Hugging Face's logo

wellecks
/
llmstep-mathlib4-pythia2.8b

Text Generation
Transformers
PyTorch
gpt_neox
text-generation-inference
Inference Endpoints
Model card Files Community
2
wellecks commited on
Commit
ed824ba
•
1 Parent(s): cbe5586

Upload folder using huggingface_hub

Browse files
Files changed (18) hide show
  1. config.json +26 -0
  2. generation_config.json +6 -0
  3. latest +1 -0
  4. pytorch_model.bin +3 -0
  5. rng_state_0.pth +3 -0
  6. rng_state_1.pth +3 -0
  7. rng_state_2.pth +3 -0
  8. rng_state_3.pth +3 -0
  9. rng_state_4.pth +3 -0
  10. rng_state_5.pth +3 -0
  11. rng_state_6.pth +3 -0
  12. rng_state_7.pth +3 -0
  13. special_tokens_map.json +6 -0
  14. tokenizer.json +0 -0
  15. tokenizer_config.json +10 -0
  16. trainer_state.json +3096 -0
  17. training_args.bin +3 -0
  18. zero_to_fp32.py +578 -0
config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "EleutherAI/pythia-2.8b-deduped",
3
+ "architectures": [
4
+ "GPTNeoXForCausalLM"
5
+ ],
6
+ "bos_token_id": 0,
7
+ "classifier_dropout": 0.1,
8
+ "eos_token_id": 0,
9
+ "hidden_act": "gelu",
10
+ "hidden_size": 2560,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 10240,
13
+ "layer_norm_eps": 1e-05,
14
+ "max_position_embeddings": 2048,
15
+ "model_type": "gpt_neox",
16
+ "num_attention_heads": 32,
17
+ "num_hidden_layers": 32,
18
+ "rotary_emb_base": 10000,
19
+ "rotary_pct": 0.25,
20
+ "tie_word_embeddings": false,
21
+ "torch_dtype": "float16",
22
+ "transformers_version": "4.30.2",
23
+ "use_cache": true,
24
+ "use_parallel_residual": true,
25
+ "vocab_size": 50278
26
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 0,
4
+ "eos_token_id": 0,
5
+ "transformers_version": "4.30.2"
6
+ }
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step5000
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b0defa8938fb24a86b5fe07bd7c37caa17fcdb57e1e133ebf330e4c0535c616
3
+ size 5684524533
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
rng_state_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7f9f7f364921228f23215c95bf20d499327f6420582a9337888df4eeb66ede
3
+ size 21579
special_tokens_map.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": "<|endoftext|>",
3
+ "eos_token": "<|endoftext|>",
4
+ "pad_token": "[PAD]",
5
+ "unk_token": "<|endoftext|>"
6
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "bos_token": "<|endoftext|>",
4
+ "clean_up_tokenization_spaces": true,
5
+ "eos_token": "<|endoftext|>",
6
+ "model_max_length": 1024,
7
+ "padding_side": "right",
8
+ "tokenizer_class": "GPTNeoXTokenizer",
9
+ "unk_token": "<|endoftext|>"
10
+ }
trainer_state.json ADDED
@@ -0,0 +1,3096 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.890625,
3
+ "best_model_checkpoint": "/net/nfs/mosaic/seanw/ntptutorial/proofstep/EleutherAI/pythia-2.8b-deduped/checkpoint-5000",
4
+ "epoch": 1.887504718761797,
5
+ "global_step": 5000,
6
+ "is_hyper_param_search": false,
7
+ "is_local_process_zero": true,
8
+ "is_world_process_zero": true,
9
+ "log_history": [
10
+ {
11
+ "epoch": 0.0,
12
+ "learning_rate": 3.4478393790900113e-06,
13
+ "loss": 2.7471,
14
+ "step": 10
15
+ },
16
+ {
17
+ "epoch": 0.01,
18
+ "learning_rate": 4.48574245242758e-06,
19
+ "loss": 2.2453,
20
+ "step": 20
21
+ },
22
+ {
23
+ "epoch": 0.01,
24
+ "learning_rate": 5.092876829713298e-06,
25
+ "loss": 2.1194,
26
+ "step": 30
27
+ },
28
+ {
29
+ "epoch": 0.02,
30
+ "learning_rate": 5.523645525765149e-06,
31
+ "loss": 1.9594,
32
+ "step": 40
33
+ },
34
+ {
35
+ "epoch": 0.02,
36
+ "learning_rate": 5.857775684842452e-06,
37
+ "loss": 1.874,
38
+ "step": 50
39
+ },
40
+ {
41
+ "epoch": 0.02,
42
+ "learning_rate": 6.130779903050867e-06,
43
+ "loss": 1.8366,
44
+ "step": 60
45
+ },
46
+ {
47
+ "epoch": 0.03,
48
+ "learning_rate": 6.361601680642952e-06,
49
+ "loss": 1.8257,
50
+ "step": 70
51
+ },
52
+ {
53
+ "epoch": 0.03,
54
+ "learning_rate": 6.561548599102718e-06,
55
+ "loss": 1.7581,
56
+ "step": 80
57
+ },
58
+ {
59
+ "epoch": 0.03,
60
+ "learning_rate": 6.737914280336586e-06,
61
+ "loss": 1.7984,
62
+ "step": 90
63
+ },
64
+ {
65
+ "epoch": 0.04,
66
+ "learning_rate": 6.895678758180023e-06,
67
+ "loss": 1.676,
68
+ "step": 100
69
+ },
70
+ {
71
+ "epoch": 0.04,
72
+ "learning_rate": 7.038394088074824e-06,
73
+ "loss": 1.7201,
74
+ "step": 110
75
+ },
76
+ {
77
+ "epoch": 0.05,
78
+ "learning_rate": 7.168682976388437e-06,
79
+ "loss": 1.6249,
80
+ "step": 120
81
+ },
82
+ {
83
+ "epoch": 0.05,
84
+ "learning_rate": 7.2885371352490585e-06,
85
+ "loss": 1.5939,
86
+ "step": 130
87
+ },
88
+ {
89
+ "epoch": 0.05,
90
+ "learning_rate": 7.39950475398052e-06,
91
+ "loss": 1.5962,
92
+ "step": 140
93
+ },
94
+ {
95
+ "epoch": 0.06,
96
+ "learning_rate": 7.502813135465739e-06,
97
+ "loss": 1.6127,
98
+ "step": 150
99
+ },
100
+ {
101
+ "epoch": 0.06,
102
+ "learning_rate": 7.599451672440288e-06,
103
+ "loss": 1.6355,
104
+ "step": 160
105
+ },
106
+ {
107
+ "epoch": 0.06,
108
+ "learning_rate": 7.690229624176853e-06,
109
+ "loss": 1.598,
110
+ "step": 170
111
+ },
112
+ {
113
+ "epoch": 0.07,
114
+ "learning_rate": 7.775817353674156e-06,
115
+ "loss": 1.5307,
116
+ "step": 180
117
+ },
118
+ {
119
+ "epoch": 0.07,
120
+ "learning_rate": 7.856776400608327e-06,
121
+ "loss": 1.5646,
122
+ "step": 190
123
+ },
124
+ {
125
+ "epoch": 0.08,
126
+ "learning_rate": 7.93358183151759e-06,
127
+ "loss": 1.5411,
128
+ "step": 200
129
+ },
130
+ {
131
+ "epoch": 0.08,
132
+ "learning_rate": 8.006639131266238e-06,
133
+ "loss": 1.5405,
134
+ "step": 210
135
+ },
136
+ {
137
+ "epoch": 0.08,
138
+ "learning_rate": 8.076297161412393e-06,
139
+ "loss": 1.5144,
140
+ "step": 220
141
+ },
142
+ {
143
+ "epoch": 0.09,
144
+ "learning_rate": 8.142858235714492e-06,
145
+ "loss": 1.485,
146
+ "step": 230
147
+ },
148
+ {
149
+ "epoch": 0.09,
150
+ "learning_rate": 8.206586049726006e-06,
151
+ "loss": 1.4476,
152
+ "step": 240
153
+ },
154
+ {
155
+ "epoch": 0.09,
156
+ "learning_rate": 8.267711990594892e-06,
157
+ "loss": 1.4511,
158
+ "step": 250
159
+ },
160
+ {
161
+ "epoch": 0.1,
162
+ "learning_rate": 8.326440208586629e-06,
163
+ "loss": 1.4577,
164
+ "step": 260
165
+ },
166
+ {
167
+ "epoch": 0.1,
168
+ "learning_rate": 8.382951730959874e-06,
169
+ "loss": 1.4848,
170
+ "step": 270
171
+ },
172
+ {
173
+ "epoch": 0.11,
174
+ "learning_rate": 8.43740782731809e-06,
175
+ "loss": 1.4021,
176
+ "step": 280
177
+ },
178
+ {
179
+ "epoch": 0.11,
180
+ "learning_rate": 8.489952784148422e-06,
181
+ "loss": 1.4138,
182
+ "step": 290
183
+ },
184
+ {
185
+ "epoch": 0.11,
186
+ "learning_rate": 8.540716208803308e-06,
187
+ "loss": 1.378,
188
+ "step": 300
189
+ },
190
+ {
191
+ "epoch": 0.12,
192
+ "learning_rate": 8.589814955558195e-06,
193
+ "loss": 1.3893,
194
+ "step": 310
195
+ },
196
+ {
197
+ "epoch": 0.12,
198
+ "learning_rate": 8.637354745777857e-06,
199
+ "loss": 1.3791,
200
+ "step": 320
201
+ },
202
+ {
203
+ "epoch": 0.12,
204
+ "learning_rate": 8.683431538698111e-06,
205
+ "loss": 1.4285,
206
+ "step": 330
207
+ },
208
+ {
209
+ "epoch": 0.13,
210
+ "learning_rate": 8.728132697514422e-06,
211
+ "loss": 1.4242,
212
+ "step": 340
213
+ },
214
+ {
215
+ "epoch": 0.13,
216
+ "learning_rate": 8.77153798639539e-06,
217
+ "loss": 1.3793,
218
+ "step": 350
219
+ },
220
+ {
221
+ "epoch": 0.14,
222
+ "learning_rate": 8.813720427011724e-06,
223
+ "loss": 1.3997,
224
+ "step": 360
225
+ },
226
+ {
227
+ "epoch": 0.14,
228
+ "learning_rate": 8.854747037685042e-06,
229
+ "loss": 1.4108,
230
+ "step": 370
231
+ },
232
+ {
233
+ "epoch": 0.14,
234
+ "learning_rate": 8.894679473945897e-06,
235
+ "loss": 1.3924,
236
+ "step": 380
237
+ },
238
+ {
239
+ "epoch": 0.15,
240
+ "learning_rate": 8.933574585872348e-06,
241
+ "loss": 1.3793,
242
+ "step": 390
243
+ },
244
+ {
245
+ "epoch": 0.15,
246
+ "learning_rate": 8.97148490485516e-06,
247
+ "loss": 1.3523,
248
+ "step": 400
249
+ },
250
+ {
251
+ "epoch": 0.15,
252
+ "learning_rate": 9.008459070248975e-06,
253
+ "loss": 1.3515,
254
+ "step": 410
255
+ },
256
+ {
257
+ "epoch": 0.16,
258
+ "learning_rate": 9.044542204603808e-06,
259
+ "loss": 1.4118,
260
+ "step": 420
261
+ },
262
+ {
263
+ "epoch": 0.16,
264
+ "learning_rate": 9.07977624473865e-06,
265
+ "loss": 1.3533,
266
+ "step": 430
267
+ },
268
+ {
269
+ "epoch": 0.17,
270
+ "learning_rate": 9.114200234749962e-06,
271
+ "loss": 1.363,
272
+ "step": 440
273
+ },
274
+ {
275
+ "epoch": 0.17,
276
+ "learning_rate": 9.147850586089026e-06,
277
+ "loss": 1.3651,
278
+ "step": 450
279
+ },
280
+ {
281
+ "epoch": 0.17,
282
+ "learning_rate": 9.180761309052062e-06,
283
+ "loss": 1.3318,
284
+ "step": 460
285
+ },
286
+ {
287
+ "epoch": 0.18,
288
+ "learning_rate": 9.21296421937283e-06,
289
+ "loss": 1.3013,
290
+ "step": 470
291
+ },
292
+ {
293
+ "epoch": 0.18,
294
+ "learning_rate": 9.244489123063576e-06,
295
+ "loss": 1.3052,
296
+ "step": 480
297
+ },
298
+ {
299
+ "epoch": 0.18,
300
+ "learning_rate": 9.27536398219589e-06,
301
+ "loss": 1.3055,
302
+ "step": 490
303
+ },
304
+ {
305
+ "epoch": 0.19,
306
+ "learning_rate": 9.305615063932462e-06,
307
