 11.2109375,
1323
+ "learning_rate": 8.241206030150754e-05,
1324
+ "loss": 1.8133,
1325
+ "step": 18000
1326
+ },
1327
+ {
1328
+ "epoch": 0.45,
1329
+ "grad_norm": 6.25,
1330
+ "learning_rate": 8.231155778894473e-05,
1331
+ "loss": 1.8227,
1332
+ "step": 18100
1333
+ },
1334
+ {
1335
+ "epoch": 0.46,
1336
+ "grad_norm": 8.46875,
1337
+ "learning_rate": 8.22110552763819e-05,
1338
+ "loss": 1.795,
1339
+ "step": 18200
1340
+ },
1341
+ {
1342
+ "epoch": 0.46,
1343
+ "grad_norm": 3.587890625,
1344
+ "learning_rate": 8.21105527638191e-05,
1345
+ "loss": 1.7832,
1346
+ "step": 18300
1347
+ },
1348
+ {
1349
+ "epoch": 0.46,
1350
+ "grad_norm": 5.296875,
1351
+ "learning_rate": 8.201005025125629e-05,
1352
+ "loss": 1.7968,
1353
+ "step": 18400
1354
+ },
1355
+ {
1356
+ "epoch": 0.46,
1357
+ "grad_norm": 4.015625,
1358
+ "learning_rate": 8.190954773869348e-05,
1359
+ "loss": 1.8195,
1360
+ "step": 18500
1361
+ },
1362
+ {
1363
+ "epoch": 0.47,
1364
+ "grad_norm": 4.75,
1365
+ "learning_rate": 8.180904522613065e-05,
1366
+ "loss": 1.782,
1367
+ "step": 18600
1368
+ },
1369
+ {
1370
+ "epoch": 0.47,
1371
+ "grad_norm": 7.046875,
1372
+ "learning_rate": 8.170854271356785e-05,
1373
+ "loss": 1.7568,
1374
+ "step": 18700
1375
+ },
1376
+ {
1377
+ "epoch": 0.47,
1378
+ "grad_norm": 6.26953125,
1379
+ "learning_rate": 8.160804020100504e-05,
1380
+ "loss": 1.8259,
1381
+ "step": 18800
1382
+ },
1383
+ {
1384
+ "epoch": 0.47,
1385
+ "grad_norm": 6.890625,
1386
+ "learning_rate": 8.150753768844221e-05,
1387
+ "loss": 1.7702,
1388
+ "step": 18900
1389
+ },
1390
+ {
1391
+ "epoch": 0.47,
1392
+ "grad_norm": 6.28515625,
1393
+ "learning_rate": 8.14070351758794e-05,
1394
+ "loss": 1.7551,
1395
+ "step": 19000
1396
+ },
1397
+ {
1398
+ "epoch": 0.48,
1399
+ "grad_norm": 5.140625,
1400
+ "learning_rate": 8.130653266331658e-05,
1401
+ "loss": 1.7674,
1402
+ "step": 19100
1403
+ },
1404
+ {
1405
+ "epoch": 0.48,
1406
+ "grad_norm": 5.94921875,
1407
+ "learning_rate": 8.120603015075378e-05,
1408
+ "loss": 1.7653,
1409
+ "step": 19200
1410
+ },
1411
+ {
1412
+ "epoch": 0.48,
1413
+ "grad_norm": 6.109375,
1414
+ "learning_rate": 8.110552763819096e-05,
1415
+ "loss": 1.7855,
1416
+ "step": 19300
1417
+ },
1418
+ {
1419
+ "epoch": 0.48,
1420
+ "grad_norm": 4.9921875,
1421
+ "learning_rate": 8.100502512562814e-05,
1422
+ "loss": 1.7272,
1423
+ "step": 19400
1424
+ },
1425
+ {
1426
+ "epoch": 0.49,
1427
+ "grad_norm": 7.34765625,
1428
+ "learning_rate": 8.090452261306533e-05,
1429
+ "loss": 1.8004,
1430
+ "step": 19500
1431
+ },
1432
+ {
1433
+ "epoch": 0.49,
1434
+ "grad_norm": 7.56640625,
1435
+ "learning_rate": 8.080402010050251e-05,
1436
+ "loss": 1.7634,
1437
+ "step": 19600
1438
+ },
1439
+ {
1440
+ "epoch": 0.49,
1441
+ "grad_norm": 8.84375,
1442
+ "learning_rate": 8.07035175879397e-05,
1443
+ "loss": 1.7221,
1444
+ "step": 19700
1445
+ },
1446
+ {
1447
+ "epoch": 0.49,
1448
+ "grad_norm": 6.00390625,
1449
+ "learning_rate": 8.060301507537689e-05,
1450
+ "loss": 1.8102,
1451
+ "step": 19800
1452
+ },
1453
+ {
1454
+ "epoch": 0.5,
1455
+ "grad_norm": 4.65625,
1456
+ "learning_rate": 8.050251256281407e-05,
1457
+ "loss": 1.7403,
1458
+ "step": 19900
1459
+ },
1460
+ {
1461
+ "epoch": 0.5,
1462
+ "grad_norm": 7.57421875,
1463
+ "learning_rate": 8.040201005025126e-05,
1464
+ "loss": 1.7996,
1465
+ "step": 20000
1466
+ },
1467
+ {
1468
+ "epoch": 0.5,
1469
+ "eval_loss": 1.774701714515686,
1470
+ "eval_runtime": 58.6217,
1471
+ "eval_samples_per_second": 17.059,
1472
+ "eval_steps_per_second": 4.265,
1473
+ "step": 20000
1474
+ },
1475
+ {
1476
+ "epoch": 0.5,
1477
+ "grad_norm": 5.9296875,
1478
+ "learning_rate": 8.030150753768845e-05,
1479
+ "loss": 1.8128,
1480
+ "step": 20100
1481
+ },
1482
+ {
1483
+ "epoch": 0.51,
1484
+ "grad_norm": 3.328125,
1485
+ "learning_rate": 8.020100502512563e-05,
1486
+ "loss": 1.822,
1487
+ "step": 20200
1488
+ },
1489
+ {
1490
+ "epoch": 0.51,
1491
+ "grad_norm": 9.859375,
1492
+ "learning_rate": 8.010050251256282e-05,
1493
+ "loss": 1.8077,
1494
+ "step": 20300
1495
+ },
1496
+ {
1497
+ "epoch": 0.51,
1498
+ "grad_norm": 6.625,
1499
+ "learning_rate": 8e-05,
1500
+ "loss": 1.759,
1501
+ "step": 20400
1502
+ },
1503
+ {
1504
+ "epoch": 0.51,
1505
+ "grad_norm": 5.71875,
1506
+ "learning_rate": 7.989949748743719e-05,
1507
+ "loss": 1.8109,
1508
+ "step": 20500
1509
+ },
1510
+ {
1511
+ "epoch": 0.52,
1512
+ "grad_norm": 5.9921875,
1513
+ "learning_rate": 7.979899497487438e-05,
1514
+ "loss": 1.7575,
1515
+ "step": 20600
1516
+ },
1517
+ {
1518
+ "epoch": 0.52,
1519
+ "grad_norm": 5.34765625,
1520
+ "learning_rate": 7.969849246231155e-05,
1521
+ "loss": 1.7939,
1522
+ "step": 20700
1523
+ },
1524
+ {
1525
+ "epoch": 0.52,
1526
+ "grad_norm": 4.98046875,
1527
+ "learning_rate": 7.959798994974875e-05,
1528
+ "loss": 1.7786,
1529
+ "step": 20800
1530
+ },
1531
+ {
1532
+ "epoch": 0.52,
1533
+ "grad_norm": 3.318359375,
1534
+ "learning_rate": 7.949748743718594e-05,
1535
+ "loss": 1.783,
1536
+ "step": 20900
1537
+ },
1538
+ {
1539
+ "epoch": 0.53,
1540
+ "grad_norm": 4.359375,
1541
+ "learning_rate": 7.939698492462313e-05,
1542
+ "loss": 1.7968,
1543
+ "step": 21000
1544
+ },
1545
+ {
1546
+ "epoch": 0.53,
1547
+ "grad_norm": 5.01171875,
1548
+ "learning_rate": 7.92964824120603e-05,
1549
+ "loss": 1.7869,
1550
+ "step": 21100
1551
+ },
1552
+ {
1553
+ "epoch": 0.53,
1554
+ "grad_norm": 7.66796875,
1555
+ "learning_rate": 7.91959798994975e-05,
1556
+ "loss": 1.8054,
1557
+ "step": 21200
1558
+ },
1559
+ {
1560
+ "epoch": 0.53,
1561
+ "grad_norm": 6.296875,
1562
+ "learning_rate": 7.909547738693468e-05,
1563
+ "loss": 1.7957,
1564
+ "step": 21300
1565
+ },
1566
+ {
1567
+ "epoch": 0.54,
1568
+ "grad_norm": 6.23828125,
1569
+ "learning_rate": 7.899497487437186e-05,
1570
+ "loss": 1.8123,
1571
+ "step": 21400
1572
+ },
1573
+ {
1574
+ "epoch": 0.54,
1575
+ "grad_norm": 4.2265625,
1576
+ "learning_rate": 7.889447236180904e-05,
1577
+ "loss": 1.79,
1578
+ "step": 21500
1579
+ },
1580
+ {
1581
+ "epoch": 0.54,
1582
+ "grad_norm": 4.39453125,
1583
+ "learning_rate": 7.879396984924623e-05,
1584
+ "loss": 1.7505,
1585
+ "step": 21600
1586
+ },
1587
+ {
1588
+ "epoch": 0.54,
1589
+ "grad_norm": 9.0546875,
1590
+ "learning_rate": 7.869346733668343e-05,
1591
+ "loss": 1.7862,
1592
+ "step": 21700
1593
+ },
1594
+ {
1595
+ "epoch": 0.55,
1596
+ "grad_norm": 9.75,
1597
+ "learning_rate": 7.85929648241206e-05,
1598
+ "loss": 1.8352,
1599
+ "step": 21800
1600
+ },
1601
+ {
1602
+ "epoch": 0.55,
1603
+ "grad_norm": 4.9765625,
1604
+ "learning_rate": 7.849246231155779e-05,
1605
+ "loss": 1.7609,
1606
+ "step": 21900
1607
+ },
1608
+ {
1609
+ "epoch": 0.55,
1610
+ "grad_norm": 10.21875,
1611
+ "learning_rate": 7.839195979899498e-05,
1612
+ "loss": 1.8483,
1613
+ "step": 22000
1614
+ },
1615
+ {
1616
+ "epoch": 0.55,
1617
+ "grad_norm": 5.8515625,
1618
+ "learning_rate": 7.829145728643216e-05,
1619
+ "loss": 1.7301,
