Hugging Face's logo

yuchenlin
/
swift_sw

Text2Text Generation
Transformers
PyTorch
t5
text-generation-inference
Inference Endpoints
Model card Files Community
1
yuchenlin commited on
Commit
bf2b3df
•
1 Parent(s): bdc1ff5

add swift

Browse files
Files changed (14) hide show
  1. config.json +32 -0
  2. latest +1 -0
  3. pytorch_model.bin +3 -0
  4. rng_state_0.pth +3 -0
  5. rng_state_1.pth +3 -0
  6. rng_state_2.pth +3 -0
  7. rng_state_3.pth +3 -0
  8. special_tokens_map.json +107 -0
  9. spiece.model +3 -0
  10. tokenizer.json +0 -0
  11. tokenizer_config.json +113 -0
  12. trainer_state.json +3056 -0
  13. training_args.bin +3 -0
  14. zero_to_fp32.py +483 -0
config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "google/flan-t5-large",
3
+ "architectures": [
4
+ "T5ForConditionalGeneration"
5
+ ],
6
+ "d_ff": 2816,
7
+ "d_kv": 64,
8
+ "d_model": 1024,
9
+ "decoder_start_token_id": 0,
10
+ "dense_act_fn": "gelu_new",
11
+ "dropout_rate": 0.1,
12
+ "eos_token_id": 1,
13
+ "feed_forward_proj": "gated-gelu",
14
+ "initializer_factor": 1.0,
15
+ "is_encoder_decoder": true,
16
+ "is_gated_act": true,
17
+ "layer_norm_epsilon": 1e-06,
18
+ "model_type": "t5",
19
+ "n_positions": 512,
20
+ "num_decoder_layers": 24,
21
+ "num_heads": 16,
22
+ "num_layers": 24,
23
+ "output_past": true,
24
+ "pad_token_id": 0,
25
+ "relative_attention_max_distance": 128,
26
+ "relative_attention_num_buckets": 32,
27
+ "tie_word_embeddings": false,
28
+ "torch_dtype": "bfloat16",
29
+ "transformers_version": "4.20.1",
30
+ "use_cache": true,
31
+ "vocab_size": 32100
32
+ }
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step500
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7e43547ce8de985d50fb374244f949110ccb7e3d3f03cf22e48b1a5d6081a117
3
+ size 1566265661
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:23c0b80749451122a0c35509de37ce46c662429441688d24da359521b642ceb0
3
+ size 14583
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d64a9dced26be7f8a65b82d693d2886088b72532e148dca01e989034289b1561
3
+ size 14583
rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7756e15760b599c2ba12e5bb394fb20bf52305994ffa46043e3217226a120b3
3
+ size 14583
rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89e66c84ba31c45cacaf0de68bf6c9430491d8136a660efe9512196b534c5cf5
3
+ size 14583
special_tokens_map.json ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<extra_id_0>",
4
+ "<extra_id_1>",
5
+ "<extra_id_2>",
6
+ "<extra_id_3>",
7
+ "<extra_id_4>",
8
+ "<extra_id_5>",
9
+ "<extra_id_6>",
10
+ "<extra_id_7>",
11
+ "<extra_id_8>",
12
+ "<extra_id_9>",
13
+ "<extra_id_10>",
14
+ "<extra_id_11>",
15
+ "<extra_id_12>",
16
+ "<extra_id_13>",
17
+ "<extra_id_14>",
18
+ "<extra_id_15>",
19
+ "<extra_id_16>",
20
+ "<extra_id_17>",
21
+ "<extra_id_18>",
22
+ "<extra_id_19>",
23
+ "<extra_id_20>",
24
+ "<extra_id_21>",
25
+ "<extra_id_22>",
26
+ "<extra_id_23>",
27
+ "<extra_id_24>",
28
+ "<extra_id_25>",
29
+ "<extra_id_26>",
30
+ "<extra_id_27>",
31
+ "<extra_id_28>",
32
+ "<extra_id_29>",
33
+ "<extra_id_30>",
34
+ "<extra_id_31>",
35
+ "<extra_id_32>",
36
+ "<extra_id_33>",
37
+ "<extra_id_34>",
38
+ "<extra_id_35>",
39
+ "<extra_id_36>",
40
+ "<extra_id_37>",
41
+ "<extra_id_38>",
42
+ "<extra_id_39>",
43
+ "<extra_id_40>",
44
+ "<extra_id_41>",
45
+ "<extra_id_42>",
46
+ "<extra_id_43>",
47
+ "<extra_id_44>",
48
+ "<extra_id_45>",
49
+ "<extra_id_46>",
50
+ "<extra_id_47>",
51
+ "<extra_id_48>",
52
+ "<extra_id_49>",
53
+ "<extra_id_50>",
54
+ "<extra_id_51>",
55
+ "<extra_id_52>",
56
+ "<extra_id_53>",
57
+ "<extra_id_54>",
58
+ "<extra_id_55>",
59
+ "<extra_id_56>",
60
+ "<extra_id_57>",
61
+ "<extra_id_58>",
62
+ "<extra_id_59>",
63
+ "<extra_id_60>",
64
+ "<extra_id_61>",
65
+ "<extra_id_62>",
66
+ "<extra_id_63>",
67
+ "<extra_id_64>",
68
+ "<extra_id_65>",
69
+ "<extra_id_66>",
70
+ "<extra_id_67>",
71
+ "<extra_id_68>",
72
+ "<extra_id_69>",
73
+ "<extra_id_70>",
74
+ "<extra_id_71>",
75
+ "<extra_id_72>",
76
+ "<extra_id_73>",
77
+ "<extra_id_74>",
78
+ "<extra_id_75>",
79
+ "<extra_id_76>",
80
+ "<extra_id_77>",
81
+ "<extra_id_78>",
82
+ "<extra_id_79>",
83
+ "<extra_id_80>",
84
+ "<extra_id_81>",
85
+ "<extra_id_82>",
86
+ "<extra_id_83>",
87
+ "<extra_id_84>",
88
+ "<extra_id_85>",
89
+ "<extra_id_86>",
90
+ "<extra_id_87>",
91
+ "<extra_id_88>",
92
+ "<extra_id_89>",
93
+ "<extra_id_90>",
94
+ "<extra_id_91>",
95
+ "<extra_id_92>",
96
+ "<extra_id_93>",
97
+ "<extra_id_94>",
98
+ "<extra_id_95>",
99
+ "<extra_id_96>",
100
+ "<extra_id_97>",
101
+ "<extra_id_98>",
102
+ "<extra_id_99>"
103
+ ],
104
+ "eos_token": "</s>",
105
+ "pad_token": "<pad>",
106
+ "unk_token": "<unk>"
107
+ }
spiece.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d60acb128cf7b7f2536e8f38a5b18a05535c9e14c7a355904270e15b0945ea86
3
+ size 791656
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,113 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<extra_id_0>",
4
+ "<extra_id_1>",
5
+ "<extra_id_2>",
6
+ "<extra_id_3>",
7
+ "<extra_id_4>",
8
+ "<extra_id_5>",
9
+ "<extra_id_6>",
10
+ "<extra_id_7>",
11
+ "<extra_id_8>",
12
+ "<extra_id_9>",
13
+ "<extra_id_10>",
14
+ "<extra_id_11>",
15
+ "<extra_id_12>",
16
+ "<extra_id_13>",
17
+ "<extra_id_14>",
18
+ "<extra_id_15>",
19
+ "<extra_id_16>",
20
+ "<extra_id_17>",
21
+ "<extra_id_18>",
22
+ "<extra_id_19>",
23
+ "<extra_id_20>",
24
+ "<extra_id_21>",
25
+ "<extra_id_22>",
26
+ "<extra_id_23>",
27
+ "<extra_id_24>",
28
+ "<extra_id_25>",
29
+ "<extra_id_26>",
30
+ "<extra_id_27>",
31
+ "<extra_id_28>",
32
+ "<extra_id_29>",
33
+ "<extra_id_30>",
34
+ "<extra_id_31>",
35
+ "<extra_id_32>",
36
+ "<extra_id_33>",
37
+ "<extra_id_34>",
38
+ "<extra_id_35>",
39
+ "<extra_id_36>",
40
+ "<extra_id_37>",
41
+ "<extra_id_38>",
42
+ "<extra_id_39>",
43
+ "<extra_id_40>",
44
+ "<extra_id_41>",
45
+ "<extra_id_42>",
46
+ "<extra_id_43>",
47
+ "<extra_id_44>",
48
+ "<extra_id_45>",
49
+ "<extra_id_46>",
50
+ "<extra_id_47>",
51
+ "<extra_id_48>",
52
+ "<extra_id_49>",
53
+ "<extra_id_50>",
54
+ "<extra_id_51>",
55
+ "<extra_id_52>",
56
+ "<extra_id_53>",
57
+ "<extra_id_54>",
58
+ "<extra_id_55>",
59
+ "<extra_id_56>",
60
+ "<extra_id_57>",
61
+ "<extra_id_58>",
62
+ "<extra_id_59>",
63
+ "<extra_id_60>",
64
+ "<extra_id_61>",
65
+ "<extra_id_62>",
66
+ "<extra_id_63>",
67
+ "<extra_id_64>",
68
+ "<extra_id_65>",
69
+ "<extra_id_66>",
70
+ "<extra_id_67>",
71
+ "<extra_id_68>",
72
+ "<extra_id_69>",
73
+ "<extra_id_70>",
74
+ "<extra_id_71>",
75
+ "<extra_id_72>",
76
+ "<extra_id_73>",
77
+ "<extra_id_74>",
78
+ "<extra_id_75>",
79
+ "<extra_id_76>",
80
+ "<extra_id_77>",
81
+ "<extra_id_78>",
82
+ "<extra_id_79>",
83
+ "<extra_id_80>",
84
+ "<extra_id_81>",
85
+ "<extra_id_82>",
86
+ "<extra_id_83>",
87
+ "<extra_id_84>",
88
+ "<extra_id_85>",
89
+ "<extra_id_86>",
90
+ "<extra_id_87>",
91
+ "<extra_id_88>",
92
+ "<extra_id_89>",
93
+ "<extra_id_90>",
94
+ "<extra_id_91>",
95
+ "<extra_id_92>",
96
+ "<extra_id_93>",
97
+ "<extra_id_94>",
98
+ "<extra_id_95>",
99
+ "<extra_id_96>",
100
+ "<extra_id_97>",
101
+ "<extra_id_98>",
102
+ "<extra_id_99>"
103
+ ],
104
+ "eos_token": "</s>",
105
+ "extra_ids": 100,
106
+ "model_max_length": 512,
107
+ "name_or_path": "google/flan-t5-large",
108
+ "pad_token": "<pad>",
109
+ "sp_model_kwargs": {},
110
+ "special_tokens_map_file": "/home/younes_huggingface_co/.cache/huggingface/hub/models--google--t5-v1_1-large/snapshots/314bc112b191ec17b625ba81438dc73d6c23659d/special_tokens_map.json",
111
+ "tokenizer_class": "T5Tokenizer",
112
+ "unk_token": "<unk>"
113
+ }
trainer_state.json ADDED
@@ -0,0 +1,3056 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.10144635289907455,
3
+ "best_model_checkpoint": "model_ckpts/flan_large_0411/checkpoint-500",
4
+ "epoch": 1.0309119010819165,
5
+ "global_step": 500,
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": 0.0,
13
+ "loss": 2.0859,
14
+ "step": 1
15
+ },
16
+ {
17
+ "epoch": 0.0,
18
+ "learning_rate": 0.0,
19
+ "loss": 1.7471,
20
+ "step": 2
21
+ },
22
+ {
23
+ "epoch": 0.01,
24
+ "learning_rate": 0.0001,
25
+ "loss": 1.509,
26
+ "step": 3
27
+ },
28
+ {
29
+ "epoch": 0.01,
30
+ "learning_rate": 0.0001,
31
+ "loss": 1.552,
32
+ "step": 4
33
+ },
34
+ {
35
+ "epoch": 0.01,
36
+ "learning_rate": 0.0001,
37
+ "loss": 1.2214,
38
+ "step": 5
39
+ },
40
+ {
41
+ "epoch": 0.01,
42
+ "learning_rate": 0.0001,
43
+ "loss": 1.1383,
44
+ "step": 6
45
+ },
46
+ {
47
+ "epoch": 0.01,
48
+ "learning_rate": 0.0001,
49
+ "loss": 1.1124,
50
+ "step": 7
51
+ },
52
+ {
53
+ "epoch": 0.02,
54
+ "learning_rate": 0.0001,
55
+ "loss": 0.9961,
56
+ "step": 8
57
+ },
58
+ {
59
+ "epoch": 0.02,
60
+ "learning_rate": 0.0001,
61
+ "loss": 0.7675,
62
+ "step": 9
63
+ },
64
+ {
65
+ "epoch": 0.02,
66
+ "learning_rate": 0.0001,
67
+ "loss": 0.8992,
68
+ "step": 10
69
+ },
70
+ {
71
+ "epoch": 0.02,
72
+ "learning_rate": 0.0001,
73
+ "loss": 0.7213,
74
+ "step": 11
75
+ },
76
+ {
77
+ "epoch": 0.02,
78
+ "learning_rate": 0.0001,
79
+ "loss": 0.8539,
80
+ "step": 12
81
+ },
82
+ {
83
+ "epoch": 0.03,
84
+ "learning_rate": 0.0001,
85
+ "loss": 0.6665,
86
+ "step": 13
87
+ },
88
+ {
89
+ "epoch": 0.03,
90
+ "learning_rate": 0.0001,
91
+ "loss": 0.6982,
92
+ "step": 14
93
+ },
94
+ {
95
+ "epoch": 0.03,
96
+ "learning_rate": 0.0001,
97
+ "loss": 0.6082,
98
+ "step": 15
99
+ },
100
+ {
101
+ "epoch": 0.03,
102
+ "learning_rate": 0.0001,