+ "loss": 1.3519,
308
+ "step": 500
309
+ },
310
+ {
311
+ "epoch": 0.19,
312
+ "eval_loss": 1.3330078125,
313
+ "eval_runtime": 65.706,
314
+ "eval_samples_per_second": 61.684,
315
+ "eval_steps_per_second": 1.933,
316
+ "step": 500
317
+ },
318
+ {
319
+ "epoch": 0.19,
320
+ "learning_rate": 9.335267074800138e-06,
321
+ "loss": 1.2696,
322
+ "step": 510
323
+ },
324
+ {
325
+ "epoch": 0.2,
326
+ "learning_rate": 9.364343281924199e-06,
327
+ "loss": 1.3356,
328
+ "step": 520
329
+ },
330
+ {
331
+ "epoch": 0.2,
332
+ "learning_rate": 9.392865622714283e-06,
333
+ "loss": 1.2978,
334
+ "step": 530
335
+ },
336
+ {
337
+ "epoch": 0.2,
338
+ "learning_rate": 9.420854804297444e-06,
339
+ "loss": 1.3221,
340
+ "step": 540
341
+ },
342
+ {
343
+ "epoch": 0.21,
344
+ "learning_rate": 9.448330393827263e-06,
345
+ "loss": 1.252,
346
+ "step": 550
347
+ },
348
+ {
349
+ "epoch": 0.21,
350
+ "learning_rate": 9.475310900655658e-06,
351
+ "loss": 1.2975,
352
+ "step": 560
353
+ },
354
+ {
355
+ "epoch": 0.22,
356
+ "learning_rate": 9.501813851231617e-06,
357
+ "loss": 1.2992,
358
+ "step": 570
359
+ },
360
+ {
361
+ "epoch": 0.22,
362
+ "learning_rate": 9.527855857485992e-06,
363
+ "loss": 1.2637,
364
+ "step": 580
365
+ },
366
+ {
367
+ "epoch": 0.22,
368
+ "learning_rate": 9.553452679370556e-06,
369
+ "loss": 1.2805,
370
+ "step": 590
371
+ },
372
+ {
373
+ "epoch": 0.23,
374
+ "learning_rate": 9.578619282140878e-06,
375
+ "loss": 1.2689,
376
+ "step": 600
377
+ },
378
+ {
379
+ "epoch": 0.23,
380
+ "learning_rate": 9.603369888904405e-06,
381
+ "loss": 1.2505,
382
+ "step": 610
383
+ },
384
+ {
385
+ "epoch": 0.23,
386
+ "learning_rate": 9.627718028895765e-06,
387
+ "loss": 1.2542,
388
+ "step": 620
389
+ },
390
+ {
391
+ "epoch": 0.24,
392
+ "learning_rate": 9.651676581889526e-06,
393
+ "loss": 1.2741,
394
+ "step": 630
395
+ },
396
+ {
397
+ "epoch": 0.24,
398
+ "learning_rate": 9.675257819115427e-06,
399
+ "loss": 1.2702,
400
+ "step": 640
401
+ },
402
+ {
403
+ "epoch": 0.25,
404
+ "learning_rate": 9.6984734410015e-06,
405
+ "loss": 1.2626,
406
+ "step": 650
407
+ },
408
+ {
409
+ "epoch": 0.25,
410
+ "learning_rate": 9.721334612035681e-06,
411
+ "loss": 1.2431,
412
+ "step": 660
413
+ },
414
+ {
415
+ "epoch": 0.25,
416
+ "learning_rate": 9.743851993005941e-06,
417
+ "loss": 1.2712,
418
+ "step": 670
419
+ },
420
+ {
421
+ "epoch": 0.26,
422
+ "learning_rate": 9.76603577085199e-06,
423
+ "loss": 1.2467,
424
+ "step": 680
425
+ },
426
+ {
427
+ "epoch": 0.26,
428
+ "learning_rate": 9.787895686337778e-06,
429
+ "loss": 1.2577,
430
+ "step": 690
431
+ },
432
+ {
433
+ "epoch": 0.26,
434
+ "learning_rate": 9.80944105973296e-06,
435
+ "loss": 1.2069,
436
+ "step": 700
437
+ },
438
+ {
439
+ "epoch": 0.27,
440
+ "learning_rate": 9.830680814672785e-06,
441
+ "loss": 1.1961,
442
+ "step": 710
443
+ },
444
+ {
445
+ "epoch": 0.27,
446
+ "learning_rate": 9.851623500349294e-06,
447
+ "loss": 1.2356,
448
+ "step": 720
449
+ },
450
+ {
451
+ "epoch": 0.28,
452
+ "learning_rate": 9.872277312171945e-06,
453
+ "loss": 1.2155,
454
+ "step": 730
455
+ },
456
+ {
457
+ "epoch": 0.28,
458
+ "learning_rate": 9.892650111022612e-06,
459
+ "loss": 1.233,
460
+ "step": 740
461
+ },
462
+ {
463
+ "epoch": 0.28,
464
+ "learning_rate": 9.91274944121818e-06,
465
+ "loss": 1.247,
466
+ "step": 750
467
+ },
468
+ {
469
+ "epoch": 0.29,
470
+ "learning_rate": 9.932582547283467e-06,
471
+ "loss": 1.2464,
472
+ "step": 760
473
+ },
474
+ {
475
+ "epoch": 0.29,
476
+ "learning_rate": 9.952156389627764e-06,
477
+ "loss": 1.2211,
478
+ "step": 770
479
+ },
480
+ {
481
+ "epoch": 0.29,
482
+ "learning_rate": 9.971477659209916e-06,
483
+ "loss": 1.2357,
484
+ "step": 780
485
+ },
486
+ {
487
+ "epoch": 0.3,
488
+ "learning_rate": 9.990552791269229e-06,
489
+ "loss": 1.2205,
490
+ "step": 790
491
+ },
492
+ {
493
+ "epoch": 0.3,
494
+ "learning_rate": 1e-05,
495
+ "loss": 1.2111,
496
+ "step": 800
497
+ },
498
+ {
499
+ "epoch": 0.31,
500
+ "learning_rate": 1e-05,
501
+ "loss": 1.1852,
502
+ "step": 810
503
+ },
504
+ {
505
+ "epoch": 0.31,
506
+ "learning_rate": 1e-05,
507
+ "loss": 1.2291,
508
+ "step": 820
509
+ },
510
+ {
511
+ "epoch": 0.31,
512
+ "learning_rate": 1e-05,
513
+ "loss": 1.217,
514
+ "step": 830
515
+ },
516
+ {
517
+ "epoch": 0.32,
518
+ "learning_rate": 1e-05,
519
+ "loss": 1.184,
520
+ "step": 840
521
+ },
522
+ {
523
+ "epoch": 0.32,
524
+ "learning_rate": 1e-05,
525
+ "loss": 1.1567,
526
+ "step": 850
527
+ },
528
+ {
529
+ "epoch": 0.32,
530
+ "learning_rate": 1e-05,
531
+ "loss": 1.1987,
532
+ "step": 860
533
+ },
534
+ {
535
+ "epoch": 0.33,
536
+ "learning_rate": 1e-05,
537
+ "loss": 1.2121,
538
+ "step": 870
539
+ },
540
+ {
541
+ "epoch": 0.33,
542
+ "learning_rate": 1e-05,
543
+ "loss": 1.191,
544
+ "step": 880
545
+ },
546
+ {
547
+ "epoch": 0.34,
548
+ "learning_rate": 1e-05,
549
+ "loss": 1.2109,
550
+ "step": 890
551
+ },
552
+ {
553
+ "epoch": 0.34,
554
+ "learning_rate": 1e-05,
555
+ "loss": 1.1724,
556
+ "step": 900
557
+ },
558
+ {
559
+ "epoch": 0.34,
560
+ "learning_rate": 1e-05,
561
+ "loss": 1.1727,
562
+ "step": 910
563
+ },
564
+ {
565
+ "epoch": 0.35,
566
+ "learning_rate": 1e-05,
567
+ "loss": 1.154,
568
+ "step": 920
569
+ },
570
+ {
571
+ "epoch": 0.35,
572
+ "learning_rate": 1e-05,
573
+ "loss": 1.168,
574
+ "step": 930
575
+ },
576
+ {
577
+ "epoch": 0.35,
578
+ "learning_rate": 1e-05,
579
+ "loss": 1.1514,
580
+ "step": 940
581
+ },
582
+ {
583
+ "epoch": 0.36,
584
+ "learning_rate": 1e-05,
585
+ "loss": 1.1677,
586
+ "step": 950
587
+ },
588
+ {
589
+ "epoch": 0.36,
590
+ "learning_rate": 1e-05,
591
+ "loss": 1.1848,
592
+ "step": 960
593
+ },
594
+ {
595
+ "epoch": 0.37,
596
+ "learning_rate": 1e-05,
597
+ "loss": 1.1555,
598
+ "step": 970
599
+ },
600
+ {
601
+ "epoch": 0.37,
602
+ "learning_rate": 1e-05,
603
+ "loss": 1.1451,
604
+ "step": 980
605
+ },
606
+ {
607
+ "epoch": 0.37,
608
+ "learning_rate": 1e-05,
609
+ "loss": 1.1323,
610
+ "step": 990
611
+ },
612
+ {
613
+ "epoch": 0.38,
614
+ "learning_rate": 1e-05,
615
+ "loss": 1.1172,
616
+ "step": 1000
617
+ },
618
+ {
619
+ "epoch": 0.38,
620
+ "eval_loss": 1.17578125,
621
+ "eval_runtime": 63.361,
622
+ "eval_samples_per_second": 63.967,
623
+ "eval_steps_per_second": 2.004,
624
+ "step": 1000
625
+ },
626
+ {
627
+ "epoch": 0.38,
628
+ "learning_rate": 1e-05,
629
+ "loss": 1.1784,
630
+ "step": 1010
631
+ },
632
+ {
633
+ "epoch": 0.39,
634
+ "learning_rate": 1e-05,
635
+ "loss": 1.1424,
636
+ "step": 1020
637
+ },
638
+ {
639
+ "epoch": 0.39,
640
+ "learning_rate": 1e-05,
641
+ "loss": 1.1858,
642
+ "step": 1030
643
+ },
644
+ {
645
+ "epoch": 0.39,
646
+ "learning_rate": 1e-05,
647
+ "loss": 1.1484,
648
+ "step": 1040
649
+ },
650
+ {
651
+ "epoch": 0.4,
652
+ "learning_rate": 1e-05,
653
+ "loss": 1.1682,
654
+ "step": 1050
655
+ },
656
+ {
657
+ "epoch": 0.4,
658
+ "learning_rate": 1e-05,
659
+ "loss": 1.1212,
660
+ "step": 1060
661
+ },
662
+ {
663
+ "epoch": 0.4,
664
+ "learning_rate": 1e-05,
665
+ "loss": 1.1324,
666
+ "step": 1070
667
+ },
668
+ {
669
+ "epoch": 0.41,
670
+ "learning_rate": 1e-05,
671
+ "loss": 1.1367,
672
+ "step": 1080
673
+ },
674
+ {
675
+ "epoch": 0.41,
676
+ "learning_rate": 1e-05,
677
+ "loss": 1.1443,
678
+ "step": 1090
679
+ },
680
+ {
681
+ "epoch": 0.42,
682
+ "learning_rate": 1e-05,
683
+ "loss": 1.1407,
684
+ "step": 1100
685
+ },
686
+ {
687
+ "epoch": 0.42,
688
+ "learning_rate": 1e-05,
689
+ "loss": 1.1193,
690
+ "step": 1110
691
+ },
692
+ {
693
+ "epoch": 0.42,
694
+ "learning_rate": 1e-05,
695
+ "loss": 1.1317,
696
+ "step": 1120
697
+ },
698
+ {
699
+ "epoch": 0.43,
700
+ "learning_rate": 1e-05,
701
+ "loss": 1.1556,
702
+ "step": 1130
703
+ },
704
+ {
705
+ "epoch": 0.43,
706
+ "learning_rate": 1e-05,
707
+ "loss": 1.1482,
708
+ "step": 1140
709
+ },
710
+ {
711
+ "epoch": 0.43,
712
+ "learning_rate": 1e-05,
713
+ "loss": 1.1255,
714
+ "step": 1150
715
+ },
716
+ {
717
+ "epoch": 0.44,
718
+ "learning_rate": 1e-05,
719
+ "loss": 1.1249,
720
+ "step": 1160
721
+ },
722
+ {
723
+ "epoch": 0.44,
724
+ "learning_rate": 1e-05,
725
+ "loss": 1.1379,
726
+ "step": 1170
727
+ },
728
+ {
729
+ "epoch": 0.45,
730
+ "learning_rate": 1e-05,
731
+ "loss": 1.1077,
732
+ "step": 1180
733
+ },
734
+ {
735
+ "epoch": 0.45,
736
+ "learning_rate": 1e-05,
737
+ "loss": 1.115,
738
+ "step": 1190
739
+ },
740
+ {
741
+ "epoch": 0.45,
742
+ "learning_rate": 1e-05,
743
+ "loss": 1.0913,
744
+ "step": 1200
745
+ },
746
+ {
747
+ "epoch": 0.46,
748
+ "learning_rate": 1e-05,
749
+ "loss": 1.0948,
750
+ "step": 1210
751
+ },
752
+ {
753
+ "epoch": 0.46,
754
+ "learning_rate": 1e-05,
755
+ "loss": 1.1593,
756
+ "step": 1220
757
+ },
758
+ {
759
+ "epoch": 0.46,
760
+ "learning_rate": 1e-05,
761
+ "loss": 1.1208,
762
+ "step": 1230
763
+ },
764
+ {
765
+ "epoch": 0.47,
766
+ "learning_rate": 1e-05,
767
+ "loss": 1.1261,
768
+ "step": 1240
769
+ },
770
+ {
771
+ "epoch": 0.47,
772
+ "learning_rate": 1e-05,
773
+ "loss": 1.0816,
774
+ "step": 1250
775
+ },
776
+ {
777
+ "epoch": 0.48,
778
