Hugging Face's logo

BrownianNotion
/
TinyLlama_v1.1_2bit_int_7B_data

TensorBoard
Safetensors
English
llama
Model card Files
xet
Metrics Community
BrownianNotion commited on
Commit
168ecc9
·
verified ·
1 Parent(s): 285b628

Add files using upload-large-folder tool

Browse files
Files changed (14) hide show
  1. added_tokens.json +3 -0
  2. config.json +28 -0
  3. generation_config.json +7 -0
  4. global_step400/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt +3 -0
  5. global_step400/mp_rank_00_model_states.pt +3 -0
  6. latest +1 -0
  7. model.safetensors +3 -0
  8. rng_state.pth +3 -0
  9. special_tokens_map.json +30 -0
  10. tokenizer.model +3 -0
  11. tokenizer_config.json +50 -0
  12. trainer_state.json +2501 -0
  13. training_args.bin +3 -0
  14. zero_to_fp32.py +760 -0
added_tokens.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {
2
+ "[PAD]": 32000
3
+ }
config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "../models/TinyLlama_v1.1/",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 5632,
14
+ "max_position_embeddings": 2048,
15
+ "model_type": "llama",
16
+ "num_attention_heads": 32,
17
+ "num_hidden_layers": 22,
18
+ "num_key_value_heads": 4,
19
+ "pretraining_tp": 1,
20
+ "rms_norm_eps": 1e-05,
21
+ "rope_scaling": null,
22
+ "rope_theta": 10000.0,
23
+ "tie_word_embeddings": false,
24
+ "torch_dtype": "bfloat16",
25
+ "transformers_version": "4.37.0",
26
+ "use_cache": true,
27
+ "vocab_size": 32001
28
+ }
generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 1,
3
+ "eos_token_id": 2,
4
+ "max_length": 2048,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.37.0"
7
+ }
global_step400/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cf0d06739eae5df8f5546aad4ae45e3cda4f342c7154ca8b372fbf1cf292bece
3
+ size 13200649212
global_step400/mp_rank_00_model_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dcdde7fea170e811adab279b0d9e1c7335fdb6c58472a953a664dfaa1e509062
3
+ size 2200176684
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step400
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aaf88a72af91ab5609aedeef108c9ce705f84be9b6b52ca9fb1d9b499654d970
3
+ size 2200128056
rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04b1b6c50b6817a190dc1417f8235d48bfa5c55010e2a039976c61e3a2670965
3
+ size 14244
special_tokens_map.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": {
3
+ "content": "<s>",
4
+ "lstrip": false,
5
+ "normalized": false,
6
+ "rstrip": false,
7
+ "single_word": false
8
+ },
9
+ "eos_token": {
10
+ "content": "</s>",
11
+ "lstrip": false,
12
+ "normalized": false,
13
+ "rstrip": false,
14
+ "single_word": false
15
+ },
16
+ "pad_token": {
17
+ "content": "[PAD]",
18
+ "lstrip": false,
19
+ "normalized": false,
20
+ "rstrip": false,
21
+ "single_word": false
22
+ },
23
+ "unk_token": {
24
+ "content": "<unk>",
25
+ "lstrip": false,
26
+ "normalized": false,
27
+ "rstrip": false,
28
+ "single_word": false
29
+ }
30
+ }
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
3
+ size 499723
tokenizer_config.json ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "added_tokens_decoder": {
5
+ "0": {
6
+ "content": "<unk>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "1": {
14
+ "content": "<s>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "2": {
22
+ "content": "</s>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "32000": {
30
+ "content": "[PAD]",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ }
37
+ },
38
+ "bos_token": "<s>",
39
+ "clean_up_tokenization_spaces": false,
40
+ "eos_token": "</s>",
41
+ "legacy": false,
42
+ "model_max_length": 1024,
43
+ "pad_token": "[PAD]",
44
+ "padding_side": "right",
45
+ "sp_model_kwargs": {},
46
+ "spaces_between_special_tokens": false,
47
+ "tokenizer_class": "LlamaTokenizer",
48
+ "unk_token": "<unk>",
49
+ "use_default_system_prompt": false
50
+ }
trainer_state.json ADDED
@@ -0,0 +1,2501 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 206.38494873046875,
3
+ "best_model_checkpoint": "./ckpts/tinyllama_v1.1/int2-g128/checkpoint-400",
4
+ "epoch": 4.0,
5
+ "eval_steps": 40,
6
+ "global_step": 400,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.01,
13
+ "learning_rate": 0.0,
14
+ "loss": 5217.7002,
15
+ "step": 1
16
+ },
17
+ {
18
+ "epoch": 0.02,
19
+ "learning_rate": 2e-05,
20
+ "loss": 5011.7588,
21
+ "step": 2
22
+ },
23
+ {
24
+ "epoch": 0.03,
25
+ "learning_rate": 2e-05,
26
+ "loss": 4812.8281,
27
+ "step": 3
28
+ },
29
+ {
30
+ "epoch": 0.04,
31
+ "learning_rate": 2e-05,
32
+ "loss": 1860.1274,
33
+ "step": 4
34
+ },
35
+ {
36
+ "epoch": 0.05,
37
+ "learning_rate": 2e-05,
38
+ "loss": 1713.5734,
39
+ "step": 5
40
+ },
41
+ {
42
+ "epoch": 0.06,
43
+ "learning_rate": 2e-05,
44
+ "loss": 1103.2999,
45
+ "step": 6
46
+ },
47
+ {
48
+ "epoch": 0.07,
49
+ "learning_rate": 2e-05,
50
+ "loss": 675.6068,
51
+ "step": 7
52
+ },
53
+ {
54
+ "epoch": 0.08,
55
+ "learning_rate": 2e-05,
56
+ "loss": 683.2965,
57
+ "step": 8
58
+ },
59
+ {
60
+ "epoch": 0.09,
61
+ "learning_rate": 2e-05,
62
+ "loss": 734.9794,
63
+ "step": 9
64
+ },
65
+ {
66
+ "epoch": 0.1,
67
+ "learning_rate": 2e-05,
68
+ "loss": 585.422,
69
+ "step": 10
70
+ },
71
+ {
72
+ "epoch": 0.11,
73
+ "learning_rate": 2e-05,
74
+ "loss": 570.1306,
75
+ "step": 11
76
+ },
77
+ {
78
+ "epoch": 0.12,
79
+ "learning_rate": 2e-05,
80
+ "loss": 599.2966,
81
+ "step": 12
82
+ },
83
+ {
84
+ "epoch": 0.13,
85
+ "learning_rate": 2e-05,
86
+ "loss": 574.2372,
87
+ "step": 13
88
+ },
89
+ {
90
+ "epoch": 0.14,
91
+ "learning_rate": 2e-05,
92
+ "loss": 481.7502,
93
+ "step": 14
94
+ },
95
+ {
96
+ "epoch": 0.15,
97
+ "learning_rate": 2e-05,
98
+ "loss": 531.738,
99
+ "step": 15
100
+ },
101
+ {
102
+ "epoch": 0.16,
103
+ "learning_rate": 2e-05,
104
+ "loss": 448.2928,
105
+ "step": 16
106
+ },
107
+ {
108
+ "epoch": 0.17,
109
+ "learning_rate": 2e-05,
110
+ "loss": 490.8231,
111
+ "step": 17
112
+ },
113
+ {
114
+ "epoch": 0.18,
115
+ "learning_rate": 2e-05,
116
+ "loss": 448.1083,
117
+ "step": 18
118
+ },
119
+ {
120
+ "epoch": 0.19,
121
+ "learning_rate": 2e-05,
122
+ "loss": 506.6299,
123
+ "step": 19
124
+ },
125
+ {
126
+ "epoch": 0.2,
127
+ "learning_rate": 2e-05,
128
+ "loss": 432.8912,
129
+ "step": 20
130
+ },
131
+ {
132
+ "epoch": 0.21,
133
+ "learning_rate": 2e-05,
134
+ "loss": 449.625,
135
+ "step": 21
136
+ },
137
+ {
138
+ "epoch": 0.22,
139
+ "learning_rate": 2e-05,
140
+ "loss": 588.6897,
141
+ "step": 22
142
+ },
143
+ {
144
+ "epoch": 0.23,
145
+ "learning_rate": 2e-05,
146
+ "loss": 405.68,
147
+ "step": 23
148
+ },
149
+ {
150