1620
+ "step": 22100
1621
+ },
1622
+ {
1623
+ "epoch": 0.56,
1624
+ "grad_norm": 5.76953125,
1625
+ "learning_rate": 7.819095477386935e-05,
1626
+ "loss": 1.7806,
1627
+ "step": 22200
1628
+ },
1629
+ {
1630
+ "epoch": 0.56,
1631
+ "grad_norm": 7.39453125,
1632
+ "learning_rate": 7.809045226130654e-05,
1633
+ "loss": 1.7654,
1634
+ "step": 22300
1635
+ },
1636
+ {
1637
+ "epoch": 0.56,
1638
+ "grad_norm": 6.19921875,
1639
+ "learning_rate": 7.798994974874372e-05,
1640
+ "loss": 1.7647,
1641
+ "step": 22400
1642
+ },
1643
+ {
1644
+ "epoch": 0.56,
1645
+ "grad_norm": 6.296875,
1646
+ "learning_rate": 7.788944723618091e-05,
1647
+ "loss": 1.7572,
1648
+ "step": 22500
1649
+ },
1650
+ {
1651
+ "epoch": 0.56,
1652
+ "eval_loss": 1.7940764427185059,
1653
+ "eval_runtime": 58.6091,
1654
+ "eval_samples_per_second": 17.062,
1655
+ "eval_steps_per_second": 4.266,
1656
+ "step": 22500
1657
+ },
1658
+ {
1659
+ "epoch": 0.56,
1660
+ "grad_norm": 9.2109375,
1661
+ "learning_rate": 7.77889447236181e-05,
1662
+ "loss": 1.8022,
1663
+ "step": 22600
1664
+ },
1665
+ {
1666
+ "epoch": 0.57,
1667
+ "grad_norm": 8.140625,
1668
+ "learning_rate": 7.768844221105528e-05,
1669
+ "loss": 1.7874,
1670
+ "step": 22700
1671
+ },
1672
+ {
1673
+ "epoch": 0.57,
1674
+ "grad_norm": 4.5078125,
1675
+ "learning_rate": 7.758793969849247e-05,
1676
+ "loss": 1.7889,
1677
+ "step": 22800
1678
+ },
1679
+ {
1680
+ "epoch": 0.57,
1681
+ "grad_norm": 6.44921875,
1682
+ "learning_rate": 7.748743718592966e-05,
1683
+ "loss": 1.7439,
1684
+ "step": 22900
1685
+ },
1686
+ {
1687
+ "epoch": 0.57,
1688
+ "grad_norm": 7.6328125,
1689
+ "learning_rate": 7.738693467336684e-05,
1690
+ "loss": 1.7804,
1691
+ "step": 23000
1692
+ },
1693
+ {
1694
+ "epoch": 0.58,
1695
+ "grad_norm": 7.578125,
1696
+ "learning_rate": 7.728643216080403e-05,
1697
+ "loss": 1.7768,
1698
+ "step": 23100
1699
+ },
1700
+ {
1701
+ "epoch": 0.58,
1702
+ "grad_norm": 8.234375,
1703
+ "learning_rate": 7.71859296482412e-05,
1704
+ "loss": 1.7694,
1705
+ "step": 23200
1706
+ },
1707
+ {
1708
+ "epoch": 0.58,
1709
+ "grad_norm": 4.140625,
1710
+ "learning_rate": 7.70854271356784e-05,
1711
+ "loss": 1.7481,
1712
+ "step": 23300
1713
+ },
1714
+ {
1715
+ "epoch": 0.58,
1716
+ "grad_norm": 4.75390625,
1717
+ "learning_rate": 7.698492462311559e-05,
1718
+ "loss": 1.7657,
1719
+ "step": 23400
1720
+ },
1721
+ {
1722
+ "epoch": 0.59,
1723
+ "grad_norm": 7.4921875,
1724
+ "learning_rate": 7.688442211055277e-05,
1725
+ "loss": 1.7717,
1726
+ "step": 23500
1727
+ },
1728
+ {
1729
+ "epoch": 0.59,
1730
+ "grad_norm": 6.15234375,
1731
+ "learning_rate": 7.678391959798995e-05,
1732
+ "loss": 1.7387,
1733
+ "step": 23600
1734
+ },
1735
+ {
1736
+ "epoch": 0.59,
1737
+ "grad_norm": 4.33203125,
1738
+ "learning_rate": 7.668341708542713e-05,
1739
+ "loss": 1.7737,
1740
+ "step": 23700
1741
+ },
1742
+ {
1743
+ "epoch": 0.59,
1744
+ "grad_norm": 4.20703125,
1745
+ "learning_rate": 7.658291457286433e-05,
1746
+ "loss": 1.7786,
1747
+ "step": 23800
1748
+ },
1749
+ {
1750
+ "epoch": 0.6,
1751
+ "grad_norm": 3.861328125,
1752
+ "learning_rate": 7.64824120603015e-05,
1753
+ "loss": 1.7953,
1754
+ "step": 23900
1755
+ },
1756
+ {
1757
+ "epoch": 0.6,
1758
+ "grad_norm": 4.7421875,
1759
+ "learning_rate": 7.638190954773869e-05,
1760
+ "loss": 1.7508,
1761
+ "step": 24000
1762
+ },
1763
+ {
1764
+ "epoch": 0.6,
1765
+ "grad_norm": 7.1484375,
1766
+ "learning_rate": 7.628140703517588e-05,
1767
+ "loss": 1.7496,
1768
+ "step": 24100
1769
+ },
1770
+ {
1771
+ "epoch": 0.6,
1772
+ "grad_norm": 4.08984375,
1773
+ "learning_rate": 7.618090452261307e-05,
1774
+ "loss": 1.7959,
1775
+ "step": 24200
1776
+ },
1777
+ {
1778
+ "epoch": 0.61,
1779
+ "grad_norm": 6.78125,
1780
+ "learning_rate": 7.608040201005025e-05,
1781
+ "loss": 1.7997,
1782
+ "step": 24300
1783
+ },
1784
+ {
1785
+ "epoch": 0.61,
1786
+ "grad_norm": 7.9140625,
1787
+ "learning_rate": 7.597989949748744e-05,
1788
+ "loss": 1.7578,
1789
+ "step": 24400
1790
+ },
1791
+ {
1792
+ "epoch": 0.61,
1793
+ "grad_norm": 4.0546875,
1794
+ "learning_rate": 7.587939698492463e-05,
1795
+ "loss": 1.7787,
1796
+ "step": 24500
1797
+ },
1798
+ {
1799
+ "epoch": 0.61,
1800
+ "grad_norm": 7.00390625,
1801
+ "learning_rate": 7.577889447236181e-05,
1802
+ "loss": 1.7563,
1803
+ "step": 24600
1804
+ },
1805
+ {
1806
+ "epoch": 0.62,
1807
+ "grad_norm": 5.078125,
1808
+ "learning_rate": 7.5678391959799e-05,
1809
+ "loss": 1.7971,
1810
+ "step": 24700
1811
+ },
1812
+ {
1813
+ "epoch": 0.62,
1814
+ "grad_norm": 7.171875,
1815
+ "learning_rate": 7.557788944723618e-05,
1816
+ "loss": 1.7662,
1817
+ "step": 24800
1818
+ },
1819
+ {
1820
+ "epoch": 0.62,
1821
+ "grad_norm": 5.6015625,
1822
+ "learning_rate": 7.547738693467337e-05,
1823
+ "loss": 1.7887,
1824
+ "step": 24900
1825
+ },
1826
+ {
1827
+ "epoch": 0.62,
1828
+ "grad_norm": 5.7109375,
1829
+ "learning_rate": 7.537688442211056e-05,
1830
+ "loss": 1.8288,
1831
+ "step": 25000
1832
+ },
1833
+ {
1834
+ "epoch": 0.62,
1835
+ "eval_loss": 1.7818756103515625,
1836
+ "eval_runtime": 58.7925,
1837
+ "eval_samples_per_second": 17.009,
1838
+ "eval_steps_per_second": 4.252,
1839
+ "step": 25000
1840
+ },
1841
+ {
1842
+ "epoch": 0.63,
1843
+ "grad_norm": 4.39453125,
1844
+ "learning_rate": 7.527638190954774e-05,
1845
+ "loss": 1.7837,
1846
+ "step": 25100
1847
+ },
1848
+ {
1849
+ "epoch": 0.63,
1850
+ "grad_norm": 4.3671875,
1851
+ "learning_rate": 7.517587939698493e-05,
1852
+ "loss": 1.767,
1853
+ "step": 25200
1854
+ },
1855
+ {
1856
+ "epoch": 0.63,
1857
+ "grad_norm": 5.31640625,
1858
+ "learning_rate": 7.507537688442212e-05,
1859
+ "loss": 1.7601,
1860
+ "step": 25300
1861
+ },
1862
+ {
1863
+ "epoch": 0.64,
1864
+ "grad_norm": 4.54296875,
1865
+ "learning_rate": 7.49748743718593e-05,
1866
+ "loss": 1.7574,
1867
+ "step": 25400
1868
+ },
1869
+ {
1870
+ "epoch": 0.64,
1871
+ "grad_norm": 4.890625,
1872
+ "learning_rate": 7.487437185929649e-05,
1873
+ "loss": 1.7523,
1874
+ "step": 25500
1875
+ },
1876
+ {
1877
+ "epoch": 0.64,
1878
+ "grad_norm": 7.71484375,
1879
+ "learning_rate": 7.477386934673368e-05,
1880
+ "loss": 1.7567,
1881
+ "step": 25600
1882
+ },
1883
+ {
1884
+ "epoch": 0.64,
1885
+ "grad_norm": 5.18359375,
1886
+ "learning_rate": 7.467336683417085e-05,
1887
+ "loss": 1.7937,
1888
+ "step": 25700
1889
+ },
1890
+ {
1891
+ "epoch": 0.65,
1892
+ "grad_norm": 5.81640625,
1893
+ "learning_rate": 7.457286432160805e-05,
1894
+ "loss": 1.7273,
1895
+ "step": 25800
1896
+ },
1897
+ {
1898
+ "epoch": 0.65,
1899
+ "grad_norm": 3.94921875,
1900
+ "learning_rate": 7.447236180904524e-05,
1901
+ "loss": 1.8141,
1902
+ "step": 25900
1903
+ },
1904
+ {
1905
+ "epoch": 0.65,
1906
+ "grad_norm": 3.439453125,
1907
+ "learning_rate": 7.437185929648241e-05,
1908
+ "loss": 1.7697,
1909
+ "step": 26000
1910
+ },
1911
+ {
1912
+ "epoch": 0.65,
1913
+ "grad_norm": 4.55859375,
1914
+ "learning_rate": 7.42713567839196e-05,