+ "learning_rate": 1e-05,
779
+ "loss": 1.0907,
780
+ "step": 1260
781
+ },
782
+ {
783
+ "epoch": 0.48,
784
+ "learning_rate": 1e-05,
785
+ "loss": 1.0547,
786
+ "step": 1270
787
+ },
788
+ {
789
+ "epoch": 0.48,
790
+ "learning_rate": 1e-05,
791
+ "loss": 1.1173,
792
+ "step": 1280
793
+ },
794
+ {
795
+ "epoch": 0.49,
796
+ "learning_rate": 1e-05,
797
+ "loss": 1.1133,
798
+ "step": 1290
799
+ },
800
+ {
801
+ "epoch": 0.49,
802
+ "learning_rate": 1e-05,
803
+ "loss": 1.1072,
804
+ "step": 1300
805
+ },
806
+ {
807
+ "epoch": 0.49,
808
+ "learning_rate": 1e-05,
809
+ "loss": 1.0862,
810
+ "step": 1310
811
+ },
812
+ {
813
+ "epoch": 0.5,
814
+ "learning_rate": 1e-05,
815
+ "loss": 1.1178,
816
+ "step": 1320
817
+ },
818
+ {
819
+ "epoch": 0.5,
820
+ "learning_rate": 1e-05,
821
+ "loss": 1.099,
822
+ "step": 1330
823
+ },
824
+ {
825
+ "epoch": 0.51,
826
+ "learning_rate": 1e-05,
827
+ "loss": 1.0853,
828
+ "step": 1340
829
+ },
830
+ {
831
+ "epoch": 0.51,
832
+ "learning_rate": 1e-05,
833
+ "loss": 1.0652,
834
+ "step": 1350
835
+ },
836
+ {
837
+ "epoch": 0.51,
838
+ "learning_rate": 1e-05,
839
+ "loss": 1.0959,
840
+ "step": 1360
841
+ },
842
+ {
843
+ "epoch": 0.52,
844
+ "learning_rate": 1e-05,
845
+ "loss": 1.0585,
846
+ "step": 1370
847
+ },
848
+ {
849
+ "epoch": 0.52,
850
+ "learning_rate": 1e-05,
851
+ "loss": 1.0644,
852
+ "step": 1380
853
+ },
854
+ {
855
+ "epoch": 0.52,
856
+ "learning_rate": 1e-05,
857
+ "loss": 1.018,
858
+ "step": 1390
859
+ },
860
+ {
861
+ "epoch": 0.53,
862
+ "learning_rate": 1e-05,
863
+ "loss": 1.0694,
864
+ "step": 1400
865
+ },
866
+ {
867
+ "epoch": 0.53,
868
+ "learning_rate": 1e-05,
869
+ "loss": 1.0861,
870
+ "step": 1410
871
+ },
872
+ {
873
+ "epoch": 0.54,
874
+ "learning_rate": 1e-05,
875
+ "loss": 1.0846,
876
+ "step": 1420
877
+ },
878
+ {
879
+ "epoch": 0.54,
880
+ "learning_rate": 1e-05,
881
+ "loss": 1.1364,
882
+ "step": 1430
883
+ },
884
+ {
885
+ "epoch": 0.54,
886
+ "learning_rate": 1e-05,
887
+ "loss": 1.0504,
888
+ "step": 1440
889
+ },
890
+ {
891
+ "epoch": 0.55,
892
+ "learning_rate": 1e-05,
893
+ "loss": 1.0912,
894
+ "step": 1450
895
+ },
896
+ {
897
+ "epoch": 0.55,
898
+ "learning_rate": 1e-05,
899
+ "loss": 1.0495,
900
+ "step": 1460
901
+ },
902
+ {
903
+ "epoch": 0.55,
904
+ "learning_rate": 1e-05,
905
+ "loss": 1.0541,
906
+ "step": 1470
907
+ },
908
+ {
909
+ "epoch": 0.56,
910
+ "learning_rate": 1e-05,
911
+ "loss": 1.0622,
912
+ "step": 1480
913
+ },
914
+ {
915
+ "epoch": 0.56,
916
+ "learning_rate": 1e-05,
917
+ "loss": 1.0501,
918
+ "step": 1490
919
+ },
920
+ {
921
+ "epoch": 0.57,
922
+ "learning_rate": 1e-05,
923
+ "loss": 1.06,
924
+ "step": 1500
925
+ },
926
+ {
927
+ "epoch": 0.57,
928
+ "eval_loss": 1.0966796875,
929
+ "eval_runtime": 63.326,
930
+ "eval_samples_per_second": 64.002,
931
+ "eval_steps_per_second": 2.005,
932
+ "step": 1500
933
+ },
934
+ {
935
+ "epoch": 0.57,
936
+ "learning_rate": 1e-05,
937
+ "loss": 1.0511,
938
+ "step": 1510
939
+ },
940
+ {
941
+ "epoch": 0.57,
942
+ "learning_rate": 1e-05,
943
+ "loss": 1.025,
944
+ "step": 1520
945
+ },
946
+ {
947
+ "epoch": 0.58,
948
+ "learning_rate": 1e-05,
949
+ "loss": 1.0679,
950
+ "step": 1530
951
+ },
952
+ {
953
+ "epoch": 0.58,
954
+ "learning_rate": 1e-05,
955
+ "loss": 1.0676,
956
+ "step": 1540
957
+ },
958
+ {
959
+ "epoch": 0.59,
960
+ "learning_rate": 1e-05,
961
+ "loss": 1.0678,
962
+ "step": 1550
963
+ },
964
+ {
965
+ "epoch": 0.59,
966
+ "learning_rate": 1e-05,
967
+ "loss": 1.0666,
968
+ "step": 1560
969
+ },
970
+ {
971
+ "epoch": 0.59,
972
+ "learning_rate": 1e-05,
973
+ "loss": 1.1123,
974
+ "step": 1570
975
+ },
976
+ {
977
+ "epoch": 0.6,
978
+ "learning_rate": 1e-05,
979
+ "loss": 1.045,
980
+ "step": 1580
981
+ },
982
+ {
983
+ "epoch": 0.6,
984
+ "learning_rate": 1e-05,
985
+ "loss": 1.0311,
986
+ "step": 1590
987
+ },
988
+ {
989
+ "epoch": 0.6,
990
+ "learning_rate": 1e-05,
991
+ "loss": 1.0675,
992
+ "step": 1600
993
+ },
994
+ {
995
+ "epoch": 0.61,
996
+ "learning_rate": 1e-05,
997
+ "loss": 1.0417,
998
+ "step": 1610
999
+ },
1000
+ {
1001
+ "epoch": 0.61,
1002
+ "learning_rate": 1e-05,
1003
+ "loss": 1.0181,
1004
+ "step": 1620
1005
+ },
1006
+ {
1007
+ "epoch": 0.62,
1008
+ "learning_rate": 1e-05,
1009
+ "loss": 1.0364,
1010
+ "step": 1630
1011
+ },
1012
+ {
1013
+ "epoch": 0.62,
1014
+ "learning_rate": 1e-05,
1015
+ "loss": 1.023,
1016
+ "step": 1640
1017
+ },
1018
+ {
1019
+ "epoch": 0.62,
1020
+ "learning_rate": 1e-05,
1021
+ "loss": 1.0502,
1022
+ "step": 1650
1023
+ },
1024
+ {
1025
+ "epoch": 0.63,
1026
+ "learning_rate": 1e-05,
1027
+ "loss": 1.0789,
1028
+ "step": 1660
1029
+ },
1030
+ {
1031
+ "epoch": 0.63,
1032
+ "learning_rate": 1e-05,
1033
+ "loss": 1.062,
1034
+ "step": 1670
1035
+ },
1036
+ {
1037
+ "epoch": 0.63,
1038
+ "learning_rate": 1e-05,
1039
+ "loss": 1.0341,
1040
+ "step": 1680
1041
+ },
1042
+ {
1043
+ "epoch": 0.64,
1044
+ "learning_rate": 1e-05,
1045
+ "loss": 1.0126,
1046
+ "step": 1690
1047
+ },
1048
+ {
1049
+ "epoch": 0.64,
1050
+ "learning_rate": 1e-05,
1051
+ "loss": 1.0094,
1052
+ "step": 1700
1053
+ },
1054
+ {
1055
+ "epoch": 0.65,
1056
+ "learning_rate": 1e-05,
1057
+ "loss": 1.0378,
1058
+ "step": 1710
1059
+ },
1060
+ {
1061
+ "epoch": 0.65,
1062
+ "learning_rate": 1e-05,
1063
+ "loss": 1.0002,
1064
+ "step": 1720
1065
+ },
1066
+ {
1067
+ "epoch": 0.65,
1068
+ "learning_rate": 1e-05,
1069
+ "loss": 1.0389,
1070
+ "step": 1730
1071
+ },
1072
+ {
1073
+ "epoch": 0.66,
1074
+ "learning_rate": 1e-05,
1075
+ "loss": 1.0541,
1076
+ "step": 1740
1077
+ },
1078
+ {
1079
+ "epoch": 0.66,
1080
+ "learning_rate": 1e-05,
1081
+ "loss": 1.0776,
1082
+ "step": 1750
1083
+ },
1084
+ {
1085
+ "epoch": 0.66,
1086
+ "learning_rate": 1e-05,
1087
+ "loss": 1.0088,
1088
+ "step": 1760
1089
+ },
1090
+ {
1091
+ "epoch": 0.67,
1092
+ "learning_rate": 1e-05,
1093
+ "loss": 1.0139,
1094
+ "step": 1770
1095
+ },
1096
+ {
1097
+ "epoch": 0.67,
1098
+ "learning_rate": 1e-05,
1099
+ "loss": 0.9772,
1100
+ "step": 1780
1101
+ },
1102
+ {
1103
+ "epoch": 0.68,
1104
+ "learning_rate": 1e-05,
1105
+ "loss": 1.0363,
1106
+ "step": 1790
1107
+ },
1108
+ {
1109
+ "epoch": 0.68,
1110
+ "learning_rate": 1e-05,
1111
+ "loss": 1.0499,
1112
+ "step": 1800
1113
+ },
1114
+ {
1115
+ "epoch": 0.68,
1116
+ "learning_rate": 1e-05,
1117
+ "loss": 1.0171,
1118
+ "step": 1810
1119
+ },
1120
+ {
1121
+ "epoch": 0.69,
1122
+ "learning_rate": 1e-05,
1123
+ "loss": 1.0138,
1124
+ "step": 1820
1125
+ },
1126
+ {
1127
+ "epoch": 0.69,
1128
+ "learning_rate": 1e-05,
1129
+ "loss": 0.973,
1130
+ "step": 1830
1131
+ },
1132
+ {
1133
+ "epoch": 0.69,
1134
+ "learning_rate": 1e-05,
1135
+ "loss": 1.0022,
1136
+ "step": 1840
1137
+ },
1138
+ {
1139
+ "epoch": 0.7,
1140
+ "learning_rate": 1e-05,
1141
+ "loss": 1.049,
1142
+ "step": 1850
1143
+ },
1144
+ {
1145
+ "epoch": 0.7,
1146
+ "learning_rate": 1e-05,
1147
+ "loss": 1.0375,
1148
+ "step": 1860
1149
+ },
1150
+ {
1151
+ "epoch": 0.71,
1152
+ "learning_rate": 1e-05,
1153
+ "loss": 1.0273,
1154
+ "step": 1870
1155
+ },
1156
+ {
1157
+ "epoch": 0.71,
1158
+ "learning_rate": 1e-05,
1159
+ "loss": 1.0266,
1160
+ "step": 1880
1161
+ },
1162
+ {
1163
+ "epoch": 0.71,
1164
+ "learning_rate": 1e-05,
1165
+ "loss": 0.9753,
1166
+ "step": 1890
1167
+ },
1168
+ {
1169
+ "epoch": 0.72,
1170
+ "learning_rate": 1e-05,
1171
+ "loss": 1.032,
1172
+ "step": 1900
1173
+ },
1174
+ {
1175
+ "epoch": 0.72,
1176
+ "learning_rate": 1e-05,
1177
+ "loss": 1.0195,
1178
+ "step": 1910
1179
+ },
1180
+ {
1181
+ "epoch": 0.72,
1182
+ "learning_rate": 1e-05,
1183
+ "loss": 1.0094,
1184
+ "step": 1920
1185
+ },
1186
+ {
1187
+ "epoch": 0.73,
1188
+ "learning_rate": 1e-05,
1189
+ "loss": 0.9572,
1190
+ "step": 1930
1191
+ },
1192
+ {
1193
+ "epoch": 0.73,
1194
+ "learning_rate": 1e-05,
1195
+ "loss": 1.0336,
1196
+ "step": 1940
1197
+ },
1198
+ {
1199
+ "epoch": 0.74,
1200
+ "learning_rate": 1e-05,
1201
+ "loss": 1.0516,
1202
+ "step": 1950
1203
+ },
1204
+ {
1205
+ "epoch": 0.74,
1206
+ "learning_rate": 1e-05,
1207
+ "loss": 1.0169,
1208
+ "step": 1960
1209
+ },
1210
+ {
1211
+ "epoch": 0.74,
1212
+ "learning_rate": 1e-05,
1213
+ "loss": 0.9965,
1214
+ "step": 1970
1215
+ },
1216
+ {
1217
+ "epoch": 0.75,
1218
+ "learning_rate": 1e-05,
1219
+ "loss": 0.9932,
1220
+ "step": 1980
1221
+ },
1222
+ {
1223
+ "epoch": 0.75,
1224
+ "learning_rate": 1e-05,
1225
+ "loss": 0.9958,
1226
+ "step": 1990
1227
+ },
1228
+ {
1229
+ "epoch": 0.76,
1230
+ "learning_rate": 1e-05,
1231
+ "loss": 1.066,
1232
+ "step": 2000
1233
+ },
1234
+ {
1235
+ "epoch": 0.76,
1236
+ "eval_loss": 1.0322265625,
1237
+ "eval_runtime": 63.4242,
1238
+ "eval_samples_per_second": 63.903,
1239
+ "eval_steps_per_second": 2.002,
1240
+ "step": 2000
1241
+ },
1242
+ {
1243
+ "epoch": 0.76,
1244
+ "learning_rate": 1e-05,
1245
+ "loss": 1.0399,
1246
+ "step": 2010
1247
+ },
1248
+ {
1249
+ "epoch": 0.76,
1250
+ "learning_rate": 1e-05,
1251
+ "loss": 1.0053,
1252
+ "step": 2020
1253
+ },
1254
+ {
1255
+ "epoch": 0.77,
1256
+ "learning_rate": 1e-05,