103
+ "loss": 0.6788,
104
+ "step": 16
105
+ },
106
+ {
107
+ "epoch": 0.04,
108
+ "learning_rate": 0.0001,
109
+ "loss": 0.6445,
110
+ "step": 17
111
+ },
112
+ {
113
+ "epoch": 0.04,
114
+ "learning_rate": 0.0001,
115
+ "loss": 0.5978,
116
+ "step": 18
117
+ },
118
+ {
119
+ "epoch": 0.04,
120
+ "learning_rate": 0.0001,
121
+ "loss": 0.6115,
122
+ "step": 19
123
+ },
124
+ {
125
+ "epoch": 0.04,
126
+ "learning_rate": 0.0001,
127
+ "loss": 0.5072,
128
+ "step": 20
129
+ },
130
+ {
131
+ "epoch": 0.04,
132
+ "learning_rate": 0.0001,
133
+ "loss": 0.5304,
134
+ "step": 21
135
+ },
136
+ {
137
+ "epoch": 0.05,
138
+ "learning_rate": 0.0001,
139
+ "loss": 0.4517,
140
+ "step": 22
141
+ },
142
+ {
143
+ "epoch": 0.05,
144
+ "learning_rate": 0.0001,
145
+ "loss": 0.4339,
146
+ "step": 23
147
+ },
148
+ {
149
+ "epoch": 0.05,
150
+ "learning_rate": 0.0001,
151
+ "loss": 0.6107,
152
+ "step": 24
153
+ },
154
+ {
155
+ "epoch": 0.05,
156
+ "learning_rate": 0.0001,
157
+ "loss": 0.4486,
158
+ "step": 25
159
+ },
160
+ {
161
+ "epoch": 0.05,
162
+ "learning_rate": 0.0001,
163
+ "loss": 0.5635,
164
+ "step": 26
165
+ },
166
+ {
167
+ "epoch": 0.06,
168
+ "learning_rate": 0.0001,
169
+ "loss": 0.517,
170
+ "step": 27
171
+ },
172
+ {
173
+ "epoch": 0.06,
174
+ "learning_rate": 0.0001,
175
+ "loss": 0.4418,
176
+ "step": 28
177
+ },
178
+ {
179
+ "epoch": 0.06,
180
+ "learning_rate": 0.0001,
181
+ "loss": 0.5549,
182
+ "step": 29
183
+ },
184
+ {
185
+ "epoch": 0.06,
186
+ "learning_rate": 0.0001,
187
+ "loss": 0.4133,
188
+ "step": 30
189
+ },
190
+ {
191
+ "epoch": 0.06,
192
+ "learning_rate": 0.0001,
193
+ "loss": 0.4221,
194
+ "step": 31
195
+ },
196
+ {
197
+ "epoch": 0.07,
198
+ "learning_rate": 0.0001,
199
+ "loss": 0.4018,
200
+ "step": 32
201
+ },
202
+ {
203
+ "epoch": 0.07,
204
+ "learning_rate": 0.0001,
205
+ "loss": 0.445,
206
+ "step": 33
207
+ },
208
+ {
209
+ "epoch": 0.07,
210
+ "learning_rate": 0.0001,
211
+ "loss": 0.3948,
212
+ "step": 34
213
+ },
214
+ {
215
+ "epoch": 0.07,
216
+ "learning_rate": 0.0001,
217
+ "loss": 0.3131,
218
+ "step": 35
219
+ },
220
+ {
221
+ "epoch": 0.07,
222
+ "learning_rate": 0.0001,
223
+ "loss": 0.2682,
224
+ "step": 36
225
+ },
226
+ {
227
+ "epoch": 0.08,
228
+ "learning_rate": 0.0001,
229
+ "loss": 0.2656,
230
+ "step": 37
231
+ },
232
+ {
233
+ "epoch": 0.08,
234
+ "learning_rate": 0.0001,
235
+ "loss": 0.3366,
236
+ "step": 38
237
+ },
238
+ {
239
+ "epoch": 0.08,
240
+ "learning_rate": 0.0001,
241
+ "loss": 0.3304,
242
+ "step": 39
243
+ },
244
+ {
245
+ "epoch": 0.08,
246
+ "learning_rate": 0.0001,
247
+ "loss": 0.3858,
248
+ "step": 40
249
+ },
250
+ {
251
+ "epoch": 0.08,
252
+ "learning_rate": 0.0001,
253
+ "loss": 0.3185,
254
+ "step": 41
255
+ },
256
+ {
257
+ "epoch": 0.09,
258
+ "learning_rate": 0.0001,
259
+ "loss": 0.3853,
260
+ "step": 42
261
+ },
262
+ {
263
+ "epoch": 0.09,
264
+ "learning_rate": 0.0001,
265
+ "loss": 0.3407,
266
+ "step": 43
267
+ },
268
+ {
269
+ "epoch": 0.09,
270
+ "learning_rate": 0.0001,
271
+ "loss": 0.398,
272
+ "step": 44
273
+ },
274
+ {
275
+ "epoch": 0.09,
276
+ "learning_rate": 0.0001,
277
+ "loss": 0.3724,
278
+ "step": 45
279
+ },
280
+ {
281
+ "epoch": 0.09,
282
+ "learning_rate": 0.0001,
283
+ "loss": 0.3071,
284
+ "step": 46
285
+ },
286
+ {
287
+ "epoch": 0.1,
288
+ "learning_rate": 0.0001,
289
+ "loss": 0.2107,
290
+ "step": 47
291
+ },
292
+ {
293
+ "epoch": 0.1,
294
+ "learning_rate": 0.0001,
295
+ "loss": 0.3249,
296
+ "step": 48
297
+ },
298
+ {
299
+ "epoch": 0.1,
300
+ "learning_rate": 0.0001,
301
+ "loss": 0.2247,
302
+ "step": 49
303
+ },
304
+ {
305
+ "epoch": 0.1,
306
+ "learning_rate": 0.0001,
307
+ "loss": 0.3205,
308
+ "step": 50
309
+ },
310
+ {
311
+ "epoch": 0.11,
312
+ "learning_rate": 0.0001,
313
+ "loss": 0.3112,
314
+ "step": 51
315
+ },
316
+ {
317
+ "epoch": 0.11,
318
+ "learning_rate": 0.0001,
319
+ "loss": 0.2215,
320
+ "step": 52
321
+ },
322
+ {
323
+ "epoch": 0.11,
324
+ "learning_rate": 0.0001,
325
+ "loss": 0.2689,
326
+ "step": 53
327
+ },
328
+ {
329
+ "epoch": 0.11,
330
+ "learning_rate": 0.0001,
331
+ "loss": 0.2726,
332
+ "step": 54
333
+ },
334
+ {
335
+ "epoch": 0.11,
336
+ "learning_rate": 0.0001,
337
+ "loss": 0.3124,
338
+ "step": 55
339
+ },
340
+ {
341
+ "epoch": 0.12,
342
+ "learning_rate": 0.0001,
343
+ "loss": 0.2395,
344
+ "step": 56
345
+ },
346
+ {
347
+ "epoch": 0.12,
348
+ "learning_rate": 0.0001,
349
+ "loss": 0.2271,
350
+ "step": 57
351
+ },
352
+ {
353
+ "epoch": 0.12,
354
+ "learning_rate": 0.0001,
355
+ "loss": 0.3099,
356
+ "step": 58
357
+ },
358
+ {
359
+ "epoch": 0.12,
360
+ "learning_rate": 0.0001,
361
+ "loss": 0.2904,
362
+ "step": 59
363
+ },
364
+ {
365
+ "epoch": 0.12,
366
+ "learning_rate": 0.0001,
367
+ "loss": 0.3311,
368
+ "step": 60
369
+ },
370
+ {
371
+ "epoch": 0.13,
372
+ "learning_rate": 0.0001,
373
+ "loss": 0.2102,
374
+ "step": 61
375
+ },
376
+ {
377
+ "epoch": 0.13,
378
+ "learning_rate": 0.0001,
379
+ "loss": 0.2088,
380
+ "step": 62
381
+ },
382
+ {
383
+ "epoch": 0.13,
384
+ "learning_rate": 0.0001,
385
+ "loss": 0.2664,
386
+ "step": 63
387
+ },
388
+ {
389
+ "epoch": 0.13,
390
+ "learning_rate": 0.0001,
391
+ "loss": 0.241,
392
+ "step": 64
393
+ },
394
+ {
395
+ "epoch": 0.13,
396
+ "learning_rate": 0.0001,
397
+ "loss": 0.2373,
398
+ "step": 65
399
+ },
400
+ {
401
+ "epoch": 0.14,
402
+ "learning_rate": 0.0001,
403
+ "loss": 0.215,
404
+ "step": 66
405
+ },
406
+ {
407
+ "epoch": 0.14,
408
+ "learning_rate": 0.0001,
409
+ "loss": 0.2384,
410
+ "step": 67
411
+ },
412
+ {
413
+ "epoch": 0.14,
414
+ "learning_rate": 0.0001,
415
+ "loss": 0.2232,
416
+ "step": 68
417
+ },
418
+ {
419
+ "epoch": 0.14,
420
+ "learning_rate": 0.0001,
421
+ "loss": 0.27,
422
+ "step": 69
423
+ },
424
+ {
425
+ "epoch": 0.14,
426
+ "learning_rate": 0.0001,
427
+ "loss": 0.2309,
428
+ "step": 70
429
+ },
430
+ {
431
+ "epoch": 0.15,
432
+ "learning_rate": 0.0001,
433
+ "loss": 0.2802,
434
+ "step": 71
435
+ },
436
+ {
437
+ "epoch": 0.15,
438
+ "learning_rate": 0.0001,
439
+ "loss": 0.1925,
440
+ "step": 72
441
+ },
442
+ {
443
+ "epoch": 0.15,
444
+ "learning_rate": 0.0001,
445
+ "loss": 0.2079,
446
+ "step": 73
447
+ },
448
+ {
449
+ "epoch": 0.15,
450
+ "learning_rate": 0.0001,
451
+ "loss": 0.2097,
452
+ "step": 74
453
+ },
454
+ {
455
+ "epoch": 0.15,
456
+ "learning_rate": 0.0001,
457
+ "loss": 0.2146,
458
+ "step": 75
459
+ },
460
+ {
461
+ "epoch": 0.16,
462
+ "learning_rate": 0.0001,
463
+ "loss": 0.1824,
464
+ "step": 76
465
+ },
466
+ {
467
+ "epoch": 0.16,
468
+ "learning_rate": 0.0001,
469
+ "loss": 0.2179,
470
+ "step": 77
471
+ },
472
+ {
473
+ "epoch": 0.16,
474
+ "learning_rate": 0.0001,
475
+ "loss": 0.203,
476
+ "step": 78
477
+ },
478
+ {
479
+ "epoch": 0.16,
480
+ "learning_rate": 0.0001,
481
+ "loss": 0.2314,
482
+ "step": 79
483
+ },
484
+ {
485
+ "epoch": 0.16,
486
+ "learning_rate": 0.0001,
487
+ "loss": 0.2374,
488
+ "step": 80
489
+ },
490
+ {
491
+ "epoch": 0.17,
492
+ "learning_rate": 0.0001,
493
+ "loss": 0.1772,
494
+ "step": 81
495
+ },
496
+ {
497
+ "epoch": 0.17,
498
+ "learning_rate": 0.0001,
499
+ "loss": 0.2744,
500
+ "step": 82
501
+ },
502
+ {
503
+ "epoch": 0.17,
504
+ "learning_rate": 0.0001,
505
+ "loss": 0.2191,
506
+ "step": 83
507
+ },
508
+ {
509
+ "epoch": 0.17,
510
+ "learning_rate": 0.0001,
511
+ "loss": 0.2669,
512
+ "step": 84
513
+ },
514
+ {
515
+ "epoch": 0.18,
516
+ "learning_rate": 0.0001,
517
+ "loss": 0.2578,
518
+ "step": 85
519
+ },
520
+ {
521
+ "epoch": 0.18,
522
+ "learning_rate": 0.0001,
523
+ "loss": 0.2093,
524
+ "step": 86
525
+ },
526
+ {
527
+ "epoch": 0.18,
528
+ "learning_rate": 0.0001,
529
+ "loss": 0.2495,
530
+ "step": 87
531
+ },
532
+ {
533
+ "epoch": 0.18,
534
+ "learning_rate": 0.0001,
535
+ "loss": 0.1716,
536
+ "step": 88
537
+ },
538
+ {
539
+ "epoch": 0.18,
540
+ "learning_rate": 0.0001,
541
+ "loss": 0.1964,
542
+ "step": 89
543
+ },
544
+ {
545
+ "epoch": 0.19,
546
+ "learning_rate": 0.0001,
547
+ "loss": 0.1601,
548
+ "step": 90
549
+ },
550
+ {
551
+ "epoch": 0.19,
552
+ "learning_rate": 0.0001,
553
+ "loss": 0.1921,
554
+ "step": 91
555
+ },
556
+ {
557
+ "epoch": 0.19,
558
+ "learning_rate": 0.0001,
559
+ "loss": 0.2021,
560
+ "step": 92
561
+ },
562
+ {
563
+ "epoch": 0.19,
564
+ "learning_rate": 0.0001,
565
+ "loss": 0.1933,
566
+ "step": 93
567
+ },
568
+ {
569
+ "epoch": 0.19,
570
+ "learning_rate": 0.0001,
571
+ "loss": 0.1747,
572
+ "step": 94
573
+ },
574
+ {
575
+ "epoch": 0.2,
576
+ "learning_rate": 0.0001,
577
+ "loss": 0.3239,
578
+ "step": 95
579
+ },
580
+ {
581
+ "epoch": 0.2,
582
+ "learning_rate": 0.0001,
583
+ "loss": 0.2016,
584
+ "step": 96
585
+ },
586
+ {
587
+ "epoch": 0.2,
588
+ "learning_rate": 0.0001,
589
+ "loss": 0.1786,
590
+ "step": 97
591
+ },
592
+ {
593
+ "epoch": 0.2,
594
+ "learning_rate": 0.0001,
595
+ "loss": 0.1484,
596
+ "step": 98
597
+ },
598
+ {
599
+ "epoch": 0.2,
600
+ "learning_rate": 0.0001,
601
+ "loss": 0.2085,
602
+ "step": 99
603
+ },
604
+ {
605
+ "epoch": 0.21,
606
+ "learning_rate": 0.0001,
607
+ "loss": 0.2037,
608
+ "step": 100
609
+ },
610
+ {
611
+ "epoch": 0.21,
612
+ "eval_loss": 0.2043856978416443,
613
+ "eval_runtime": 949.0923,
614
+ "eval_samples_per_second": 34.016,
615
+ "eval_steps_per_second": 0.267,
616
+ "step": 100
617
+ },
618
+ {
619
+ "epoch": 0.21,
620
+ "learning_rate": 0.0001,
621
+ "loss": 0.209,