1915
+ "loss": 1.7679,
1916
+ "step": 26100
1917
+ },
1918
+ {
1919
+ "epoch": 0.66,
1920
+ "grad_norm": 6.6640625,
1921
+ "learning_rate": 7.417085427135678e-05,
1922
+ "loss": 1.8057,
1923
+ "step": 26200
1924
+ },
1925
+ {
1926
+ "epoch": 0.66,
1927
+ "grad_norm": 3.77734375,
1928
+ "learning_rate": 7.407035175879398e-05,
1929
+ "loss": 1.7748,
1930
+ "step": 26300
1931
+ },
1932
+ {
1933
+ "epoch": 0.66,
1934
+ "grad_norm": 5.78515625,
1935
+ "learning_rate": 7.396984924623115e-05,
1936
+ "loss": 1.7825,
1937
+ "step": 26400
1938
+ },
1939
+ {
1940
+ "epoch": 0.66,
1941
+ "grad_norm": 4.49609375,
1942
+ "learning_rate": 7.386934673366834e-05,
1943
+ "loss": 1.7735,
1944
+ "step": 26500
1945
+ },
1946
+ {
1947
+ "epoch": 0.67,
1948
+ "grad_norm": 6.53515625,
1949
+ "learning_rate": 7.376884422110553e-05,
1950
+ "loss": 1.7663,
1951
+ "step": 26600
1952
+ },
1953
+ {
1954
+ "epoch": 0.67,
1955
+ "grad_norm": 3.6875,
1956
+ "learning_rate": 7.366834170854271e-05,
1957
+ "loss": 1.7844,
1958
+ "step": 26700
1959
+ },
1960
+ {
1961
+ "epoch": 0.67,
1962
+ "grad_norm": 4.16015625,
1963
+ "learning_rate": 7.35678391959799e-05,
1964
+ "loss": 1.797,
1965
+ "step": 26800
1966
+ },
1967
+ {
1968
+ "epoch": 0.67,
1969
+ "grad_norm": 7.1640625,
1970
+ "learning_rate": 7.346733668341709e-05,
1971
+ "loss": 1.8147,
1972
+ "step": 26900
1973
+ },
1974
+ {
1975
+ "epoch": 0.68,
1976
+ "grad_norm": 3.8671875,
1977
+ "learning_rate": 7.336683417085427e-05,
1978
+ "loss": 1.7814,
1979
+ "step": 27000
1980
+ },
1981
+ {
1982
+ "epoch": 0.68,
1983
+ "grad_norm": 7.89453125,
1984
+ "learning_rate": 7.326633165829146e-05,
1985
+ "loss": 1.7911,
1986
+ "step": 27100
1987
+ },
1988
+ {
1989
+ "epoch": 0.68,
1990
+ "grad_norm": 4.95703125,
1991
+ "learning_rate": 7.316582914572865e-05,
1992
+ "loss": 1.7882,
1993
+ "step": 27200
1994
+ },
1995
+ {
1996
+ "epoch": 0.68,
1997
+ "grad_norm": 5.84765625,
1998
+ "learning_rate": 7.306532663316583e-05,
1999
+ "loss": 1.7886,
2000
+ "step": 27300
2001
+ },
2002
+ {
2003
+ "epoch": 0.69,
2004
+ "grad_norm": 7.3046875,
2005
+ "learning_rate": 7.296482412060302e-05,
2006
+ "loss": 1.7752,
2007
+ "step": 27400
2008
+ },
2009
+ {
2010
+ "epoch": 0.69,
2011
+ "grad_norm": 5.328125,
2012
+ "learning_rate": 7.28643216080402e-05,
2013
+ "loss": 1.7561,
2014
+ "step": 27500
2015
+ },
2016
+ {
2017
+ "epoch": 0.69,
2018
+ "eval_loss": 1.7702152729034424,
2019
+ "eval_runtime": 58.6028,
2020
+ "eval_samples_per_second": 17.064,
2021
+ "eval_steps_per_second": 4.266,
2022
+ "step": 27500
2023
+ },
2024
+ {
2025
+ "epoch": 0.69,
2026
+ "grad_norm": 5.05078125,
2027
+ "learning_rate": 7.276381909547739e-05,
2028
+ "loss": 1.8037,
2029
+ "step": 27600
2030
+ },
2031
+ {
2032
+ "epoch": 0.69,
2033
+ "grad_norm": 5.90625,
2034
+ "learning_rate": 7.266331658291458e-05,
2035
+ "loss": 1.7701,
2036
+ "step": 27700
2037
+ },
2038
+ {
2039
+ "epoch": 0.69,
2040
+ "grad_norm": 6.7265625,
2041
+ "learning_rate": 7.256281407035177e-05,
2042
+ "loss": 1.7891,
2043
+ "step": 27800
2044
+ },
2045
+ {
2046
+ "epoch": 0.7,
2047
+ "grad_norm": 5.3515625,
2048
+ "learning_rate": 7.246231155778895e-05,
2049
+ "loss": 1.7706,
2050
+ "step": 27900
2051
+ },
2052
+ {
2053
+ "epoch": 0.7,
2054
+ "grad_norm": 5.18359375,
2055
+ "learning_rate": 7.236180904522614e-05,
2056
+ "loss": 1.7655,
2057
+ "step": 28000
2058
+ },
2059
+ {
2060
+ "epoch": 0.7,
2061
+ "grad_norm": 5.33984375,
2062
+ "learning_rate": 7.226130653266333e-05,
2063
+ "loss": 1.7846,
2064
+ "step": 28100
2065
+ },
2066
+ {
2067
+ "epoch": 0.7,
2068
+ "grad_norm": 8.046875,
2069
+ "learning_rate": 7.21608040201005e-05,
2070
+ "loss": 1.7997,
2071
+ "step": 28200
2072
+ },
2073
+ {
2074
+ "epoch": 0.71,
2075
+ "grad_norm": 4.5,
2076
+ "learning_rate": 7.206030150753768e-05,
2077
+ "loss": 1.7938,
2078
+ "step": 28300
2079
+ },
2080
+ {
2081
+ "epoch": 0.71,
2082
+ "grad_norm": 3.861328125,
2083
+ "learning_rate": 7.195979899497488e-05,
2084
+ "loss": 1.7676,
2085
+ "step": 28400
2086
+ },
2087
+ {
2088
+ "epoch": 0.71,
2089
+ "grad_norm": 4.6484375,
2090
+ "learning_rate": 7.185929648241206e-05,
2091
+ "loss": 1.7469,
2092
+ "step": 28500
2093
+ },
2094
+ {
2095
+ "epoch": 0.71,
2096
+ "grad_norm": 5.90625,
2097
+ "learning_rate": 7.175879396984924e-05,
2098
+ "loss": 1.7756,
2099
+ "step": 28600
2100
+ },
2101
+ {
2102
+ "epoch": 0.72,
2103
+ "grad_norm": 3.4765625,
2104
+ "learning_rate": 7.165829145728643e-05,
2105
+ "loss": 1.8028,
2106
+ "step": 28700
2107
+ },
2108
+ {
2109
+ "epoch": 0.72,
2110
+ "grad_norm": 4.734375,
2111
+ "learning_rate": 7.155778894472363e-05,
2112
+ "loss": 1.7702,
2113
+ "step": 28800
2114
+ },
2115
+ {
2116
+ "epoch": 0.72,
2117
+ "grad_norm": 5.33203125,
2118
+ "learning_rate": 7.14572864321608e-05,
2119
+ "loss": 1.7662,
2120
+ "step": 28900
2121
+ },
2122
+ {
2123
+ "epoch": 0.72,
2124
+ "grad_norm": 3.783203125,
2125
+ "learning_rate": 7.135678391959799e-05,
2126
+ "loss": 1.7767,
2127
+ "step": 29000
2128
+ },
2129
+ {
2130
+ "epoch": 0.73,
2131
+ "grad_norm": 5.734375,
2132
+ "learning_rate": 7.125628140703518e-05,
2133
+ "loss": 1.7829,
2134
+ "step": 29100
2135
+ },
2136
+ {
2137
+ "epoch": 0.73,
2138
+ "grad_norm": 4.79296875,
2139
+ "learning_rate": 7.115577889447236e-05,
2140
+ "loss": 1.7596,
2141
+ "step": 29200
2142
+ },
2143
+ {
2144
+ "epoch": 0.73,
2145
+ "grad_norm": 5.20703125,
2146
+ "learning_rate": 7.105527638190955e-05,
2147
+ "loss": 1.7136,
2148
+ "step": 29300
2149
+ },
2150
+ {
2151
+ "epoch": 0.73,
2152
+ "grad_norm": 7.37109375,
2153
+ "learning_rate": 7.095477386934674e-05,
2154
+ "loss": 1.7612,
2155
+ "step": 29400
2156
+ },
2157
+ {
2158
+ "epoch": 0.74,
2159
+ "grad_norm": 3.88671875,
2160
+ "learning_rate": 7.085427135678392e-05,
2161
+ "loss": 1.768,
2162
+ "step": 29500
2163
+ },
2164
+ {
2165
+ "epoch": 0.74,
2166
+ "grad_norm": 3.248046875,
2167
+ "learning_rate": 7.075376884422111e-05,
2168
+ "loss": 1.7706,
2169
+ "step": 29600
2170
+ },
2171
+ {
2172
+ "epoch": 0.74,
2173
+ "grad_norm": 4.6171875,
2174
+ "learning_rate": 7.06532663316583e-05,
2175
+ "loss": 1.7715,
2176
+ "step": 29700
2177
+ },
2178
+ {
2179
+ "epoch": 0.74,
2180
+ "grad_norm": 5.38671875,
2181
+ "learning_rate": 7.055276381909548e-05,
2182
+ "loss": 1.7381,
2183
+ "step": 29800
2184
+ },
2185
+ {
2186
+ "epoch": 0.75,
2187
+ "grad_norm": 6.9375,
2188
+ "learning_rate": 7.045226130653267e-05,
2189
+ "loss": 1.7993,
2190
+ "step": 29900
2191
+ },
2192
+ {
2193
+ "epoch": 0.75,
2194
+ "grad_norm": 6.6171875,
2195
+ "learning_rate": 7.035175879396985e-05,
2196
+ "loss": 1.7474,
2197
+ "step": 30000
2198
+ },
2199
+ {
2200
+ "epoch": 0.75,
2201
+ "eval_loss": 1.79494309425354,
2202
+ "eval_runtime": 58.6639,
2203
+ "eval_samples_per_second": 17.046,
2204
+ "eval_steps_per_second": 4.262,
2205
+ "step": 30000
2206
+ },
2207
+ {
2208
+ "epoch": 0.75,
2209
+ "grad_norm": 6.90234375,
2210
+ "learning_rate": 7.025125628140704e-05,