+ "epoch": 0.24,
151
+ "learning_rate": 2e-05,
152
+ "loss": 516.0303,
153
+ "step": 24
154
+ },
155
+ {
156
+ "epoch": 0.25,
157
+ "learning_rate": 2e-05,
158
+ "loss": 401.7353,
159
+ "step": 25
160
+ },
161
+ {
162
+ "epoch": 0.26,
163
+ "learning_rate": 2e-05,
164
+ "loss": 399.9238,
165
+ "step": 26
166
+ },
167
+ {
168
+ "epoch": 0.27,
169
+ "learning_rate": 2e-05,
170
+ "loss": 351.5388,
171
+ "step": 27
172
+ },
173
+ {
174
+ "epoch": 0.28,
175
+ "learning_rate": 2e-05,
176
+ "loss": 401.7846,
177
+ "step": 28
178
+ },
179
+ {
180
+ "epoch": 0.29,
181
+ "learning_rate": 2e-05,
182
+ "loss": 385.5488,
183
+ "step": 29
184
+ },
185
+ {
186
+ "epoch": 0.3,
187
+ "learning_rate": 2e-05,
188
+ "loss": 378.391,
189
+ "step": 30
190
+ },
191
+ {
192
+ "epoch": 0.31,
193
+ "learning_rate": 2e-05,
194
+ "loss": 446.2048,
195
+ "step": 31
196
+ },
197
+ {
198
+ "epoch": 0.32,
199
+ "learning_rate": 2e-05,
200
+ "loss": 354.5841,
201
+ "step": 32
202
+ },
203
+ {
204
+ "epoch": 0.33,
205
+ "learning_rate": 2e-05,
206
+ "loss": 370.7014,
207
+ "step": 33
208
+ },
209
+ {
210
+ "epoch": 0.34,
211
+ "learning_rate": 2e-05,
212
+ "loss": 397.8215,
213
+ "step": 34
214
+ },
215
+ {
216
+ "epoch": 0.35,
217
+ "learning_rate": 2e-05,
218
+ "loss": 400.33,
219
+ "step": 35
220
+ },
221
+ {
222
+ "epoch": 0.36,
223
+ "learning_rate": 2e-05,
224
+ "loss": 379.8638,
225
+ "step": 36
226
+ },
227
+ {
228
+ "epoch": 0.37,
229
+ "learning_rate": 2e-05,
230
+ "loss": 282.1729,
231
+ "step": 37
232
+ },
233
+ {
234
+ "epoch": 0.38,
235
+ "learning_rate": 2e-05,
236
+ "loss": 288.713,
237
+ "step": 38
238
+ },
239
+ {
240
+ "epoch": 0.39,
241
+ "learning_rate": 2e-05,
242
+ "loss": 362.3285,
243
+ "step": 39
244
+ },
245
+ {
246
+ "epoch": 0.4,
247
+ "learning_rate": 2e-05,
248
+ "loss": 345.252,
249
+ "step": 40
250
+ },
251
+ {
252
+ "epoch": 0.4,
253
+ "eval_loss": 349.94061279296875,
254
+ "eval_runtime": 91.4064,
255
+ "eval_samples_per_second": 17.504,
256
+ "eval_steps_per_second": 1.094,
257
+ "step": 40
258
+ },
259
+ {
260
+ "epoch": 0.41,
261
+ "learning_rate": 2e-05,
262
+ "loss": 391.5078,
263
+ "step": 41
264
+ },
265
+ {
266
+ "epoch": 0.42,
267
+ "learning_rate": 2e-05,
268
+ "loss": 332.1484,
269
+ "step": 42
270
+ },
271
+ {
272
+ "epoch": 0.43,
273
+ "learning_rate": 2e-05,
274
+ "loss": 352.9743,
275
+ "step": 43
276
+ },
277
+ {
278
+ "epoch": 0.44,
279
+ "learning_rate": 2e-05,
280
+ "loss": 314.8037,
281
+ "step": 44
282
+ },
283
+ {
284
+ "epoch": 0.45,
285
+ "learning_rate": 2e-05,
286
+ "loss": 386.3977,
287
+ "step": 45
288
+ },
289
+ {
290
+ "epoch": 0.46,
291
+ "learning_rate": 2e-05,
292
+ "loss": 359.9244,
293
+ "step": 46
294
+ },
295
+ {
296
+ "epoch": 0.47,
297
+ "learning_rate": 2e-05,
298
+ "loss": 376.9478,
299
+ "step": 47
300
+ },
301
+ {
302
+ "epoch": 0.48,
303
+ "learning_rate": 2e-05,
304
+ "loss": 307.694,
305
+ "step": 48
306
+ },
307
+ {
308
+ "epoch": 0.49,
309
+ "learning_rate": 2e-05,
310
+ "loss": 359.4525,
311
+ "step": 49
312
+ },
313
+ {
314
+ "epoch": 0.5,
315
+ "learning_rate": 2e-05,
316
+ "loss": 319.51,
317
+ "step": 50
318
+ },
319
+ {
320
+ "epoch": 0.51,
321
+ "learning_rate": 2e-05,
322
+ "loss": 349.2659,
323
+ "step": 51
324
+ },
325
+ {
326
+ "epoch": 0.52,
327
+ "learning_rate": 2e-05,
328
+ "loss": 332.9238,
329
+ "step": 52
330
+ },
331
+ {
332
+ "epoch": 0.53,
333
+ "learning_rate": 2e-05,
334
+ "loss": 324.871,
335
+ "step": 53
336
+ },
337
+ {
338
+ "epoch": 0.54,
339
+ "learning_rate": 2e-05,
340
+ "loss": 305.993,
341
+ "step": 54
342
+ },
343
+ {
344
+ "epoch": 0.55,
345
+ "learning_rate": 2e-05,
346
+ "loss": 334.1832,
347
+ "step": 55
348
+ },
349
+ {
350
+ "epoch": 0.56,
351
+ "learning_rate": 2e-05,
352
+ "loss": 393.5037,
353
+ "step": 56
354
+ },
355
+ {
356
+ "epoch": 0.57,
357
+ "learning_rate": 2e-05,
358
+ "loss": 453.1027,
359
+ "step": 57
360
+ },
361
+ {
362
+ "epoch": 0.58,
363
+ "learning_rate": 2e-05,
364
+ "loss": 306.5744,
365
+ "step": 58
366
+ },
367
+ {
368
+ "epoch": 0.59,
369
+ "learning_rate": 2e-05,
370
+ "loss": 343.3282,
371
+ "step": 59
372
+ },
373
+ {
374
+ "epoch": 0.6,
375
+ "learning_rate": 2e-05,
376
+ "loss": 367.3992,
377
+ "step": 60
378
+ },
379
+ {
380
+ "epoch": 0.61,
381
+ "learning_rate": 2e-05,
382
+ "loss": 252.5841,
383
+ "step": 61
384
+ },
385
+ {
386
+ "epoch": 0.62,
387
+ "learning_rate": 2e-05,
388
+ "loss": 332.6815,
389
+ "step": 62
390
+ },
391
+ {
392
+ "epoch": 0.63,
393
+ "learning_rate": 2e-05,
394
+ "loss": 260.7815,
395
+ "step": 63
396
+ },
397
+ {
398
+ "epoch": 0.64,
399
+ "learning_rate": 2e-05,
400
+ "loss": 322.11,
401
+ "step": 64
402
+ },
403
+ {
404
+ "epoch": 0.65,
405
+ "learning_rate": 2e-05,
406
+ "loss": 316.3943,
407
+ "step": 65
408
+ },
409
+ {
410
+ "epoch": 0.66,
411
+ "learning_rate": 2e-05,
412
+ "loss": 282.9459,
413
+ "step": 66
414
+ },
415
+ {
416
+ "epoch": 0.67,
417
+ "learning_rate": 2e-05,
418
+ "loss": 389.4407,
419
+ "step": 67
420
+ },
421
+ {
422
+ "epoch": 0.68,
423
+ "learning_rate": 2e-05,
424
+ "loss": 266.971,
425
+ "step": 68
426
+ },
427
+ {
428
+ "epoch": 0.69,
429
+ "learning_rate": 2e-05,
430
+ "loss": 309.2886,
431
+ "step": 69
432
+ },
433
+ {
434
+ "epoch": 0.7,
435
+ "learning_rate": 2e-05,
436
+ "loss": 248.2584,
437
+ "step": 70
438
+ },
439
+ {
440
+ "epoch": 0.71,
441
+ "learning_rate": 2e-05,
442
+ "loss": 311.3903,
443
+ "step": 71
444
+ },
445
+ {
446
+ "epoch": 0.72,
447
+ "learning_rate": 2e-05,
448
+ "loss": 281.4355,
449
+ "step": 72
450
+ },
451
+ {
452
+ "epoch": 0.73,
453
+ "learning_rate": 2e-05,
454
+ "loss": 348.8126,
455
+ "step": 73
456
+ },
457
+ {
458
+ "epoch": 0.74,
459
+ "learning_rate": 2e-05,
460
+ "loss": 335.4451,
461
+ "step": 74
462
+ },
463
+ {
464
+ "epoch": 0.75,
465
+ "learning_rate": 2e-05,
466
+ "loss": 252.2587,
467
+ "step": 75
468
+ },
469
+ {
470
+ "epoch": 0.76,
471
+ "learning_rate": 2e-05,
472
+ "loss": 280.6372,
473
+ "step": 76
474
+ },
475
+ {
476
+ "epoch": 0.77,
477
+ "learning_rate": 2e-05,
478
+ "loss": 276.1567,
479
+ "step": 77
480
+ },
481
+ {
482
+ "epoch": 0.78,
483
+ "learning_rate": 2e-05,
484
+ "loss": 295.8489,
485
+ "step": 78
486
+ },
487
+ {
488
+ "epoch": 0.79,
489
+ "learning_rate": 2e-05,
490
+ "loss": 258.2193,
491
+ "step": 79
492
+ },
493
+ {
494
+ "epoch": 0.8,
495
+ "learning_rate": 2e-05,
496
+ "loss": 303.292,
497
+ "step": 80
498
+ },
499
+ {
500
+ "epoch": 0.8,
501
+ "eval_loss": 290.92047119140625,
502
+ "eval_runtime": 91.4002,
503
+ "eval_samples_per_second": 17.505,
504
+ "eval_steps_per_second": 1.094,
505