1257
+ "loss": 0.9866,
1258
+ "step": 2030
1259
+ },
1260
+ {
1261
+ "epoch": 0.77,
1262
+ "learning_rate": 1e-05,
1263
+ "loss": 0.9962,
1264
+ "step": 2040
1265
+ },
1266
+ {
1267
+ "epoch": 0.77,
1268
+ "learning_rate": 1e-05,
1269
+ "loss": 1.003,
1270
+ "step": 2050
1271
+ },
1272
+ {
1273
+ "epoch": 0.78,
1274
+ "learning_rate": 1e-05,
1275
+ "loss": 0.9685,
1276
+ "step": 2060
1277
+ },
1278
+ {
1279
+ "epoch": 0.78,
1280
+ "learning_rate": 1e-05,
1281
+ "loss": 0.9901,
1282
+ "step": 2070
1283
+ },
1284
+ {
1285
+ "epoch": 0.79,
1286
+ "learning_rate": 1e-05,
1287
+ "loss": 1.0062,
1288
+ "step": 2080
1289
+ },
1290
+ {
1291
+ "epoch": 0.79,
1292
+ "learning_rate": 1e-05,
1293
+ "loss": 0.9667,
1294
+ "step": 2090
1295
+ },
1296
+ {
1297
+ "epoch": 0.79,
1298
+ "learning_rate": 1e-05,
1299
+ "loss": 1.0048,
1300
+ "step": 2100
1301
+ },
1302
+ {
1303
+ "epoch": 0.8,
1304
+ "learning_rate": 1e-05,
1305
+ "loss": 0.9898,
1306
+ "step": 2110
1307
+ },
1308
+ {
1309
+ "epoch": 0.8,
1310
+ "learning_rate": 1e-05,
1311
+ "loss": 0.989,
1312
+ "step": 2120
1313
+ },
1314
+ {
1315
+ "epoch": 0.8,
1316
+ "learning_rate": 1e-05,
1317
+ "loss": 0.9959,
1318
+ "step": 2130
1319
+ },
1320
+ {
1321
+ "epoch": 0.81,
1322
+ "learning_rate": 1e-05,
1323
+ "loss": 0.9936,
1324
+ "step": 2140
1325
+ },
1326
+ {
1327
+ "epoch": 0.81,
1328
+ "learning_rate": 1e-05,
1329
+ "loss": 0.9809,
1330
+ "step": 2150
1331
+ },
1332
+ {
1333
+ "epoch": 0.82,
1334
+ "learning_rate": 1e-05,
1335
+ "loss": 0.9878,
1336
+ "step": 2160
1337
+ },
1338
+ {
1339
+ "epoch": 0.82,
1340
+ "learning_rate": 1e-05,
1341
+ "loss": 0.9786,
1342
+ "step": 2170
1343
+ },
1344
+ {
1345
+ "epoch": 0.82,
1346
+ "learning_rate": 1e-05,
1347
+ "loss": 0.9745,
1348
+ "step": 2180
1349
+ },
1350
+ {
1351
+ "epoch": 0.83,
1352
+ "learning_rate": 1e-05,
1353
+ "loss": 0.9598,
1354
+ "step": 2190
1355
+ },
1356
+ {
1357
+ "epoch": 0.83,
1358
+ "learning_rate": 1e-05,
1359
+ "loss": 0.9861,
1360
+ "step": 2200
1361
+ },
1362
+ {
1363
+ "epoch": 0.83,
1364
+ "learning_rate": 1e-05,
1365
+ "loss": 1.0198,
1366
+ "step": 2210
1367
+ },
1368
+ {
1369
+ "epoch": 0.84,
1370
+ "learning_rate": 1e-05,
1371
+ "loss": 0.958,
1372
+ "step": 2220
1373
+ },
1374
+ {
1375
+ "epoch": 0.84,
1376
+ "learning_rate": 1e-05,
1377
+ "loss": 1.0112,
1378
+ "step": 2230
1379
+ },
1380
+ {
1381
+ "epoch": 0.85,
1382
+ "learning_rate": 1e-05,
1383
+ "loss": 0.9824,
1384
+ "step": 2240
1385
+ },
1386
+ {
1387
+ "epoch": 0.85,
1388
+ "learning_rate": 1e-05,
1389
+ "loss": 1.016,
1390
+ "step": 2250
1391
+ },
1392
+ {
1393
+ "epoch": 0.85,
1394
+ "learning_rate": 1e-05,
1395
+ "loss": 0.9787,
1396
+ "step": 2260
1397
+ },
1398
+ {
1399
+ "epoch": 0.86,
1400
+ "learning_rate": 1e-05,
1401
+ "loss": 0.9343,
1402
+ "step": 2270
1403
+ },
1404
+ {
1405
+ "epoch": 0.86,
1406
+ "learning_rate": 1e-05,
1407
+ "loss": 0.985,
1408
+ "step": 2280
1409
+ },
1410
+ {
1411
+ "epoch": 0.86,
1412
+ "learning_rate": 1e-05,
1413
+ "loss": 1.0047,
1414
+ "step": 2290
1415
+ },
1416
+ {
1417
+ "epoch": 0.87,
1418
+ "learning_rate": 1e-05,
1419
+ "loss": 0.9547,
1420
+ "step": 2300
1421
+ },
1422
+ {
1423
+ "epoch": 0.87,
1424
+ "learning_rate": 1e-05,
1425
+ "loss": 0.9782,
1426
+ "step": 2310
1427
+ },
1428
+ {
1429
+ "epoch": 0.88,
1430
+ "learning_rate": 1e-05,
1431
+ "loss": 0.9645,
1432
+ "step": 2320
1433
+ },
1434
+ {
1435
+ "epoch": 0.88,
1436
+ "learning_rate": 1e-05,
1437
+ "loss": 0.9736,
1438
+ "step": 2330
1439
+ },
1440
+ {
1441
+ "epoch": 0.88,
1442
+ "learning_rate": 1e-05,
1443
+ "loss": 0.98,
1444
+ "step": 2340
1445
+ },
1446
+ {
1447
+ "epoch": 0.89,
1448
+ "learning_rate": 1e-05,
1449
+ "loss": 0.9407,
1450
+ "step": 2350
1451
+ },
1452
+ {
1453
+ "epoch": 0.89,
1454
+ "learning_rate": 1e-05,
1455
+ "loss": 0.9727,
1456
+ "step": 2360
1457
+ },
1458
+ {
1459
+ "epoch": 0.89,
1460
+ "learning_rate": 1e-05,
1461
+ "loss": 0.9996,
1462
+ "step": 2370
1463
+ },
1464
+ {
1465
+ "epoch": 0.9,
1466
+ "learning_rate": 1e-05,
1467
+ "loss": 0.9139,
1468
+ "step": 2380
1469
+ },
1470
+ {
1471
+ "epoch": 0.9,
1472
+ "learning_rate": 1e-05,
1473
+ "loss": 0.9455,
1474
+ "step": 2390
1475
+ },
1476
+ {
1477
+ "epoch": 0.91,
1478
+ "learning_rate": 1e-05,
1479
+ "loss": 0.9754,
1480
+ "step": 2400
1481
+ },
1482
+ {
1483
+ "epoch": 0.91,
1484
+ "learning_rate": 1e-05,
1485
+ "loss": 0.9237,
1486
+ "step": 2410
1487
+ },
1488
+ {
1489
+ "epoch": 0.91,
1490
+ "learning_rate": 1e-05,
1491
+ "loss": 0.9409,
1492
+ "step": 2420
1493
+ },
1494
+ {
1495
+ "epoch": 0.92,
1496
+ "learning_rate": 1e-05,
1497
+ "loss": 0.9335,
1498
+ "step": 2430
1499
+ },
1500
+ {
1501
+ "epoch": 0.92,
1502
+ "learning_rate": 1e-05,
1503
+ "loss": 0.97,
1504
+ "step": 2440
1505
+ },
1506
+ {
1507
+ "epoch": 0.92,
1508
+ "learning_rate": 1e-05,
1509
+ "loss": 0.9355,
1510
+ "step": 2450
1511
+ },
1512
+ {
1513
+ "epoch": 0.93,
1514
+ "learning_rate": 1e-05,
1515
+ "loss": 0.9547,
1516
+ "step": 2460
1517
+ },
1518
+ {
1519
+ "epoch": 0.93,
1520
+ "learning_rate": 1e-05,
1521
+ "loss": 0.978,
1522
+ "step": 2470
1523
+ },
1524
+ {
1525
+ "epoch": 0.94,
1526
+ "learning_rate": 1e-05,
1527
+ "loss": 0.9312,
1528
+ "step": 2480
1529
+ },
1530
+ {
1531
+ "epoch": 0.94,
1532
+ "learning_rate": 1e-05,
1533
+ "loss": 0.9677,
1534
+ "step": 2490
1535
+ },
1536
+ {
1537
+ "epoch": 0.94,
1538
+ "learning_rate": 1e-05,
1539
+ "loss": 0.9637,
1540
+ "step": 2500
1541
+ },
1542
+ {
1543
+ "epoch": 0.94,
1544
+ "eval_loss": 0.986328125,
1545
+ "eval_runtime": 63.3465,
1546
+ "eval_samples_per_second": 63.981,
1547
+ "eval_steps_per_second": 2.005,
1548
+ "step": 2500
1549
+ },
1550
+ {
1551
+ "epoch": 0.95,
1552
+ "learning_rate": 1e-05,
1553
+ "loss": 0.9618,
1554
+ "step": 2510
1555
+ },
1556
+ {
1557
+ "epoch": 0.95,
1558
+ "learning_rate": 1e-05,
1559
+ "loss": 0.9371,
1560
+ "step": 2520
1561
+ },
1562
+ {
1563
+ "epoch": 0.96,
1564
+ "learning_rate": 1e-05,
1565
+ "loss": 0.9065,
1566
+ "step": 2530
1567
+ },
1568
+ {
1569
+ "epoch": 0.96,
1570
+ "learning_rate": 1e-05,
1571
+ "loss": 0.9718,
1572
+ "step": 2540
1573
+ },
1574
+ {
1575
+ "epoch": 0.96,
1576
+ "learning_rate": 1e-05,
1577
+ "loss": 0.9376,
1578
+ "step": 2550
1579
+ },
1580
+ {
1581
+ "epoch": 0.97,
1582
+ "learning_rate": 1e-05,
1583
+ "loss": 0.9367,
1584
+ "step": 2560
1585
+ },
1586
+ {
1587
+ "epoch": 0.97,
1588
+ "learning_rate": 1e-05,
1589
+ "loss": 0.9576,
1590
+ "step": 2570
1591
+ },
1592
+ {
1593
+ "epoch": 0.97,
1594
+ "learning_rate": 1e-05,
1595
+ "loss": 0.9476,
1596
+ "step": 2580
1597
+ },
1598
+ {
1599
+ "epoch": 0.98,
1600
+ "learning_rate": 1e-05,
1601
+ "loss": 0.941,
1602
+ "step": 2590
1603
+ },
1604
+ {
1605
+ "epoch": 0.98,
1606
+ "learning_rate": 1e-05,
1607
+ "loss": 0.9305,
1608
+ "step": 2600
1609
+ },
1610
+ {
1611
+ "epoch": 0.99,
1612
+ "learning_rate": 1e-05,
1613
+ "loss": 0.9501,
1614
+ "step": 2610
1615
+ },
1616
+ {
1617
+ "epoch": 0.99,
1618
+ "learning_rate": 1e-05,
1619
+ "loss": 0.9287,
1620
+ "step": 2620
1621
+ },
1622
+ {
1623
+ "epoch": 0.99,
1624
+ "learning_rate": 1e-05,
1625
+ "loss": 0.9217,
1626
+ "step": 2630
1627
+ },
1628
+ {
1629
+ "epoch": 1.0,
1630
+ "learning_rate": 1e-05,
1631
+ "loss": 0.9236,
1632
+ "step": 2640
1633
+ },
1634
+ {
1635
+ "epoch": 1.0,
1636
+ "learning_rate": 1e-05,
1637
+ "loss": 0.9539,
1638
+ "step": 2650
1639
+ },
1640
+ {
1641
+ "epoch": 1.0,
1642
+ "learning_rate": 1e-05,
1643
+ "loss": 0.7743,
1644
+ "step": 2660
1645
+ },
1646
+ {
1647
+ "epoch": 1.01,
1648
+ "learning_rate": 1e-05,
1649
+ "loss": 0.7539,
1650
+ "step": 2670
1651
+ },
1652
+ {
1653
+ "epoch": 1.01,
1654
+ "learning_rate": 1e-05,
1655
+ "loss": 0.7568,
1656
+ "step": 2680
1657
+ },
1658
+ {
1659
+ "epoch": 1.02,
1660
+ "learning_rate": 1e-05,
1661
+ "loss": 0.7495,
1662
+ "step": 2690
1663
+ },
1664
+ {
1665
+ "epoch": 1.02,
1666
+ "learning_rate": 1e-05,
1667
+ "loss": 0.7418,
1668
+ "step": 2700
1669
+ },
1670
+ {
1671
+ "epoch": 1.02,
1672
+ "learning_rate": 1e-05,
1673
+ "loss": 0.7482,
1674
+ "step": 2710
1675
+ },
1676
+ {
1677
+ "epoch": 1.03,
1678
+ "learning_rate": 1e-05,
1679
+ "loss": 0.7582,
1680
+ "step": 2720
1681
+ },
1682
+ {
1683
+ "epoch": 1.03,
1684
+ "learning_rate": 1e-05,
1685
+ "loss": 0.7262,
1686
+ "step": 2730
1687
+ },
1688
+ {
1689
+ "epoch": 1.03,
1690
+ "learning_rate": 1e-05,
1691
+ "loss": 0.7819,
1692
+ "step": 2740
1693
+ },
1694
+ {
1695
+ "epoch": 1.04,
1696
+ "learning_rate": 1e-05,
1697
+ "loss": 0.7121,
1698
+ "step": 2750
1699
+ },
1700
+ {
1701
+ "epoch": 1.04,
1702
+ "learning_rate": 1e-05,
1703
+ "loss": 0.7488,
1704
+ "step": 2760
1705
+ },
1706
+ {
1707
+ "epoch": 1.05,
1708
+ "learning_rate": 1e-05,
1709
+ "loss": 0.7519,
1710
+ "step": 2770
1711
+ },
1712
+ {
1713
+ "epoch": 1.05,
1714
+ "learning_rate": 1e-05,
1715
+ "loss": 0.749,
1716
+ "step": 2780
1717
+ },
1718
+ {
1719
+ "epoch": 1.05,
1720
+ "learning_rate": 1e-05,
1721
+ "loss": 0.7465,
1722
+ "step": 2790
1723
+ },
1724
+ {
1725
+ "epoch": 1.06,
1726
+ "learning_rate": 1e-05,