622
+ "step": 101
623
+ },
624
+ {
625
+ "epoch": 0.21,
626
+ "learning_rate": 0.0001,
627
+ "loss": 0.2336,
628
+ "step": 102
629
+ },
630
+ {
631
+ "epoch": 0.21,
632
+ "learning_rate": 0.0001,
633
+ "loss": 0.2064,
634
+ "step": 103
635
+ },
636
+ {
637
+ "epoch": 0.21,
638
+ "learning_rate": 0.0001,
639
+ "loss": 0.1811,
640
+ "step": 104
641
+ },
642
+ {
643
+ "epoch": 0.22,
644
+ "learning_rate": 0.0001,
645
+ "loss": 0.2142,
646
+ "step": 105
647
+ },
648
+ {
649
+ "epoch": 0.22,
650
+ "learning_rate": 0.0001,
651
+ "loss": 0.1959,
652
+ "step": 106
653
+ },
654
+ {
655
+ "epoch": 0.22,
656
+ "learning_rate": 0.0001,
657
+ "loss": 0.1612,
658
+ "step": 107
659
+ },
660
+ {
661
+ "epoch": 0.22,
662
+ "learning_rate": 0.0001,
663
+ "loss": 0.1505,
664
+ "step": 108
665
+ },
666
+ {
667
+ "epoch": 0.22,
668
+ "learning_rate": 0.0001,
669
+ "loss": 0.2053,
670
+ "step": 109
671
+ },
672
+ {
673
+ "epoch": 0.23,
674
+ "learning_rate": 0.0001,
675
+ "loss": 0.14,
676
+ "step": 110
677
+ },
678
+ {
679
+ "epoch": 0.23,
680
+ "learning_rate": 0.0001,
681
+ "loss": 0.2863,
682
+ "step": 111
683
+ },
684
+ {
685
+ "epoch": 0.23,
686
+ "learning_rate": 0.0001,
687
+ "loss": 0.1826,
688
+ "step": 112
689
+ },
690
+ {
691
+ "epoch": 0.23,
692
+ "learning_rate": 0.0001,
693
+ "loss": 0.146,
694
+ "step": 113
695
+ },
696
+ {
697
+ "epoch": 0.23,
698
+ "learning_rate": 0.0001,
699
+ "loss": 0.1208,
700
+ "step": 114
701
+ },
702
+ {
703
+ "epoch": 0.24,
704
+ "learning_rate": 0.0001,
705
+ "loss": 0.1573,
706
+ "step": 115
707
+ },
708
+ {
709
+ "epoch": 0.24,
710
+ "learning_rate": 0.0001,
711
+ "loss": 0.2216,
712
+ "step": 116
713
+ },
714
+ {
715
+ "epoch": 0.24,
716
+ "learning_rate": 0.0001,
717
+ "loss": 0.1939,
718
+ "step": 117
719
+ },
720
+ {
721
+ "epoch": 0.24,
722
+ "learning_rate": 0.0001,
723
+ "loss": 0.1872,
724
+ "step": 118
725
+ },
726
+ {
727
+ "epoch": 0.25,
728
+ "learning_rate": 0.0001,
729
+ "loss": 0.1737,
730
+ "step": 119
731
+ },
732
+ {
733
+ "epoch": 0.25,
734
+ "learning_rate": 0.0001,
735
+ "loss": 0.1509,
736
+ "step": 120
737
+ },
738
+ {
739
+ "epoch": 0.25,
740
+ "learning_rate": 0.0001,
741
+ "loss": 0.1952,
742
+ "step": 121
743
+ },
744
+ {
745
+ "epoch": 0.25,
746
+ "learning_rate": 0.0001,
747
+ "loss": 0.1518,
748
+ "step": 122
749
+ },
750
+ {
751
+ "epoch": 0.25,
752
+ "learning_rate": 0.0001,
753
+ "loss": 0.1514,
754
+ "step": 123
755
+ },
756
+ {
757
+ "epoch": 0.26,
758
+ "learning_rate": 0.0001,
759
+ "loss": 0.166,
760
+ "step": 124
761
+ },
762
+ {
763
+ "epoch": 0.26,
764
+ "learning_rate": 0.0001,
765
+ "loss": 0.1327,
766
+ "step": 125
767
+ },
768
+ {
769
+ "epoch": 0.26,
770
+ "learning_rate": 0.0001,
771
+ "loss": 0.1821,
772
+ "step": 126
773
+ },
774
+ {
775
+ "epoch": 0.26,
776
+ "learning_rate": 0.0001,
777
+ "loss": 0.2404,
778
+ "step": 127
779
+ },
780
+ {
781
+ "epoch": 0.26,
782
+ "learning_rate": 0.0001,
783
+ "loss": 0.1416,
784
+ "step": 128
785
+ },
786
+ {
787
+ "epoch": 0.27,
788
+ "learning_rate": 0.0001,
789
+ "loss": 0.1641,
790
+ "step": 129
791
+ },
792
+ {
793
+ "epoch": 0.27,
794
+ "learning_rate": 0.0001,
795
+ "loss": 0.1298,
796
+ "step": 130
797
+ },
798
+ {
799
+ "epoch": 0.27,
800
+ "learning_rate": 0.0001,
801
+ "loss": 0.1486,
802
+ "step": 131
803
+ },
804
+ {
805
+ "epoch": 0.27,
806
+ "learning_rate": 0.0001,
807
+ "loss": 0.1599,
808
+ "step": 132
809
+ },
810
+ {
811
+ "epoch": 0.27,
812
+ "learning_rate": 0.0001,
813
+ "loss": 0.0898,
814
+ "step": 133
815
+ },
816
+ {
817
+ "epoch": 0.28,
818
+ "learning_rate": 0.0001,
819
+ "loss": 0.1458,
820
+ "step": 134
821
+ },
822
+ {
823
+ "epoch": 0.28,
824
+ "learning_rate": 0.0001,
825
+ "loss": 0.107,
826
+ "step": 135
827
+ },
828
+ {
829
+ "epoch": 0.28,
830
+ "learning_rate": 0.0001,
831
+ "loss": 0.1175,
832
+ "step": 136
833
+ },
834
+ {
835
+ "epoch": 0.28,
836
+ "learning_rate": 0.0001,
837
+ "loss": 0.1248,
838
+ "step": 137
839
+ },
840
+ {
841
+ "epoch": 0.28,
842
+ "learning_rate": 0.0001,
843
+ "loss": 0.1983,
844
+ "step": 138
845
+ },
846
+ {
847
+ "epoch": 0.29,
848
+ "learning_rate": 0.0001,
849
+ "loss": 0.1772,
850
+ "step": 139
851
+ },
852
+ {
853
+ "epoch": 0.29,
854
+ "learning_rate": 0.0001,
855
+ "loss": 0.0884,
856
+ "step": 140
857
+ },
858
+ {
859
+ "epoch": 0.29,
860
+ "learning_rate": 0.0001,
861
+ "loss": 0.1383,
862
+ "step": 141
863
+ },
864
+ {
865
+ "epoch": 0.29,
866
+ "learning_rate": 0.0001,
867
+ "loss": 0.1082,
868
+ "step": 142
869
+ },
870
+ {
871
+ "epoch": 0.29,
872
+ "learning_rate": 0.0001,
873
+ "loss": 0.1572,
874
+ "step": 143
875
+ },
876
+ {
877
+ "epoch": 0.3,
878
+ "learning_rate": 0.0001,
879
+ "loss": 0.1179,
880
+ "step": 144
881
+ },
882
+ {
883
+ "epoch": 0.3,
884
+ "learning_rate": 0.0001,
885
+ "loss": 0.1147,
886
+ "step": 145
887
+ },
888
+ {
889
+ "epoch": 0.3,
890
+ "learning_rate": 0.0001,
891
+ "loss": 0.1277,
892
+ "step": 146
893
+ },
894
+ {
895
+ "epoch": 0.3,
896
+ "learning_rate": 0.0001,
897
+ "loss": 0.2077,
898
+ "step": 147
899
+ },
900
+ {
901
+ "epoch": 0.3,
902
+ "learning_rate": 0.0001,
903
+ "loss": 0.1235,
904
+ "step": 148
905
+ },
906
+ {
907
+ "epoch": 0.31,
908
+ "learning_rate": 0.0001,
909
+ "loss": 0.1499,
910
+ "step": 149
911
+ },
912
+ {
913
+ "epoch": 0.31,
914
+ "learning_rate": 0.0001,
915
+ "loss": 0.1552,
916
+ "step": 150
917
+ },
918
+ {
919
+ "epoch": 0.31,
920
+ "learning_rate": 0.0001,
921
+ "loss": 0.1877,
922
+ "step": 151
923
+ },
924
+ {
925
+ "epoch": 0.31,
926
+ "learning_rate": 0.0001,
927
+ "loss": 0.1677,
928
+ "step": 152
929
+ },
930
+ {
931
+ "epoch": 0.32,
932
+ "learning_rate": 0.0001,
933
+ "loss": 0.1878,
934
+ "step": 153
935
+ },
936
+ {
937
+ "epoch": 0.32,
938
+ "learning_rate": 0.0001,
939
+ "loss": 0.1817,
940
+ "step": 154
941
+ },
942
+ {
943
+ "epoch": 0.32,
944
+ "learning_rate": 0.0001,
945
+ "loss": 0.1055,
946
+ "step": 155
947
+ },
948
+ {
949
+ "epoch": 0.32,
950
+ "learning_rate": 0.0001,
951
+ "loss": 0.162,
952
+ "step": 156
953
+ },
954
+ {
955
+ "epoch": 0.32,
956
+ "learning_rate": 0.0001,
957
+ "loss": 0.1303,
958
+ "step": 157
959
+ },
960
+ {
961
+ "epoch": 0.33,
962
+ "learning_rate": 0.0001,
963
+ "loss": 0.0981,
964
+ "step": 158
965
+ },
966
+ {
967
+ "epoch": 0.33,
968
+ "learning_rate": 0.0001,
969
+ "loss": 0.0963,
970
+ "step": 159
971
+ },
972
+ {
973
+ "epoch": 0.33,
974
+ "learning_rate": 0.0001,
975
+ "loss": 0.1327,
976
+ "step": 160
977
+ },
978
+ {
979
+ "epoch": 0.33,
980
+ "learning_rate": 0.0001,
981
+ "loss": 0.167,
982
+ "step": 161
983
+ },
984
+ {
985
+ "epoch": 0.33,
986
+ "learning_rate": 0.0001,
987
+ "loss": 0.1431,
988
+ "step": 162
989
+ },
990
+ {
991
+ "epoch": 0.34,
992
+ "learning_rate": 0.0001,
993
+ "loss": 0.1583,
994
+ "step": 163
995
+ },
996
+ {
997
+ "epoch": 0.34,
998
+ "learning_rate": 0.0001,
999
+ "loss": 0.1395,
1000
+ "step": 164
1001
+ },
1002
+ {
1003
+ "epoch": 0.34,
1004
+ "learning_rate": 0.0001,
1005
+ "loss": 0.0974,
1006
+ "step": 165
1007
+ },
1008
+ {
1009
+ "epoch": 0.34,
1010
+ "learning_rate": 0.0001,
1011
+ "loss": 0.2083,
1012
+ "step": 166
1013
+ },
1014
+ {
1015
+ "epoch": 0.34,
1016
+ "learning_rate": 0.0001,
1017
+ "loss": 0.1538,
1018
+ "step": 167
1019
+ },
1020
+ {
1021
+ "epoch": 0.35,
1022
+ "learning_rate": 0.0001,
1023
+ "loss": 0.1298,
1024
+ "step": 168
1025
+ },
1026
+ {
1027
+ "epoch": 0.35,
1028
+ "learning_rate": 0.0001,
1029
+ "loss": 0.1947,
1030
+ "step": 169
1031
+ },
1032
+ {
1033
+ "epoch": 0.35,
1034
+ "learning_rate": 0.0001,
1035
+ "loss": 0.1031,
1036
+ "step": 170
1037
+ },
1038
+ {
1039
+ "epoch": 0.35,
1040
+ "learning_rate": 0.0001,
1041
+ "loss": 0.1545,
1042
+ "step": 171
1043
+ },
1044
+ {
1045
+ "epoch": 0.35,
1046
+ "learning_rate": 0.0001,
1047
+ "loss": 0.1194,
1048
+ "step": 172
1049
+ },
1050
+ {
1051
+ "epoch": 0.36,
1052
+ "learning_rate": 0.0001,
1053
+ "loss": 0.0988,
1054
+ "step": 173
1055
+ },
1056
+ {
1057
+ "epoch": 0.36,
1058
+ "learning_rate": 0.0001,
1059
+ "loss": 0.1526,
1060
+ "step": 174
1061
+ },
1062
+ {
1063
+ "epoch": 0.36,
1064
+ "learning_rate": 0.0001,
1065
+ "loss": 0.1673,
1066
+ "step": 175
1067
+ },
1068
+ {
1069
+ "epoch": 0.36,
1070
+ "learning_rate": 0.0001,
1071
+ "loss": 0.0977,
1072
+ "step": 176
1073
+ },
1074
+ {
1075
+ "epoch": 0.36,
1076
+ "learning_rate": 0.0001,
1077
+ "loss": 0.1249,
1078
+ "step": 177
1079
+ },
1080
+ {
1081
+ "epoch": 0.37,
1082
+ "learning_rate": 0.0001,
1083
+ "loss": 0.1572,
1084
+ "step": 178
1085
+ },
1086
+ {
1087
+ "epoch": 0.37,
1088
+ "learning_rate": 0.0001,
1089
+ "loss": 0.1437,
1090
+ "step": 179
1091
+ },
1092
+ {
1093
+ "epoch": 0.37,
1094
+ "learning_rate": 0.0001,
1095
+ "loss": 0.1001,
1096
+ "step": 180
1097
+ },
1098
+ {
1099
+ "epoch": 0.37,
1100
+ "learning_rate": 0.0001,
1101
+ "loss": 0.1167,
1102
+ "step": 181
1103
+ },
1104
+ {
1105
+ "epoch": 0.38,
1106
+ "learning_rate": 0.0001,
1107
+ "loss": 0.2169,
1108
+ "step": 182
1109
+ },
1110
+ {
1111
+ "epoch": 0.38,
1112
+ "learning_rate": 0.0001,
1113
+ "loss": 0.1151,
1114
+ "step": 183
1115
+ },
1116
+ {
1117
+ "epoch": 0.38,
1118
+ "learning_rate": 0.0001,
1119
+ "loss": 0.1024,
1120
+ "step": 184
1121
+ },
1122
+ {
1123
+ "epoch": 0.38,
1124
+ "learning_rate": 0.0001,
1125
+ "loss": 0.144,
1126
+ "step": 185
1127
+ },
1128
+ {
1129
+ "epoch": 0.38,
1130
+ "learning_rate": 0.0001,
1131
+ "loss": 0.2502,