+ "step": 80
506
+ },
507
+ {
508
+ "epoch": 0.81,
509
+ "learning_rate": 2e-05,
510
+ "loss": 292.2969,
511
+ "step": 81
512
+ },
513
+ {
514
+ "epoch": 0.82,
515
+ "learning_rate": 2e-05,
516
+ "loss": 292.5847,
517
+ "step": 82
518
+ },
519
+ {
520
+ "epoch": 0.83,
521
+ "learning_rate": 2e-05,
522
+ "loss": 287.2512,
523
+ "step": 83
524
+ },
525
+ {
526
+ "epoch": 0.84,
527
+ "learning_rate": 2e-05,
528
+ "loss": 274.7219,
529
+ "step": 84
530
+ },
531
+ {
532
+ "epoch": 0.85,
533
+ "learning_rate": 2e-05,
534
+ "loss": 259.4895,
535
+ "step": 85
536
+ },
537
+ {
538
+ "epoch": 0.86,
539
+ "learning_rate": 2e-05,
540
+ "loss": 308.7735,
541
+ "step": 86
542
+ },
543
+ {
544
+ "epoch": 0.87,
545
+ "learning_rate": 2e-05,
546
+ "loss": 358.5125,
547
+ "step": 87
548
+ },
549
+ {
550
+ "epoch": 0.88,
551
+ "learning_rate": 2e-05,
552
+ "loss": 228.2059,
553
+ "step": 88
554
+ },
555
+ {
556
+ "epoch": 0.89,
557
+ "learning_rate": 2e-05,
558
+ "loss": 221.5,
559
+ "step": 89
560
+ },
561
+ {
562
+ "epoch": 0.9,
563
+ "learning_rate": 2e-05,
564
+ "loss": 256.0698,
565
+ "step": 90
566
+ },
567
+ {
568
+ "epoch": 0.91,
569
+ "learning_rate": 2e-05,
570
+ "loss": 280.6437,
571
+ "step": 91
572
+ },
573
+ {
574
+ "epoch": 0.92,
575
+ "learning_rate": 2e-05,
576
+ "loss": 265.0312,
577
+ "step": 92
578
+ },
579
+ {
580
+ "epoch": 0.93,
581
+ "learning_rate": 2e-05,
582
+ "loss": 276.1062,
583
+ "step": 93
584
+ },
585
+ {
586
+ "epoch": 0.94,
587
+ "learning_rate": 2e-05,
588
+ "loss": 332.605,
589
+ "step": 94
590
+ },
591
+ {
592
+ "epoch": 0.95,
593
+ "learning_rate": 2e-05,
594
+ "loss": 241.6841,
595
+ "step": 95
596
+ },
597
+ {
598
+ "epoch": 0.96,
599
+ "learning_rate": 2e-05,
600
+ "loss": 290.7553,
601
+ "step": 96
602
+ },
603
+ {
604
+ "epoch": 0.97,
605
+ "learning_rate": 2e-05,
606
+ "loss": 242.8853,
607
+ "step": 97
608
+ },
609
+ {
610
+ "epoch": 0.98,
611
+ "learning_rate": 2e-05,
612
+ "loss": 267.5484,
613
+ "step": 98
614
+ },
615
+ {
616
+ "epoch": 0.99,
617
+ "learning_rate": 2e-05,
618
+ "loss": 263.2162,
619
+ "step": 99
620
+ },
621
+ {
622
+ "epoch": 1.0,
623
+ "learning_rate": 2e-05,
624
+ "loss": 226.8806,
625
+ "step": 100
626
+ },
627
+ {
628
+ "epoch": 1.01,
629
+ "learning_rate": 2e-05,
630
+ "loss": 175.9632,
631
+ "step": 101
632
+ },
633
+ {
634
+ "epoch": 1.02,
635
+ "learning_rate": 2e-05,
636
+ "loss": 293.8885,
637
+ "step": 102
638
+ },
639
+ {
640
+ "epoch": 1.03,
641
+ "learning_rate": 2e-05,
642
+ "loss": 264.4063,
643
+ "step": 103
644
+ },
645
+ {
646
+ "epoch": 1.04,
647
+ "learning_rate": 2e-05,
648
+ "loss": 268.8441,
649
+ "step": 104
650
+ },
651
+ {
652
+ "epoch": 1.05,
653
+ "learning_rate": 2e-05,
654
+ "loss": 251.1409,
655
+ "step": 105
656
+ },
657
+ {
658
+ "epoch": 1.06,
659
+ "learning_rate": 2e-05,
660
+ "loss": 222.4796,
661
+ "step": 106
662
+ },
663
+ {
664
+ "epoch": 1.07,
665
+ "learning_rate": 2e-05,
666
+ "loss": 259.7393,
667
+ "step": 107
668
+ },
669
+ {
670
+ "epoch": 1.08,
671
+ "learning_rate": 2e-05,
672
+ "loss": 247.0995,
673
+ "step": 108
674
+ },
675
+ {
676
+ "epoch": 1.09,
677
+ "learning_rate": 2e-05,
678
+ "loss": 228.6188,
679
+ "step": 109
680
+ },
681
+ {
682
+ "epoch": 1.1,
683
+ "learning_rate": 2e-05,
684
+ "loss": 238.4029,
685
+ "step": 110
686
+ },
687
+ {
688
+ "epoch": 1.11,
689
+ "learning_rate": 2e-05,
690
+ "loss": 249.7835,
691
+ "step": 111
692
+ },
693
+ {
694
+ "epoch": 1.12,
695
+ "learning_rate": 2e-05,
696
+ "loss": 255.0745,
697
+ "step": 112
698
+ },
699
+ {
700
+ "epoch": 1.13,
701
+ "learning_rate": 2e-05,
702
+ "loss": 281.3386,
703
+ "step": 113
704
+ },
705
+ {
706
+ "epoch": 1.14,
707
+ "learning_rate": 2e-05,
708
+ "loss": 258.1128,
709
+ "step": 114
710
+ },
711
+ {
712
+ "epoch": 1.15,
713
+ "learning_rate": 2e-05,
714
+ "loss": 258.487,
715
+ "step": 115
716
+ },
717
+ {
718
+ "epoch": 1.16,
719
+ "learning_rate": 2e-05,
720
+ "loss": 252.1913,
721
+ "step": 116
722
+ },
723
+ {
724
+ "epoch": 1.17,
725
+ "learning_rate": 2e-05,
726
+ "loss": 222.6366,
727
+ "step": 117
728
+ },
729
+ {
730
+ "epoch": 1.18,
731
+ "learning_rate": 2e-05,
732
+ "loss": 247.7612,
733
+ "step": 118
734
+ },
735
+ {
736
+ "epoch": 1.19,
737
+ "learning_rate": 2e-05,
738
+ "loss": 212.2664,
739
+ "step": 119
740
+ },
741
+ {
742
+ "epoch": 1.2,
743
+ "learning_rate": 2e-05,
744
+ "loss": 260.1885,
745
+ "step": 120
746
+ },
747
+ {
748
+ "epoch": 1.2,
749
+ "eval_loss": 268.6912841796875,
750
+ "eval_runtime": 91.4506,
751
+ "eval_samples_per_second": 17.496,
752
+ "eval_steps_per_second": 1.093,
753
+ "step": 120
754
+ },
755
+ {
756
+ "epoch": 1.21,
757
+ "learning_rate": 2e-05,
758
+ "loss": 225.2303,
759
+ "step": 121
760
+ },
761
+ {
762
+ "epoch": 1.22,
763
+ "learning_rate": 2e-05,
764
+ "loss": 208.434,
765
+ "step": 122
766
+ },
767
+ {
768
+ "epoch": 1.23,
769
+ "learning_rate": 2e-05,
770
+ "loss": 260.5778,
771
+ "step": 123
772
+ },
773
+ {
774
+ "epoch": 1.24,
775
+ "learning_rate": 2e-05,
776
+ "loss": 256.1859,
777
+ "step": 124
778
+ },
779
+ {
780
+ "epoch": 1.25,
781
+ "learning_rate": 2e-05,
782
+ "loss": 225.3799,
783
+ "step": 125
784
+ },
785
+ {
786
+ "epoch": 1.26,
787
+ "learning_rate": 2e-05,
788
+ "loss": 242.6659,
789
+ "step": 126
790
+ },
791
+ {
792
+ "epoch": 1.27,
793
+ "learning_rate": 2e-05,
794
+ "loss": 218.2521,
795
+ "step": 127
796
+ },
797
+ {
798
+ "epoch": 1.28,
799
+ "learning_rate": 2e-05,
800
+ "loss": 237.711,
801
+ "step": 128
802
+ },
803
+ {
804
+ "epoch": 1.29,
805
+ "learning_rate": 2e-05,
806
+ "loss": 228.6392,
807
+ "step": 129
808
+ },
809
+ {
810
+ "epoch": 1.3,
811
+ "learning_rate": 2e-05,
812
+ "loss": 257.0567,
813
+ "step": 130
814
+ },
815
+ {
816
+ "epoch": 1.31,
817
+ "learning_rate": 2e-05,
818
+ "loss": 225.4318,
819
+ "step": 131
820
+ },
821
+ {
822
+ "epoch": 1.32,
823
+ "learning_rate": 2e-05,
824
+ "loss": 255.6358,
825
+ "step": 132
826
+ },
827
+ {
828
+ "epoch": 1.33,
829
+ "learning_rate": 2e-05,
830
+ "loss": 243.6262,
831
+ "step": 133
832
+ },
833
+ {
834
+ "epoch": 1.34,
835
+ "learning_rate": 2e-05,
836
+ "loss": 235.9305,
837
+ "step": 134
838
+ },
839
+ {
840
+ "epoch": 1.35,
841
+ "learning_rate": 2e-05,
842
+ "loss": 238.0324,
843
+ "step": 135
844
+ },
845
+ {
846
+ "epoch": 1.36,
847
+ "learning_rate": 2e-05,
848
+ "loss": 239.2688,
849
+ "step": 136
850
+ },
851
+ {
852
+ "epoch": 1.37,
853
+ "learning_rate": 2e-05,
854
+ "loss": 234.8799,
855
+ "step": 137
856
+ },
857
+ {
858
+ "epoch": 1.38,
859
+ "learning_rate": 2e-05,
860
+ "loss": 249.6847,
861
+ "step": 138
862