2211
+ "loss": 1.7628,
2212
+ "step": 30100
2213
+ },
2214
+ {
2215
+ "epoch": 0.76,
2216
+ "grad_norm": 4.24609375,
2217
+ "learning_rate": 7.015075376884423e-05,
2218
+ "loss": 1.7704,
2219
+ "step": 30200
2220
+ },
2221
+ {
2222
+ "epoch": 0.76,
2223
+ "grad_norm": 6.03125,
2224
+ "learning_rate": 7.00502512562814e-05,
2225
+ "loss": 1.7551,
2226
+ "step": 30300
2227
+ },
2228
+ {
2229
+ "epoch": 0.76,
2230
+ "grad_norm": 8.1328125,
2231
+ "learning_rate": 6.99497487437186e-05,
2232
+ "loss": 1.7783,
2233
+ "step": 30400
2234
+ },
2235
+ {
2236
+ "epoch": 0.76,
2237
+ "grad_norm": 4.6328125,
2238
+ "learning_rate": 6.984924623115579e-05,
2239
+ "loss": 1.7669,
2240
+ "step": 30500
2241
+ },
2242
+ {
2243
+ "epoch": 0.77,
2244
+ "grad_norm": 3.54296875,
2245
+ "learning_rate": 6.974874371859297e-05,
2246
+ "loss": 1.7497,
2247
+ "step": 30600
2248
+ },
2249
+ {
2250
+ "epoch": 0.77,
2251
+ "grad_norm": 3.130859375,
2252
+ "learning_rate": 6.964824120603015e-05,
2253
+ "loss": 1.7455,
2254
+ "step": 30700
2255
+ },
2256
+ {
2257
+ "epoch": 0.77,
2258
+ "grad_norm": 6.28515625,
2259
+ "learning_rate": 6.954773869346733e-05,
2260
+ "loss": 1.7537,
2261
+ "step": 30800
2262
+ },
2263
+ {
2264
+ "epoch": 0.77,
2265
+ "grad_norm": 6.5390625,
2266
+ "learning_rate": 6.944723618090453e-05,
2267
+ "loss": 1.7651,
2268
+ "step": 30900
2269
+ },
2270
+ {
2271
+ "epoch": 0.78,
2272
+ "grad_norm": 7.15234375,
2273
+ "learning_rate": 6.93467336683417e-05,
2274
+ "loss": 1.7822,
2275
+ "step": 31000
2276
+ },
2277
+ {
2278
+ "epoch": 0.78,
2279
+ "grad_norm": 5.40625,
2280
+ "learning_rate": 6.924623115577889e-05,
2281
+ "loss": 1.7987,
2282
+ "step": 31100
2283
+ },
2284
+ {
2285
+ "epoch": 0.78,
2286
+ "grad_norm": 5.0546875,
2287
+ "learning_rate": 6.914572864321608e-05,
2288
+ "loss": 1.7787,
2289
+ "step": 31200
2290
+ },
2291
+ {
2292
+ "epoch": 0.78,
2293
+ "grad_norm": 6.83203125,
2294
+ "learning_rate": 6.904522613065328e-05,
2295
+ "loss": 1.7629,
2296
+ "step": 31300
2297
+ },
2298
+ {
2299
+ "epoch": 0.79,
2300
+ "grad_norm": 8.703125,
2301
+ "learning_rate": 6.894472361809045e-05,
2302
+ "loss": 1.7253,
2303
+ "step": 31400
2304
+ },
2305
+ {
2306
+ "epoch": 0.79,
2307
+ "grad_norm": 6.3359375,
2308
+ "learning_rate": 6.884422110552764e-05,
2309
+ "loss": 1.7446,
2310
+ "step": 31500
2311
+ },
2312
+ {
2313
+ "epoch": 0.79,
2314
+ "grad_norm": 7.13671875,
2315
+ "learning_rate": 6.874371859296482e-05,
2316
+ "loss": 1.7621,
2317
+ "step": 31600
2318
+ },
2319
+ {
2320
+ "epoch": 0.79,
2321
+ "grad_norm": 6.875,
2322
+ "learning_rate": 6.864321608040201e-05,
2323
+ "loss": 1.7461,
2324
+ "step": 31700
2325
+ },
2326
+ {
2327
+ "epoch": 0.8,
2328
+ "grad_norm": 6.109375,
2329
+ "learning_rate": 6.85427135678392e-05,
2330
+ "loss": 1.795,
2331
+ "step": 31800
2332
+ },
2333
+ {
2334
+ "epoch": 0.8,
2335
+ "grad_norm": 4.515625,
2336
+ "learning_rate": 6.844221105527638e-05,
2337
+ "loss": 1.7707,
2338
+ "step": 31900
2339
+ },
2340
+ {
2341
+ "epoch": 0.8,
2342
+ "grad_norm": 6.69140625,
2343
+ "learning_rate": 6.834170854271357e-05,
2344
+ "loss": 1.7683,
2345
+ "step": 32000
2346
+ },
2347
+ {
2348
+ "epoch": 0.8,
2349
+ "grad_norm": 3.7734375,
2350
+ "learning_rate": 6.824120603015076e-05,
2351
+ "loss": 1.7407,
2352
+ "step": 32100
2353
+ },
2354
+ {
2355
+ "epoch": 0.81,
2356
+ "grad_norm": 7.03515625,
2357
+ "learning_rate": 6.814070351758794e-05,
2358
+ "loss": 1.7581,
2359
+ "step": 32200
2360
+ },
2361
+ {
2362
+ "epoch": 0.81,
2363
+ "grad_norm": 7.68359375,
2364
+ "learning_rate": 6.804020100502513e-05,
2365
+ "loss": 1.7809,
2366
+ "step": 32300
2367
+ },
2368
+ {
2369
+ "epoch": 0.81,
2370
+ "grad_norm": 6.17578125,
2371
+ "learning_rate": 6.793969849246232e-05,
2372
+ "loss": 1.7748,
2373
+ "step": 32400
2374
+ },
2375
+ {
2376
+ "epoch": 0.81,
2377
+ "grad_norm": 7.1484375,
2378
+ "learning_rate": 6.78391959798995e-05,
2379
+ "loss": 1.7814,
2380
+ "step": 32500
2381
+ },
2382
+ {
2383
+ "epoch": 0.81,
2384
+ "eval_loss": 1.7562289237976074,
2385
+ "eval_runtime": 58.6717,
2386
+ "eval_samples_per_second": 17.044,
2387
+ "eval_steps_per_second": 4.261,
2388
+ "step": 32500
2389
+ },
2390
+ {
2391
+ "epoch": 0.81,
2392
+ "grad_norm": 5.6796875,
2393
+ "learning_rate": 6.773869346733669e-05,
2394
+ "loss": 1.7753,
2395
+ "step": 32600
2396
+ },
2397
+ {
2398
+ "epoch": 0.82,
2399
+ "grad_norm": 6.26171875,
2400
+ "learning_rate": 6.763819095477388e-05,
2401
+ "loss": 1.7479,
2402
+ "step": 32700
2403
+ },
2404
+ {
2405
+ "epoch": 0.82,
2406
+ "grad_norm": 4.37109375,
2407
+ "learning_rate": 6.753768844221105e-05,
2408
+ "loss": 1.8101,
2409
+ "step": 32800
2410
+ },
2411
+ {
2412
+ "epoch": 0.82,
2413
+ "grad_norm": 4.2421875,
2414
+ "learning_rate": 6.743718592964824e-05,
2415
+ "loss": 1.7543,
2416
+ "step": 32900
2417
+ },
2418
+ {
2419
+ "epoch": 0.82,
2420
+ "grad_norm": 3.560546875,
2421
+ "learning_rate": 6.733668341708544e-05,
2422
+ "loss": 1.7124,
2423
+ "step": 33000
2424
+ },
2425
+ {
2426
+ "epoch": 0.83,
2427
+ "grad_norm": 6.15625,
2428
+ "learning_rate": 6.723618090452262e-05,
2429
+ "loss": 1.7396,
2430
+ "step": 33100
2431
+ },
2432
+ {
2433
+ "epoch": 0.83,
2434
+ "grad_norm": 6.2890625,
2435
+ "learning_rate": 6.71356783919598e-05,
2436
+ "loss": 1.7732,
2437
+ "step": 33200
2438
+ },
2439
+ {
2440
+ "epoch": 0.83,
2441
+ "grad_norm": 3.71875,
2442
+ "learning_rate": 6.703517587939698e-05,
2443
+ "loss": 1.7662,
2444
+ "step": 33300
2445
+ },
2446
+ {
2447
+ "epoch": 0.83,
2448
+ "grad_norm": 4.69140625,
2449
+ "learning_rate": 6.693467336683418e-05,
2450
+ "loss": 1.7547,
2451
+ "step": 33400
2452
+ },
2453
+ {
2454
+ "epoch": 0.84,
2455
+ "grad_norm": 6.50390625,
2456
+ "learning_rate": 6.683417085427135e-05,
2457
+ "loss": 1.7678,
2458
+ "step": 33500
2459
+ },
2460
+ {
2461
+ "epoch": 0.84,
2462
+ "grad_norm": 3.205078125,
2463
+ "learning_rate": 6.673366834170854e-05,
2464
+ "loss": 1.7886,
2465
+ "step": 33600
2466
+ },
2467
+ {
2468
+ "epoch": 0.84,
2469
+ "grad_norm": 4.8125,
2470
+ "learning_rate": 6.663316582914573e-05,
2471
+ "loss": 1.7663,
2472
+ "step": 33700
2473
+ },
2474
+ {
2475
+ "epoch": 0.84,
2476
+ "grad_norm": 6.640625,
2477
+ "learning_rate": 6.653266331658293e-05,
2478
+ "loss": 1.7383,
2479
+ "step": 33800
2480
+ },
2481
+ {
2482
+ "epoch": 0.85,
2483
+ "grad_norm": 4.8515625,
2484
+ "learning_rate": 6.64321608040201e-05,
2485
+ "loss": 1.7691,
2486
+ "step": 33900
2487
+ },
2488
+ {
2489
+ "epoch": 0.85,
2490
+ "grad_norm": 4.2578125,
2491
+ "learning_rate": 6.633165829145729e-05,
2492
+ "loss": 1.7573,
2493
+ "step": 34000
2494
+ },
2495
+ {
2496
+ "epoch": 0.85,
2497
+ "grad_norm": 2.880859375,
2498
+ "learning_rate": 6.623115577889447e-05,
2499
+ "loss": 1.7703,
2500
+ "step": 34100
2501
+ },
2502
+ {
2503
+ "epoch": 0.85,
2504
+ "grad_norm": 5.46875,
2505
+ "learning_rate": 6.613065326633166e-05,
2506
+ "loss": 1.781,