1132
+ "step": 186
1133
+ },
1134
+ {
1135
+ "epoch": 0.39,
1136
+ "learning_rate": 0.0001,
1137
+ "loss": 0.1616,
1138
+ "step": 187
1139
+ },
1140
+ {
1141
+ "epoch": 0.39,
1142
+ "learning_rate": 0.0001,
1143
+ "loss": 0.0828,
1144
+ "step": 188
1145
+ },
1146
+ {
1147
+ "epoch": 0.39,
1148
+ "learning_rate": 0.0001,
1149
+ "loss": 0.1487,
1150
+ "step": 189
1151
+ },
1152
+ {
1153
+ "epoch": 0.39,
1154
+ "learning_rate": 0.0001,
1155
+ "loss": 0.1605,
1156
+ "step": 190
1157
+ },
1158
+ {
1159
+ "epoch": 0.39,
1160
+ "learning_rate": 0.0001,
1161
+ "loss": 0.1711,
1162
+ "step": 191
1163
+ },
1164
+ {
1165
+ "epoch": 0.4,
1166
+ "learning_rate": 0.0001,
1167
+ "loss": 0.0971,
1168
+ "step": 192
1169
+ },
1170
+ {
1171
+ "epoch": 0.4,
1172
+ "learning_rate": 0.0001,
1173
+ "loss": 0.1173,
1174
+ "step": 193
1175
+ },
1176
+ {
1177
+ "epoch": 0.4,
1178
+ "learning_rate": 0.0001,
1179
+ "loss": 0.2035,
1180
+ "step": 194
1181
+ },
1182
+ {
1183
+ "epoch": 0.4,
1184
+ "learning_rate": 0.0001,
1185
+ "loss": 0.0935,
1186
+ "step": 195
1187
+ },
1188
+ {
1189
+ "epoch": 0.4,
1190
+ "learning_rate": 0.0001,
1191
+ "loss": 0.1642,
1192
+ "step": 196
1193
+ },
1194
+ {
1195
+ "epoch": 0.41,
1196
+ "learning_rate": 0.0001,
1197
+ "loss": 0.0919,
1198
+ "step": 197
1199
+ },
1200
+ {
1201
+ "epoch": 0.41,
1202
+ "learning_rate": 0.0001,
1203
+ "loss": 0.1287,
1204
+ "step": 198
1205
+ },
1206
+ {
1207
+ "epoch": 0.41,
1208
+ "learning_rate": 0.0001,
1209
+ "loss": 0.104,
1210
+ "step": 199
1211
+ },
1212
+ {
1213
+ "epoch": 0.41,
1214
+ "learning_rate": 0.0001,
1215
+ "loss": 0.1304,
1216
+ "step": 200
1217
+ },
1218
+ {
1219
+ "epoch": 0.41,
1220
+ "eval_loss": 0.1713656336069107,
1221
+ "eval_runtime": 941.1442,
1222
+ "eval_samples_per_second": 34.303,
1223
+ "eval_steps_per_second": 0.269,
1224
+ "step": 200
1225
+ },
1226
+ {
1227
+ "epoch": 0.41,
1228
+ "learning_rate": 0.0001,
1229
+ "loss": 0.1692,
1230
+ "step": 201
1231
+ },
1232
+ {
1233
+ "epoch": 0.42,
1234
+ "learning_rate": 0.0001,
1235
+ "loss": 0.1555,
1236
+ "step": 202
1237
+ },
1238
+ {
1239
+ "epoch": 0.42,
1240
+ "learning_rate": 0.0001,
1241
+ "loss": 0.0741,
1242
+ "step": 203
1243
+ },
1244
+ {
1245
+ "epoch": 0.42,
1246
+ "learning_rate": 0.0001,
1247
+ "loss": 0.0975,
1248
+ "step": 204
1249
+ },
1250
+ {
1251
+ "epoch": 0.42,
1252
+ "learning_rate": 0.0001,
1253
+ "loss": 0.1215,
1254
+ "step": 205
1255
+ },
1256
+ {
1257
+ "epoch": 0.42,
1258
+ "learning_rate": 0.0001,
1259
+ "loss": 0.1174,
1260
+ "step": 206
1261
+ },
1262
+ {
1263
+ "epoch": 0.43,
1264
+ "learning_rate": 0.0001,
1265
+ "loss": 0.1581,
1266
+ "step": 207
1267
+ },
1268
+ {
1269
+ "epoch": 0.43,
1270
+ "learning_rate": 0.0001,
1271
+ "loss": 0.16,
1272
+ "step": 208
1273
+ },
1274
+ {
1275
+ "epoch": 0.43,
1276
+ "learning_rate": 0.0001,
1277
+ "loss": 0.1579,
1278
+ "step": 209
1279
+ },
1280
+ {
1281
+ "epoch": 0.43,
1282
+ "learning_rate": 0.0001,
1283
+ "loss": 0.0981,
1284
+ "step": 210
1285
+ },
1286
+ {
1287
+ "epoch": 0.43,
1288
+ "learning_rate": 0.0001,
1289
+ "loss": 0.0843,
1290
+ "step": 211
1291
+ },
1292
+ {
1293
+ "epoch": 0.44,
1294
+ "learning_rate": 0.0001,
1295
+ "loss": 0.124,
1296
+ "step": 212
1297
+ },
1298
+ {
1299
+ "epoch": 0.44,
1300
+ "learning_rate": 0.0001,
1301
+ "loss": 0.1287,
1302
+ "step": 213
1303
+ },
1304
+ {
1305
+ "epoch": 0.44,
1306
+ "learning_rate": 0.0001,
1307
+ "loss": 0.1344,
1308
+ "step": 214
1309
+ },
1310
+ {
1311
+ "epoch": 0.44,
1312
+ "learning_rate": 0.0001,
1313
+ "loss": 0.1168,
1314
+ "step": 215
1315
+ },
1316
+ {
1317
+ "epoch": 0.45,
1318
+ "learning_rate": 0.0001,
1319
+ "loss": 0.1141,
1320
+ "step": 216
1321
+ },
1322
+ {
1323
+ "epoch": 0.45,
1324
+ "learning_rate": 0.0001,
1325
+ "loss": 0.1051,
1326
+ "step": 217
1327
+ },
1328
+ {
1329
+ "epoch": 0.45,
1330
+ "learning_rate": 0.0001,
1331
+ "loss": 0.0952,
1332
+ "step": 218
1333
+ },
1334
+ {
1335
+ "epoch": 0.45,
1336
+ "learning_rate": 0.0001,
1337
+ "loss": 0.0949,
1338
+ "step": 219
1339
+ },
1340
+ {
1341
+ "epoch": 0.45,
1342
+ "learning_rate": 0.0001,
1343
+ "loss": 0.0679,
1344
+ "step": 220
1345
+ },
1346
+ {
1347
+ "epoch": 0.46,
1348
+ "learning_rate": 0.0001,
1349
+ "loss": 0.1158,
1350
+ "step": 221
1351
+ },
1352
+ {
1353
+ "epoch": 0.46,
1354
+ "learning_rate": 0.0001,
1355
+ "loss": 0.0817,
1356
+ "step": 222
1357
+ },
1358
+ {
1359
+ "epoch": 0.46,
1360
+ "learning_rate": 0.0001,
1361
+ "loss": 0.1035,
1362
+ "step": 223
1363
+ },
1364
+ {
1365
+ "epoch": 0.46,
1366
+ "learning_rate": 0.0001,
1367
+ "loss": 0.1039,
1368
+ "step": 224
1369
+ },
1370
+ {
1371
+ "epoch": 0.46,
1372
+ "learning_rate": 0.0001,
1373
+ "loss": 0.1031,
1374
+ "step": 225
1375
+ },
1376
+ {
1377
+ "epoch": 0.47,
1378
+ "learning_rate": 0.0001,
1379
+ "loss": 0.1152,
1380
+ "step": 226
1381
+ },
1382
+ {
1383
+ "epoch": 0.47,
1384
+ "learning_rate": 0.0001,
1385
+ "loss": 0.0968,
1386
+ "step": 227
1387
+ },
1388
+ {
1389
+ "epoch": 0.47,
1390
+ "learning_rate": 0.0001,
1391
+ "loss": 0.0682,
1392
+ "step": 228
1393
+ },
1394
+ {
1395
+ "epoch": 0.47,
1396
+ "learning_rate": 0.0001,
1397
+ "loss": 0.0822,
1398
+ "step": 229
1399
+ },
1400
+ {
1401
+ "epoch": 0.47,
1402
+ "learning_rate": 0.0001,
1403
+ "loss": 0.1364,
1404
+ "step": 230
1405
+ },
1406
+ {
1407
+ "epoch": 0.48,
1408
+ "learning_rate": 0.0001,
1409
+ "loss": 0.0945,
1410
+ "step": 231
1411
+ },
1412
+ {
1413
+ "epoch": 0.48,
1414
+ "learning_rate": 0.0001,
1415
+ "loss": 0.1614,
1416
+ "step": 232
1417
+ },
1418
+ {
1419
+ "epoch": 0.48,
1420
+ "learning_rate": 0.0001,
1421
+ "loss": 0.0957,
1422
+ "step": 233
1423
+ },
1424
+ {
1425
+ "epoch": 0.48,
1426
+ "learning_rate": 0.0001,
1427
+ "loss": 0.1053,
1428
+ "step": 234
1429
+ },
1430
+ {
1431
+ "epoch": 0.48,
1432
+ "learning_rate": 0.0001,
1433
+ "loss": 0.0719,
1434
+ "step": 235
1435
+ },
1436
+ {
1437
+ "epoch": 0.49,
1438
+ "learning_rate": 0.0001,
1439
+ "loss": 0.1377,
1440
+ "step": 236
1441
+ },
1442
+ {
1443
+ "epoch": 0.49,
1444
+ "learning_rate": 0.0001,
1445
+ "loss": 0.0988,
1446
+ "step": 237
1447
+ },
1448
+ {
1449
+ "epoch": 0.49,
1450
+ "learning_rate": 0.0001,
1451
+ "loss": 0.1374,
1452
+ "step": 238
1453
+ },
1454
+ {
1455
+ "epoch": 0.49,
1456
+ "learning_rate": 0.0001,
1457
+ "loss": 0.0913,
1458
+ "step": 239
1459
+ },
1460
+ {
1461
+ "epoch": 0.49,
1462
+ "learning_rate": 0.0001,
1463
+ "loss": 0.1398,
1464
+ "step": 240
1465
+ },
1466
+ {
1467
+ "epoch": 0.5,
1468
+ "learning_rate": 0.0001,
1469
+ "loss": 0.1,
1470
+ "step": 241
1471
+ },
1472
+ {
1473
+ "epoch": 0.5,
1474
+ "learning_rate": 0.0001,
1475
+ "loss": 0.0954,
1476
+ "step": 242
1477
+ },
1478
+ {
1479
+ "epoch": 0.5,
1480
+ "learning_rate": 0.0001,
1481
+ "loss": 0.0851,
1482
+ "step": 243
1483
+ },
1484
+ {
1485
+ "epoch": 0.5,
1486
+ "learning_rate": 0.0001,
1487
+ "loss": 0.0924,
1488
+ "step": 244
1489
+ },
1490
+ {
1491
+ "epoch": 0.5,
1492
+ "learning_rate": 0.0001,
1493
+ "loss": 0.112,
1494
+ "step": 245
1495
+ },
1496
+ {
1497
+ "epoch": 0.51,
1498
+ "learning_rate": 0.0001,
1499
+ "loss": 0.0862,
1500
+ "step": 246
1501
+ },
1502
+ {
1503
+ "epoch": 0.51,
1504
+ "learning_rate": 0.0001,
1505
+ "loss": 0.1043,
1506
+ "step": 247
1507
+ },
1508
+ {
1509
+ "epoch": 0.51,
1510
+ "learning_rate": 0.0001,
1511
+ "loss": 0.2192,
1512
+ "step": 248
1513
+ },
1514
+ {
1515
+ "epoch": 0.51,
1516
+ "learning_rate": 0.0001,
1517
+ "loss": 0.1003,
1518
+ "step": 249
1519
+ },
1520
+ {
1521
+ "epoch": 0.52,
1522
+ "learning_rate": 0.0001,
1523
+ "loss": 0.0825,
1524
+ "step": 250
1525
+ },
1526
+ {
1527
+ "epoch": 0.52,
1528
+ "learning_rate": 0.0001,
1529
+ "loss": 0.0998,
1530
+ "step": 251
1531
+ },
1532
+ {
1533
+ "epoch": 0.52,
1534
+ "learning_rate": 0.0001,
1535
+ "loss": 0.1333,
1536
+ "step": 252
1537
+ },
1538
+ {
1539
+ "epoch": 0.52,
1540
+ "learning_rate": 0.0001,
1541
+ "loss": 0.1,
1542
+ "step": 253
1543
+ },
1544
+ {
1545
+ "epoch": 0.52,
1546
+ "learning_rate": 0.0001,
1547
+ "loss": 0.1085,
1548
+ "step": 254
1549
+ },
1550
+ {
1551
+ "epoch": 0.53,
1552
+ "learning_rate": 0.0001,
1553
+ "loss": 0.1291,
1554
+ "step": 255
1555
+ },
1556
+ {
1557
+ "epoch": 0.53,
1558
+ "learning_rate": 0.0001,
1559
+ "loss": 0.121,
1560
+ "step": 256
1561
+ },
1562
+ {
1563
+ "epoch": 0.53,
1564
+ "learning_rate": 0.0001,
1565
+ "loss": 0.0972,
1566
+ "step": 257
1567
+ },
1568
+ {
1569
+ "epoch": 0.53,
1570
+ "learning_rate": 0.0001,
1571
+ "loss": 0.0779,
1572
+ "step": 258
1573
+ },
1574
+ {
1575
+ "epoch": 0.53,
1576
+ "learning_rate": 0.0001,
1577
+ "loss": 0.0833,
1578
+ "step": 259
1579
+ },
1580
+ {
1581
+ "epoch": 0.54,
1582
+ "learning_rate": 0.0001,
1583
+ "loss": 0.0992,
1584
+ "step": 260
1585
+ },
1586
+ {
1587
+ "epoch": 0.54,
1588
+ "learning_rate": 0.0001,
1589
+ "loss": 0.1175,
1590
+ "step": 261
1591
+ },
1592
+ {
1593
+ "epoch": 0.54,
1594
+ "learning_rate": 0.0001,
1595
+ "loss": 0.1271,
1596
+ "step": 262
1597
+ },
1598
+ {
1599
+ "epoch": 0.54,
1600
+ "learning_rate": 0.0001,
1601
+ "loss": 0.1154,
1602
+ "step": 263
1603
+ },
1604
+ {
1605
+ "epoch": 0.54,
1606
+ "learning_rate": 0.0001,
1607
+ "loss": 0.0816,
1608
+ "step": 264
1609
+ },
1610
+ {
1611
+ "epoch": 0.55,
1612
+ "learning_rate": 0.0001,
1613
+ "loss": 0.0985,
1614
+ "step": 265
1615
+ },
1616
+ {
1617
+ "epoch": 0.55,
1618
+ "learning_rate": 0.0001,
1619
+ "loss": 0.1091,