+ },
863
+ {
864
+ "epoch": 1.39,
865
+ "learning_rate": 2e-05,
866
+ "loss": 259.0303,
867
+ "step": 139
868
+ },
869
+ {
870
+ "epoch": 1.4,
871
+ "learning_rate": 2e-05,
872
+ "loss": 230.0663,
873
+ "step": 140
874
+ },
875
+ {
876
+ "epoch": 1.41,
877
+ "learning_rate": 2e-05,
878
+ "loss": 312.8887,
879
+ "step": 141
880
+ },
881
+ {
882
+ "epoch": 1.42,
883
+ "learning_rate": 2e-05,
884
+ "loss": 214.6919,
885
+ "step": 142
886
+ },
887
+ {
888
+ "epoch": 1.43,
889
+ "learning_rate": 2e-05,
890
+ "loss": 204.0403,
891
+ "step": 143
892
+ },
893
+ {
894
+ "epoch": 1.44,
895
+ "learning_rate": 2e-05,
896
+ "loss": 219.8406,
897
+ "step": 144
898
+ },
899
+ {
900
+ "epoch": 1.45,
901
+ "learning_rate": 2e-05,
902
+ "loss": 229.476,
903
+ "step": 145
904
+ },
905
+ {
906
+ "epoch": 1.46,
907
+ "learning_rate": 2e-05,
908
+ "loss": 222.8145,
909
+ "step": 146
910
+ },
911
+ {
912
+ "epoch": 1.47,
913
+ "learning_rate": 2e-05,
914
+ "loss": 257.3806,
915
+ "step": 147
916
+ },
917
+ {
918
+ "epoch": 1.48,
919
+ "learning_rate": 2e-05,
920
+ "loss": 206.661,
921
+ "step": 148
922
+ },
923
+ {
924
+ "epoch": 1.49,
925
+ "learning_rate": 2e-05,
926
+ "loss": 244.2539,
927
+ "step": 149
928
+ },
929
+ {
930
+ "epoch": 1.5,
931
+ "learning_rate": 2e-05,
932
+ "loss": 219.9999,
933
+ "step": 150
934
+ },
935
+ {
936
+ "epoch": 1.51,
937
+ "learning_rate": 2e-05,
938
+ "loss": 186.9665,
939
+ "step": 151
940
+ },
941
+ {
942
+ "epoch": 1.52,
943
+ "learning_rate": 2e-05,
944
+ "loss": 246.9571,
945
+ "step": 152
946
+ },
947
+ {
948
+ "epoch": 1.53,
949
+ "learning_rate": 2e-05,
950
+ "loss": 296.5907,
951
+ "step": 153
952
+ },
953
+ {
954
+ "epoch": 1.54,
955
+ "learning_rate": 2e-05,
956
+ "loss": 235.987,
957
+ "step": 154
958
+ },
959
+ {
960
+ "epoch": 1.55,
961
+ "learning_rate": 2e-05,
962
+ "loss": 232.2841,
963
+ "step": 155
964
+ },
965
+ {
966
+ "epoch": 1.56,
967
+ "learning_rate": 2e-05,
968
+ "loss": 257.2687,
969
+ "step": 156
970
+ },
971
+ {
972
+ "epoch": 1.57,
973
+ "learning_rate": 2e-05,
974
+ "loss": 229.2959,
975
+ "step": 157
976
+ },
977
+ {
978
+ "epoch": 1.58,
979
+ "learning_rate": 2e-05,
980
+ "loss": 204.7547,
981
+ "step": 158
982
+ },
983
+ {
984
+ "epoch": 1.59,
985
+ "learning_rate": 2e-05,
986
+ "loss": 229.0461,
987
+ "step": 159
988
+ },
989
+ {
990
+ "epoch": 1.6,
991
+ "learning_rate": 2e-05,
992
+ "loss": 208.6121,
993
+ "step": 160
994
+ },
995
+ {
996
+ "epoch": 1.6,
997
+ "eval_loss": 251.8457794189453,
998
+ "eval_runtime": 91.4542,
999
+ "eval_samples_per_second": 17.495,
1000
+ "eval_steps_per_second": 1.093,
1001
+ "step": 160
1002
+ },
1003
+ {
1004
+ "epoch": 1.61,
1005
+ "learning_rate": 2e-05,
1006
+ "loss": 213.7581,
1007
+ "step": 161
1008
+ },
1009
+ {
1010
+ "epoch": 1.62,
1011
+ "learning_rate": 2e-05,
1012
+ "loss": 250.1387,
1013
+ "step": 162
1014
+ },
1015
+ {
1016
+ "epoch": 1.63,
1017
+ "learning_rate": 2e-05,
1018
+ "loss": 216.0783,
1019
+ "step": 163
1020
+ },
1021
+ {
1022
+ "epoch": 1.64,
1023
+ "learning_rate": 2e-05,
1024
+ "loss": 223.1108,
1025
+ "step": 164
1026
+ },
1027
+ {
1028
+ "epoch": 1.65,
1029
+ "learning_rate": 2e-05,
1030
+ "loss": 223.4337,
1031
+ "step": 165
1032
+ },
1033
+ {
1034
+ "epoch": 1.66,
1035
+ "learning_rate": 2e-05,
1036
+ "loss": 216.0298,
1037
+ "step": 166
1038
+ },
1039
+ {
1040
+ "epoch": 1.67,
1041
+ "learning_rate": 2e-05,
1042
+ "loss": 210.1397,
1043
+ "step": 167
1044
+ },
1045
+ {
1046
+ "epoch": 1.68,
1047
+ "learning_rate": 2e-05,
1048
+ "loss": 255.6102,
1049
+ "step": 168
1050
+ },
1051
+ {
1052
+ "epoch": 1.69,
1053
+ "learning_rate": 2e-05,
1054
+ "loss": 206.8196,
1055
+ "step": 169
1056
+ },
1057
+ {
1058
+ "epoch": 1.7,
1059
+ "learning_rate": 2e-05,
1060
+ "loss": 225.3016,
1061
+ "step": 170
1062
+ },
1063
+ {
1064
+ "epoch": 1.71,
1065
+ "learning_rate": 2e-05,
1066
+ "loss": 204.423,
1067
+ "step": 171
1068
+ },
1069
+ {
1070
+ "epoch": 1.72,
1071
+ "learning_rate": 2e-05,
1072
+ "loss": 200.3793,
1073
+ "step": 172
1074
+ },
1075
+ {
1076
+ "epoch": 1.73,
1077
+ "learning_rate": 2e-05,
1078
+ "loss": 254.3165,
1079
+ "step": 173
1080
+ },
1081
+ {
1082
+ "epoch": 1.74,
1083
+ "learning_rate": 2e-05,
1084
+ "loss": 228.116,
1085
+ "step": 174
1086
+ },
1087
+ {
1088
+ "epoch": 1.75,
1089
+ "learning_rate": 2e-05,
1090
+ "loss": 215.9781,
1091
+ "step": 175
1092
+ },
1093
+ {
1094
+ "epoch": 1.76,
1095
+ "learning_rate": 2e-05,
1096
+ "loss": 240.427,
1097
+ "step": 176
1098
+ },
1099
+ {
1100
+ "epoch": 1.77,
1101
+ "learning_rate": 2e-05,
1102
+ "loss": 285.4974,
1103
+ "step": 177
1104
+ },
1105
+ {
1106
+ "epoch": 1.78,
1107
+ "learning_rate": 2e-05,
1108
+ "loss": 241.3725,
1109
+ "step": 178
1110
+ },
1111
+ {
1112
+ "epoch": 1.79,
1113
+ "learning_rate": 2e-05,
1114
+ "loss": 208.2607,
1115
+ "step": 179
1116
+ },
1117
+ {
1118
+ "epoch": 1.8,
1119
+ "learning_rate": 2e-05,
1120
+ "loss": 189.7236,
1121
+ "step": 180
1122
+ },
1123
+ {
1124
+ "epoch": 1.81,
1125
+ "learning_rate": 2e-05,
1126
+ "loss": 251.2979,
1127
+ "step": 181
1128
+ },
1129
+ {
1130
+ "epoch": 1.82,
1131
+ "learning_rate": 2e-05,
1132
+ "loss": 221.9034,
1133
+ "step": 182
1134
+ },
1135
+ {
1136
+ "epoch": 1.83,
1137
+ "learning_rate": 2e-05,
1138
+ "loss": 212.9315,
1139
+ "step": 183
1140
+ },
1141
+ {
1142
+ "epoch": 1.84,
1143
+ "learning_rate": 2e-05,
1144
+ "loss": 269.8028,
1145
+ "step": 184
1146
+ },
1147
+ {
1148
+ "epoch": 1.85,
1149
+ "learning_rate": 2e-05,
1150
+ "loss": 234.7929,
1151
+ "step": 185
1152
+ },
1153
+ {
1154
+ "epoch": 1.86,
1155
+ "learning_rate": 2e-05,
1156
+ "loss": 232.632,
1157
+ "step": 186
1158
+ },
1159
+ {
1160
+ "epoch": 1.87,
1161
+ "learning_rate": 2e-05,
1162
+ "loss": 234.5318,
1163
+ "step": 187
1164
+ },
1165
+ {
1166
+ "epoch": 1.88,
1167
+ "learning_rate": 2e-05,
1168
+ "loss": 217.2522,
1169
+ "step": 188
1170
+ },
1171
+ {
1172
+ "epoch": 1.89,
1173
+ "learning_rate": 2e-05,
1174
+ "loss": 212.185,
1175
+ "step": 189
1176
+ },
1177
+ {
1178
+ "epoch": 1.9,
1179
+ "learning_rate": 2e-05,
1180
+ "loss": 193.634,
1181
+ "step": 190
1182
+ },
1183
+ {
1184
+ "epoch": 1.91,
1185
+ "learning_rate": 2e-05,
1186
+ "loss": 194.4389,
1187
+ "step": 191
1188
+ },
1189
+ {
1190
+ "epoch": 1.92,
1191
+ "learning_rate": 2e-05,
1192
+ "loss": 239.7684,
1193
+ "step": 192
1194
+ },
1195
+ {
1196
+ "epoch": 1.93,
1197
+ "learning_rate": 2e-05,
1198
+ "loss": 214.6728,
1199
+ "step": 193
1200
+ },
1201
+ {
1202
+ "epoch": 1.94,
1203
+ "learning_rate": 2e-05,
1204
+ "loss": 238.8546,
1205
+ "step": 194
1206
+ },
1207