2507
+ "step": 34200
2508
+ },
2509
+ {
2510
+ "epoch": 0.86,
2511
+ "grad_norm": 4.01953125,
2512
+ "learning_rate": 6.603015075376885e-05,
2513
+ "loss": 1.728,
2514
+ "step": 34300
2515
+ },
2516
+ {
2517
+ "epoch": 0.86,
2518
+ "grad_norm": 6.93359375,
2519
+ "learning_rate": 6.592964824120603e-05,
2520
+ "loss": 1.7715,
2521
+ "step": 34400
2522
+ },
2523
+ {
2524
+ "epoch": 0.86,
2525
+ "grad_norm": 4.55078125,
2526
+ "learning_rate": 6.582914572864322e-05,
2527
+ "loss": 1.7755,
2528
+ "step": 34500
2529
+ },
2530
+ {
2531
+ "epoch": 0.86,
2532
+ "grad_norm": 3.46875,
2533
+ "learning_rate": 6.57286432160804e-05,
2534
+ "loss": 1.7887,
2535
+ "step": 34600
2536
+ },
2537
+ {
2538
+ "epoch": 0.87,
2539
+ "grad_norm": 7.51953125,
2540
+ "learning_rate": 6.562814070351759e-05,
2541
+ "loss": 1.7574,
2542
+ "step": 34700
2543
+ },
2544
+ {
2545
+ "epoch": 0.87,
2546
+ "grad_norm": 5.30859375,
2547
+ "learning_rate": 6.552763819095478e-05,
2548
+ "loss": 1.7129,
2549
+ "step": 34800
2550
+ },
2551
+ {
2552
+ "epoch": 0.87,
2553
+ "grad_norm": 5.5859375,
2554
+ "learning_rate": 6.542713567839197e-05,
2555
+ "loss": 1.7443,
2556
+ "step": 34900
2557
+ },
2558
+ {
2559
+ "epoch": 0.88,
2560
+ "grad_norm": 4.28125,
2561
+ "learning_rate": 6.532663316582915e-05,
2562
+ "loss": 1.7575,
2563
+ "step": 35000
2564
+ },
2565
+ {
2566
+ "epoch": 0.88,
2567
+ "eval_loss": 1.7717657089233398,
2568
+ "eval_runtime": 58.5137,
2569
+ "eval_samples_per_second": 17.09,
2570
+ "eval_steps_per_second": 4.273,
2571
+ "step": 35000
2572
+ },
2573
+ {
2574
+ "epoch": 0.88,
2575
+ "grad_norm": 4.7890625,
2576
+ "learning_rate": 6.522613065326634e-05,
2577
+ "loss": 1.7768,
2578
+ "step": 35100
2579
+ },
2580
+ {
2581
+ "epoch": 0.88,
2582
+ "grad_norm": 6.58984375,
2583
+ "learning_rate": 6.512562814070352e-05,
2584
+ "loss": 1.7784,
2585
+ "step": 35200
2586
+ },
2587
+ {
2588
+ "epoch": 0.88,
2589
+ "grad_norm": 4.98046875,
2590
+ "learning_rate": 6.50251256281407e-05,
2591
+ "loss": 1.7156,
2592
+ "step": 35300
2593
+ },
2594
+ {
2595
+ "epoch": 0.89,
2596
+ "grad_norm": 7.83203125,
2597
+ "learning_rate": 6.492462311557788e-05,
2598
+ "loss": 1.7136,
2599
+ "step": 35400
2600
+ },
2601
+ {
2602
+ "epoch": 0.89,
2603
+ "grad_norm": 4.390625,
2604
+ "learning_rate": 6.482412060301508e-05,
2605
+ "loss": 1.7739,
2606
+ "step": 35500
2607
+ },
2608
+ {
2609
+ "epoch": 0.89,
2610
+ "grad_norm": 4.9375,
2611
+ "learning_rate": 6.472361809045227e-05,
2612
+ "loss": 1.7445,
2613
+ "step": 35600
2614
+ },
2615
+ {
2616
+ "epoch": 0.89,
2617
+ "grad_norm": 5.24609375,
2618
+ "learning_rate": 6.462311557788944e-05,
2619
+ "loss": 1.7398,
2620
+ "step": 35700
2621
+ },
2622
+ {
2623
+ "epoch": 0.9,
2624
+ "grad_norm": 5.81640625,
2625
+ "learning_rate": 6.452261306532663e-05,
2626
+ "loss": 1.7367,
2627
+ "step": 35800
2628
+ },
2629
+ {
2630
+ "epoch": 0.9,
2631
+ "grad_norm": 9.03125,
2632
+ "learning_rate": 6.442211055276383e-05,
2633
+ "loss": 1.7496,
2634
+ "step": 35900
2635
+ },
2636
+ {
2637
+ "epoch": 0.9,
2638
+ "grad_norm": 10.4609375,
2639
+ "learning_rate": 6.4321608040201e-05,
2640
+ "loss": 1.7613,
2641
+ "step": 36000
2642
+ },
2643
+ {
2644
+ "epoch": 0.9,
2645
+ "grad_norm": 6.48046875,
2646
+ "learning_rate": 6.422110552763819e-05,
2647
+ "loss": 1.7279,
2648
+ "step": 36100
2649
+ },
2650
+ {
2651
+ "epoch": 0.91,
2652
+ "grad_norm": 5.78125,
2653
+ "learning_rate": 6.412060301507538e-05,
2654
+ "loss": 1.7665,
2655
+ "step": 36200
2656
+ },
2657
+ {
2658
+ "epoch": 0.91,
2659
+ "grad_norm": 5.59765625,
2660
+ "learning_rate": 6.402010050251258e-05,
2661
+ "loss": 1.7672,
2662
+ "step": 36300
2663
+ },
2664
+ {
2665
+ "epoch": 0.91,
2666
+ "grad_norm": 4.73828125,
2667
+ "learning_rate": 6.391959798994975e-05,
2668
+ "loss": 1.7223,
2669
+ "step": 36400
2670
+ },
2671
+ {
2672
+ "epoch": 0.91,
2673
+ "grad_norm": 4.0625,
2674
+ "learning_rate": 6.381909547738694e-05,
2675
+ "loss": 1.7493,
2676
+ "step": 36500
2677
+ },
2678
+ {
2679
+ "epoch": 0.92,
2680
+ "grad_norm": 6.0546875,
2681
+ "learning_rate": 6.371859296482412e-05,
2682
+ "loss": 1.7542,
2683
+ "step": 36600
2684
+ },
2685
+ {
2686
+ "epoch": 0.92,
2687
+ "grad_norm": 4.54296875,
2688
+ "learning_rate": 6.361809045226131e-05,
2689
+ "loss": 1.7434,
2690
+ "step": 36700
2691
+ },
2692
+ {
2693
+ "epoch": 0.92,
2694
+ "grad_norm": 3.837890625,
2695
+ "learning_rate": 6.35175879396985e-05,
2696
+ "loss": 1.7627,
2697
+ "step": 36800
2698
+ },
2699
+ {
2700
+ "epoch": 0.92,
2701
+ "grad_norm": 4.46484375,
2702
+ "learning_rate": 6.341708542713568e-05,
2703
+ "loss": 1.759,
2704
+ "step": 36900
2705
+ },
2706
+ {
2707
+ "epoch": 0.93,
2708
+ "grad_norm": 4.89453125,
2709
+ "learning_rate": 6.331658291457287e-05,
2710
+ "loss": 1.7065,
2711
+ "step": 37000
2712
+ },
2713
+ {
2714
+ "epoch": 0.93,
2715
+ "grad_norm": 4.9921875,
2716
+ "learning_rate": 6.321608040201005e-05,
2717
+ "loss": 1.7447,
2718
+ "step": 37100
2719
+ },
2720
+ {
2721
+ "epoch": 0.93,
2722
+ "grad_norm": 5.95703125,
2723
+ "learning_rate": 6.311557788944724e-05,
2724
+ "loss": 1.7706,
2725
+ "step": 37200
2726
+ },
2727
+ {
2728
+ "epoch": 0.93,
2729
+ "grad_norm": 4.89453125,
2730
+ "learning_rate": 6.301507537688443e-05,
2731
+ "loss": 1.8006,
2732
+ "step": 37300
2733
+ },
2734
+ {
2735
+ "epoch": 0.94,
2736
+ "grad_norm": 4.66015625,
2737
+ "learning_rate": 6.291457286432161e-05,
2738
+ "loss": 1.7311,
2739
+ "step": 37400
2740
+ },
2741
+ {
2742
+ "epoch": 0.94,
2743
+ "grad_norm": 3.96875,
2744
+ "learning_rate": 6.28140703517588e-05,
2745
+ "loss": 1.756,
2746
+ "step": 37500
2747
+ },
2748
+ {
2749
+ "epoch": 0.94,
2750
+ "eval_loss": 1.722901701927185,
2751
+ "eval_runtime": 58.6285,
2752
+ "eval_samples_per_second": 17.057,
2753
+ "eval_steps_per_second": 4.264,
2754
+ "step": 37500
2755
+ },
2756
+ {
2757
+ "epoch": 0.94,
2758
+ "grad_norm": 3.828125,
2759
+ "learning_rate": 6.271356783919599e-05,
2760
+ "loss": 1.7585,
2761
+ "step": 37600
2762
+ },
2763
+ {
2764
+ "epoch": 0.94,
2765
+ "grad_norm": 4.26953125,
2766
+ "learning_rate": 6.261306532663317e-05,
2767
+ "loss": 1.7415,
2768
+ "step": 37700
2769
+ },
2770
+ {
2771
+ "epoch": 0.94,
2772
+ "grad_norm": 4.6640625,
2773
+ "learning_rate": 6.251256281407035e-05,
2774
+ "loss": 1.8039,
2775
+ "step": 37800
2776
+ },
2777
+ {
2778
+ "epoch": 0.95,
2779
+ "grad_norm": 5.43359375,
2780
+ "learning_rate": 6.241206030150753e-05,
2781
+ "loss": 1.7832,
2782
+ "step": 37900
2783
+ },
2784
+ {
2785
+ "epoch": 0.95,
2786
+ "grad_norm": 4.63671875,
2787
+ "learning_rate": 6.231155778894473e-05,
2788
+ "loss": 1.7409,
2789
+ "step": 38000
2790
+ },
2791
+ {
2792
+ "epoch": 0.95,
2793
+ "grad_norm": 4.0546875,
2794
+ "learning_rate": 6.221105527638192e-05,
2795
+ "loss": 1.7652,
2796
+ "step": 38100
2797
+ },
2798
+ {
2799
+ "epoch": 0.95,
2800
+ "grad_norm": 7.16015625,
2801
+ "learning_rate": 6.211055276381909e-05,
2802