1620
+ "step": 266
1621
+ },
1622
+ {
1623
+ "epoch": 0.55,
1624
+ "learning_rate": 0.0001,
1625
+ "loss": 0.114,
1626
+ "step": 267
1627
+ },
1628
+ {
1629
+ "epoch": 0.55,
1630
+ "learning_rate": 0.0001,
1631
+ "loss": 0.0989,
1632
+ "step": 268
1633
+ },
1634
+ {
1635
+ "epoch": 0.55,
1636
+ "learning_rate": 0.0001,
1637
+ "loss": 0.1541,
1638
+ "step": 269
1639
+ },
1640
+ {
1641
+ "epoch": 0.56,
1642
+ "learning_rate": 0.0001,
1643
+ "loss": 0.0868,
1644
+ "step": 270
1645
+ },
1646
+ {
1647
+ "epoch": 0.56,
1648
+ "learning_rate": 0.0001,
1649
+ "loss": 0.0645,
1650
+ "step": 271
1651
+ },
1652
+ {
1653
+ "epoch": 0.56,
1654
+ "learning_rate": 0.0001,
1655
+ "loss": 0.0927,
1656
+ "step": 272
1657
+ },
1658
+ {
1659
+ "epoch": 0.56,
1660
+ "learning_rate": 0.0001,
1661
+ "loss": 0.1146,
1662
+ "step": 273
1663
+ },
1664
+ {
1665
+ "epoch": 0.56,
1666
+ "learning_rate": 0.0001,
1667
+ "loss": 0.157,
1668
+ "step": 274
1669
+ },
1670
+ {
1671
+ "epoch": 0.57,
1672
+ "learning_rate": 0.0001,
1673
+ "loss": 0.0823,
1674
+ "step": 275
1675
+ },
1676
+ {
1677
+ "epoch": 0.57,
1678
+ "learning_rate": 0.0001,
1679
+ "loss": 0.1343,
1680
+ "step": 276
1681
+ },
1682
+ {
1683
+ "epoch": 0.57,
1684
+ "learning_rate": 0.0001,
1685
+ "loss": 0.1015,
1686
+ "step": 277
1687
+ },
1688
+ {
1689
+ "epoch": 0.57,
1690
+ "learning_rate": 0.0001,
1691
+ "loss": 0.0746,
1692
+ "step": 278
1693
+ },
1694
+ {
1695
+ "epoch": 0.57,
1696
+ "learning_rate": 0.0001,
1697
+ "loss": 0.0918,
1698
+ "step": 279
1699
+ },
1700
+ {
1701
+ "epoch": 0.58,
1702
+ "learning_rate": 0.0001,
1703
+ "loss": 0.1038,
1704
+ "step": 280
1705
+ },
1706
+ {
1707
+ "epoch": 0.58,
1708
+ "learning_rate": 0.0001,
1709
+ "loss": 0.1023,
1710
+ "step": 281
1711
+ },
1712
+ {
1713
+ "epoch": 0.58,
1714
+ "learning_rate": 0.0001,
1715
+ "loss": 0.0809,
1716
+ "step": 282
1717
+ },
1718
+ {
1719
+ "epoch": 0.58,
1720
+ "learning_rate": 0.0001,
1721
+ "loss": 0.0715,
1722
+ "step": 283
1723
+ },
1724
+ {
1725
+ "epoch": 0.59,
1726
+ "learning_rate": 0.0001,
1727
+ "loss": 0.0806,
1728
+ "step": 284
1729
+ },
1730
+ {
1731
+ "epoch": 0.59,
1732
+ "learning_rate": 0.0001,
1733
+ "loss": 0.0997,
1734
+ "step": 285
1735
+ },
1736
+ {
1737
+ "epoch": 0.59,
1738
+ "learning_rate": 0.0001,
1739
+ "loss": 0.0793,
1740
+ "step": 286
1741
+ },
1742
+ {
1743
+ "epoch": 0.59,
1744
+ "learning_rate": 0.0001,
1745
+ "loss": 0.0658,
1746
+ "step": 287
1747
+ },
1748
+ {
1749
+ "epoch": 0.59,
1750
+ "learning_rate": 0.0001,
1751
+ "loss": 0.1345,
1752
+ "step": 288
1753
+ },
1754
+ {
1755
+ "epoch": 0.6,
1756
+ "learning_rate": 0.0001,
1757
+ "loss": 0.0867,
1758
+ "step": 289
1759
+ },
1760
+ {
1761
+ "epoch": 0.6,
1762
+ "learning_rate": 0.0001,
1763
+ "loss": 0.0642,
1764
+ "step": 290
1765
+ },
1766
+ {
1767
+ "epoch": 0.6,
1768
+ "learning_rate": 0.0001,
1769
+ "loss": 0.09,
1770
+ "step": 291
1771
+ },
1772
+ {
1773
+ "epoch": 0.6,
1774
+ "learning_rate": 0.0001,
1775
+ "loss": 0.0836,
1776
+ "step": 292
1777
+ },
1778
+ {
1779
+ "epoch": 0.6,
1780
+ "learning_rate": 0.0001,
1781
+ "loss": 0.056,
1782
+ "step": 293
1783
+ },
1784
+ {
1785
+ "epoch": 0.61,
1786
+ "learning_rate": 0.0001,
1787
+ "loss": 0.1309,
1788
+ "step": 294
1789
+ },
1790
+ {
1791
+ "epoch": 0.61,
1792
+ "learning_rate": 0.0001,
1793
+ "loss": 0.0825,
1794
+ "step": 295
1795
+ },
1796
+ {
1797
+ "epoch": 0.61,
1798
+ "learning_rate": 0.0001,
1799
+ "loss": 0.1343,
1800
+ "step": 296
1801
+ },
1802
+ {
1803
+ "epoch": 0.61,
1804
+ "learning_rate": 0.0001,
1805
+ "loss": 0.0874,
1806
+ "step": 297
1807
+ },
1808
+ {
1809
+ "epoch": 0.61,
1810
+ "learning_rate": 0.0001,
1811
+ "loss": 0.1005,
1812
+ "step": 298
1813
+ },
1814
+ {
1815
+ "epoch": 0.62,
1816
+ "learning_rate": 0.0001,
1817
+ "loss": 0.0804,
1818
+ "step": 299
1819
+ },
1820
+ {
1821
+ "epoch": 0.62,
1822
+ "learning_rate": 0.0001,
1823
+ "loss": 0.0688,
1824
+ "step": 300
1825
+ },
1826
+ {
1827
+ "epoch": 0.62,
1828
+ "eval_loss": 0.11704447865486145,
1829
+ "eval_runtime": 967.1425,
1830
+ "eval_samples_per_second": 33.381,
1831
+ "eval_steps_per_second": 0.262,
1832
+ "step": 300
1833
+ },
1834
+ {
1835
+ "epoch": 0.62,
1836
+ "learning_rate": 0.0001,
1837
+ "loss": 0.0868,
1838
+ "step": 301
1839
+ },
1840
+ {
1841
+ "epoch": 0.62,
1842
+ "learning_rate": 0.0001,
1843
+ "loss": 0.0682,
1844
+ "step": 302
1845
+ },
1846
+ {
1847
+ "epoch": 0.62,
1848
+ "learning_rate": 0.0001,
1849
+ "loss": 0.1118,
1850
+ "step": 303
1851
+ },
1852
+ {
1853
+ "epoch": 0.63,
1854
+ "learning_rate": 0.0001,
1855
+ "loss": 0.0908,
1856
+ "step": 304
1857
+ },
1858
+ {
1859
+ "epoch": 0.63,
1860
+ "learning_rate": 0.0001,
1861
+ "loss": 0.0633,
1862
+ "step": 305
1863
+ },
1864
+ {
1865
+ "epoch": 0.63,
1866
+ "learning_rate": 0.0001,
1867
+ "loss": 0.0839,
1868
+ "step": 306
1869
+ },
1870
+ {
1871
+ "epoch": 0.63,
1872
+ "learning_rate": 0.0001,
1873
+ "loss": 0.0796,
1874
+ "step": 307
1875
+ },
1876
+ {
1877
+ "epoch": 0.63,
1878
+ "learning_rate": 0.0001,
1879
+ "loss": 0.0631,
1880
+ "step": 308
1881
+ },
1882
+ {
1883
+ "epoch": 0.64,
1884
+ "learning_rate": 0.0001,
1885
+ "loss": 0.0797,
1886
+ "step": 309
1887
+ },
1888
+ {
1889
+ "epoch": 0.64,
1890
+ "learning_rate": 0.0001,
1891
+ "loss": 0.1014,
1892
+ "step": 310
1893
+ },
1894
+ {
1895
+ "epoch": 0.64,
1896
+ "learning_rate": 0.0001,
1897
+ "loss": 0.1586,
1898
+ "step": 311
1899
+ },
1900
+ {
1901
+ "epoch": 0.64,
1902
+ "learning_rate": 0.0001,
1903
+ "loss": 0.0789,
1904
+ "step": 312
1905
+ },
1906
+ {
1907
+ "epoch": 0.65,
1908
+ "learning_rate": 0.0001,
1909
+ "loss": 0.0722,
1910
+ "step": 313
1911
+ },
1912
+ {
1913
+ "epoch": 0.65,
1914
+ "learning_rate": 0.0001,
1915
+ "loss": 0.12,
1916
+ "step": 314
1917
+ },
1918
+ {
1919
+ "epoch": 0.65,
1920
+ "learning_rate": 0.0001,
1921
+ "loss": 0.0665,
1922
+ "step": 315
1923
+ },
1924
+ {
1925
+ "epoch": 0.65,
1926
+ "learning_rate": 0.0001,
1927
+ "loss": 0.0963,
1928
+ "step": 316
1929
+ },
1930
+ {
1931
+ "epoch": 0.65,
1932
+ "learning_rate": 0.0001,
1933
+ "loss": 0.1095,
1934
+ "step": 317
1935
+ },
1936
+ {
1937
+ "epoch": 0.66,
1938
+ "learning_rate": 0.0001,
1939
+ "loss": 0.0728,
1940
+ "step": 318
1941
+ },
1942
+ {
1943
+ "epoch": 0.66,
1944
+ "learning_rate": 0.0001,
1945
+ "loss": 0.0795,
1946
+ "step": 319
1947
+ },
1948
+ {
1949
+ "epoch": 0.66,
1950
+ "learning_rate": 0.0001,
1951
+ "loss": 0.0729,
1952
+ "step": 320
1953
+ },
1954
+ {
1955
+ "epoch": 0.66,
1956
+ "learning_rate": 0.0001,
1957
+ "loss": 0.0599,
1958
+ "step": 321
1959
+ },
1960
+ {
1961
+ "epoch": 0.66,
1962
+ "learning_rate": 0.0001,
1963
+ "loss": 0.0803,
1964
+ "step": 322
1965
+ },
1966
+ {
1967
+ "epoch": 0.67,
1968
+ "learning_rate": 0.0001,
1969
+ "loss": 0.0511,
1970
+ "step": 323
1971
+ },
1972
+ {
1973
+ "epoch": 0.67,
1974
+ "learning_rate": 0.0001,
1975
+ "loss": 0.0783,
1976
+ "step": 324
1977
+ },
1978
+ {
1979
+ "epoch": 0.67,
1980
+ "learning_rate": 0.0001,
1981
+ "loss": 0.1076,
1982
+ "step": 325
1983
+ },
1984
+ {
1985
+ "epoch": 0.67,
1986
+ "learning_rate": 0.0001,
1987
+ "loss": 0.0917,
1988
+ "step": 326
1989
+ },
1990
+ {
1991
+ "epoch": 0.67,
1992
+ "learning_rate": 0.0001,
1993
+ "loss": 0.0581,
1994
+ "step": 327
1995
+ },
1996
+ {
1997
+ "epoch": 0.68,
1998
+ "learning_rate": 0.0001,
1999
+ "loss": 0.0665,
2000
+ "step": 328
2001
+ },
2002
+ {
2003
+ "epoch": 0.68,
2004
+ "learning_rate": 0.0001,
2005
+ "loss": 0.0707,
2006
+ "step": 329
2007
+ },
2008
+ {
2009
+ "epoch": 0.68,
2010
+ "learning_rate": 0.0001,
2011
+ "loss": 0.0926,
2012
+ "step": 330
2013
+ },
2014
+ {
2015
+ "epoch": 0.68,
2016
+ "learning_rate": 0.0001,
2017
+ "loss": 0.0544,
2018
+ "step": 331
2019
+ },
2020
+ {
2021
+ "epoch": 0.68,
2022
+ "learning_rate": 0.0001,
2023
+ "loss": 0.1131,
2024
+ "step": 332
2025
+ },
2026
+ {
2027
+ "epoch": 0.69,
2028
+ "learning_rate": 0.0001,
2029
+ "loss": 0.1424,
2030
+ "step": 333
2031
+ },
2032
+ {
2033
+ "epoch": 0.69,
2034
+ "learning_rate": 0.0001,
2035
+ "loss": 0.0938,
2036
+ "step": 334
2037
+ },
2038
+ {
2039
+ "epoch": 0.69,
2040
+ "learning_rate": 0.0001,
2041
+ "loss": 0.1193,
2042
+ "step": 335
2043
+ },
2044
+ {
2045
+ "epoch": 0.69,
2046
+ "learning_rate": 0.0001,
2047
+ "loss": 0.0688,
2048
+ "step": 336
2049
+ },
2050
+ {
2051
+ "epoch": 0.69,
2052
+ "learning_rate": 0.0001,
2053
+ "loss": 0.1052,
2054
+ "step": 337
2055
+ },
2056
+ {
2057
+ "epoch": 0.7,
2058
+ "learning_rate": 0.0001,
2059
+ "loss": 0.1174,
2060
+ "step": 338
2061
+ },
2062
+ {
2063
+ "epoch": 0.7,
2064
+ "learning_rate": 0.0001,
2065
+ "loss": 0.085,
2066
+ "step": 339
2067
+ },
2068
+ {
2069
+ "epoch": 0.7,
2070
+ "learning_rate": 0.0001,
2071
+ "loss": 0.0913,
2072
+ "step": 340
2073
+ },
2074
+ {
2075
+ "epoch": 0.7,
2076
+ "learning_rate": 0.0001,
2077
+ "loss": 0.0902,
2078
+ "step": 341
2079
+ },
2080
+ {
2081
+ "epoch": 0.7,
2082
+ "learning_rate": 0.0001,
2083
+ "loss": 0.0591,
2084
+ "step": 342
2085
+ },
2086
+ {
2087
+ "epoch": 0.71,
2088
+ "learning_rate": 0.0001,
2089
+ "loss": 0.0858,
2090
+ "step": 343
2091
+ },
2092
+ {
2093
+ "epoch": 0.71,
2094
+ "learning_rate": 0.0001,
2095
+ "loss": 0.0681,
2096
+ "step": 344
2097
+ },
2098
+ {
2099
+ "epoch": 0.71,
2100
+ "learning_rate": 0.0001,
2101
+ "loss": 0.0756,
2102
+ "step": 345
2103
+ },
2104
+ {
2105
+ "epoch": 0.71,
2106
+ "learning_rate": 0.0001,
2107