+ {
1208
+ "epoch": 1.95,
1209
+ "learning_rate": 2e-05,
1210
+ "loss": 252.4766,
1211
+ "step": 195
1212
+ },
1213
+ {
1214
+ "epoch": 1.96,
1215
+ "learning_rate": 2e-05,
1216
+ "loss": 213.9979,
1217
+ "step": 196
1218
+ },
1219
+ {
1220
+ "epoch": 1.97,
1221
+ "learning_rate": 2e-05,
1222
+ "loss": 226.8183,
1223
+ "step": 197
1224
+ },
1225
+ {
1226
+ "epoch": 1.98,
1227
+ "learning_rate": 2e-05,
1228
+ "loss": 178.3002,
1229
+ "step": 198
1230
+ },
1231
+ {
1232
+ "epoch": 1.99,
1233
+ "learning_rate": 2e-05,
1234
+ "loss": 226.1548,
1235
+ "step": 199
1236
+ },
1237
+ {
1238
+ "epoch": 2.0,
1239
+ "learning_rate": 2e-05,
1240
+ "loss": 197.3228,
1241
+ "step": 200
1242
+ },
1243
+ {
1244
+ "epoch": 2.0,
1245
+ "eval_loss": 234.64894104003906,
1246
+ "eval_runtime": 91.475,
1247
+ "eval_samples_per_second": 17.491,
1248
+ "eval_steps_per_second": 1.093,
1249
+ "step": 200
1250
+ },
1251
+ {
1252
+ "epoch": 2.01,
1253
+ "learning_rate": 2e-05,
1254
+ "loss": 177.7395,
1255
+ "step": 201
1256
+ },
1257
+ {
1258
+ "epoch": 2.02,
1259
+ "learning_rate": 2e-05,
1260
+ "loss": 207.6397,
1261
+ "step": 202
1262
+ },
1263
+ {
1264
+ "epoch": 2.03,
1265
+ "learning_rate": 2e-05,
1266
+ "loss": 234.7,
1267
+ "step": 203
1268
+ },
1269
+ {
1270
+ "epoch": 2.04,
1271
+ "learning_rate": 2e-05,
1272
+ "loss": 161.9427,
1273
+ "step": 204
1274
+ },
1275
+ {
1276
+ "epoch": 2.05,
1277
+ "learning_rate": 2e-05,
1278
+ "loss": 152.5065,
1279
+ "step": 205
1280
+ },
1281
+ {
1282
+ "epoch": 2.06,
1283
+ "learning_rate": 2e-05,
1284
+ "loss": 209.7177,
1285
+ "step": 206
1286
+ },
1287
+ {
1288
+ "epoch": 2.07,
1289
+ "learning_rate": 2e-05,
1290
+ "loss": 180.4935,
1291
+ "step": 207
1292
+ },
1293
+ {
1294
+ "epoch": 2.08,
1295
+ "learning_rate": 2e-05,
1296
+ "loss": 189.8789,
1297
+ "step": 208
1298
+ },
1299
+ {
1300
+ "epoch": 2.09,
1301
+ "learning_rate": 2e-05,
1302
+ "loss": 179.2588,
1303
+ "step": 209
1304
+ },
1305
+ {
1306
+ "epoch": 2.1,
1307
+ "learning_rate": 2e-05,
1308
+ "loss": 199.4266,
1309
+ "step": 210
1310
+ },
1311
+ {
1312
+ "epoch": 2.11,
1313
+ "learning_rate": 2e-05,
1314
+ "loss": 217.2014,
1315
+ "step": 211
1316
+ },
1317
+ {
1318
+ "epoch": 2.12,
1319
+ "learning_rate": 2e-05,
1320
+ "loss": 194.8955,
1321
+ "step": 212
1322
+ },
1323
+ {
1324
+ "epoch": 2.13,
1325
+ "learning_rate": 2e-05,
1326
+ "loss": 199.1684,
1327
+ "step": 213
1328
+ },
1329
+ {
1330
+ "epoch": 2.14,
1331
+ "learning_rate": 2e-05,
1332
+ "loss": 184.8403,
1333
+ "step": 214
1334
+ },
1335
+ {
1336
+ "epoch": 2.15,
1337
+ "learning_rate": 2e-05,
1338
+ "loss": 181.7052,
1339
+ "step": 215
1340
+ },
1341
+ {
1342
+ "epoch": 2.16,
1343
+ "learning_rate": 2e-05,
1344
+ "loss": 178.4005,
1345
+ "step": 216
1346
+ },
1347
+ {
1348
+ "epoch": 2.17,
1349
+ "learning_rate": 2e-05,
1350
+ "loss": 166.0668,
1351
+ "step": 217
1352
+ },
1353
+ {
1354
+ "epoch": 2.18,
1355
+ "learning_rate": 2e-05,
1356
+ "loss": 200.5868,
1357
+ "step": 218
1358
+ },
1359
+ {
1360
+ "epoch": 2.19,
1361
+ "learning_rate": 2e-05,
1362
+ "loss": 164.0996,
1363
+ "step": 219
1364
+ },
1365
+ {
1366
+ "epoch": 2.2,
1367
+ "learning_rate": 2e-05,
1368
+ "loss": 215.2086,
1369
+ "step": 220
1370
+ },
1371
+ {
1372
+ "epoch": 2.21,
1373
+ "learning_rate": 2e-05,
1374
+ "loss": 182.4766,
1375
+ "step": 221
1376
+ },
1377
+ {
1378
+ "epoch": 2.22,
1379
+ "learning_rate": 2e-05,
1380
+ "loss": 265.9225,
1381
+ "step": 222
1382
+ },
1383
+ {
1384
+ "epoch": 2.23,
1385
+ "learning_rate": 2e-05,
1386
+ "loss": 170.2314,
1387
+ "step": 223
1388
+ },
1389
+ {
1390
+ "epoch": 2.24,
1391
+ "learning_rate": 2e-05,
1392
+ "loss": 168.8915,
1393
+ "step": 224
1394
+ },
1395
+ {
1396
+ "epoch": 2.25,
1397
+ "learning_rate": 2e-05,
1398
+ "loss": 202.6422,
1399
+ "step": 225
1400
+ },
1401
+ {
1402
+ "epoch": 2.26,
1403
+ "learning_rate": 2e-05,
1404
+ "loss": 193.6124,
1405
+ "step": 226
1406
+ },
1407
+ {
1408
+ "epoch": 2.27,
1409
+ "learning_rate": 2e-05,
1410
+ "loss": 191.7759,
1411
+ "step": 227
1412
+ },
1413
+ {
1414
+ "epoch": 2.28,
1415
+ "learning_rate": 2e-05,
1416
+ "loss": 172.5128,
1417
+ "step": 228
1418
+ },
1419
+ {
1420
+ "epoch": 2.29,
1421
+ "learning_rate": 2e-05,
1422
+ "loss": 174.589,
1423
+ "step": 229
1424
+ },
1425
+ {
1426
+ "epoch": 2.3,
1427
+ "learning_rate": 2e-05,
1428
+ "loss": 190.2146,
1429
+ "step": 230
1430
+ },
1431
+ {
1432
+ "epoch": 2.31,
1433
+ "learning_rate": 2e-05,
1434
+ "loss": 206.5455,
1435
+ "step": 231
1436
+ },
1437
+ {
1438
+ "epoch": 2.32,
1439
+ "learning_rate": 2e-05,
1440
+ "loss": 212.3613,
1441
+ "step": 232
1442
+ },
1443
+ {
1444
+ "epoch": 2.33,
1445
+ "learning_rate": 2e-05,
1446
+ "loss": 196.8155,
1447
+ "step": 233
1448
+ },
1449
+ {
1450
+ "epoch": 2.34,
1451
+ "learning_rate": 2e-05,
1452
+ "loss": 175.7169,
1453
+ "step": 234
1454
+ },
1455
+ {
1456
+ "epoch": 2.35,
1457
+ "learning_rate": 2e-05,
1458
+ "loss": 246.1433,
1459
+ "step": 235
1460
+ },
1461
+ {
1462
+ "epoch": 2.36,
1463
+ "learning_rate": 2e-05,
1464
+ "loss": 273.7065,
1465
+ "step": 236
1466
+ },
1467
+ {
1468
+ "epoch": 2.37,
1469
+ "learning_rate": 2e-05,
1470
+ "loss": 158.33,
1471
+ "step": 237
1472
+ },
1473
+ {
1474
+ "epoch": 2.38,
1475
+ "learning_rate": 2e-05,
1476
+ "loss": 159.6902,
1477
+ "step": 238
1478
+ },
1479
+ {
1480
+ "epoch": 2.39,
1481
+ "learning_rate": 2e-05,
1482
+ "loss": 260.6693,
1483
+ "step": 239
1484
+ },
1485
+ {
1486
+ "epoch": 2.4,
1487
+ "learning_rate": 2e-05,
1488
+ "loss": 191.4345,
1489
+ "step": 240
1490
+ },
1491
+ {
1492
+ "epoch": 2.4,
1493
+ "eval_loss": 228.75088500976562,
1494
+ "eval_runtime": 91.7222,
1495
+ "eval_samples_per_second": 17.444,
1496
+ "eval_steps_per_second": 1.09,
1497
+ "step": 240
1498
+ },
1499
+ {
1500
+ "epoch": 2.41,
1501
+ "learning_rate": 2e-05,
1502
+ "loss": 194.6978,
1503
+ "step": 241
1504
+ },
1505
+ {
1506
+ "epoch": 2.42,
1507
+ "learning_rate": 2e-05,
1508
+ "loss": 170.113,
1509
+ "step": 242
1510
+ },
1511
+ {
1512
+ "epoch": 2.43,
1513
+ "learning_rate": 2e-05,
1514
+ "loss": 206.9311,
1515
+ "step": 243
1516
+ },
1517
+ {
1518
+ "epoch": 2.44,
1519
+ "learning_rate": 2e-05,
1520
+ "loss": 195.2319,
1521
+ "step": 244
1522
+ },
1523
+ {
1524
+ "epoch": 2.45,
1525
+ "learning_rate": 2e-05,
1526
+ "loss": 163.4541,
1527
+ "step": 245
1528
+ },
1529
+ {
1530
+ "epoch": 2.46,
1531
+ "learning_rate": 2e-05,
1532
+ "loss": 194.2114,
1533
+ "step": 246
1534
+ },
1535
+ {
1536
+ "epoch": 2.47,
1537
+ "learning_rate": 2e-05,
1538
+ "loss": 204.1492,
1539
+ "step": 247
1540
+ },
1541
+ {