1727
+ "loss": 0.763,
1728
+ "step": 2800
1729
+ },
1730
+ {
1731
+ "epoch": 1.06,
1732
+ "learning_rate": 1e-05,
1733
+ "loss": 0.7538,
1734
+ "step": 2810
1735
+ },
1736
+ {
1737
+ "epoch": 1.06,
1738
+ "learning_rate": 1e-05,
1739
+ "loss": 0.767,
1740
+ "step": 2820
1741
+ },
1742
+ {
1743
+ "epoch": 1.07,
1744
+ "learning_rate": 1e-05,
1745
+ "loss": 0.7657,
1746
+ "step": 2830
1747
+ },
1748
+ {
1749
+ "epoch": 1.07,
1750
+ "learning_rate": 1e-05,
1751
+ "loss": 0.7488,
1752
+ "step": 2840
1753
+ },
1754
+ {
1755
+ "epoch": 1.08,
1756
+ "learning_rate": 1e-05,
1757
+ "loss": 0.7788,
1758
+ "step": 2850
1759
+ },
1760
+ {
1761
+ "epoch": 1.08,
1762
+ "learning_rate": 1e-05,
1763
+ "loss": 0.7324,
1764
+ "step": 2860
1765
+ },
1766
+ {
1767
+ "epoch": 1.08,
1768
+ "learning_rate": 1e-05,
1769
+ "loss": 0.7477,
1770
+ "step": 2870
1771
+ },
1772
+ {
1773
+ "epoch": 1.09,
1774
+ "learning_rate": 1e-05,
1775
+ "loss": 0.7515,
1776
+ "step": 2880
1777
+ },
1778
+ {
1779
+ "epoch": 1.09,
1780
+ "learning_rate": 1e-05,
1781
+ "loss": 0.7256,
1782
+ "step": 2890
1783
+ },
1784
+ {
1785
+ "epoch": 1.09,
1786
+ "learning_rate": 1e-05,
1787
+ "loss": 0.762,
1788
+ "step": 2900
1789
+ },
1790
+ {
1791
+ "epoch": 1.1,
1792
+ "learning_rate": 1e-05,
1793
+ "loss": 0.7495,
1794
+ "step": 2910
1795
+ },
1796
+ {
1797
+ "epoch": 1.1,
1798
+ "learning_rate": 1e-05,
1799
+ "loss": 0.756,
1800
+ "step": 2920
1801
+ },
1802
+ {
1803
+ "epoch": 1.11,
1804
+ "learning_rate": 1e-05,
1805
+ "loss": 0.7124,
1806
+ "step": 2930
1807
+ },
1808
+ {
1809
+ "epoch": 1.11,
1810
+ "learning_rate": 1e-05,
1811
+ "loss": 0.7457,
1812
+ "step": 2940
1813
+ },
1814
+ {
1815
+ "epoch": 1.11,
1816
+ "learning_rate": 1e-05,
1817
+ "loss": 0.7372,
1818
+ "step": 2950
1819
+ },
1820
+ {
1821
+ "epoch": 1.12,
1822
+ "learning_rate": 1e-05,
1823
+ "loss": 0.7532,
1824
+ "step": 2960
1825
+ },
1826
+ {
1827
+ "epoch": 1.12,
1828
+ "learning_rate": 1e-05,
1829
+ "loss": 0.7486,
1830
+ "step": 2970
1831
+ },
1832
+ {
1833
+ "epoch": 1.12,
1834
+ "learning_rate": 1e-05,
1835
+ "loss": 0.7631,
1836
+ "step": 2980
1837
+ },
1838
+ {
1839
+ "epoch": 1.13,
1840
+ "learning_rate": 1e-05,
1841
+ "loss": 0.7308,
1842
+ "step": 2990
1843
+ },
1844
+ {
1845
+ "epoch": 1.13,
1846
+ "learning_rate": 1e-05,
1847
+ "loss": 0.7603,
1848
+ "step": 3000
1849
+ },
1850
+ {
1851
+ "epoch": 1.13,
1852
+ "eval_loss": 0.984375,
1853
+ "eval_runtime": 63.3332,
1854
+ "eval_samples_per_second": 63.995,
1855
+ "eval_steps_per_second": 2.005,
1856
+ "step": 3000
1857
+ },
1858
+ {
1859
+ "epoch": 1.14,
1860
+ "learning_rate": 1e-05,
1861
+ "loss": 0.731,
1862
+ "step": 3010
1863
+ },
1864
+ {
1865
+ "epoch": 1.14,
1866
+ "learning_rate": 1e-05,
1867
+ "loss": 0.7598,
1868
+ "step": 3020
1869
+ },
1870
+ {
1871
+ "epoch": 1.14,
1872
+ "learning_rate": 1e-05,
1873
+ "loss": 0.7506,
1874
+ "step": 3030
1875
+ },
1876
+ {
1877
+ "epoch": 1.15,
1878
+ "learning_rate": 1e-05,
1879
+ "loss": 0.772,
1880
+ "step": 3040
1881
+ },
1882
+ {
1883
+ "epoch": 1.15,
1884
+ "learning_rate": 1e-05,
1885
+ "loss": 0.7408,
1886
+ "step": 3050
1887
+ },
1888
+ {
1889
+ "epoch": 1.16,
1890
+ "learning_rate": 1e-05,
1891
+ "loss": 0.7467,
1892
+ "step": 3060
1893
+ },
1894
+ {
1895
+ "epoch": 1.16,
1896
+ "learning_rate": 1e-05,
1897
+ "loss": 0.7571,
1898
+ "step": 3070
1899
+ },
1900
+ {
1901
+ "epoch": 1.16,
1902
+ "learning_rate": 1e-05,
1903
+ "loss": 0.7578,
1904
+ "step": 3080
1905
+ },
1906
+ {
1907
+ "epoch": 1.17,
1908
+ "learning_rate": 1e-05,
1909
+ "loss": 0.7561,
1910
+ "step": 3090
1911
+ },
1912
+ {
1913
+ "epoch": 1.17,
1914
+ "learning_rate": 1e-05,
1915
+ "loss": 0.7461,
1916
+ "step": 3100
1917
+ },
1918
+ {
1919
+ "epoch": 1.17,
1920
+ "learning_rate": 1e-05,
1921
+ "loss": 0.7609,
1922
+ "step": 3110
1923
+ },
1924
+ {
1925
+ "epoch": 1.18,
1926
+ "learning_rate": 1e-05,
1927
+ "loss": 0.7579,
1928
+ "step": 3120
1929
+ },
1930
+ {
1931
+ "epoch": 1.18,
1932
+ "learning_rate": 1e-05,
1933
+ "loss": 0.7346,
1934
+ "step": 3130
1935
+ },
1936
+ {
1937
+ "epoch": 1.19,
1938
+ "learning_rate": 1e-05,
1939
+ "loss": 0.7582,
1940
+ "step": 3140
1941
+ },
1942
+ {
1943
+ "epoch": 1.19,
1944
+ "learning_rate": 1e-05,
1945
+ "loss": 0.7515,
1946
+ "step": 3150
1947
+ },
1948
+ {
1949
+ "epoch": 1.19,
1950
+ "learning_rate": 1e-05,
1951
+ "loss": 0.7084,
1952
+ "step": 3160
1953
+ },
1954
+ {
1955
+ "epoch": 1.2,
1956
+ "learning_rate": 1e-05,
1957
+ "loss": 0.7534,
1958
+ "step": 3170
1959
+ },
1960
+ {
1961
+ "epoch": 1.2,
1962
+ "learning_rate": 1e-05,
1963
+ "loss": 0.722,
1964
+ "step": 3180
1965
+ },
1966
+ {
1967
+ "epoch": 1.2,
1968
+ "learning_rate": 1e-05,
1969
+ "loss": 0.7267,
1970
+ "step": 3190
1971
+ },
1972
+ {
1973
+ "epoch": 1.21,
1974
+ "learning_rate": 1e-05,
1975
+ "loss": 0.7094,
1976
+ "step": 3200
1977
+ },
1978
+ {
1979
+ "epoch": 1.21,
1980
+ "learning_rate": 1e-05,
1981
+ "loss": 0.7181,
1982
+ "step": 3210
1983
+ },
1984
+ {
1985
+ "epoch": 1.22,
1986
+ "learning_rate": 1e-05,
1987
+ "loss": 0.7269,
1988
+ "step": 3220
1989
+ },
1990
+ {
1991
+ "epoch": 1.22,
1992
+ "learning_rate": 1e-05,
1993
+ "loss": 0.7156,
1994
+ "step": 3230
1995
+ },
1996
+ {
1997
+ "epoch": 1.22,
1998
+ "learning_rate": 1e-05,
1999
+ "loss": 0.7522,
2000
+ "step": 3240
2001
+ },
2002
+ {
2003
+ "epoch": 1.23,
2004
+ "learning_rate": 1e-05,
2005
+ "loss": 0.7427,
2006
+ "step": 3250
2007
+ },
2008
+ {
2009
+ "epoch": 1.23,
2010
+ "learning_rate": 1e-05,
2011
+ "loss": 0.7311,
2012
+ "step": 3260
2013
+ },
2014
+ {
2015
+ "epoch": 1.23,
2016
+ "learning_rate": 1e-05,
2017
+ "loss": 0.7845,
2018
+ "step": 3270
2019
+ },
2020
+ {
2021
+ "epoch": 1.24,
2022
+ "learning_rate": 1e-05,
2023
+ "loss": 0.7514,
2024
+ "step": 3280
2025
+ },
2026
+ {
2027
+ "epoch": 1.24,
2028
+ "learning_rate": 1e-05,
2029
+ "loss": 0.7294,
2030
+ "step": 3290
2031
+ },
2032
+ {
2033
+ "epoch": 1.25,
2034
+ "learning_rate": 1e-05,
2035
+ "loss": 0.7567,
2036
+ "step": 3300
2037
+ },
2038
+ {
2039
+ "epoch": 1.25,
2040
+ "learning_rate": 1e-05,
2041
+ "loss": 0.7449,
2042
+ "step": 3310
2043
+ },
2044
+ {
2045
+ "epoch": 1.25,
2046
+ "learning_rate": 1e-05,
2047
+ "loss": 0.7431,
2048
+ "step": 3320
2049
+ },
2050
+ {
2051
+ "epoch": 1.26,
2052
+ "learning_rate": 1e-05,
2053
+ "loss": 0.7834,
2054
+ "step": 3330
2055
+ },
2056
+ {
2057
+ "epoch": 1.26,
2058
+ "learning_rate": 1e-05,
2059
+ "loss": 0.7338,
2060
+ "step": 3340
2061
+ },
2062
+ {
2063
+ "epoch": 1.26,
2064
+ "learning_rate": 1e-05,
2065
+ "loss": 0.7162,
2066
+ "step": 3350
2067
+ },
2068
+ {
2069
+ "epoch": 1.27,
2070
+ "learning_rate": 1e-05,
2071
+ "loss": 0.7502,
2072
+ "step": 3360
2073
+ },
2074
+ {
2075
+ "epoch": 1.27,
2076
+ "learning_rate": 1e-05,
2077
+ "loss": 0.7363,
2078
+ "step": 3370
2079
+ },
2080
+ {
2081
+ "epoch": 1.28,
2082
+ "learning_rate": 1e-05,
2083
+ "loss": 0.7367,
2084
+ "step": 3380
2085
+ },
2086
+ {
2087
+ "epoch": 1.28,
2088
+ "learning_rate": 1e-05,
2089
+ "loss": 0.7237,
2090
+ "step": 3390
2091
+ },
2092
+ {
2093
+ "epoch": 1.28,
2094
+ "learning_rate": 1e-05,
2095
+ "loss": 0.7386,
2096
+ "step": 3400
2097
+ },
2098
+ {
2099
+ "epoch": 1.29,
2100
+ "learning_rate": 1e-05,
2101
+ "loss": 0.7276,
2102
+ "step": 3410
2103
+ },
2104
+ {
2105
+ "epoch": 1.29,
2106
+ "learning_rate": 1e-05,
2107
+ "loss": 0.7408,
2108
+ "step": 3420
2109
+ },
2110
+ {
2111
+ "epoch": 1.29,
2112
+ "learning_rate": 1e-05,
2113
+ "loss": 0.7531,
2114
+ "step": 3430
2115
+ },
2116
+ {
2117
+ "epoch": 1.3,
2118
+ "learning_rate": 1e-05,
2119
+ "loss": 0.754,
2120
+ "step": 3440
2121
+ },
2122
+ {
2123
+ "epoch": 1.3,
2124
+ "learning_rate": 1e-05,
2125
+ "loss": 0.7348,
2126
+ "step": 3450
2127
+ },
2128
+ {
2129
+ "epoch": 1.31,
2130
+ "learning_rate": 1e-05,
2131
+ "loss": 0.7325,
2132
+ "step": 3460
2133
+ },
2134
+ {
2135
+ "epoch": 1.31,
2136
+ "learning_rate": 1e-05,
2137
+ "loss": 0.7191,
2138
+ "step": 3470
2139
+ },
2140
+ {
2141
+ "epoch": 1.31,
2142
+ "learning_rate": 1e-05,
2143
+ "loss": 0.7066,
2144
+ "step": 3480
2145
+ },
2146
+ {
2147
+ "epoch": 1.32,
2148
+ "learning_rate": 1e-05,
2149
+ "loss": 0.7459,
2150
+ "step": 3490
2151
+ },
2152
+ {
2153
+ "epoch": 1.32,
2154
+ "learning_rate": 1e-05,
2155
+ "loss": 0.7565,
2156
+ "step": 3500
2157
+ },
2158
+ {
2159
+ "epoch": 1.32,
2160
+ "eval_loss": 0.955078125,
2161
+ "eval_runtime": 63.3357,
2162
+ "eval_samples_per_second": 63.992,
2163
+ "eval_steps_per_second": 2.005,
2164
+ "step": 3500
2165
+ },
2166
+ {
2167
+ "epoch": 1.33,
2168
+ "learning_rate": 1e-05,
2169
+ "loss": 0.7043,
2170
+ "step": 3510
2171
+ },
2172
+ {
2173
+ "epoch": 1.33,
2174
+ "learning_rate": 1e-05,
2175
+ "loss": 0.7165,
2176
+ "step": 3520
2177
+ },
2178
+ {
2179
+ "epoch": 1.33,
2180
+ "learning_rate": 1e-05,
2181
+ "loss": 0.7536,
2182
+ "step": 3530
2183
+ },
2184
+ {
2185
+ "epoch": 1.34,
2186
+ "learning_rate": 1e-05,
2187
+ "loss": 0.7346,
2188
+ "step": 3540
2189
+ },
2190
+ {
2191
+ "epoch": 1.34,
2192
+ "learning_rate": 1e-05,
2193
+ "loss": 0.7011,