+ "loss": 0.0842,
2108
+ "step": 346
2109
+ },
2110
+ {
2111
+ "epoch": 0.72,
2112
+ "learning_rate": 0.0001,
2113
+ "loss": 0.1008,
2114
+ "step": 347
2115
+ },
2116
+ {
2117
+ "epoch": 0.72,
2118
+ "learning_rate": 0.0001,
2119
+ "loss": 0.1254,
2120
+ "step": 348
2121
+ },
2122
+ {
2123
+ "epoch": 0.72,
2124
+ "learning_rate": 0.0001,
2125
+ "loss": 0.0795,
2126
+ "step": 349
2127
+ },
2128
+ {
2129
+ "epoch": 0.72,
2130
+ "learning_rate": 0.0001,
2131
+ "loss": 0.0869,
2132
+ "step": 350
2133
+ },
2134
+ {
2135
+ "epoch": 0.72,
2136
+ "learning_rate": 0.0001,
2137
+ "loss": 0.1,
2138
+ "step": 351
2139
+ },
2140
+ {
2141
+ "epoch": 0.73,
2142
+ "learning_rate": 0.0001,
2143
+ "loss": 0.0925,
2144
+ "step": 352
2145
+ },
2146
+ {
2147
+ "epoch": 0.73,
2148
+ "learning_rate": 0.0001,
2149
+ "loss": 0.0609,
2150
+ "step": 353
2151
+ },
2152
+ {
2153
+ "epoch": 0.73,
2154
+ "learning_rate": 0.0001,
2155
+ "loss": 0.0579,
2156
+ "step": 354
2157
+ },
2158
+ {
2159
+ "epoch": 0.73,
2160
+ "learning_rate": 0.0001,
2161
+ "loss": 0.1045,
2162
+ "step": 355
2163
+ },
2164
+ {
2165
+ "epoch": 0.73,
2166
+ "learning_rate": 0.0001,
2167
+ "loss": 0.082,
2168
+ "step": 356
2169
+ },
2170
+ {
2171
+ "epoch": 0.74,
2172
+ "learning_rate": 0.0001,
2173
+ "loss": 0.1129,
2174
+ "step": 357
2175
+ },
2176
+ {
2177
+ "epoch": 0.74,
2178
+ "learning_rate": 0.0001,
2179
+ "loss": 0.0776,
2180
+ "step": 358
2181
+ },
2182
+ {
2183
+ "epoch": 0.74,
2184
+ "learning_rate": 0.0001,
2185
+ "loss": 0.0747,
2186
+ "step": 359
2187
+ },
2188
+ {
2189
+ "epoch": 0.74,
2190
+ "learning_rate": 0.0001,
2191
+ "loss": 0.1089,
2192
+ "step": 360
2193
+ },
2194
+ {
2195
+ "epoch": 0.74,
2196
+ "learning_rate": 0.0001,
2197
+ "loss": 0.0581,
2198
+ "step": 361
2199
+ },
2200
+ {
2201
+ "epoch": 0.75,
2202
+ "learning_rate": 0.0001,
2203
+ "loss": 0.0613,
2204
+ "step": 362
2205
+ },
2206
+ {
2207
+ "epoch": 0.75,
2208
+ "learning_rate": 0.0001,
2209
+ "loss": 0.0679,
2210
+ "step": 363
2211
+ },
2212
+ {
2213
+ "epoch": 0.75,
2214
+ "learning_rate": 0.0001,
2215
+ "loss": 0.0817,
2216
+ "step": 364
2217
+ },
2218
+ {
2219
+ "epoch": 0.75,
2220
+ "learning_rate": 0.0001,
2221
+ "loss": 0.0609,
2222
+ "step": 365
2223
+ },
2224
+ {
2225
+ "epoch": 0.75,
2226
+ "learning_rate": 0.0001,
2227
+ "loss": 0.0688,
2228
+ "step": 366
2229
+ },
2230
+ {
2231
+ "epoch": 0.76,
2232
+ "learning_rate": 0.0001,
2233
+ "loss": 0.126,
2234
+ "step": 367
2235
+ },
2236
+ {
2237
+ "epoch": 0.76,
2238
+ "learning_rate": 0.0001,
2239
+ "loss": 0.0956,
2240
+ "step": 368
2241
+ },
2242
+ {
2243
+ "epoch": 0.76,
2244
+ "learning_rate": 0.0001,
2245
+ "loss": 0.0566,
2246
+ "step": 369
2247
+ },
2248
+ {
2249
+ "epoch": 0.76,
2250
+ "learning_rate": 0.0001,
2251
+ "loss": 0.0407,
2252
+ "step": 370
2253
+ },
2254
+ {
2255
+ "epoch": 0.76,
2256
+ "learning_rate": 0.0001,
2257
+ "loss": 0.0685,
2258
+ "step": 371
2259
+ },
2260
+ {
2261
+ "epoch": 0.77,
2262
+ "learning_rate": 0.0001,
2263
+ "loss": 0.0404,
2264
+ "step": 372
2265
+ },
2266
+ {
2267
+ "epoch": 0.77,
2268
+ "learning_rate": 0.0001,
2269
+ "loss": 0.1011,
2270
+ "step": 373
2271
+ },
2272
+ {
2273
+ "epoch": 0.77,
2274
+ "learning_rate": 0.0001,
2275
+ "loss": 0.0549,
2276
+ "step": 374
2277
+ },
2278
+ {
2279
+ "epoch": 0.77,
2280
+ "learning_rate": 0.0001,
2281
+ "loss": 0.0562,
2282
+ "step": 375
2283
+ },
2284
+ {
2285
+ "epoch": 0.77,
2286
+ "learning_rate": 0.0001,
2287
+ "loss": 0.0477,
2288
+ "step": 376
2289
+ },
2290
+ {
2291
+ "epoch": 0.78,
2292
+ "learning_rate": 0.0001,
2293
+ "loss": 0.1497,
2294
+ "step": 377
2295
+ },
2296
+ {
2297
+ "epoch": 0.78,
2298
+ "learning_rate": 0.0001,
2299
+ "loss": 0.0704,
2300
+ "step": 378
2301
+ },
2302
+ {
2303
+ "epoch": 0.78,
2304
+ "learning_rate": 0.0001,
2305
+ "loss": 0.097,
2306
+ "step": 379
2307
+ },
2308
+ {
2309
+ "epoch": 0.78,
2310
+ "learning_rate": 0.0001,
2311
+ "loss": 0.0732,
2312
+ "step": 380
2313
+ },
2314
+ {
2315
+ "epoch": 0.79,
2316
+ "learning_rate": 0.0001,
2317
+ "loss": 0.0572,
2318
+ "step": 381
2319
+ },
2320
+ {
2321
+ "epoch": 0.79,
2322
+ "learning_rate": 0.0001,
2323
+ "loss": 0.0902,
2324
+ "step": 382
2325
+ },
2326
+ {
2327
+ "epoch": 0.79,
2328
+ "learning_rate": 0.0001,
2329
+ "loss": 0.0783,
2330
+ "step": 383
2331
+ },
2332
+ {
2333
+ "epoch": 0.79,
2334
+ "learning_rate": 0.0001,
2335
+ "loss": 0.1082,
2336
+ "step": 384
2337
+ },
2338
+ {
2339
+ "epoch": 0.79,
2340
+ "learning_rate": 0.0001,
2341
+ "loss": 0.057,
2342
+ "step": 385
2343
+ },
2344
+ {
2345
+ "epoch": 0.8,
2346
+ "learning_rate": 0.0001,
2347
+ "loss": 0.0636,
2348
+ "step": 386
2349
+ },
2350
+ {
2351
+ "epoch": 0.8,
2352
+ "learning_rate": 0.0001,
2353
+ "loss": 0.1116,
2354
+ "step": 387
2355
+ },
2356
+ {
2357
+ "epoch": 0.8,
2358
+ "learning_rate": 0.0001,
2359
+ "loss": 0.0442,
2360
+ "step": 388
2361
+ },
2362
+ {
2363
+ "epoch": 0.8,
2364
+ "learning_rate": 0.0001,
2365
+ "loss": 0.1222,
2366
+ "step": 389
2367
+ },
2368
+ {
2369
+ "epoch": 0.8,
2370
+ "learning_rate": 0.0001,
2371
+ "loss": 0.0978,
2372
+ "step": 390
2373
+ },
2374
+ {
2375
+ "epoch": 0.81,
2376
+ "learning_rate": 0.0001,
2377
+ "loss": 0.0529,
2378
+ "step": 391
2379
+ },
2380
+ {
2381
+ "epoch": 0.81,
2382
+ "learning_rate": 0.0001,
2383
+ "loss": 0.0465,
2384
+ "step": 392
2385
+ },
2386
+ {
2387
+ "epoch": 0.81,
2388
+ "learning_rate": 0.0001,
2389
+ "loss": 0.0378,
2390
+ "step": 393
2391
+ },
2392
+ {
2393
+ "epoch": 0.81,
2394
+ "learning_rate": 0.0001,
2395
+ "loss": 0.094,
2396
+ "step": 394
2397
+ },
2398
+ {
2399
+ "epoch": 0.81,
2400
+ "learning_rate": 0.0001,
2401
+ "loss": 0.106,
2402
+ "step": 395
2403
+ },
2404
+ {
2405
+ "epoch": 0.82,
2406
+ "learning_rate": 0.0001,
2407
+ "loss": 0.0707,
2408
+ "step": 396
2409
+ },
2410
+ {
2411
+ "epoch": 0.82,
2412
+ "learning_rate": 0.0001,
2413
+ "loss": 0.0667,
2414
+ "step": 397
2415
+ },
2416
+ {
2417
+ "epoch": 0.82,
2418
+ "learning_rate": 0.0001,
2419
+ "loss": 0.0662,
2420
+ "step": 398
2421
+ },
2422
+ {
2423
+ "epoch": 0.82,
2424
+ "learning_rate": 0.0001,
2425
+ "loss": 0.0952,
2426
+ "step": 399
2427
+ },
2428
+ {
2429
+ "epoch": 0.82,
2430
+ "learning_rate": 0.0001,
2431
+ "loss": 0.055,
2432
+ "step": 400
2433
+ },
2434
+ {
2435
+ "epoch": 0.82,
2436
+ "eval_loss": 0.10834690928459167,
2437
+ "eval_runtime": 964.0919,
2438
+ "eval_samples_per_second": 33.486,
2439
+ "eval_steps_per_second": 0.262,
2440
+ "step": 400
2441
+ },
2442
+ {
2443
+ "epoch": 0.83,
2444
+ "learning_rate": 0.0001,
2445
+ "loss": 0.088,
2446
+ "step": 401
2447
+ },
2448
+ {
2449
+ "epoch": 0.83,
2450
+ "learning_rate": 0.0001,
2451
+ "loss": 0.084,
2452
+ "step": 402
2453
+ },
2454
+ {
2455
+ "epoch": 0.83,
2456
+ "learning_rate": 0.0001,
2457
+ "loss": 0.0543,
2458
+ "step": 403
2459
+ },
2460
+ {
2461
+ "epoch": 0.83,
2462
+ "learning_rate": 0.0001,
2463
+ "loss": 0.0969,
2464
+ "step": 404
2465
+ },
2466
+ {
2467
+ "epoch": 0.83,
2468
+ "learning_rate": 0.0001,
2469
+ "loss": 0.0494,
2470
+ "step": 405
2471
+ },
2472
+ {
2473
+ "epoch": 0.84,
2474
+ "learning_rate": 0.0001,
2475
+ "loss": 0.1015,
2476
+ "step": 406
2477
+ },
2478
+ {
2479
+ "epoch": 0.84,
2480
+ "learning_rate": 0.0001,
2481
+ "loss": 0.0843,
2482
+ "step": 407
2483
+ },
2484
+ {
2485
+ "epoch": 0.84,
2486
+ "learning_rate": 0.0001,
2487
+ "loss": 0.0427,
2488
+ "step": 408
2489
+ },
2490
+ {
2491
+ "epoch": 0.84,
2492
+ "learning_rate": 0.0001,
2493
+ "loss": 0.0777,
2494
+ "step": 409
2495
+ },
2496
+ {
2497
+ "epoch": 0.84,
2498
+ "learning_rate": 0.0001,
2499
+ "loss": 0.0499,
2500
+ "step": 410
2501
+ },
2502
+ {
2503
+ "epoch": 0.85,
2504
+ "learning_rate": 0.0001,
2505
+ "loss": 0.1067,
2506
+ "step": 411
2507
+ },
2508
+ {
2509
+ "epoch": 0.85,
2510
+ "learning_rate": 0.0001,
2511
+ "loss": 0.1234,
2512
+ "step": 412
2513
+ },
2514
+ {
2515
+ "epoch": 0.85,
2516
+ "learning_rate": 0.0001,
2517
+ "loss": 0.0464,
2518
+ "step": 413
2519
+ },
2520
+ {
2521
+ "epoch": 0.85,
2522
+ "learning_rate": 0.0001,
2523
+ "loss": 0.0744,
2524
+ "step": 414
2525
+ },
2526
+ {
2527
+ "epoch": 0.86,
2528
+ "learning_rate": 0.0001,
2529
+ "loss": 0.1134,
2530
+ "step": 415
2531
+ },
2532
+ {
2533
+ "epoch": 0.86,
2534
+ "learning_rate": 0.0001,
2535
+ "loss": 0.0643,
2536
+ "step": 416
2537
+ },
2538
+ {
2539
+ "epoch": 0.86,
2540
+ "learning_rate": 0.0001,
2541
+ "loss": 0.0636,
2542
+ "step": 417
2543
+ },
2544
+ {
2545
+ "epoch": 0.86,
2546
+ "learning_rate": 0.0001,
2547
+ "loss": 0.0602,
2548
+ "step": 418
2549
+ },
2550
+ {
2551
+ "epoch": 0.86,
2552
+ "learning_rate": 0.0001,
2553
+ "loss": 0.0845,
2554
+ "step": 419
2555
+ },
2556
+ {
2557
+ "epoch": 0.87,
2558
+ "learning_rate": 0.0001,
2559
+ "loss": 0.0637,
2560
+ "step": 420
2561
+ },
2562
+ {
2563
+ "epoch": 0.87,
2564
+ "learning_rate": 0.0001,
2565
+ "loss": 0.0806,
2566
+ "step": 421
2567
+ },
2568
+ {
2569
+ "epoch": 0.87,
2570
+ "learning_rate": 0.0001,
2571
+ "loss": 0.1163,
2572
+ "step": 422
2573
+ },
2574
+ {
2575
+ "epoch": 0.87,
2576
+ "learning_rate": 0.0001,
2577
+ "loss": 0.0887,
2578
+ "step": 423
2579
+ },
2580
+ {
2581
+ "epoch": 0.87,
2582
+ "learning_rate": 0.0001,
2583
+ "loss": 0.0852,
2584
+ "step": 424
2585
+ },
2586
+ {
2587
+ "epoch": 0.88,
2588
+ "learning_rate": 0.0001,
2589
+ "loss": 0.0468,
2590
+ "step": 425
2591
+ },
2592
+ {
2593
+ "epoch": 0.88,
2594
+ "learning_rate": 0.0001,
2595