1542
+ "epoch": 2.48,
1543
+ "learning_rate": 2e-05,
1544
+ "loss": 202.1168,
1545
+ "step": 248
1546
+ },
1547
+ {
1548
+ "epoch": 2.49,
1549
+ "learning_rate": 2e-05,
1550
+ "loss": 188.9232,
1551
+ "step": 249
1552
+ },
1553
+ {
1554
+ "epoch": 2.5,
1555
+ "learning_rate": 2e-05,
1556
+ "loss": 183.1904,
1557
+ "step": 250
1558
+ },
1559
+ {
1560
+ "epoch": 2.51,
1561
+ "learning_rate": 2e-05,
1562
+ "loss": 171.6944,
1563
+ "step": 251
1564
+ },
1565
+ {
1566
+ "epoch": 2.52,
1567
+ "learning_rate": 2e-05,
1568
+ "loss": 218.1628,
1569
+ "step": 252
1570
+ },
1571
+ {
1572
+ "epoch": 2.53,
1573
+ "learning_rate": 2e-05,
1574
+ "loss": 178.1614,
1575
+ "step": 253
1576
+ },
1577
+ {
1578
+ "epoch": 2.54,
1579
+ "learning_rate": 2e-05,
1580
+ "loss": 175.8137,
1581
+ "step": 254
1582
+ },
1583
+ {
1584
+ "epoch": 2.55,
1585
+ "learning_rate": 2e-05,
1586
+ "loss": 176.3016,
1587
+ "step": 255
1588
+ },
1589
+ {
1590
+ "epoch": 2.56,
1591
+ "learning_rate": 2e-05,
1592
+ "loss": 195.611,
1593
+ "step": 256
1594
+ },
1595
+ {
1596
+ "epoch": 2.57,
1597
+ "learning_rate": 2e-05,
1598
+ "loss": 154.6473,
1599
+ "step": 257
1600
+ },
1601
+ {
1602
+ "epoch": 2.58,
1603
+ "learning_rate": 2e-05,
1604
+ "loss": 175.7625,
1605
+ "step": 258
1606
+ },
1607
+ {
1608
+ "epoch": 2.59,
1609
+ "learning_rate": 2e-05,
1610
+ "loss": 180.9702,
1611
+ "step": 259
1612
+ },
1613
+ {
1614
+ "epoch": 2.6,
1615
+ "learning_rate": 2e-05,
1616
+ "loss": 172.006,
1617
+ "step": 260
1618
+ },
1619
+ {
1620
+ "epoch": 2.61,
1621
+ "learning_rate": 2e-05,
1622
+ "loss": 166.616,
1623
+ "step": 261
1624
+ },
1625
+ {
1626
+ "epoch": 2.62,
1627
+ "learning_rate": 2e-05,
1628
+ "loss": 205.9087,
1629
+ "step": 262
1630
+ },
1631
+ {
1632
+ "epoch": 2.63,
1633
+ "learning_rate": 2e-05,
1634
+ "loss": 195.5401,
1635
+ "step": 263
1636
+ },
1637
+ {
1638
+ "epoch": 2.64,
1639
+ "learning_rate": 2e-05,
1640
+ "loss": 182.7327,
1641
+ "step": 264
1642
+ },
1643
+ {
1644
+ "epoch": 2.65,
1645
+ "learning_rate": 2e-05,
1646
+ "loss": 187.6268,
1647
+ "step": 265
1648
+ },
1649
+ {
1650
+ "epoch": 2.66,
1651
+ "learning_rate": 2e-05,
1652
+ "loss": 150.9506,
1653
+ "step": 266
1654
+ },
1655
+ {
1656
+ "epoch": 2.67,
1657
+ "learning_rate": 2e-05,
1658
+ "loss": 187.1612,
1659
+ "step": 267
1660
+ },
1661
+ {
1662
+ "epoch": 2.68,
1663
+ "learning_rate": 2e-05,
1664
+ "loss": 199.4861,
1665
+ "step": 268
1666
+ },
1667
+ {
1668
+ "epoch": 2.69,
1669
+ "learning_rate": 2e-05,
1670
+ "loss": 197.6736,
1671
+ "step": 269
1672
+ },
1673
+ {
1674
+ "epoch": 2.7,
1675
+ "learning_rate": 2e-05,
1676
+ "loss": 204.7334,
1677
+ "step": 270
1678
+ },
1679
+ {
1680
+ "epoch": 2.71,
1681
+ "learning_rate": 2e-05,
1682
+ "loss": 186.2923,
1683
+ "step": 271
1684
+ },
1685
+ {
1686
+ "epoch": 2.72,
1687
+ "learning_rate": 2e-05,
1688
+ "loss": 191.4558,
1689
+ "step": 272
1690
+ },
1691
+ {
1692
+ "epoch": 2.73,
1693
+ "learning_rate": 2e-05,
1694
+ "loss": 195.1405,
1695
+ "step": 273
1696
+ },
1697
+ {
1698
+ "epoch": 2.74,
1699
+ "learning_rate": 2e-05,
1700
+ "loss": 193.3551,
1701
+ "step": 274
1702
+ },
1703
+ {
1704
+ "epoch": 2.75,
1705
+ "learning_rate": 2e-05,
1706
+ "loss": 191.0934,
1707
+ "step": 275
1708
+ },
1709
+ {
1710
+ "epoch": 2.76,
1711
+ "learning_rate": 2e-05,
1712
+ "loss": 181.389,
1713
+ "step": 276
1714
+ },
1715
+ {
1716
+ "epoch": 2.77,
1717
+ "learning_rate": 2e-05,
1718
+ "loss": 175.3716,
1719
+ "step": 277
1720
+ },
1721
+ {
1722
+ "epoch": 2.78,
1723
+ "learning_rate": 2e-05,
1724
+ "loss": 172.3194,
1725
+ "step": 278
1726
+ },
1727
+ {
1728
+ "epoch": 2.79,
1729
+ "learning_rate": 2e-05,
1730
+ "loss": 210.0355,
1731
+ "step": 279
1732
+ },
1733
+ {
1734
+ "epoch": 2.8,
1735
+ "learning_rate": 2e-05,
1736
+ "loss": 151.5427,
1737
+ "step": 280
1738
+ },
1739
+ {
1740
+ "epoch": 2.8,
1741
+ "eval_loss": 220.8135528564453,
1742
+ "eval_runtime": 91.7827,
1743
+ "eval_samples_per_second": 17.432,
1744
+ "eval_steps_per_second": 1.09,
1745
+ "step": 280
1746
+ },
1747
+ {
1748
+ "epoch": 2.81,
1749
+ "learning_rate": 2e-05,
1750
+ "loss": 216.8114,
1751
+ "step": 281
1752
+ },
1753
+ {
1754
+ "epoch": 2.82,
1755
+ "learning_rate": 2e-05,
1756
+ "loss": 204.5617,
1757
+ "step": 282
1758
+ },
1759
+ {
1760
+ "epoch": 2.83,
1761
+ "learning_rate": 2e-05,
1762
+ "loss": 170.5889,
1763
+ "step": 283
1764
+ },
1765
+ {
1766
+ "epoch": 2.84,
1767
+ "learning_rate": 2e-05,
1768
+ "loss": 207.3868,
1769
+ "step": 284
1770
+ },
1771
+ {
1772
+ "epoch": 2.85,
1773
+ "learning_rate": 2e-05,
1774
+ "loss": 181.0243,
1775
+ "step": 285
1776
+ },
1777
+ {
1778
+ "epoch": 2.86,
1779
+ "learning_rate": 2e-05,
1780
+ "loss": 181.3605,
1781
+ "step": 286
1782
+ },
1783
+ {
1784
+ "epoch": 2.87,
1785
+ "learning_rate": 2e-05,
1786
+ "loss": 151.9068,
1787
+ "step": 287
1788
+ },
1789
+ {
1790
+ "epoch": 2.88,
1791
+ "learning_rate": 2e-05,
1792
+ "loss": 181.1088,
1793
+ "step": 288
1794
+ },
1795
+ {
1796
+ "epoch": 2.89,
1797
+ "learning_rate": 2e-05,
1798
+ "loss": 168.5044,
1799
+ "step": 289
1800
+ },
1801
+ {
1802
+ "epoch": 2.9,
1803
+ "learning_rate": 2e-05,
1804
+ "loss": 169.9193,
1805
+ "step": 290
1806
+ },
1807
+ {
1808
+ "epoch": 2.91,
1809
+ "learning_rate": 2e-05,
1810
+ "loss": 163.9978,
1811
+ "step": 291
1812
+ },
1813
+ {
1814
+ "epoch": 2.92,
1815
+ "learning_rate": 2e-05,
1816
+ "loss": 160.5624,
1817
+ "step": 292
1818
+ },
1819
+ {
1820
+ "epoch": 2.93,
1821
+ "learning_rate": 2e-05,
1822
+ "loss": 168.6609,
1823
+ "step": 293
1824
+ },
1825
+ {
1826
+ "epoch": 2.94,
1827
+ "learning_rate": 2e-05,
1828
+ "loss": 176.1334,
1829
+ "step": 294
1830
+ },
1831
+ {
1832
+ "epoch": 2.95,
1833
+ "learning_rate": 2e-05,
1834
+ "loss": 183.0596,
1835
+ "step": 295
1836
+ },
1837
+ {
1838
+ "epoch": 2.96,
1839
+ "learning_rate": 2e-05,
1840
+ "loss": 137.6645,
1841
+ "step": 296
1842
+ },
1843
+ {
1844
+ "epoch": 2.97,
1845
+ "learning_rate": 2e-05,
1846
+ "loss": 207.355,
1847
+ "step": 297
1848
+ },
1849
+ {
1850
+ "epoch": 2.98,
1851
+ "learning_rate": 2e-05,
1852
+ "loss": 142.0311,
1853
+ "step": 298
1854
+ },
1855
+ {
1856
+ "epoch": 2.99,
1857
+ "learning_rate": 2e-05,
1858
+ "loss": 220.0689,
1859
+ "step": 299
1860
+ },
1861
+ {
1862
+ "epoch": 3.0,
1863
+ "learning_rate": 2e-05,
1864
+ "loss": 188.8789,
1865
+ "step": 300
1866
+ },
1867
+ {
1868
+ "epoch": 3.01,
1869
+ "learning_rate": 2e-05,
1870
+ "loss": 125.5,
1871
+ "step": 301
1872
+ },
1873
+ {
1874
+ "epoch": 3.02,
1875
+ "learning_rate": 2e-05,
1876
+ "loss": 135.3729,
1877
+ "step": 302
1878
+ },