2194
+ "step": 3550
2195
+ },
2196
+ {
2197
+ "epoch": 1.34,
2198
+ "learning_rate": 1e-05,
2199
+ "loss": 0.7522,
2200
+ "step": 3560
2201
+ },
2202
+ {
2203
+ "epoch": 1.35,
2204
+ "learning_rate": 1e-05,
2205
+ "loss": 0.7458,
2206
+ "step": 3570
2207
+ },
2208
+ {
2209
+ "epoch": 1.35,
2210
+ "learning_rate": 1e-05,
2211
+ "loss": 0.7432,
2212
+ "step": 3580
2213
+ },
2214
+ {
2215
+ "epoch": 1.36,
2216
+ "learning_rate": 1e-05,
2217
+ "loss": 0.7417,
2218
+ "step": 3590
2219
+ },
2220
+ {
2221
+ "epoch": 1.36,
2222
+ "learning_rate": 1e-05,
2223
+ "loss": 0.7519,
2224
+ "step": 3600
2225
+ },
2226
+ {
2227
+ "epoch": 1.36,
2228
+ "learning_rate": 1e-05,
2229
+ "loss": 0.7754,
2230
+ "step": 3610
2231
+ },
2232
+ {
2233
+ "epoch": 1.37,
2234
+ "learning_rate": 1e-05,
2235
+ "loss": 0.7256,
2236
+ "step": 3620
2237
+ },
2238
+ {
2239
+ "epoch": 1.37,
2240
+ "learning_rate": 1e-05,
2241
+ "loss": 0.7529,
2242
+ "step": 3630
2243
+ },
2244
+ {
2245
+ "epoch": 1.37,
2246
+ "learning_rate": 1e-05,
2247
+ "loss": 0.7545,
2248
+ "step": 3640
2249
+ },
2250
+ {
2251
+ "epoch": 1.38,
2252
+ "learning_rate": 1e-05,
2253
+ "loss": 0.7778,
2254
+ "step": 3650
2255
+ },
2256
+ {
2257
+ "epoch": 1.38,
2258
+ "learning_rate": 1e-05,
2259
+ "loss": 0.6999,
2260
+ "step": 3660
2261
+ },
2262
+ {
2263
+ "epoch": 1.39,
2264
+ "learning_rate": 1e-05,
2265
+ "loss": 0.741,
2266
+ "step": 3670
2267
+ },
2268
+ {
2269
+ "epoch": 1.39,
2270
+ "learning_rate": 1e-05,
2271
+ "loss": 0.7464,
2272
+ "step": 3680
2273
+ },
2274
+ {
2275
+ "epoch": 1.39,
2276
+ "learning_rate": 1e-05,
2277
+ "loss": 0.7568,
2278
+ "step": 3690
2279
+ },
2280
+ {
2281
+ "epoch": 1.4,
2282
+ "learning_rate": 1e-05,
2283
+ "loss": 0.7132,
2284
+ "step": 3700
2285
+ },
2286
+ {
2287
+ "epoch": 1.4,
2288
+ "learning_rate": 1e-05,
2289
+ "loss": 0.7391,
2290
+ "step": 3710
2291
+ },
2292
+ {
2293
+ "epoch": 1.4,
2294
+ "learning_rate": 1e-05,
2295
+ "loss": 0.7498,
2296
+ "step": 3720
2297
+ },
2298
+ {
2299
+ "epoch": 1.41,
2300
+ "learning_rate": 1e-05,
2301
+ "loss": 0.7508,
2302
+ "step": 3730
2303
+ },
2304
+ {
2305
+ "epoch": 1.41,
2306
+ "learning_rate": 1e-05,
2307
+ "loss": 0.7193,
2308
+ "step": 3740
2309
+ },
2310
+ {
2311
+ "epoch": 1.42,
2312
+ "learning_rate": 1e-05,
2313
+ "loss": 0.74,
2314
+ "step": 3750
2315
+ },
2316
+ {
2317
+ "epoch": 1.42,
2318
+ "learning_rate": 1e-05,
2319
+ "loss": 0.7091,
2320
+ "step": 3760
2321
+ },
2322
+ {
2323
+ "epoch": 1.42,
2324
+ "learning_rate": 1e-05,
2325
+ "loss": 0.7419,
2326
+ "step": 3770
2327
+ },
2328
+ {
2329
+ "epoch": 1.43,
2330
+ "learning_rate": 1e-05,
2331
+ "loss": 0.7241,
2332
+ "step": 3780
2333
+ },
2334
+ {
2335
+ "epoch": 1.43,
2336
+ "learning_rate": 1e-05,
2337
+ "loss": 0.7318,
2338
+ "step": 3790
2339
+ },
2340
+ {
2341
+ "epoch": 1.43,
2342
+ "learning_rate": 1e-05,
2343
+ "loss": 0.716,
2344
+ "step": 3800
2345
+ },
2346
+ {
2347
+ "epoch": 1.44,
2348
+ "learning_rate": 1e-05,
2349
+ "loss": 0.7247,
2350
+ "step": 3810
2351
+ },
2352
+ {
2353
+ "epoch": 1.44,
2354
+ "learning_rate": 1e-05,
2355
+ "loss": 0.7435,
2356
+ "step": 3820
2357
+ },
2358
+ {
2359
+ "epoch": 1.45,
2360
+ "learning_rate": 1e-05,
2361
+ "loss": 0.7483,
2362
+ "step": 3830
2363
+ },
2364
+ {
2365
+ "epoch": 1.45,
2366
+ "learning_rate": 1e-05,
2367
+ "loss": 0.7394,
2368
+ "step": 3840
2369
+ },
2370
+ {
2371
+ "epoch": 1.45,
2372
+ "learning_rate": 1e-05,
2373
+ "loss": 0.7231,
2374
+ "step": 3850
2375
+ },
2376
+ {
2377
+ "epoch": 1.46,
2378
+ "learning_rate": 1e-05,
2379
+ "loss": 0.7005,
2380
+ "step": 3860
2381
+ },
2382
+ {
2383
+ "epoch": 1.46,
2384
+ "learning_rate": 1e-05,
2385
+ "loss": 0.7177,
2386
+ "step": 3870
2387
+ },
2388
+ {
2389
+ "epoch": 1.46,
2390
+ "learning_rate": 1e-05,
2391
+ "loss": 0.7449,
2392
+ "step": 3880
2393
+ },
2394
+ {
2395
+ "epoch": 1.47,
2396
+ "learning_rate": 1e-05,
2397
+ "loss": 0.7093,
2398
+ "step": 3890
2399
+ },
2400
+ {
2401
+ "epoch": 1.47,
2402
+ "learning_rate": 1e-05,
2403
+ "loss": 0.7262,
2404
+ "step": 3900
2405
+ },
2406
+ {
2407
+ "epoch": 1.48,
2408
+ "learning_rate": 1e-05,
2409
+ "loss": 0.7039,
2410
+ "step": 3910
2411
+ },
2412
+ {
2413
+ "epoch": 1.48,
2414
+ "learning_rate": 1e-05,
2415
+ "loss": 0.7563,
2416
+ "step": 3920
2417
+ },
2418
+ {
2419
+ "epoch": 1.48,
2420
+ "learning_rate": 1e-05,
2421
+ "loss": 0.7385,
2422
+ "step": 3930
2423
+ },
2424
+ {
2425
+ "epoch": 1.49,
2426
+ "learning_rate": 1e-05,
2427
+ "loss": 0.7416,
2428
+ "step": 3940
2429
+ },
2430
+ {
2431
+ "epoch": 1.49,
2432
+ "learning_rate": 1e-05,
2433
+ "loss": 0.725,
2434
+ "step": 3950
2435
+ },
2436
+ {
2437
+ "epoch": 1.49,
2438
+ "learning_rate": 1e-05,
2439
+ "loss": 0.7311,
2440
+ "step": 3960
2441
+ },
2442
+ {
2443
+ "epoch": 1.5,
2444
+ "learning_rate": 1e-05,
2445
+ "loss": 0.7245,
2446
+ "step": 3970
2447
+ },
2448
+ {
2449
+ "epoch": 1.5,
2450
+ "learning_rate": 1e-05,
2451
+ "loss": 0.7307,
2452
+ "step": 3980
2453
+ },
2454
+ {
2455
+ "epoch": 1.51,
2456
+ "learning_rate": 1e-05,
2457
+ "loss": 0.7105,
2458
+ "step": 3990
2459
+ },
2460
+ {
2461
+ "epoch": 1.51,
2462
+ "learning_rate": 1e-05,
2463
+ "loss": 0.7363,
2464
+ "step": 4000
2465
+ },
2466
+ {
2467
+ "epoch": 1.51,
2468
+ "eval_loss": 0.93310546875,
2469
+ "eval_runtime": 63.3035,
2470
+ "eval_samples_per_second": 64.025,
2471
+ "eval_steps_per_second": 2.006,
2472
+ "step": 4000
2473
+ },
2474
+ {
2475
+ "epoch": 1.51,
2476
+ "learning_rate": 1e-05,
2477
+ "loss": 0.726,
2478
+ "step": 4010
2479
+ },
2480
+ {
2481
+ "epoch": 1.52,
2482
+ "learning_rate": 1e-05,
2483
+ "loss": 0.7045,
2484
+ "step": 4020
2485
+ },
2486
+ {
2487
+ "epoch": 1.52,
2488
+ "learning_rate": 1e-05,
2489
+ "loss": 0.7705,
2490
+ "step": 4030
2491
+ },
2492
+ {
2493
+ "epoch": 1.53,
2494
+ "learning_rate": 1e-05,
2495
+ "loss": 0.7275,
2496
+ "step": 4040
2497
+ },
2498
+ {
2499
+ "epoch": 1.53,
2500
+ "learning_rate": 1e-05,
2501
+ "loss": 0.7203,
2502
+ "step": 4050
2503
+ },
2504
+ {
2505
+ "epoch": 1.53,
2506
+ "learning_rate": 1e-05,
2507
+ "loss": 0.7247,
2508
+ "step": 4060
2509
+ },
2510
+ {
2511
+ "epoch": 1.54,
2512
+ "learning_rate": 1e-05,
2513
+ "loss": 0.7406,
2514
+ "step": 4070
2515
+ },
2516
+ {
2517
+ "epoch": 1.54,
2518
+ "learning_rate": 1e-05,
2519
+ "loss": 0.7232,
2520
+ "step": 4080
2521
+ },
2522
+ {
2523
+ "epoch": 1.54,
2524
+ "learning_rate": 1e-05,
2525
+ "loss": 0.7175,
2526
+ "step": 4090
2527
+ },
2528
+ {
2529
+ "epoch": 1.55,
2530
+ "learning_rate": 1e-05,
2531
+ "loss": 0.7253,
2532
+ "step": 4100
2533
+ },
2534
+ {
2535
+ "epoch": 1.55,
2536
+ "learning_rate": 1e-05,
2537
+ "loss": 0.711,
2538
+ "step": 4110
2539
+ },
2540
+ {
2541
+ "epoch": 1.56,
2542
+ "learning_rate": 1e-05,
2543
+ "loss": 0.7011,
2544
+ "step": 4120
2545
+ },
2546
+ {
2547
+ "epoch": 1.56,
2548
+ "learning_rate": 1e-05,
2549
+ "loss": 0.698,
2550
+ "step": 4130
2551
+ },
2552
+ {
2553
+ "epoch": 1.56,
2554
+ "learning_rate": 1e-05,
2555
+ "loss": 0.7018,
2556
+ "step": 4140
2557
+ },
2558
+ {
2559
+ "epoch": 1.57,
2560
+ "learning_rate": 1e-05,
2561
+ "loss": 0.7196,
2562
+ "step": 4150
2563
+ },
2564
+ {
2565
+ "epoch": 1.57,
2566
+ "learning_rate": 1e-05,
2567
+ "loss": 0.7241,
2568
+ "step": 4160
2569
+ },
2570
+ {
2571
+ "epoch": 1.57,
2572
+ "learning_rate": 1e-05,
2573
+ "loss": 0.7109,
2574
+ "step": 4170
2575
+ },
2576
+ {
2577
+ "epoch": 1.58,
2578
+ "learning_rate": 1e-05,
2579
+ "loss": 0.7066,
2580
+ "step": 4180
2581
+ },
2582
+ {
2583
+ "epoch": 1.58,
2584
+ "learning_rate": 1e-05,
2585
+ "loss": 0.7301,
2586
+ "step": 4190
2587
+ },
2588
+ {
2589
+ "epoch": 1.59,
2590
+ "learning_rate": 1e-05,
2591
+ "loss": 0.7094,
2592
+ "step": 4200
2593
+ },
2594
+ {
2595
+ "epoch": 1.59,
2596
+ "learning_rate": 1e-05,
2597
+ "loss": 0.7125,
2598
+ "step": 4210
2599
+ },
2600
+ {
2601
+ "epoch": 1.59,
2602
+ "learning_rate": 1e-05,
2603
+ "loss": 0.7067,
2604
+ "step": 4220
2605
+ },
2606
+ {
2607
+ "epoch": 1.6,
2608
+ "learning_rate": 1e-05,
2609
+ "loss": 0.7244,
2610
+ "step": 4230
2611
+ },
2612
+ {
2613
+ "epoch": 1.6,
2614
+ "learning_rate": 1e-05,
2615
+ "loss": 0.7316,
2616
+ "step": 4240
2617
+ },
2618
+ {
2619
+ "epoch": 1.6,
2620
+ "learning_rate": 1e-05,
2621
+ "loss": 0.7565,
2622
+ "step": 4250
2623
+ },
2624
+ {
2625
+ "epoch": 1.61,
2626
+ "learning_rate": 1e-05,
2627
+ "loss": 0.7009,
2628
+ "step": 4260
2629
+ },
2630
+ {
2631
+ "epoch": 1.61,
2632
+ "learning_rate": 1e-05,
2633
+ "loss": 0.71,
2634
+ "step": 4270
2635
+ },
2636
+ {
2637
+ "epoch": 1.62,
2638
+ "learning_rate": 1e-05,
2639
+ "loss": 0.706,
2640
+ "step": 4280
2641
+ },
2642
+ {
2643
+ "epoch": 1.62,
2644
+ "learning_rate": 1e-05,
2645
+ "loss": 0.688,
2646
+ "step": 4290
2647
+ },
2648
+ {
2649
+ "epoch": 1.62,
2650
+ "learning_rate": 1e-05,
2651
+ "loss": 0.7051,
2652
+ "step": 4300
2653
+ },
2654
+ {
2655
+ "epoch": 1.63,
2656
+ "learning_rate": 1e-05,
2657
+ "loss": 0.7253,
2658
+ "step": 4310
2659
+ },
2660
+ {
2661
+ "epoch": 1.63,
2662
+ "learning_rate": 1e-05,
2663