+ "loss": 1.7609,
2803
+ "step": 38200
2804
+ },
2805
+ {
2806
+ "epoch": 0.96,
2807
+ "grad_norm": 6.53125,
2808
+ "learning_rate": 6.201005025125628e-05,
2809
+ "loss": 1.7524,
2810
+ "step": 38300
2811
+ },
2812
+ {
2813
+ "epoch": 0.96,
2814
+ "grad_norm": 5.3359375,
2815
+ "learning_rate": 6.190954773869348e-05,
2816
+ "loss": 1.7825,
2817
+ "step": 38400
2818
+ },
2819
+ {
2820
+ "epoch": 0.96,
2821
+ "grad_norm": 4.17578125,
2822
+ "learning_rate": 6.180904522613065e-05,
2823
+ "loss": 1.773,
2824
+ "step": 38500
2825
+ },
2826
+ {
2827
+ "epoch": 0.96,
2828
+ "grad_norm": 3.9375,
2829
+ "learning_rate": 6.170854271356784e-05,
2830
+ "loss": 1.7343,
2831
+ "step": 38600
2832
+ },
2833
+ {
2834
+ "epoch": 0.97,
2835
+ "grad_norm": 5.09765625,
2836
+ "learning_rate": 6.160804020100502e-05,
2837
+ "loss": 1.7402,
2838
+ "step": 38700
2839
+ },
2840
+ {
2841
+ "epoch": 0.97,
2842
+ "grad_norm": 5.07421875,
2843
+ "learning_rate": 6.150753768844222e-05,
2844
+ "loss": 1.7091,
2845
+ "step": 38800
2846
+ },
2847
+ {
2848
+ "epoch": 0.97,
2849
+ "grad_norm": 4.8671875,
2850
+ "learning_rate": 6.14070351758794e-05,
2851
+ "loss": 1.7008,
2852
+ "step": 38900
2853
+ },
2854
+ {
2855
+ "epoch": 0.97,
2856
+ "grad_norm": 4.67578125,
2857
+ "learning_rate": 6.130653266331658e-05,
2858
+ "loss": 1.7596,
2859
+ "step": 39000
2860
+ },
2861
+ {
2862
+ "epoch": 0.98,
2863
+ "grad_norm": 5.1015625,
2864
+ "learning_rate": 6.120603015075377e-05,
2865
+ "loss": 1.7664,
2866
+ "step": 39100
2867
+ },
2868
+ {
2869
+ "epoch": 0.98,
2870
+ "grad_norm": 5.96875,
2871
+ "learning_rate": 6.110552763819096e-05,
2872
+ "loss": 1.7649,
2873
+ "step": 39200
2874
+ },
2875
+ {
2876
+ "epoch": 0.98,
2877
+ "grad_norm": 5.12890625,
2878
+ "learning_rate": 6.1005025125628143e-05,
2879
+ "loss": 1.763,
2880
+ "step": 39300
2881
+ },
2882
+ {
2883
+ "epoch": 0.98,
2884
+ "grad_norm": 4.3046875,
2885
+ "learning_rate": 6.090452261306533e-05,
2886
+ "loss": 1.734,
2887
+ "step": 39400
2888
+ },
2889
+ {
2890
+ "epoch": 0.99,
2891
+ "grad_norm": 3.650390625,
2892
+ "learning_rate": 6.080402010050251e-05,
2893
+ "loss": 1.6931,
2894
+ "step": 39500
2895
+ },
2896
+ {
2897
+ "epoch": 0.99,
2898
+ "grad_norm": 5.3125,
2899
+ "learning_rate": 6.070351758793971e-05,
2900
+ "loss": 1.7677,
2901
+ "step": 39600
2902
+ },
2903
+ {
2904
+ "epoch": 0.99,
2905
+ "grad_norm": 4.58984375,
2906
+ "learning_rate": 6.060301507537689e-05,
2907
+ "loss": 1.7173,
2908
+ "step": 39700
2909
+ },
2910
+ {
2911
+ "epoch": 0.99,
2912
+ "grad_norm": 1.9599609375,
2913
+ "learning_rate": 6.0502512562814076e-05,
2914
+ "loss": 1.7413,
2915
+ "step": 39800
2916
+ },
2917
+ {
2918
+ "epoch": 1.0,
2919
+ "grad_norm": 6.78125,
2920
+ "learning_rate": 6.0402010050251256e-05,
2921
+ "loss": 1.7572,
2922
+ "step": 39900
2923
+ },
2924
+ {
2925
+ "epoch": 1.0,
2926
+ "grad_norm": 7.78125,
2927
+ "learning_rate": 6.030150753768844e-05,
2928
+ "loss": 1.7379,
2929
+ "step": 40000
2930
+ },
2931
+ {
2932
+ "epoch": 1.0,
2933
+ "eval_loss": 1.736067295074463,
2934
+ "eval_runtime": 58.5709,
2935
+ "eval_samples_per_second": 17.073,
2936
+ "eval_steps_per_second": 4.268,
2937
+ "step": 40000
2938
+ },
2939
+ {
2940
+ "epoch": 1.0,
2941
+ "grad_norm": 5.5546875,
2942
+ "learning_rate": 6.0201005025125635e-05,
2943
+ "loss": 1.7405,
2944
+ "step": 40100
2945
+ },
2946
+ {
2947
+ "epoch": 1.0,
2948
+ "grad_norm": 4.4609375,
2949
+ "learning_rate": 6.0100502512562815e-05,
2950
+ "loss": 1.7797,
2951
+ "step": 40200
2952
+ },
2953
+ {
2954
+ "epoch": 1.01,
2955
+ "grad_norm": 7.5703125,
2956
+ "learning_rate": 6e-05,
2957
+ "loss": 1.7335,
2958
+ "step": 40300
2959
+ },
2960
+ {
2961
+ "epoch": 1.01,
2962
+ "grad_norm": 6.08203125,
2963
+ "learning_rate": 5.989949748743718e-05,
2964
+ "loss": 1.7507,
2965
+ "step": 40400
2966
+ },
2967
+ {
2968
+ "epoch": 1.01,
2969
+ "grad_norm": 4.37890625,
2970
+ "learning_rate": 5.979899497487438e-05,
2971
+ "loss": 1.7076,
2972
+ "step": 40500
2973
+ },
2974
+ {
2975
+ "epoch": 1.01,
2976
+ "grad_norm": 4.1171875,
2977
+ "learning_rate": 5.969849246231156e-05,
2978
+ "loss": 1.7574,
2979
+ "step": 40600
2980
+ },
2981
+ {
2982
+ "epoch": 1.02,
2983
+ "grad_norm": 3.7890625,
2984
+ "learning_rate": 5.959798994974875e-05,
2985
+ "loss": 1.75,
2986
+ "step": 40700
2987
+ },
2988
+ {
2989
+ "epoch": 1.02,
2990
+ "grad_norm": 5.9609375,
2991
+ "learning_rate": 5.949748743718593e-05,
2992
+ "loss": 1.7911,
2993
+ "step": 40800
2994
+ },
2995
+ {
2996
+ "epoch": 1.02,
2997
+ "grad_norm": 6.52734375,
2998
+ "learning_rate": 5.939698492462312e-05,
2999
+ "loss": 1.7292,
3000
+ "step": 40900
3001
+ },
3002
+ {
3003
+ "epoch": 1.02,
3004
+ "grad_norm": 4.48828125,
3005
+ "learning_rate": 5.929648241206031e-05,
3006
+ "loss": 1.7336,
3007
+ "step": 41000
3008
+ },
3009
+ {
3010
+ "epoch": 1.03,
3011
+ "grad_norm": 8.0625,
3012
+ "learning_rate": 5.9195979899497487e-05,
3013
+ "loss": 1.7322,
3014
+ "step": 41100
3015
+ },
3016
+ {
3017
+ "epoch": 1.03,
3018
+ "grad_norm": 7.1875,
3019
+ "learning_rate": 5.909547738693467e-05,
3020
+ "loss": 1.7528,
3021
+ "step": 41200
3022
+ },
3023
+ {
3024
+ "epoch": 1.03,
3025
+ "grad_norm": 4.54296875,
3026
+ "learning_rate": 5.8994974874371866e-05,
3027
+ "loss": 1.7232,
3028
+ "step": 41300
3029
+ },
3030
+ {
3031
+ "epoch": 1.03,
3032
+ "grad_norm": 6.11328125,
3033
+ "learning_rate": 5.889447236180905e-05,
3034
+ "loss": 1.6972,
3035
+ "step": 41400
3036
+ },
3037
+ {
3038
+ "epoch": 1.04,
3039
+ "grad_norm": 6.6640625,
3040
+ "learning_rate": 5.879396984924623e-05,
3041
+ "loss": 1.7163,
3042
+ "step": 41500
3043
+ },
3044
+ {
3045
+ "epoch": 1.04,
3046
+ "grad_norm": 3.345703125,
3047
+ "learning_rate": 5.869346733668342e-05,
3048
+ "loss": 1.7457,
3049
+ "step": 41600
3050
+ },
3051
+ {
3052
+ "epoch": 1.04,
3053
+ "grad_norm": 3.603515625,
3054
+ "learning_rate": 5.859296482412061e-05,
3055
+ "loss": 1.7534,
3056
+ "step": 41700
3057
+ },
3058
+ {
3059
+ "epoch": 1.04,
3060
+ "grad_norm": 6.234375,
3061
+ "learning_rate": 5.849246231155779e-05,
3062
+ "loss": 1.7467,
3063
+ "step": 41800
3064
+ },
3065
+ {
3066
+ "epoch": 1.05,
3067
+ "grad_norm": 4.12109375,
3068
+ "learning_rate": 5.839195979899498e-05,
3069
+ "loss": 1.7112,
3070
+ "step": 41900
3071
+ },
3072
+ {
3073
+ "epoch": 1.05,
3074
+ "grad_norm": 3.712890625,
3075
+ "learning_rate": 5.829145728643216e-05,
3076
+ "loss": 1.7927,
3077
+ "step": 42000
3078
+ },
3079
+ {
3080
+ "epoch": 1.05,
3081
+ "grad_norm": 3.259765625,
3082
+ "learning_rate": 5.819095477386936e-05,
3083
+ "loss": 1.7359,
3084
+ "step": 42100
3085
+ },
3086
+ {
3087
+ "epoch": 1.05,
3088
+ "grad_norm": 7.0703125,
3089
+ "learning_rate": 5.809045226130654e-05,
3090
+ "loss": 1.7463,
3091
+ "step": 42200
3092
+ },
3093
+ {
3094
+ "epoch": 1.06,
3095
+ "grad_norm": 5.9140625,
3096
+ "learning_rate": 5.7989949748743724e-05,
3097
+ "loss": 1.7415,
3098
+ "step": 42300
3099