+ "loss": 0.035,
2596
+ "step": 426
2597
+ },
2598
+ {
2599
+ "epoch": 0.88,
2600
+ "learning_rate": 0.0001,
2601
+ "loss": 0.0691,
2602
+ "step": 427
2603
+ },
2604
+ {
2605
+ "epoch": 0.88,
2606
+ "learning_rate": 0.0001,
2607
+ "loss": 0.0598,
2608
+ "step": 428
2609
+ },
2610
+ {
2611
+ "epoch": 0.88,
2612
+ "learning_rate": 0.0001,
2613
+ "loss": 0.0653,
2614
+ "step": 429
2615
+ },
2616
+ {
2617
+ "epoch": 0.89,
2618
+ "learning_rate": 0.0001,
2619
+ "loss": 0.0783,
2620
+ "step": 430
2621
+ },
2622
+ {
2623
+ "epoch": 0.89,
2624
+ "learning_rate": 0.0001,
2625
+ "loss": 0.0858,
2626
+ "step": 431
2627
+ },
2628
+ {
2629
+ "epoch": 0.89,
2630
+ "learning_rate": 0.0001,
2631
+ "loss": 0.0845,
2632
+ "step": 432
2633
+ },
2634
+ {
2635
+ "epoch": 0.89,
2636
+ "learning_rate": 0.0001,
2637
+ "loss": 0.0677,
2638
+ "step": 433
2639
+ },
2640
+ {
2641
+ "epoch": 0.89,
2642
+ "learning_rate": 0.0001,
2643
+ "loss": 0.0609,
2644
+ "step": 434
2645
+ },
2646
+ {
2647
+ "epoch": 0.9,
2648
+ "learning_rate": 0.0001,
2649
+ "loss": 0.0679,
2650
+ "step": 435
2651
+ },
2652
+ {
2653
+ "epoch": 0.9,
2654
+ "learning_rate": 0.0001,
2655
+ "loss": 0.0898,
2656
+ "step": 436
2657
+ },
2658
+ {
2659
+ "epoch": 0.9,
2660
+ "learning_rate": 0.0001,
2661
+ "loss": 0.118,
2662
+ "step": 437
2663
+ },
2664
+ {
2665
+ "epoch": 0.9,
2666
+ "learning_rate": 0.0001,
2667
+ "loss": 0.0654,
2668
+ "step": 438
2669
+ },
2670
+ {
2671
+ "epoch": 0.9,
2672
+ "learning_rate": 0.0001,
2673
+ "loss": 0.0688,
2674
+ "step": 439
2675
+ },
2676
+ {
2677
+ "epoch": 0.91,
2678
+ "learning_rate": 0.0001,
2679
+ "loss": 0.0776,
2680
+ "step": 440
2681
+ },
2682
+ {
2683
+ "epoch": 0.91,
2684
+ "learning_rate": 0.0001,
2685
+ "loss": 0.0471,
2686
+ "step": 441
2687
+ },
2688
+ {
2689
+ "epoch": 0.91,
2690
+ "learning_rate": 0.0001,
2691
+ "loss": 0.0637,
2692
+ "step": 442
2693
+ },
2694
+ {
2695
+ "epoch": 0.91,
2696
+ "learning_rate": 0.0001,
2697
+ "loss": 0.0776,
2698
+ "step": 443
2699
+ },
2700
+ {
2701
+ "epoch": 0.91,
2702
+ "learning_rate": 0.0001,
2703
+ "loss": 0.0877,
2704
+ "step": 444
2705
+ },
2706
+ {
2707
+ "epoch": 0.92,
2708
+ "learning_rate": 0.0001,
2709
+ "loss": 0.0739,
2710
+ "step": 445
2711
+ },
2712
+ {
2713
+ "epoch": 0.92,
2714
+ "learning_rate": 0.0001,
2715
+ "loss": 0.0605,
2716
+ "step": 446
2717
+ },
2718
+ {
2719
+ "epoch": 0.92,
2720
+ "learning_rate": 0.0001,
2721
+ "loss": 0.0581,
2722
+ "step": 447
2723
+ },
2724
+ {
2725
+ "epoch": 0.92,
2726
+ "learning_rate": 0.0001,
2727
+ "loss": 0.0601,
2728
+ "step": 448
2729
+ },
2730
+ {
2731
+ "epoch": 0.93,
2732
+ "learning_rate": 0.0001,
2733
+ "loss": 0.0386,
2734
+ "step": 449
2735
+ },
2736
+ {
2737
+ "epoch": 0.93,
2738
+ "learning_rate": 0.0001,
2739
+ "loss": 0.0696,
2740
+ "step": 450
2741
+ },
2742
+ {
2743
+ "epoch": 0.93,
2744
+ "learning_rate": 0.0001,
2745
+ "loss": 0.1212,
2746
+ "step": 451
2747
+ },
2748
+ {
2749
+ "epoch": 0.93,
2750
+ "learning_rate": 0.0001,
2751
+ "loss": 0.1015,
2752
+ "step": 452
2753
+ },
2754
+ {
2755
+ "epoch": 0.93,
2756
+ "learning_rate": 0.0001,
2757
+ "loss": 0.0549,
2758
+ "step": 453
2759
+ },
2760
+ {
2761
+ "epoch": 0.94,
2762
+ "learning_rate": 0.0001,
2763
+ "loss": 0.0789,
2764
+ "step": 454
2765
+ },
2766
+ {
2767
+ "epoch": 0.94,
2768
+ "learning_rate": 0.0001,
2769
+ "loss": 0.0589,
2770
+ "step": 455
2771
+ },
2772
+ {
2773
+ "epoch": 0.94,
2774
+ "learning_rate": 0.0001,
2775
+ "loss": 0.0517,
2776
+ "step": 456
2777
+ },
2778
+ {
2779
+ "epoch": 0.94,
2780
+ "learning_rate": 0.0001,
2781
+ "loss": 0.1175,
2782
+ "step": 457
2783
+ },
2784
+ {
2785
+ "epoch": 0.94,
2786
+ "learning_rate": 0.0001,
2787
+ "loss": 0.1051,
2788
+ "step": 458
2789
+ },
2790
+ {
2791
+ "epoch": 0.95,
2792
+ "learning_rate": 0.0001,
2793
+ "loss": 0.0799,
2794
+ "step": 459
2795
+ },
2796
+ {
2797
+ "epoch": 0.95,
2798
+ "learning_rate": 0.0001,
2799
+ "loss": 0.061,
2800
+ "step": 460
2801
+ },
2802
+ {
2803
+ "epoch": 0.95,
2804
+ "learning_rate": 0.0001,
2805
+ "loss": 0.102,
2806
+ "step": 461
2807
+ },
2808
+ {
2809
+ "epoch": 0.95,
2810
+ "learning_rate": 0.0001,
2811
+ "loss": 0.076,
2812
+ "step": 462
2813
+ },
2814
+ {
2815
+ "epoch": 0.95,
2816
+ "learning_rate": 0.0001,
2817
+ "loss": 0.0885,
2818
+ "step": 463
2819
+ },
2820
+ {
2821
+ "epoch": 0.96,
2822
+ "learning_rate": 0.0001,
2823
+ "loss": 0.1001,
2824
+ "step": 464
2825
+ },
2826
+ {
2827
+ "epoch": 0.96,
2828
+ "learning_rate": 0.0001,
2829
+ "loss": 0.0541,
2830
+ "step": 465
2831
+ },
2832
+ {
2833
+ "epoch": 0.96,
2834
+ "learning_rate": 0.0001,
2835
+ "loss": 0.0625,
2836
+ "step": 466
2837
+ },
2838
+ {
2839
+ "epoch": 0.96,
2840
+ "learning_rate": 0.0001,
2841
+ "loss": 0.0762,
2842
+ "step": 467
2843
+ },
2844
+ {
2845
+ "epoch": 0.96,
2846
+ "learning_rate": 0.0001,
2847
+ "loss": 0.0836,
2848
+ "step": 468
2849
+ },
2850
+ {
2851
+ "epoch": 0.97,
2852
+ "learning_rate": 0.0001,
2853
+ "loss": 0.0593,
2854
+ "step": 469
2855
+ },
2856
+ {
2857
+ "epoch": 0.97,
2858
+ "learning_rate": 0.0001,
2859
+ "loss": 0.0892,
2860
+ "step": 470
2861
+ },
2862
+ {
2863
+ "epoch": 0.97,
2864
+ "learning_rate": 0.0001,
2865
+ "loss": 0.0841,
2866
+ "step": 471
2867
+ },
2868
+ {
2869
+ "epoch": 0.97,
2870
+ "learning_rate": 0.0001,
2871
+ "loss": 0.0647,
2872
+ "step": 472
2873
+ },
2874
+ {
2875
+ "epoch": 0.97,
2876
+ "learning_rate": 0.0001,
2877
+ "loss": 0.0566,
2878
+ "step": 473
2879
+ },
2880
+ {
2881
+ "epoch": 0.98,
2882
+ "learning_rate": 0.0001,
2883
+ "loss": 0.105,
2884
+ "step": 474
2885
+ },
2886
+ {
2887
+ "epoch": 0.98,
2888
+ "learning_rate": 0.0001,
2889
+ "loss": 0.0926,
2890
+ "step": 475
2891
+ },
2892
+ {
2893
+ "epoch": 0.98,
2894
+ "learning_rate": 0.0001,
2895
+ "loss": 0.0564,
2896
+ "step": 476
2897
+ },
2898
+ {
2899
+ "epoch": 0.98,
2900
+ "learning_rate": 0.0001,
2901
+ "loss": 0.0504,
2902
+ "step": 477
2903
+ },
2904
+ {
2905
+ "epoch": 0.99,
2906
+ "learning_rate": 0.0001,
2907
+ "loss": 0.0474,
2908
+ "step": 478
2909
+ },
2910
+ {
2911
+ "epoch": 0.99,
2912
+ "learning_rate": 0.0001,
2913
+ "loss": 0.0477,
2914
+ "step": 479
2915
+ },
2916
+ {
2917
+ "epoch": 0.99,
2918
+ "learning_rate": 0.0001,
2919
+ "loss": 0.0577,
2920
+ "step": 480
2921
+ },
2922
+ {
2923
+ "epoch": 0.99,
2924
+ "learning_rate": 0.0001,
2925
+ "loss": 0.0782,
2926
+ "step": 481
2927
+ },
2928
+ {
2929
+ "epoch": 0.99,
2930
+ "learning_rate": 0.0001,
2931
+ "loss": 0.059,
2932
+ "step": 482
2933
+ },
2934
+ {
2935
+ "epoch": 1.0,
2936
+ "learning_rate": 0.0001,
2937
+ "loss": 0.0467,
2938
+ "step": 483
2939
+ },
2940
+ {
2941
+ "epoch": 1.0,
2942
+ "learning_rate": 0.0001,
2943
+ "loss": 0.0392,
2944
+ "step": 484
2945
+ },
2946
+ {
2947
+ "epoch": 1.0,
2948
+ "learning_rate": 0.0001,
2949
+ "loss": 0.079,
2950
+ "step": 485
2951
+ },
2952
+ {
2953
+ "epoch": 1.0,
2954
+ "learning_rate": 0.0001,
2955
+ "loss": 0.125,
2956
+ "step": 486
2957
+ },
2958
+ {
2959
+ "epoch": 1.0,
2960
+ "learning_rate": 0.0001,
2961
+ "loss": 0.0938,
2962
+ "step": 487
2963
+ },
2964
+ {
2965
+ "epoch": 1.01,
2966
+ "learning_rate": 0.0001,
2967
+ "loss": 0.0671,
2968
+ "step": 488
2969
+ },
2970
+ {
2971
+ "epoch": 1.01,
2972
+ "learning_rate": 0.0001,
2973
+ "loss": 0.0634,
2974
+ "step": 489
2975
+ },
2976
+ {
2977
+ "epoch": 1.01,
2978
+ "learning_rate": 0.0001,
2979
+ "loss": 0.066,
2980
+ "step": 490
2981
+ },
2982
+ {
2983
+ "epoch": 1.01,
2984
+ "learning_rate": 0.0001,
2985
+ "loss": 0.0636,
2986
+ "step": 491
2987
+ },
2988
+ {
2989
+ "epoch": 1.01,
2990
+ "learning_rate": 0.0001,
2991
+ "loss": 0.0463,
2992
+ "step": 492
2993
+ },
2994
+ {
2995
+ "epoch": 1.02,
2996
+ "learning_rate": 0.0001,
2997
+ "loss": 0.0723,
2998
+ "step": 493
2999
+ },
3000
+ {
3001
+ "epoch": 1.02,
3002
+ "learning_rate": 0.0001,
3003
+ "loss": 0.0808,
3004
+ "step": 494
3005
+ },
3006
+ {
3007
+ "epoch": 1.02,
3008
+ "learning_rate": 0.0001,
3009
+ "loss": 0.0636,
3010
+ "step": 495
3011
+ },
3012
+ {
3013
+ "epoch": 1.02,
3014
+ "learning_rate": 0.0001,
3015
+ "loss": 0.0643,
3016
+ "step": 496
3017
+ },
3018
+ {
3019
+ "epoch": 1.02,
3020
+ "learning_rate": 0.0001,
3021
+ "loss": 0.0511,
3022
+ "step": 497
3023
+ },
3024
+ {
3025
+ "epoch": 1.03,
3026
+ "learning_rate": 0.0001,
3027
+ "loss": 0.0624,
3028
+ "step": 498
3029
+ },
3030
+ {
3031
+ "epoch": 1.03,
3032
+ "learning_rate": 0.0001,
3033
+ "loss": 0.0528,
3034
+ "step": 499
3035
+ },
3036
+ {
3037
+ "epoch": 1.03,
3038
+ "learning_rate": 0.0001,
3039
+ "loss": 0.046,
3040
+ "step": 500
3041
+ },
3042
+ {
3043
+ "epoch": 1.03,
3044
+ "eval_loss": 0.10144635289907455,
3045
+ "eval_runtime": 971.1805,
3046
+ "eval_samples_per_second": 33.242,
3047
+ "eval_steps_per_second": 0.261,
3048
+ "step": 500
3049
+ }
3050
+ ],
3051
+ "max_steps": 4850,
3052
+ "num_train_epochs": 10,
3053
+ "total_flos": 2.229441347410985e+17,
3054
+ "trial_name": null,
3055
+ "trial_params": null
3056
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f3afca07744a9198d8ab8532516ebb3d52a262ffc7ff1324ade6888688d3bcb
3
+ size 4603
zero_to_fp32.py ADDED
@@ -0,0 +1,483 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+ '''Copyright The Microsoft DeepSpeed Team'''
3
+
4
+ # This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
5
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
6
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
7
+ # application.