+ "loss": 0.7152,
2664
+ "step": 4320
2665
+ },
2666
+ {
2667
+ "epoch": 1.63,
2668
+ "learning_rate": 1e-05,
2669
+ "loss": 0.7489,
2670
+ "step": 4330
2671
+ },
2672
+ {
2673
+ "epoch": 1.64,
2674
+ "learning_rate": 1e-05,
2675
+ "loss": 0.7165,
2676
+ "step": 4340
2677
+ },
2678
+ {
2679
+ "epoch": 1.64,
2680
+ "learning_rate": 1e-05,
2681
+ "loss": 0.7563,
2682
+ "step": 4350
2683
+ },
2684
+ {
2685
+ "epoch": 1.65,
2686
+ "learning_rate": 1e-05,
2687
+ "loss": 0.7121,
2688
+ "step": 4360
2689
+ },
2690
+ {
2691
+ "epoch": 1.65,
2692
+ "learning_rate": 1e-05,
2693
+ "loss": 0.6982,
2694
+ "step": 4370
2695
+ },
2696
+ {
2697
+ "epoch": 1.65,
2698
+ "learning_rate": 1e-05,
2699
+ "loss": 0.7037,
2700
+ "step": 4380
2701
+ },
2702
+ {
2703
+ "epoch": 1.66,
2704
+ "learning_rate": 1e-05,
2705
+ "loss": 0.7059,
2706
+ "step": 4390
2707
+ },
2708
+ {
2709
+ "epoch": 1.66,
2710
+ "learning_rate": 1e-05,
2711
+ "loss": 0.7096,
2712
+ "step": 4400
2713
+ },
2714
+ {
2715
+ "epoch": 1.66,
2716
+ "learning_rate": 1e-05,
2717
+ "loss": 0.7103,
2718
+ "step": 4410
2719
+ },
2720
+ {
2721
+ "epoch": 1.67,
2722
+ "learning_rate": 1e-05,
2723
+ "loss": 0.7088,
2724
+ "step": 4420
2725
+ },
2726
+ {
2727
+ "epoch": 1.67,
2728
+ "learning_rate": 1e-05,
2729
+ "loss": 0.7573,
2730
+ "step": 4430
2731
+ },
2732
+ {
2733
+ "epoch": 1.68,
2734
+ "learning_rate": 1e-05,
2735
+ "loss": 0.7192,
2736
+ "step": 4440
2737
+ },
2738
+ {
2739
+ "epoch": 1.68,
2740
+ "learning_rate": 1e-05,
2741
+ "loss": 0.7296,
2742
+ "step": 4450
2743
+ },
2744
+ {
2745
+ "epoch": 1.68,
2746
+ "learning_rate": 1e-05,
2747
+ "loss": 0.7115,
2748
+ "step": 4460
2749
+ },
2750
+ {
2751
+ "epoch": 1.69,
2752
+ "learning_rate": 1e-05,
2753
+ "loss": 0.7153,
2754
+ "step": 4470
2755
+ },
2756
+ {
2757
+ "epoch": 1.69,
2758
+ "learning_rate": 1e-05,
2759
+ "loss": 0.6649,
2760
+ "step": 4480
2761
+ },
2762
+ {
2763
+ "epoch": 1.69,
2764
+ "learning_rate": 1e-05,
2765
+ "loss": 0.6822,
2766
+ "step": 4490
2767
+ },
2768
+ {
2769
+ "epoch": 1.7,
2770
+ "learning_rate": 1e-05,
2771
+ "loss": 0.7234,
2772
+ "step": 4500
2773
+ },
2774
+ {
2775
+ "epoch": 1.7,
2776
+ "eval_loss": 0.908203125,
2777
+ "eval_runtime": 63.3469,
2778
+ "eval_samples_per_second": 63.981,
2779
+ "eval_steps_per_second": 2.005,
2780
+ "step": 4500
2781
+ },
2782
+ {
2783
+ "epoch": 1.7,
2784
+ "learning_rate": 1e-05,
2785
+ "loss": 0.7053,
2786
+ "step": 4510
2787
+ },
2788
+ {
2789
+ "epoch": 1.71,
2790
+ "learning_rate": 1e-05,
2791
+ "loss": 0.7222,
2792
+ "step": 4520
2793
+ },
2794
+ {
2795
+ "epoch": 1.71,
2796
+ "learning_rate": 1e-05,
2797
+ "loss": 0.6748,
2798
+ "step": 4530
2799
+ },
2800
+ {
2801
+ "epoch": 1.71,
2802
+ "learning_rate": 1e-05,
2803
+ "loss": 0.7166,
2804
+ "step": 4540
2805
+ },
2806
+ {
2807
+ "epoch": 1.72,
2808
+ "learning_rate": 1e-05,
2809
+ "loss": 0.7074,
2810
+ "step": 4550
2811
+ },
2812
+ {
2813
+ "epoch": 1.72,
2814
+ "learning_rate": 1e-05,
2815
+ "loss": 0.7214,
2816
+ "step": 4560
2817
+ },
2818
+ {
2819
+ "epoch": 1.73,
2820
+ "learning_rate": 1e-05,
2821
+ "loss": 0.6938,
2822
+ "step": 4570
2823
+ },
2824
+ {
2825
+ "epoch": 1.73,
2826
+ "learning_rate": 1e-05,
2827
+ "loss": 0.7237,
2828
+ "step": 4580
2829
+ },
2830
+ {
2831
+ "epoch": 1.73,
2832
+ "learning_rate": 1e-05,
2833
+ "loss": 0.6958,
2834
+ "step": 4590
2835
+ },
2836
+ {
2837
+ "epoch": 1.74,
2838
+ "learning_rate": 1e-05,
2839
+ "loss": 0.6849,
2840
+ "step": 4600
2841
+ },
2842
+ {
2843
+ "epoch": 1.74,
2844
+ "learning_rate": 1e-05,
2845
+ "loss": 0.6959,
2846
+ "step": 4610
2847
+ },
2848
+ {
2849
+ "epoch": 1.74,
2850
+ "learning_rate": 1e-05,
2851
+ "loss": 0.7126,
2852
+ "step": 4620
2853
+ },
2854
+ {
2855
+ "epoch": 1.75,
2856
+ "learning_rate": 1e-05,
2857
+ "loss": 0.7266,
2858
+ "step": 4630
2859
+ },
2860
+ {
2861
+ "epoch": 1.75,
2862
+ "learning_rate": 1e-05,
2863
+ "loss": 0.6683,
2864
+ "step": 4640
2865
+ },
2866
+ {
2867
+ "epoch": 1.76,
2868
+ "learning_rate": 1e-05,
2869
+ "loss": 0.7248,
2870
+ "step": 4650
2871
+ },
2872
+ {
2873
+ "epoch": 1.76,
2874
+ "learning_rate": 1e-05,
2875
+ "loss": 0.7117,
2876
+ "step": 4660
2877
+ },
2878
+ {
2879
+ "epoch": 1.76,
2880
+ "learning_rate": 1e-05,
2881
+ "loss": 0.6998,
2882
+ "step": 4670
2883
+ },
2884
+ {
2885
+ "epoch": 1.77,
2886
+ "learning_rate": 1e-05,
2887
+ "loss": 0.6986,
2888
+ "step": 4680
2889
+ },
2890
+ {
2891
+ "epoch": 1.77,
2892
+ "learning_rate": 1e-05,
2893
+ "loss": 0.6631,
2894
+ "step": 4690
2895
+ },
2896
+ {
2897
+ "epoch": 1.77,
2898
+ "learning_rate": 1e-05,
2899
+ "loss": 0.6857,
2900
+ "step": 4700
2901
+ },
2902
+ {
2903
+ "epoch": 1.78,
2904
+ "learning_rate": 1e-05,
2905
+ "loss": 0.6649,
2906
+ "step": 4710
2907
+ },
2908
+ {
2909
+ "epoch": 1.78,
2910
+ "learning_rate": 1e-05,
2911
+ "loss": 0.699,
2912
+ "step": 4720
2913
+ },
2914
+ {
2915
+ "epoch": 1.79,
2916
+ "learning_rate": 1e-05,
2917
+ "loss": 0.6991,
2918
+ "step": 4730
2919
+ },
2920
+ {
2921
+ "epoch": 1.79,
2922
+ "learning_rate": 1e-05,
2923
+ "loss": 0.703,
2924
+ "step": 4740
2925
+ },
2926
+ {
2927
+ "epoch": 1.79,
2928
+ "learning_rate": 1e-05,
2929
+ "loss": 0.7259,
2930
+ "step": 4750
2931
+ },
2932
+ {
2933
+ "epoch": 1.8,
2934
+ "learning_rate": 1e-05,
2935
+ "loss": 0.7023,
2936
+ "step": 4760
2937
+ },
2938
+ {
2939
+ "epoch": 1.8,
2940
+ "learning_rate": 1e-05,
2941
+ "loss": 0.7281,
2942
+ "step": 4770
2943
+ },
2944
+ {
2945
+ "epoch": 1.8,
2946
+ "learning_rate": 1e-05,
2947
+ "loss": 0.7213,
2948
+ "step": 4780
2949
+ },
2950
+ {
2951
+ "epoch": 1.81,
2952
+ "learning_rate": 1e-05,
2953
+ "loss": 0.6808,
2954
+ "step": 4790
2955
+ },
2956
+ {
2957
+ "epoch": 1.81,
2958
+ "learning_rate": 1e-05,
2959
+ "loss": 0.7253,
2960
+ "step": 4800
2961
+ },
2962
+ {
2963
+ "epoch": 1.82,
2964
+ "learning_rate": 1e-05,
2965
+ "loss": 0.7199,
2966
+ "step": 4810
2967
+ },
2968
+ {
2969
+ "epoch": 1.82,
2970
+ "learning_rate": 1e-05,
2971
+ "loss": 0.745,
2972
+ "step": 4820
2973
+ },
2974
+ {
2975
+ "epoch": 1.82,
2976
+ "learning_rate": 1e-05,
2977
+ "loss": 0.6928,
2978
+ "step": 4830
2979
+ },
2980
+ {
2981
+ "epoch": 1.83,
2982
+ "learning_rate": 1e-05,
2983
+ "loss": 0.6942,
2984
+ "step": 4840
2985
+ },
2986
+ {
2987
+ "epoch": 1.83,
2988
+ "learning_rate": 1e-05,
2989
+ "loss": 0.6963,
2990
+ "step": 4850
2991
+ },
2992
+ {
2993
+ "epoch": 1.83,
2994
+ "learning_rate": 1e-05,
2995
+ "loss": 0.7025,
2996
+ "step": 4860
2997
+ },
2998
+ {
2999
+ "epoch": 1.84,
3000
+ "learning_rate": 1e-05,
3001
+ "loss": 0.6947,
3002
+ "step": 4870
3003
+ },
3004
+ {
3005
+ "epoch": 1.84,
3006
+ "learning_rate": 1e-05,
3007
+ "loss": 0.6959,
3008
+ "step": 4880
3009
+ },
3010
+ {
3011
+ "epoch": 1.85,
3012
+ "learning_rate": 1e-05,
3013
+ "loss": 0.6984,
3014
+ "step": 4890
3015
+ },
3016
+ {
3017
+ "epoch": 1.85,
3018
+ "learning_rate": 1e-05,
3019
+ "loss": 0.7351,
3020
+ "step": 4900
3021
+ },
3022
+ {
3023
+ "epoch": 1.85,
3024
+ "learning_rate": 1e-05,
3025
+ "loss": 0.7095,
3026
+ "step": 4910
3027
+ },
3028
+ {
3029
+ "epoch": 1.86,
3030
+ "learning_rate": 1e-05,
3031
+ "loss": 0.6915,
3032
+ "step": 4920
3033
+ },
3034
+ {
3035
+ "epoch": 1.86,
3036
+ "learning_rate": 1e-05,
3037
+ "loss": 0.694,
3038
+ "step": 4930
3039
+ },
3040
+ {
3041
+ "epoch": 1.86,
3042
+ "learning_rate": 1e-05,
3043
+ "loss": 0.6907,
3044
+ "step": 4940
3045
+ },
3046
+ {
3047
+ "epoch": 1.87,
3048
+ "learning_rate": 1e-05,
3049
+ "loss": 0.7157,
3050
+ "step": 4950
3051
+ },
3052
+ {
3053
+ "epoch": 1.87,
3054
+ "learning_rate": 1e-05,
3055
+ "loss": 0.6906,
3056
+ "step": 4960
3057
+ },
3058
+ {
3059
+ "epoch": 1.88,
3060
+ "learning_rate": 1e-05,
3061
+ "loss": 0.675,
3062
+ "step": 4970
3063
+ },
3064
+ {
3065
+ "epoch": 1.88,
3066
+ "learning_rate": 1e-05,
3067
+ "loss": 0.6764,
3068
+ "step": 4980
3069
+ },
3070
+ {
3071
+ "epoch": 1.88,
3072
+ "learning_rate": 1e-05,
3073
+ "loss": 0.6821,
3074
+ "step": 4990
3075
+ },
3076
+ {
3077
+ "epoch": 1.89,
3078
+ "learning_rate": 1e-05,
3079
+ "loss": 0.7015,
3080
+ "step": 5000
3081
+ },
3082
+ {
3083
+ "epoch": 1.89,
3084
+ "eval_loss": 0.890625,
3085
+ "eval_runtime": 63.2179,
3086
+ "eval_samples_per_second": 64.112,
3087
+ "eval_steps_per_second": 2.009,
3088
+ "step": 5000
3089
+ }
3090
+ ],
3091
+ "max_steps": 26490,
3092
+ "num_train_epochs": 10,
3093
+ "total_flos": 851208285061120.0,
3094
+ "trial_name": null,
3095
+ "trial_params": null
3096
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:918b90617c1ee72624f25a530d44250c3c4b74ca71b34d9874c93debc1b7cbe3
3
+ size 6255
zero_to_fp32.py ADDED
@@ -0,0 +1,578 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example: python zero_to_fp32.py . pytorch_model.bin
14
+
15
+ import argparse
16
+ import torch
17
+ import glob
18
+ import math
19
+ import os
20
+ import re
21
+ from collections import OrderedDict
22
+ from dataclasses import dataclass
23
+
24
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
25
+ # DeepSpeed data structures it has to be available in the current python environment.