+ },
3100
+ {
3101
+ "epoch": 1.06,
3102
+ "grad_norm": 6.02734375,
3103
+ "learning_rate": 5.7889447236180904e-05,
3104
+ "loss": 1.7429,
3105
+ "step": 42400
3106
+ },
3107
+ {
3108
+ "epoch": 1.06,
3109
+ "grad_norm": 3.603515625,
3110
+ "learning_rate": 5.778894472361809e-05,
3111
+ "loss": 1.7432,
3112
+ "step": 42500
3113
+ },
3114
+ {
3115
+ "epoch": 1.06,
3116
+ "eval_loss": 1.719661831855774,
3117
+ "eval_runtime": 58.5951,
3118
+ "eval_samples_per_second": 17.066,
3119
+ "eval_steps_per_second": 4.267,
3120
+ "step": 42500
3121
+ },
3122
+ {
3123
+ "epoch": 1.06,
3124
+ "grad_norm": 3.814453125,
3125
+ "learning_rate": 5.7688442211055284e-05,
3126
+ "loss": 1.7588,
3127
+ "step": 42600
3128
+ },
3129
+ {
3130
+ "epoch": 1.07,
3131
+ "grad_norm": 3.80859375,
3132
+ "learning_rate": 5.7587939698492464e-05,
3133
+ "loss": 1.7605,
3134
+ "step": 42700
3135
+ },
3136
+ {
3137
+ "epoch": 1.07,
3138
+ "grad_norm": 4.453125,
3139
+ "learning_rate": 5.748743718592965e-05,
3140
+ "loss": 1.733,
3141
+ "step": 42800
3142
+ },
3143
+ {
3144
+ "epoch": 1.07,
3145
+ "grad_norm": 5.4765625,
3146
+ "learning_rate": 5.738693467336683e-05,
3147
+ "loss": 1.702,
3148
+ "step": 42900
3149
+ },
3150
+ {
3151
+ "epoch": 1.07,
3152
+ "grad_norm": 4.71484375,
3153
+ "learning_rate": 5.728643216080403e-05,
3154
+ "loss": 1.7601,
3155
+ "step": 43000
3156
+ },
3157
+ {
3158
+ "epoch": 1.08,
3159
+ "grad_norm": 3.859375,
3160
+ "learning_rate": 5.718592964824121e-05,
3161
+ "loss": 1.7724,
3162
+ "step": 43100
3163
+ },
3164
+ {
3165
+ "epoch": 1.08,
3166
+ "grad_norm": 5.125,
3167
+ "learning_rate": 5.7085427135678396e-05,
3168
+ "loss": 1.7171,
3169
+ "step": 43200
3170
+ },
3171
+ {
3172
+ "epoch": 1.08,
3173
+ "grad_norm": 4.109375,
3174
+ "learning_rate": 5.6984924623115576e-05,
3175
+ "loss": 1.7536,
3176
+ "step": 43300
3177
+ },
3178
+ {
3179
+ "epoch": 1.08,
3180
+ "grad_norm": 5.15234375,
3181
+ "learning_rate": 5.688442211055277e-05,
3182
+ "loss": 1.7804,
3183
+ "step": 43400
3184
+ },
3185
+ {
3186
+ "epoch": 1.09,
3187
+ "grad_norm": 4.98828125,
3188
+ "learning_rate": 5.6783919597989955e-05,
3189
+ "loss": 1.7804,
3190
+ "step": 43500
3191
+ },
3192
+ {
3193
+ "epoch": 1.09,
3194
+ "grad_norm": 4.6484375,
3195
+ "learning_rate": 5.6683417085427135e-05,
3196
+ "loss": 1.7586,
3197
+ "step": 43600
3198
+ },
3199
+ {
3200
+ "epoch": 1.09,
3201
+ "grad_norm": 6.4140625,
3202
+ "learning_rate": 5.658291457286432e-05,
3203
+ "loss": 1.7365,
3204
+ "step": 43700
3205
+ },
3206
+ {
3207
+ "epoch": 1.09,
3208
+ "grad_norm": 3.33984375,
3209
+ "learning_rate": 5.6482412060301515e-05,
3210
+ "loss": 1.7381,
3211
+ "step": 43800
3212
+ },
3213
+ {
3214
+ "epoch": 1.1,
3215
+ "grad_norm": 4.29296875,
3216
+ "learning_rate": 5.63819095477387e-05,
3217
+ "loss": 1.7308,
3218
+ "step": 43900
3219
+ },
3220
+ {
3221
+ "epoch": 1.1,
3222
+ "grad_norm": 5.703125,
3223
+ "learning_rate": 5.628140703517588e-05,
3224
+ "loss": 1.7421,
3225
+ "step": 44000
3226
+ },
3227
+ {
3228
+ "epoch": 1.1,
3229
+ "grad_norm": 5.48828125,
3230
+ "learning_rate": 5.618090452261307e-05,
3231
+ "loss": 1.7377,
3232
+ "step": 44100
3233
+ },
3234
+ {
3235
+ "epoch": 1.1,
3236
+ "grad_norm": 6.125,
3237
+ "learning_rate": 5.608040201005026e-05,
3238
+ "loss": 1.7009,
3239
+ "step": 44200
3240
+ },
3241
+ {
3242
+ "epoch": 1.11,
3243
+ "grad_norm": 4.26953125,
3244
+ "learning_rate": 5.597989949748744e-05,
3245
+ "loss": 1.7114,
3246
+ "step": 44300
3247
+ },
3248
+ {
3249
+ "epoch": 1.11,
3250
+ "grad_norm": 6.7578125,
3251
+ "learning_rate": 5.587939698492463e-05,
3252
+ "loss": 1.7266,
3253
+ "step": 44400
3254
+ },
3255
+ {
3256
+ "epoch": 1.11,
3257
+ "grad_norm": 4.55078125,
3258
+ "learning_rate": 5.577889447236181e-05,
3259
+ "loss": 1.7497,
3260
+ "step": 44500
3261
+ },
3262
+ {
3263
+ "epoch": 1.11,
3264
+ "grad_norm": 4.0703125,
3265
+ "learning_rate": 5.567839195979899e-05,
3266
+ "loss": 1.7285,
3267
+ "step": 44600
3268
+ },
3269
+ {
3270
+ "epoch": 1.12,
3271
+ "grad_norm": 5.0390625,
3272
+ "learning_rate": 5.5577889447236186e-05,
3273
+ "loss": 1.7311,
3274
+ "step": 44700
3275
+ },
3276
+ {
3277
+ "epoch": 1.12,
3278
+ "grad_norm": 4.8046875,
3279
+ "learning_rate": 5.547738693467337e-05,
3280
+ "loss": 1.7544,
3281
+ "step": 44800
3282
+ },
3283
+ {
3284
+ "epoch": 1.12,
3285
+ "grad_norm": 5.578125,
3286
+ "learning_rate": 5.537688442211055e-05,
3287
+ "loss": 1.706,
3288
+ "step": 44900
3289
+ },
3290
+ {
3291
+ "epoch": 1.12,
3292
+ "grad_norm": 3.46875,
3293
+ "learning_rate": 5.527638190954774e-05,
3294
+ "loss": 1.6826,
3295
+ "step": 45000
3296
+ },
3297
+ {
3298
+ "epoch": 1.12,
3299
+ "eval_loss": 1.756588101387024,
3300
+ "eval_runtime": 58.5831,
3301
+ "eval_samples_per_second": 17.07,
3302
+ "eval_steps_per_second": 4.267,
3303
+ "step": 45000
3304
+ },
3305
+ {
3306
+ "epoch": 1.13,
3307
+ "grad_norm": 5.4140625,
3308
+ "learning_rate": 5.517587939698493e-05,
3309
+ "loss": 1.7379,
3310
+ "step": 45100
3311
+ },
3312
+ {
3313
+ "epoch": 1.13,
3314
+ "grad_norm": 3.04296875,
3315
+ "learning_rate": 5.507537688442211e-05,
3316
+ "loss": 1.7533,
3317
+ "step": 45200
3318
+ },
3319
+ {
3320
+ "epoch": 1.13,
3321
+ "grad_norm": 4.421875,
3322
+ "learning_rate": 5.49748743718593e-05,
3323
+ "loss": 1.7273,
3324
+ "step": 45300
3325
+ },
3326
+ {
3327
+ "epoch": 1.14,
3328
+ "grad_norm": 3.1875,
3329
+ "learning_rate": 5.487437185929648e-05,
3330
+ "loss": 1.7534,
3331
+ "step": 45400
3332
+ },
3333
+ {
3334
+ "epoch": 1.14,
3335
+ "grad_norm": 5.34375,
3336
+ "learning_rate": 5.477386934673368e-05,
3337
+ "loss": 1.7731,
3338
+ "step": 45500
3339
+ },
3340
+ {
3341
+ "epoch": 1.14,
3342
+ "grad_norm": 5.34375,
3343
+ "learning_rate": 5.467336683417086e-05,
3344
+ "loss": 1.7151,
3345
+ "step": 45600
3346
+ },
3347
+ {
3348
+ "epoch": 1.14,
3349
+ "grad_norm": 4.171875,
3350
+ "learning_rate": 5.4572864321608045e-05,
3351
+ "loss": 1.7643,
3352
+ "step": 45700
3353
+ },
3354
+ {
3355
+ "epoch": 1.15,
3356
+ "grad_norm": 5.21875,
3357
+ "learning_rate": 5.4472361809045224e-05,
3358
+ "loss": 1.74,
3359
+ "step": 45800
3360
+ },
3361
+ {
3362
+ "epoch": 1.15,
3363
+ "grad_norm": 4.8515625,
3364
+ "learning_rate": 5.437185929648242e-05,
3365
+ "loss": 1.7517,
3366
+ "step": 45900
3367
+ },
3368
+ {
3369
+ "epoch": 1.15,
3370
+ "grad_norm": 4.109375,
3371
+ "learning_rate": 5.4271356783919604e-05,
3372
+ "loss": 1.7648,
3373
+ "step": 46000
3374
+ },
3375
+ {
3376
+ "epoch": 1.15,
3377
+ "grad_norm": 4.51171875,
3378
+ "learning_rate": 5.4170854271356784e-05,
3379
+ "loss": 1.7025,
3380
+ "step": 46100
3381
+ },
3382
+ {
3383
+ "epoch": 1.16,
3384
+ "grad_norm": 5.63671875,
3385
+ "learning_rate": 5.407035175879397e-05,
3386
+ "loss": 1.6961,
3387
+ "step": 46200
3388
+ },
3389
+ {
3390
+ "epoch": 1.16,
3391
+ "grad_norm": 3.171875,
3392
+ "learning_rate": 5.3969849246231163e-05,
3393
+ "loss": 1.74,
3394
+ "step": 46300