8
+ #
9
+ # example: python zero_to_fp32.py . pytorch_model.bin
10
+
11
+ import argparse
12
+ import torch
13
+ import glob
14
+ import math
15
+ import os
16
+ import re
17
+ from collections import OrderedDict
18
+
19
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
20
+ # DeepSpeed data structures it has to be available in the current python environment.
21
+ from deepspeed.utils import logger
22
+ from deepspeed.checkpoint.constants import (DS_VERSION,
23
+ OPTIMIZER_STATE_DICT,
24
+ SINGLE_PARTITION_OF_FP32_GROUPS,
25
+ FP32_FLAT_GROUPS,
26
+ ZERO_STAGE,
27
+ PARTITION_COUNT,
28
+ PARAM_SHAPES,
29
+ BUFFER_NAMES)
30
+
31
+ debug = 0
32
+
33
+ # load to cpu
34
+ device = torch.device('cpu')
35
+
36
+
37
+ def atoi(text):
38
+ return int(text) if text.isdigit() else text
39
+
40
+
41
+ def natural_keys(text):
42
+ '''
43
+ alist.sort(key=natural_keys) sorts in human order
44
+ http://nedbatchelder.com/blog/200712/human_sorting.html
45
+ (See Toothy's implementation in the comments)
46
+ '''
47
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
48
+
49
+
50
+ def get_model_state_file(checkpoint_dir, zero_stage):
51
+ if not os.path.isdir(checkpoint_dir):
52
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
53
+
54
+ # there should be only one file
55
+ if zero_stage == 2:
56
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
57
+ elif zero_stage == 3:
58
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
59
+
60
+ if not os.path.exists(file):
61
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
62
+
63
+ return file
64
+
65
+
66
+ def get_optim_files(checkpoint_dir):
67
+ # XXX: need to test that this simple glob rule works for multi-node setup too
68
+ optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
69
+ "*_optim_states.pt")),
70
+ key=natural_keys)
71
+
72
+ if len(optim_files) == 0:
73
+ raise FileNotFoundError(
74
+ f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
75
+
76
+ return optim_files
77
+
78
+
79
+ def parse_model_state(file):
80
+ state_dict = torch.load(file, map_location=device)
81
+
82
+ if BUFFER_NAMES not in state_dict:
83
+ raise ValueError(f"{file} is not a model state checkpoint")
84
+ buffer_names = state_dict[BUFFER_NAMES]
85
+ if debug:
86
+ print("Found buffers:", buffer_names)
87
+
88
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
89
+ buffers = {
90
+ k: v.float()
91
+ for k,
92
+ v in state_dict["module"].items() if k in buffer_names
93
+ }
94
+ param_shapes = state_dict[PARAM_SHAPES]
95
+
96
+ ds_version = state_dict.get(DS_VERSION, None)
97
+
98
+ return buffers, param_shapes, ds_version
99
+
100
+
101
+ def parse_optim_states(files, ds_checkpoint_dir):
102
+
103
+ total_files = len(files)
104
+ state_dicts = []
105
+ for f in files:
106
+ state_dicts.append(torch.load(f, map_location=device))
107
+
108
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
109
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
110
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
111
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
112
+
113
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
114
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
115
+ # use the max of the partition_count to get the dp world_size.
116
+
117
+ if type(world_size) is list:
118
+ world_size = max(world_size)
119
+
120
+ if world_size != total_files:
121
+ raise ValueError(
122
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
123
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
124
+ )
125
+
126
+ # the groups are named differently in each stage
127
+ if zero_stage == 2:
128
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
129
+ elif zero_stage == 3:
130
+ fp32_groups_key = FP32_FLAT_GROUPS
131
+ else:
132
+ raise ValueError(f"unknown zero stage {zero_stage}")
133
+
134
+ if zero_stage == 2:
135
+ fp32_flat_groups = [
136
+ state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
137
+ for i in range(len(state_dicts))
138
+ ]
139
+ elif zero_stage == 3:
140
+ # if there is more than one param group, there will be multiple flattened tensors - one
141
+ # flattened tensor per group - for simplicity merge them into a single tensor
142
+ #
143
+ # XXX: could make the script more memory efficient for when there are multiple groups - it
144
+ # will require matching the sub-lists of param_shapes for each param group flattened tensor
145
+
146
+ fp32_flat_groups = [
147
+ torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
148
+ 0) for i in range(len(state_dicts))
149
+ ]
150
+
151
+ return zero_stage, world_size, fp32_flat_groups
152
+
153
+
154
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
155
+ """
156
+ Returns fp32 state_dict reconstructed from ds checkpoint
157
+
158
+ Args:
159
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
160
+
161
+ """
162
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
163
+
164
+ optim_files = get_optim_files(ds_checkpoint_dir)
165
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
166
+ print(
167
+ f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
168
+
169
+ model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
170
+ buffers, param_shapes, ds_version = parse_model_state(model_file)
171
+ print(f'Parsing checkpoint created by deepspeed=={ds_version}')
172
+
173
+ if zero_stage == 2:
174
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
175
+ param_shapes,
176
+ fp32_flat_groups,
177
+ buffers)
178
+ elif zero_stage == 3:
179
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
180
+ param_shapes,
181
+ fp32_flat_groups,
182
+ buffers)
183
+
184
+
185
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
186
+ param_shapes,
187
+ fp32_flat_groups,
188
+ buffers):
189
+
190
+ # Reconstruction protocol:
191
+ #
192
+ # XXX: document this
193
+
194
+ if debug:
195
+ for i in range(world_size):
196
+ for j in range(len(fp32_flat_groups[0])):
197
+ print(
198
+ f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
199
+
200
+ # XXX: memory usage doubles here (zero2)
201
+ num_param_groups = len(fp32_flat_groups[0])
202
+ merged_single_partition_of_fp32_groups = []
203
+ for i in range(num_param_groups):
204
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
205
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
206
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
207
+ avail_numel = sum([
208
+ full_single_fp32_vector.numel()
209
+ for full_single_fp32_vector in merged_single_partition_of_fp32_groups
210
+ ])
211
+
212
+ if debug:
213
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
214
+ wanted_numel = sum(
215
+ [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
216
+ # not asserting if there is a mismatch due to possible padding
217
+ print(f"Have {avail_numel} numels to process.")
218
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
219
+
220
+ state_dict = OrderedDict()
221
+
222
+ # buffers
223
+ state_dict.update(buffers)
224
+ if debug:
225
+ print(f"added {len(buffers)} buffers")
226
+
227
+ # params
228
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
229
+ # out-of-core computing solution
230
+ total_numel = 0
231
+ total_params = 0
232
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
233
+ offset = 0
234
+ avail_numel = full_single_fp32_vector.numel()
235
+ for name, shape in shapes.items():
236
+
237
+ unpartitioned_numel = shape.numel()
238
+ total_numel += unpartitioned_numel
239
+ total_params += 1
240
+
241
+ if debug:
242
+ print(
243
+ f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
244
+ )
245
+ state_dict[name] = full_single_fp32_vector.narrow(
246
+ 0,
247
+ offset,
248
+ unpartitioned_numel).view(shape)
249
+ offset += unpartitioned_numel
250
+
251
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
252
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
253
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
254
+ # live optimizer object, so we are checking that the numbers are within the right range
255
+ align_to = 2 * world_size
256
+
257
+ def zero2_align(x):
258
+ return align_to * math.ceil(x / align_to)
259
+
260
+ if debug:
261
+ print(f"original offset={offset}, avail_numel={avail_numel}")
262
+
263
+ offset = zero2_align(offset)
264
+ avail_numel = zero2_align(avail_numel)
265
+
266
+ if debug:
267
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
268
+
269
+ # Sanity check
270
+ if offset != avail_numel:
271
+ raise ValueError(
272
+ f"consumed {offset} numels out of {avail_numel} - something is wrong")
273
+
274
+ print(
275
+ f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
276
+ )
277
+
278
+ return state_dict
279
+
280
+
281
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
282
+ remainder = unpartitioned_numel % world_size
283
+ padding_numel = (world_size - remainder) if remainder else 0
284
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
285
+ return partitioned_numel, padding_numel
286
+
287
+
288
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
289
+ param_shapes,
290
+ fp32_flat_groups,
291
+ buffers):
292
+
293
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
294
+ # param, re-consolidating each param, while dealing with padding if any
295
+
296
+ avail_numel = fp32_flat_groups[0].numel() * world_size
297
+ # merge list of dicts, preserving order
298
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
299
+
300
+ if debug:
301
+ for i in range(world_size):
302
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
303
+
304
+ wanted_params = len(param_shapes)
305
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
306
+ # not asserting if there is a mismatch due to possible padding
307
+ print(f"Have {avail_numel} numels to process.")
308
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
309
+
310
+ state_dict = OrderedDict()
311
+
312
+ # buffers
313
+ state_dict.update(buffers)
314
+ if debug:
315
+ print(f"added {len(buffers)} buffers")
316
+
317
+ # params
318
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
319
+ # out-of-core computing solution
320
+ offset = 0
321
+ total_numel = 0
322
+ total_params = 0
323
+ for name, shape in param_shapes.items():
324
+
325
+ unpartitioned_numel = shape.numel()
326
+ total_numel += unpartitioned_numel
327
+ total_params += 1
328
+
329
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
330
+
331
+ if debug:
332
+ print(
333
+ f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
334
+ )
335
+
336
+ # XXX: memory usage doubles here
337
+ state_dict[name] = torch.cat(
338
+ tuple(fp32_flat_groups[i].narrow(0,
339
+ offset,
340
+ partitioned_numel)
341
+ for i in range(world_size)),
342
+ 0).narrow(0,
343
+ 0,
344
+ unpartitioned_numel).view(shape)
345
+ offset += partitioned_numel
346
+
347
+ offset *= world_size
348
+
349
+ # Sanity check
350
+ if offset != avail_numel:
351
+ raise ValueError(
352
+ f"consumed {offset} numels out of {avail_numel} - something is wrong")
353
+
354
+ print(
355
+ f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
356
+ )
357
+
358
+ return state_dict
359
+
360
+
361
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
362
+ """
363
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
364
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
365
+ via a model hub.
366
+
367
+ Args:
368
+ - ``checkpoint_dir``: path to the desired checkpoint folder
369
+ - ``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``
370
+
371
+ Returns:
372
+ - pytorch ``state_dict``
373
+
374
+ Note: this approach may not work if your application doesn't have sufficient free CPU memory and
375
+ you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
376
+ the checkpoint.
377
+
378
+ A typical usage might be ::
379
+
380
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
381
+ # do the training and checkpoint saving
382
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
383
+ model = model.cpu() # move to cpu
384
+ model.load_state_dict(state_dict)
385
+ # submit to model hub or save the model to share with others
386
+
387
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
388
+ application. i.e. you will need to re-initialize the deepspeed engine, since
389
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
390
+
391
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
392
+
393
+ """
394
+ if tag is None:
395
+ latest_path = os.path.join(checkpoint_dir, 'latest')
396
+ if os.path.isfile(latest_path):
397
+ with open(latest_path, 'r') as fd:
398
+ tag = fd.read().strip()
399
+ else:
400
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
401
+
402
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
403
+
404
+ if not os.path.isdir(ds_checkpoint_dir):
405
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
406
+
407
+ return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
408
+
409
+
410
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
411
+ """
412
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
413
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
414
+
415
+ Args:
416
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
417
+ - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
418
+ - ``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``
419
+ """
420
+
421
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
422
+ print(f"Saving fp32 state dict to {output_file}")
423
+ torch.save(state_dict, output_file)
424
+
425
+
426
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
427
+ """
428
+ 1. Put the provided model to cpu
429
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
430
+ 3. Load it into the provided model
431
+
432
+ Args:
433
+ - ``model``: the model object to update
434
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
435
+ - ``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``
436
+
437
+ Returns:
438
+ - ``model`: modified model
439
+
440
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
441
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
442
+ conveniently placed for you in the checkpoint folder.
443
+
444
+ A typical usage might be ::
445
+
446
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
447
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
448
+ # submit to model hub or save the model to share with others
449
+
450
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
451
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
452
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
453
+
454
+ """
455
+ logger.info(f"Extracting fp32 weights")
456
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
457
+
458
+ logger.info(f"Overwriting model with fp32 weights")
459
+ model = model.cpu()
460
+ model.load_state_dict(state_dict, strict=False)
461
+
462
+ return model
463
+
464
+
465
+ if __name__ == "__main__":
466
+
467
+ parser = argparse.ArgumentParser()
468
+ parser.add_argument(
469
+ "checkpoint_dir",
470
+ type=str,
471
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
472
+ parser.add_argument(
473
+ "output_file",
474
+ type=str,
475
+ help=
476
+ "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
477
+ )
478
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
479
+ args = parser.parse_args()
480
+
481
+ debug = args.debug
482
+
483
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)