1879
+ {
1880
+ "epoch": 3.03,
1881
+ "learning_rate": 2e-05,
1882
+ "loss": 138.8647,
1883
+ "step": 303
1884
+ },
1885
+ {
1886
+ "epoch": 3.04,
1887
+ "learning_rate": 2e-05,
1888
+ "loss": 179.7612,
1889
+ "step": 304
1890
+ },
1891
+ {
1892
+ "epoch": 3.05,
1893
+ "learning_rate": 2e-05,
1894
+ "loss": 155.7191,
1895
+ "step": 305
1896
+ },
1897
+ {
1898
+ "epoch": 3.06,
1899
+ "learning_rate": 2e-05,
1900
+ "loss": 142.4713,
1901
+ "step": 306
1902
+ },
1903
+ {
1904
+ "epoch": 3.07,
1905
+ "learning_rate": 2e-05,
1906
+ "loss": 153.5572,
1907
+ "step": 307
1908
+ },
1909
+ {
1910
+ "epoch": 3.08,
1911
+ "learning_rate": 2e-05,
1912
+ "loss": 158.0588,
1913
+ "step": 308
1914
+ },
1915
+ {
1916
+ "epoch": 3.09,
1917
+ "learning_rate": 2e-05,
1918
+ "loss": 188.1761,
1919
+ "step": 309
1920
+ },
1921
+ {
1922
+ "epoch": 3.1,
1923
+ "learning_rate": 2e-05,
1924
+ "loss": 149.4881,
1925
+ "step": 310
1926
+ },
1927
+ {
1928
+ "epoch": 3.11,
1929
+ "learning_rate": 2e-05,
1930
+ "loss": 143.047,
1931
+ "step": 311
1932
+ },
1933
+ {
1934
+ "epoch": 3.12,
1935
+ "learning_rate": 2e-05,
1936
+ "loss": 149.3641,
1937
+ "step": 312
1938
+ },
1939
+ {
1940
+ "epoch": 3.13,
1941
+ "learning_rate": 2e-05,
1942
+ "loss": 157.5219,
1943
+ "step": 313
1944
+ },
1945
+ {
1946
+ "epoch": 3.14,
1947
+ "learning_rate": 2e-05,
1948
+ "loss": 160.7546,
1949
+ "step": 314
1950
+ },
1951
+ {
1952
+ "epoch": 3.15,
1953
+ "learning_rate": 2e-05,
1954
+ "loss": 204.2928,
1955
+ "step": 315
1956
+ },
1957
+ {
1958
+ "epoch": 3.16,
1959
+ "learning_rate": 2e-05,
1960
+ "loss": 162.4894,
1961
+ "step": 316
1962
+ },
1963
+ {
1964
+ "epoch": 3.17,
1965
+ "learning_rate": 2e-05,
1966
+ "loss": 142.8077,
1967
+ "step": 317
1968
+ },
1969
+ {
1970
+ "epoch": 3.18,
1971
+ "learning_rate": 2e-05,
1972
+ "loss": 189.9061,
1973
+ "step": 318
1974
+ },
1975
+ {
1976
+ "epoch": 3.19,
1977
+ "learning_rate": 2e-05,
1978
+ "loss": 160.648,
1979
+ "step": 319
1980
+ },
1981
+ {
1982
+ "epoch": 3.2,
1983
+ "learning_rate": 2e-05,
1984
+ "loss": 195.4514,
1985
+ "step": 320
1986
+ },
1987
+ {
1988
+ "epoch": 3.2,
1989
+ "eval_loss": 214.12860107421875,
1990
+ "eval_runtime": 91.6917,
1991
+ "eval_samples_per_second": 17.45,
1992
+ "eval_steps_per_second": 1.091,
1993
+ "step": 320
1994
+ },
1995
+ {
1996
+ "epoch": 3.21,
1997
+ "learning_rate": 2e-05,
1998
+ "loss": 165.5777,
1999
+ "step": 321
2000
+ },
2001
+ {
2002
+ "epoch": 3.22,
2003
+ "learning_rate": 2e-05,
2004
+ "loss": 157.4871,
2005
+ "step": 322
2006
+ },
2007
+ {
2008
+ "epoch": 3.23,
2009
+ "learning_rate": 2e-05,
2010
+ "loss": 156.9734,
2011
+ "step": 323
2012
+ },
2013
+ {
2014
+ "epoch": 3.24,
2015
+ "learning_rate": 2e-05,
2016
+ "loss": 171.0179,
2017
+ "step": 324
2018
+ },
2019
+ {
2020
+ "epoch": 3.25,
2021
+ "learning_rate": 2e-05,
2022
+ "loss": 158.758,
2023
+ "step": 325
2024
+ },
2025
+ {
2026
+ "epoch": 3.26,
2027
+ "learning_rate": 2e-05,
2028
+ "loss": 180.9356,
2029
+ "step": 326
2030
+ },
2031
+ {
2032
+ "epoch": 3.27,
2033
+ "learning_rate": 2e-05,
2034
+ "loss": 132.6712,
2035
+ "step": 327
2036
+ },
2037
+ {
2038
+ "epoch": 3.28,
2039
+ "learning_rate": 2e-05,
2040
+ "loss": 142.0085,
2041
+ "step": 328
2042
+ },
2043
+ {
2044
+ "epoch": 3.29,
2045
+ "learning_rate": 2e-05,
2046
+ "loss": 122.8798,
2047
+ "step": 329
2048
+ },
2049
+ {
2050
+ "epoch": 3.3,
2051
+ "learning_rate": 2e-05,
2052
+ "loss": 166.3639,
2053
+ "step": 330
2054
+ },
2055
+ {
2056
+ "epoch": 3.31,
2057
+ "learning_rate": 2e-05,
2058
+ "loss": 126.7612,
2059
+ "step": 331
2060
+ },
2061
+ {
2062
+ "epoch": 3.32,
2063
+ "learning_rate": 2e-05,
2064
+ "loss": 188.9171,
2065
+ "step": 332
2066
+ },
2067
+ {
2068
+ "epoch": 3.33,
2069
+ "learning_rate": 2e-05,
2070
+ "loss": 151.0345,
2071
+ "step": 333
2072
+ },
2073
+ {
2074
+ "epoch": 3.34,
2075
+ "learning_rate": 2e-05,
2076
+ "loss": 159.7177,
2077
+ "step": 334
2078
+ },
2079
+ {
2080
+ "epoch": 3.35,
2081
+ "learning_rate": 2e-05,
2082
+ "loss": 150.0818,
2083
+ "step": 335
2084
+ },
2085
+ {
2086
+ "epoch": 3.36,
2087
+ "learning_rate": 2e-05,
2088
+ "loss": 162.671,
2089
+ "step": 336
2090
+ },
2091
+ {
2092
+ "epoch": 3.37,
2093
+ "learning_rate": 2e-05,
2094
+ "loss": 129.0101,
2095
+ "step": 337
2096
+ },
2097
+ {
2098
+ "epoch": 3.38,
2099
+ "learning_rate": 2e-05,
2100
+ "loss": 187.8155,
2101
+ "step": 338
2102
+ },
2103
+ {
2104
+ "epoch": 3.39,
2105
+ "learning_rate": 2e-05,
2106
+ "loss": 136.3083,
2107
+ "step": 339
2108
+ },
2109
+ {
2110
+ "epoch": 3.4,
2111
+ "learning_rate": 2e-05,
2112
+ "loss": 139.6308,
2113
+ "step": 340
2114
+ },
2115
+ {
2116
+ "epoch": 3.41,
2117
+ "learning_rate": 2e-05,
2118
+ "loss": 158.1506,
2119
+ "step": 341
2120
+ },
2121
+ {
2122
+ "epoch": 3.42,
2123
+ "learning_rate": 2e-05,
2124
+ "loss": 152.7077,
2125
+ "step": 342
2126
+ },
2127
+ {
2128
+ "epoch": 3.43,
2129
+ "learning_rate": 2e-05,
2130
+ "loss": 173.8158,
2131
+ "step": 343
2132
+ },
2133
+ {
2134
+ "epoch": 3.44,
2135
+ "learning_rate": 2e-05,
2136
+ "loss": 128.805,
2137
+ "step": 344
2138
+ },
2139
+ {
2140
+ "epoch": 3.45,
2141
+ "learning_rate": 2e-05,
2142
+ "loss": 152.3222,
2143
+ "step": 345
2144
+ },
2145
+ {
2146
+ "epoch": 3.46,
2147
+ "learning_rate": 2e-05,
2148
+ "loss": 163.3487,
2149
+ "step": 346
2150
+ },
2151
+ {
2152
+ "epoch": 3.47,
2153
+ "learning_rate": 2e-05,
2154
+ "loss": 169.0825,
2155
+ "step": 347
2156
+ },
2157
+ {
2158
+ "epoch": 3.48,
2159
+ "learning_rate": 2e-05,
2160
+ "loss": 156.5232,
2161
+ "step": 348
2162
+ },
2163
+ {
2164
+ "epoch": 3.49,
2165
+ "learning_rate": 2e-05,
2166
+ "loss": 188.6721,
2167
+ "step": 349
2168
+ },
2169
+ {
2170
+ "epoch": 3.5,
2171
+ "learning_rate": 2e-05,
2172
+ "loss": 201.6223,
2173
+ "step": 350
2174
+ },
2175
+ {
2176
+ "epoch": 3.51,
2177
+ "learning_rate": 2e-05,
2178
+ "loss": 294.7936,
2179
+ "step": 351
2180
+ },
2181
+ {
2182
+ "epoch": 3.52,
2183
+ "learning_rate": 2e-05,
2184
+ "loss": 155.2639,
2185
+ "step": 352
2186
+ },
2187
+ {
2188
+ "epoch": 3.53,
2189
+ "learning_rate": 2e-05,
2190
+ "loss": 148.6182,
2191
+ "step": 353
2192
+ },
2193
+ {
2194
+ "epoch": 3.54,
2195
+ "learning_rate": 2e-05,
2196
+ "loss": 207.8028,
2197
+ "step": 354
2198
+ },
2199
+ {
2200
+ "epoch": 3.55,
2201
+ "learning_rate": 2e-05,
2202
+ "loss": 163.1711,
2203
+ "step": 355
2204
+ },
2205
+ {
2206
+ "epoch": 3.56,
2207
+ "learning_rate": 2e-05,
2208
+ "loss": 162.5552,
2209
+ "step": 356
2210
+ },
2211
+ {
2212
+ "epoch": 3.57,
2213
+ "learning_rate": 2e-05,
2214
+ "loss": 167.8712,
2215
+ "step": 357