3395
+ },
3396
+ {
3397
+ "epoch": 1.16,
3398
+ "grad_norm": 3.279296875,
3399
+ "learning_rate": 5.386934673366835e-05,
3400
+ "loss": 1.7582,
3401
+ "step": 46400
3402
+ },
3403
+ {
3404
+ "epoch": 1.16,
3405
+ "grad_norm": 4.80859375,
3406
+ "learning_rate": 5.376884422110553e-05,
3407
+ "loss": 1.7057,
3408
+ "step": 46500
3409
+ },
3410
+ {
3411
+ "epoch": 1.17,
3412
+ "grad_norm": 2.744140625,
3413
+ "learning_rate": 5.3668341708542716e-05,
3414
+ "loss": 1.7378,
3415
+ "step": 46600
3416
+ },
3417
+ {
3418
+ "epoch": 1.17,
3419
+ "grad_norm": 4.34765625,
3420
+ "learning_rate": 5.356783919597991e-05,
3421
+ "loss": 1.7267,
3422
+ "step": 46700
3423
+ },
3424
+ {
3425
+ "epoch": 1.17,
3426
+ "grad_norm": 3.021484375,
3427
+ "learning_rate": 5.346733668341709e-05,
3428
+ "loss": 1.7465,
3429
+ "step": 46800
3430
+ },
3431
+ {
3432
+ "epoch": 1.17,
3433
+ "grad_norm": 6.91796875,
3434
+ "learning_rate": 5.3366834170854276e-05,
3435
+ "loss": 1.7399,
3436
+ "step": 46900
3437
+ },
3438
+ {
3439
+ "epoch": 1.18,
3440
+ "grad_norm": 7.6484375,
3441
+ "learning_rate": 5.3266331658291455e-05,
3442
+ "loss": 1.7236,
3443
+ "step": 47000
3444
+ },
3445
+ {
3446
+ "epoch": 1.18,
3447
+ "grad_norm": 4.51953125,
3448
+ "learning_rate": 5.316582914572864e-05,
3449
+ "loss": 1.7199,
3450
+ "step": 47100
3451
+ },
3452
+ {
3453
+ "epoch": 1.18,
3454
+ "grad_norm": 2.994140625,
3455
+ "learning_rate": 5.3065326633165835e-05,
3456
+ "loss": 1.7073,
3457
+ "step": 47200
3458
+ },
3459
+ {
3460
+ "epoch": 1.18,
3461
+ "grad_norm": 3.33984375,
3462
+ "learning_rate": 5.296482412060302e-05,
3463
+ "loss": 1.7313,
3464
+ "step": 47300
3465
+ },
3466
+ {
3467
+ "epoch": 1.19,
3468
+ "grad_norm": 3.443359375,
3469
+ "learning_rate": 5.28643216080402e-05,
3470
+ "loss": 1.702,
3471
+ "step": 47400
3472
+ },
3473
+ {
3474
+ "epoch": 1.19,
3475
+ "grad_norm": 3.958984375,
3476
+ "learning_rate": 5.276381909547739e-05,
3477
+ "loss": 1.7091,
3478
+ "step": 47500
3479
+ },
3480
+ {
3481
+ "epoch": 1.19,
3482
+ "eval_loss": 1.7188657522201538,
3483
+ "eval_runtime": 58.5581,
3484
+ "eval_samples_per_second": 17.077,
3485
+ "eval_steps_per_second": 4.269,
3486
+ "step": 47500
3487
+ },
3488
+ {
3489
+ "epoch": 1.19,
3490
+ "grad_norm": 3.6640625,
3491
+ "learning_rate": 5.266331658291458e-05,
3492
+ "loss": 1.7334,
3493
+ "step": 47600
3494
+ },
3495
+ {
3496
+ "epoch": 1.19,
3497
+ "grad_norm": 4.73046875,
3498
+ "learning_rate": 5.256281407035176e-05,
3499
+ "loss": 1.7369,
3500
+ "step": 47700
3501
+ },
3502
+ {
3503
+ "epoch": 1.2,
3504
+ "grad_norm": 6.57421875,
3505
+ "learning_rate": 5.246231155778895e-05,
3506
+ "loss": 1.6972,
3507
+ "step": 47800
3508
+ },
3509
+ {
3510
+ "epoch": 1.2,
3511
+ "grad_norm": 4.62890625,
3512
+ "learning_rate": 5.236180904522613e-05,
3513
+ "loss": 1.75,
3514
+ "step": 47900
3515
+ },
3516
+ {
3517
+ "epoch": 1.2,
3518
+ "grad_norm": 3.353515625,
3519
+ "learning_rate": 5.226130653266332e-05,
3520
+ "loss": 1.7259,
3521
+ "step": 48000
3522
+ },
3523
+ {
3524
+ "epoch": 1.2,
3525
+ "grad_norm": 5.00390625,
3526
+ "learning_rate": 5.2160804020100507e-05,
3527
+ "loss": 1.7427,
3528
+ "step": 48100
3529
+ },
3530
+ {
3531
+ "epoch": 1.21,
3532
+ "grad_norm": 3.01953125,
3533
+ "learning_rate": 5.206030150753769e-05,
3534
+ "loss": 1.6955,
3535
+ "step": 48200
3536
+ },
3537
+ {
3538
+ "epoch": 1.21,
3539
+ "grad_norm": 4.51953125,
3540
+ "learning_rate": 5.195979899497487e-05,
3541
+ "loss": 1.7535,
3542
+ "step": 48300
3543
+ },
3544
+ {
3545
+ "epoch": 1.21,
3546
+ "grad_norm": 4.23046875,
3547
+ "learning_rate": 5.1859296482412066e-05,
3548
+ "loss": 1.7482,
3549
+ "step": 48400
3550
+ },
3551
+ {
3552
+ "epoch": 1.21,
3553
+ "grad_norm": 5.6875,
3554
+ "learning_rate": 5.175879396984925e-05,
3555
+ "loss": 1.6747,
3556
+ "step": 48500
3557
+ },
3558
+ {
3559
+ "epoch": 1.22,
3560
+ "grad_norm": 6.8125,
3561
+ "learning_rate": 5.165829145728643e-05,
3562
+ "loss": 1.6639,
3563
+ "step": 48600
3564
+ },
3565
+ {
3566
+ "epoch": 1.22,
3567
+ "grad_norm": 3.064453125,
3568
+ "learning_rate": 5.155778894472362e-05,
3569
+ "loss": 1.7246,
3570
+ "step": 48700
3571
+ },
3572
+ {
3573
+ "epoch": 1.22,
3574
+ "grad_norm": 5.59375,
3575
+ "learning_rate": 5.145728643216081e-05,
3576
+ "loss": 1.7406,
3577
+ "step": 48800
3578
+ },
3579
+ {
3580
+ "epoch": 1.22,
3581
+ "grad_norm": 3.166015625,
3582
+ "learning_rate": 5.135678391959799e-05,
3583
+ "loss": 1.7114,
3584
+ "step": 48900
3585
+ },
3586
+ {
3587
+ "epoch": 1.23,
3588
+ "grad_norm": 5.05859375,
3589
+ "learning_rate": 5.125628140703518e-05,
3590
+ "loss": 1.7422,
3591
+ "step": 49000
3592
+ },
3593
+ {
3594
+ "epoch": 1.23,
3595
+ "grad_norm": 4.33984375,
3596
+ "learning_rate": 5.1155778894472365e-05,
3597
+ "loss": 1.7329,
3598
+ "step": 49100
3599
+ },
3600
+ {
3601
+ "epoch": 1.23,
3602
+ "grad_norm": 6.41796875,
3603
+ "learning_rate": 5.1055276381909544e-05,
3604
+ "loss": 1.7284,
3605
+ "step": 49200
3606
+ },
3607
+ {
3608
+ "epoch": 1.23,
3609
+ "grad_norm": 4.53125,
3610
+ "learning_rate": 5.095477386934674e-05,
3611
+ "loss": 1.6935,
3612
+ "step": 49300
3613
+ },
3614
+ {
3615
+ "epoch": 1.23,
3616
+ "grad_norm": 5.109375,
3617
+ "learning_rate": 5.0854271356783924e-05,
3618
+ "loss": 1.7171,
3619
+ "step": 49400
3620
+ },
3621
+ {
3622
+ "epoch": 1.24,
3623
+ "grad_norm": 4.97265625,
3624
+ "learning_rate": 5.0753768844221104e-05,
3625
+ "loss": 1.7192,
3626
+ "step": 49500
3627
+ },
3628
+ {
3629
+ "epoch": 1.24,
3630
+ "grad_norm": 4.36328125,
3631
+ "learning_rate": 5.065326633165829e-05,
3632
+ "loss": 1.775,
3633
+ "step": 49600
3634
+ },
3635
+ {
3636
+ "epoch": 1.24,
3637
+ "grad_norm": 4.40625,
3638
+ "learning_rate": 5.0552763819095484e-05,
3639
+ "loss": 1.7524,
3640
+ "step": 49700
3641
+ },
3642
+ {
3643
+ "epoch": 1.25,
3644
+ "grad_norm": 3.193359375,
3645
+ "learning_rate": 5.045226130653266e-05,
3646
+ "loss": 1.7293,
3647
+ "step": 49800
3648
+ },
3649
+ {
3650
+ "epoch": 1.25,
3651
+ "grad_norm": 3.62890625,
3652
+ "learning_rate": 5.035175879396985e-05,
3653
+ "loss": 1.7557,
3654
+ "step": 49900
3655
+ },
3656
+ {
3657
+ "epoch": 1.25,
3658
+ "grad_norm": 5.51171875,
3659
+ "learning_rate": 5.0251256281407036e-05,
3660
+ "loss": 1.7435,
3661
+ "step": 50000
3662
+ },
3663
+ {
3664
+ "epoch": 1.25,
3665
+ "eval_loss": 1.704933762550354,
3666
+ "eval_runtime": 58.601,
3667
+ "eval_samples_per_second": 17.065,
3668
+ "eval_steps_per_second": 4.266,
3669
+ "step": 50000
3670
+ }
3671
+ ],
3672
+ "logging_steps": 100,
3673
+ "max_steps": 100000,
3674
+ "num_input_tokens_seen": 0,
3675
+ "num_train_epochs": 3,
3676
+ "save_steps": 2500,
3677
+ "total_flos": 7.866849165312e+17,
3678
+ "train_batch_size": 4,
3679
+ "trial_name": null,
3680
+ "trial_params": null
3681
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:841cb7e6f9fc4f19db87ce9d0bb0834d6ed9f6df3d14ff1ec46778f5841bf8d1
3
+ size 4856