26
+ from deepspeed.utils import logger
27
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
28
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
29
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
30
+
31
+
32
+ @dataclass
33
+ class zero_model_state:
34
+ buffers: dict()
35
+ param_shapes: dict()
36
+ shared_params: list
37
+ ds_version: int
38
+ frozen_param_shapes: dict()
39
+ frozen_param_fragments: dict()
40
+
41
+
42
+ debug = 0
43
+
44
+ # load to cpu
45
+ device = torch.device('cpu')
46
+
47
+
48
+ def atoi(text):
49
+ return int(text) if text.isdigit() else text
50
+
51
+
52
+ def natural_keys(text):
53
+ '''
54
+ alist.sort(key=natural_keys) sorts in human order
55
+ http://nedbatchelder.com/blog/200712/human_sorting.html
56
+ (See Toothy's implementation in the comments)
57
+ '''
58
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
59
+
60
+
61
+ def get_model_state_file(checkpoint_dir, zero_stage):
62
+ if not os.path.isdir(checkpoint_dir):
63
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
64
+
65
+ # there should be only one file
66
+ if zero_stage == 2:
67
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
68
+ elif zero_stage == 3:
69
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
70
+
71
+ if not os.path.exists(file):
72
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
73
+
74
+ return file
75
+
76
+
77
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
78
+ # XXX: need to test that this simple glob rule works for multi-node setup too
79
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
80
+
81
+ if len(ckpt_files) == 0:
82
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
83
+
84
+ return ckpt_files
85
+
86
+
87
+ def get_optim_files(checkpoint_dir):
88
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
89
+
90
+
91
+ def get_model_state_files(checkpoint_dir):
92
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
93
+
94
+
95
+ def parse_model_states(files):
96
+ zero_model_states = []
97
+ for file in files:
98
+ state_dict = torch.load(file, map_location=device)
99
+
100
+ if BUFFER_NAMES not in state_dict:
101
+ raise ValueError(f"{file} is not a model state checkpoint")
102
+ buffer_names = state_dict[BUFFER_NAMES]
103
+ if debug:
104
+ print("Found buffers:", buffer_names)
105
+
106
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
107
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
108
+ param_shapes = state_dict[PARAM_SHAPES]
109
+
110
+ # collect parameters that are included in param_shapes
111
+ param_names = []
112
+ for s in param_shapes:
113
+ for name in s.keys():
114
+ param_names.append(name)
115
+
116
+ # update with frozen parameters
117
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
118
+ if frozen_param_shapes is not None:
119
+ if debug:
120
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
121
+ param_names += list(frozen_param_shapes.keys())
122
+
123
+ # handle shared params
124
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
125
+
126
+ ds_version = state_dict.get(DS_VERSION, None)
127
+
128
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
129
+
130
+ z_model_state = zero_model_state(buffers=buffers,
131
+ param_shapes=param_shapes,
132
+ shared_params=shared_params,
133
+ ds_version=ds_version,
134
+ frozen_param_shapes=frozen_param_shapes,
135
+ frozen_param_fragments=frozen_param_fragments)
136
+ zero_model_states.append(z_model_state)
137
+
138
+ return zero_model_states
139
+
140
+
141
+ def parse_optim_states(files, ds_checkpoint_dir):
142
+
143
+ total_files = len(files)
144
+ state_dicts = []
145
+ for f in files:
146
+ state_dicts.append(torch.load(f, map_location=device))
147
+
148
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
149
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
150
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
151
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
152
+
153
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
154
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
155
+ # use the max of the partition_count to get the dp world_size.
156
+
157
+ if type(world_size) is list:
158
+ world_size = max(world_size)
159
+
160
+ if world_size != total_files:
161
+ raise ValueError(
162
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
163
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
164
+ )
165
+
166
+ # the groups are named differently in each stage
167
+ if zero_stage == 2:
168
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
169
+ elif zero_stage == 3:
170
+ fp32_groups_key = FP32_FLAT_GROUPS
171
+ else:
172
+ raise ValueError(f"unknown zero stage {zero_stage}")
173
+
174
+ if zero_stage == 2:
175
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
176
+ elif zero_stage == 3:
177
+ # if there is more than one param group, there will be multiple flattened tensors - one
178
+ # flattened tensor per group - for simplicity merge them into a single tensor
179
+ #
180
+ # XXX: could make the script more memory efficient for when there are multiple groups - it
181
+ # will require matching the sub-lists of param_shapes for each param group flattened tensor
182
+
183
+ fp32_flat_groups = [
184
+ torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
185
+ ]
186
+
187
+ return zero_stage, world_size, fp32_flat_groups
188
+
189
+
190
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
191
+ """
192
+ Returns fp32 state_dict reconstructed from ds checkpoint
193
+
194
+ Args:
195
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
196
+
197
+ """
198
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
199
+
200
+ optim_files = get_optim_files(ds_checkpoint_dir)
201
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
202
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
203
+
204
+ model_files = get_model_state_files(ds_checkpoint_dir)
205
+
206
+ zero_model_states = parse_model_states(model_files)
207
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
208
+
209
+ if zero_stage == 2:
210
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
211
+ elif zero_stage == 3:
212
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
248
+ param_shapes = zero_model_states[0].param_shapes
249
+
250
+ # Reconstruction protocol:
251
+ #
252
+ # XXX: document this
253
+
254
+ if debug:
255
+ for i in range(world_size):
256
+ for j in range(len(fp32_flat_groups[0])):
257
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
258
+
259
+ # XXX: memory usage doubles here (zero2)
260
+ num_param_groups = len(fp32_flat_groups[0])
261
+ merged_single_partition_of_fp32_groups = []
262
+ for i in range(num_param_groups):
263
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
264
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
265
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
266
+ avail_numel = sum(
267
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
268
+
269
+ if debug:
270
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
271
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
272
+ # not asserting if there is a mismatch due to possible padding
273
+ print(f"Have {avail_numel} numels to process.")
274
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
275
+
276
+ # params
277
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
278
+ # out-of-core computing solution
279
+ total_numel = 0
280
+ total_params = 0
281
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
282
+ offset = 0
283
+ avail_numel = full_single_fp32_vector.numel()
284
+ for name, shape in shapes.items():
285
+
286
+ unpartitioned_numel = shape.numel()
287
+ total_numel += unpartitioned_numel
288
+ total_params += 1
289
+
290
+ if debug:
291
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
292
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
293
+ offset += unpartitioned_numel
294
+
295
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
296
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
297
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
298
+ # live optimizer object, so we are checking that the numbers are within the right range
299
+ align_to = 2 * world_size
300
+
301
+ def zero2_align(x):
302
+ return align_to * math.ceil(x / align_to)
303
+
304
+ if debug:
305
+ print(f"original offset={offset}, avail_numel={avail_numel}")
306
+
307
+ offset = zero2_align(offset)
308
+ avail_numel = zero2_align(avail_numel)
309
+
310
+ if debug:
311
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
312
+
313
+ # Sanity check
314
+ if offset != avail_numel:
315
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
316
+
317
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
318
+
319
+
320
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
321
+ state_dict = OrderedDict()
322
+
323
+ # buffers
324
+ buffers = zero_model_states[0].buffers
325
+ state_dict.update(buffers)
326
+ if debug:
327
+ print(f"added {len(buffers)} buffers")
328
+
329
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
330
+
331
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
332
+
333
+ # recover shared parameters
334
+ for pair in zero_model_states[0].shared_params:
335
+ if pair[1] in state_dict:
336
+ state_dict[pair[0]] = state_dict[pair[1]]
337
+
338
+ return state_dict
339
+
340
+
341
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
342
+ remainder = unpartitioned_numel % world_size
343
+ padding_numel = (world_size - remainder) if remainder else 0
344
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
345
+ return partitioned_numel, padding_numel
346
+
347
+
348
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
349
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
350
+ return
351
+
352
+ if debug:
353
+ for i in range(world_size):
354
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
355
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
356
+
357
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
358
+ wanted_params = len(frozen_param_shapes)
359
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
360
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
361
+ print(f'Frozen params: Have {avail_numel} numels to process.')
362
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
363
+
364
+ total_params = 0
365
+ total_numel = 0
366
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
367
+ total_params += 1
368
+ unpartitioned_numel = shape.numel()
369
+ total_numel += unpartitioned_numel
370
+
371
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
372
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
373
+
374
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
375
+
376
+ if debug:
377
+ print(
378
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
379
+ )
380
+
381
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
382
+
383
+
384
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
385
+ param_shapes = zero_model_states[0].param_shapes
386
+ avail_numel = fp32_flat_groups[0].numel() * world_size
387
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
388
+ # param, re-consolidating each param, while dealing with padding if any
389
+
390
+ # merge list of dicts, preserving order
391
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
392
+
393
+ if debug:
394
+ for i in range(world_size):
395
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
396
+
397
+ wanted_params = len(param_shapes)
398
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
399
+ # not asserting if there is a mismatch due to possible padding
400
+ avail_numel = fp32_flat_groups[0].numel() * world_size
401
+ print(f"Trainable params: Have {avail_numel} numels to process.")
402
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
403
+
404
+ # params
405
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
406
+ # out-of-core computing solution
407
+ offset = 0
408
+ total_numel = 0
409
+ total_params = 0
410
+ for name, shape in param_shapes.items():
411
+
412
+ unpartitioned_numel = shape.numel()
413
+ total_numel += unpartitioned_numel
414
+ total_params += 1
415
+
416
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
417
+
418
+ if debug:
419
+ print(
420
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
421
+ )
422
+
423
+ # XXX: memory usage doubles here
424
+ state_dict[name] = torch.cat(
425
+ tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
426
+ 0).narrow(0, 0, unpartitioned_numel).view(shape)
427
+ offset += partitioned_numel
428
+
429
+ offset *= world_size
430
+
431
+ # Sanity check
432
+ if offset != avail_numel:
433
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
434
+
435
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
436
+
437
+
438
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
439
+ state_dict = OrderedDict()
440
+
441
+ # buffers
442
+ buffers = zero_model_states[0].buffers
443
+ state_dict.update(buffers)
444
+ if debug:
445
+ print(f"added {len(buffers)} buffers")
446
+
447
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
448
+
449
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
450
+
451
+ # recover shared parameters
452
+ for pair in zero_model_states[0].shared_params:
453
+ if pair[1] in state_dict:
454
+ state_dict[pair[0]] = state_dict[pair[1]]
455
+
456
+ return state_dict
457
+
458
+
459
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
460
+ """
461
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
462
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
463
+ via a model hub.
464
+
465
+ Args:
466
+ - ``checkpoint_dir``: path to the desired checkpoint folder
467
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
468
+
469
+ Returns:
470
+ - pytorch ``state_dict``
471
+
472
+ Note: this approach may not work if your application doesn't have sufficient free CPU memory and
473
+ you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
474
+ the checkpoint.
475
+
476
+ A typical usage might be ::
477
+
478
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
479
+ # do the training and checkpoint saving
480
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
481
+ model = model.cpu() # move to cpu
482
+ model.load_state_dict(state_dict)
483
+ # submit to model hub or save the model to share with others
484
+
485
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
486
+ application. i.e. you will need to re-initialize the deepspeed engine, since
487
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
488
+
489
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
490
+
491
+ """
492
+ if tag is None:
493
+ latest_path = os.path.join(checkpoint_dir, 'latest')
494
+ if os.path.isfile(latest_path):
495
+ with open(latest_path, 'r') as fd:
496
+ tag = fd.read().strip()
497
+ else:
498
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
499
+
500
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
501
+
502
+ if not os.path.isdir(ds_checkpoint_dir):
503
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
504
+
505
+ return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
506
+
507
+
508
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
509
+ """
510
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
511
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
512
+
513
+ Args:
514
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
515
+ - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
516
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
517
+ """
518
+
519
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
520
+ print(f"Saving fp32 state dict to {output_file}")
521
+ torch.save(state_dict, output_file)
522
+
523
+
524
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
525
+ """
526
+ 1. Put the provided model to cpu
527
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
528
+ 3. Load it into the provided model
529
+
530
+ Args:
531
+ - ``model``: the model object to update
532
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
533
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
534
+
535
+ Returns:
536
+ - ``model`: modified model
537
+
538
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
539
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
540
+ conveniently placed for you in the checkpoint folder.
541
+
542
+ A typical usage might be ::
543
+
544
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
545
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
546
+ # submit to model hub or save the model to share with others
547
+
548
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
549
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
550
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
551
+
552
+ """
553
+ logger.info(f"Extracting fp32 weights")
554
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
555
+
556
+ logger.info(f"Overwriting model with fp32 weights")
557
+ model = model.cpu()
558
+ model.load_state_dict(state_dict, strict=False)
559
+
560
+ return model
561
+
562
+
563
+ if __name__ == "__main__":
564
+
565
+ parser = argparse.ArgumentParser()
566
+ parser.add_argument("checkpoint_dir",
567
+ type=str,
568
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
569
+ parser.add_argument(
570
+ "output_file",
571
+ type=str,
572
+ help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
573
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
574
+ args = parser.parse_args()
575
+
576
+ debug = args.debug
577
+
578
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)