2216
+ },
2217
+ {
2218
+ "epoch": 3.58,
2219
+ "learning_rate": 2e-05,
2220
+ "loss": 155.6208,
2221
+ "step": 358
2222
+ },
2223
+ {
2224
+ "epoch": 3.59,
2225
+ "learning_rate": 2e-05,
2226
+ "loss": 178.2028,
2227
+ "step": 359
2228
+ },
2229
+ {
2230
+ "epoch": 3.6,
2231
+ "learning_rate": 2e-05,
2232
+ "loss": 174.9905,
2233
+ "step": 360
2234
+ },
2235
+ {
2236
+ "epoch": 3.6,
2237
+ "eval_loss": 211.39349365234375,
2238
+ "eval_runtime": 91.4666,
2239
+ "eval_samples_per_second": 17.493,
2240
+ "eval_steps_per_second": 1.093,
2241
+ "step": 360
2242
+ },
2243
+ {
2244
+ "epoch": 3.61,
2245
+ "learning_rate": 2e-05,
2246
+ "loss": 149.9229,
2247
+ "step": 361
2248
+ },
2249
+ {
2250
+ "epoch": 3.62,
2251
+ "learning_rate": 2e-05,
2252
+ "loss": 166.3071,
2253
+ "step": 362
2254
+ },
2255
+ {
2256
+ "epoch": 3.63,
2257
+ "learning_rate": 2e-05,
2258
+ "loss": 124.036,
2259
+ "step": 363
2260
+ },
2261
+ {
2262
+ "epoch": 3.64,
2263
+ "learning_rate": 2e-05,
2264
+ "loss": 172.7505,
2265
+ "step": 364
2266
+ },
2267
+ {
2268
+ "epoch": 3.65,
2269
+ "learning_rate": 2e-05,
2270
+ "loss": 148.0122,
2271
+ "step": 365
2272
+ },
2273
+ {
2274
+ "epoch": 3.66,
2275
+ "learning_rate": 2e-05,
2276
+ "loss": 185.0846,
2277
+ "step": 366
2278
+ },
2279
+ {
2280
+ "epoch": 3.67,
2281
+ "learning_rate": 2e-05,
2282
+ "loss": 153.4166,
2283
+ "step": 367
2284
+ },
2285
+ {
2286
+ "epoch": 3.68,
2287
+ "learning_rate": 2e-05,
2288
+ "loss": 144.2338,
2289
+ "step": 368
2290
+ },
2291
+ {
2292
+ "epoch": 3.69,
2293
+ "learning_rate": 2e-05,
2294
+ "loss": 158.5771,
2295
+ "step": 369
2296
+ },
2297
+ {
2298
+ "epoch": 3.7,
2299
+ "learning_rate": 2e-05,
2300
+ "loss": 163.8886,
2301
+ "step": 370
2302
+ },
2303
+ {
2304
+ "epoch": 3.71,
2305
+ "learning_rate": 2e-05,
2306
+ "loss": 151.2742,
2307
+ "step": 371
2308
+ },
2309
+ {
2310
+ "epoch": 3.72,
2311
+ "learning_rate": 2e-05,
2312
+ "loss": 169.2691,
2313
+ "step": 372
2314
+ },
2315
+ {
2316
+ "epoch": 3.73,
2317
+ "learning_rate": 2e-05,
2318
+ "loss": 125.0493,
2319
+ "step": 373
2320
+ },
2321
+ {
2322
+ "epoch": 3.74,
2323
+ "learning_rate": 2e-05,
2324
+ "loss": 144.3527,
2325
+ "step": 374
2326
+ },
2327
+ {
2328
+ "epoch": 3.75,
2329
+ "learning_rate": 2e-05,
2330
+ "loss": 210.2006,
2331
+ "step": 375
2332
+ },
2333
+ {
2334
+ "epoch": 3.76,
2335
+ "learning_rate": 2e-05,
2336
+ "loss": 162.8882,
2337
+ "step": 376
2338
+ },
2339
+ {
2340
+ "epoch": 3.77,
2341
+ "learning_rate": 2e-05,
2342
+ "loss": 163.0425,
2343
+ "step": 377
2344
+ },
2345
+ {
2346
+ "epoch": 3.78,
2347
+ "learning_rate": 2e-05,
2348
+ "loss": 144.6404,
2349
+ "step": 378
2350
+ },
2351
+ {
2352
+ "epoch": 3.79,
2353
+ "learning_rate": 2e-05,
2354
+ "loss": 169.7259,
2355
+ "step": 379
2356
+ },
2357
+ {
2358
+ "epoch": 3.8,
2359
+ "learning_rate": 2e-05,
2360
+ "loss": 117.309,
2361
+ "step": 380
2362
+ },
2363
+ {
2364
+ "epoch": 3.81,
2365
+ "learning_rate": 2e-05,
2366
+ "loss": 179.2435,
2367
+ "step": 381
2368
+ },
2369
+ {
2370
+ "epoch": 3.82,
2371
+ "learning_rate": 2e-05,
2372
+ "loss": 156.4202,
2373
+ "step": 382
2374
+ },
2375
+ {
2376
+ "epoch": 3.83,
2377
+ "learning_rate": 2e-05,
2378
+ "loss": 212.8055,
2379
+ "step": 383
2380
+ },
2381
+ {
2382
+ "epoch": 3.84,
2383
+ "learning_rate": 2e-05,
2384
+ "loss": 130.1424,
2385
+ "step": 384
2386
+ },
2387
+ {
2388
+ "epoch": 3.85,
2389
+ "learning_rate": 2e-05,
2390
+ "loss": 143.6542,
2391
+ "step": 385
2392
+ },
2393
+ {
2394
+ "epoch": 3.86,
2395
+ "learning_rate": 2e-05,
2396
+ "loss": 193.6444,
2397
+ "step": 386
2398
+ },
2399
+ {
2400
+ "epoch": 3.87,
2401
+ "learning_rate": 2e-05,
2402
+ "loss": 176.6723,
2403
+ "step": 387
2404
+ },
2405
+ {
2406
+ "epoch": 3.88,
2407
+ "learning_rate": 2e-05,
2408
+ "loss": 150.6032,
2409
+ "step": 388
2410
+ },
2411
+ {
2412
+ "epoch": 3.89,
2413
+ "learning_rate": 2e-05,
2414
+ "loss": 146.1843,
2415
+ "step": 389
2416
+ },
2417
+ {
2418
+ "epoch": 3.9,
2419
+ "learning_rate": 2e-05,
2420
+ "loss": 152.6586,
2421
+ "step": 390
2422
+ },
2423
+ {
2424
+ "epoch": 3.91,
2425
+ "learning_rate": 2e-05,
2426
+ "loss": 162.0343,
2427
+ "step": 391
2428
+ },
2429
+ {
2430
+ "epoch": 3.92,
2431
+ "learning_rate": 2e-05,
2432
+ "loss": 157.9043,
2433
+ "step": 392
2434
+ },
2435
+ {
2436
+ "epoch": 3.93,
2437
+ "learning_rate": 2e-05,
2438
+ "loss": 140.6674,
2439
+ "step": 393
2440
+ },
2441
+ {
2442
+ "epoch": 3.94,
2443
+ "learning_rate": 2e-05,
2444
+ "loss": 186.9754,
2445
+ "step": 394
2446
+ },
2447
+ {
2448
+ "epoch": 3.95,
2449
+ "learning_rate": 2e-05,
2450
+ "loss": 157.4324,
2451
+ "step": 395
2452
+ },
2453
+ {
2454
+ "epoch": 3.96,
2455
+ "learning_rate": 2e-05,
2456
+ "loss": 151.3968,
2457
+ "step": 396
2458
+ },
2459
+ {
2460
+ "epoch": 3.97,
2461
+ "learning_rate": 2e-05,
2462
+ "loss": 182.1434,
2463
+ "step": 397
2464
+ },
2465
+ {
2466
+ "epoch": 3.98,
2467
+ "learning_rate": 2e-05,
2468
+ "loss": 164.4491,
2469
+ "step": 398
2470
+ },
2471
+ {
2472
+ "epoch": 3.99,
2473
+ "learning_rate": 2e-05,
2474
+ "loss": 151.253,
2475
+ "step": 399
2476
+ },
2477
+ {
2478
+ "epoch": 4.0,
2479
+ "learning_rate": 2e-05,
2480
+ "loss": 136.0101,
2481
+ "step": 400
2482
+ },
2483
+ {
2484
+ "epoch": 4.0,
2485
+ "eval_loss": 206.38494873046875,
2486
+ "eval_runtime": 91.5142,
2487
+ "eval_samples_per_second": 17.484,
2488
+ "eval_steps_per_second": 1.093,
2489
+ "step": 400
2490
+ }
2491
+ ],
2492
+ "logging_steps": 1.0,
2493
+ "max_steps": 400,
2494
+ "num_input_tokens_seen": 0,
2495
+ "num_train_epochs": 4,
2496
+ "save_steps": 40,
2497
+ "total_flos": 1.627150507573248e+17,
2498
+ "train_batch_size": 16,
2499
+ "trial_name": null,
2500
+ "trial_params": null
2501
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fba6d1b6db4cd757ebccd472bdb8f133d6cf1722737cf67f0403e4040c3f41c
3
+ size 6392
zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 1, 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:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
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 _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``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``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``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``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``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``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info(f"Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info(f"Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)