Commit
•
c185259
1
Parent(s):
196f07f
init
Browse files- config.json +30 -0
- latest +1 -0
- pytorch_model.bin +3 -0
- rng_state_0.pth +3 -0
- rng_state_1.pth +3 -0
- rng_state_2.pth +3 -0
- rng_state_3.pth +3 -0
- special_tokens_map.json +1 -0
- tokenizer.json +0 -0
- tokenizer_config.json +1 -0
- trainer_state.json +1248 -0
- training_args.bin +3 -0
- zero_to_fp32.py +348 -0
config.json
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "google/t5-large-lm-adapt",
|
3 |
+
"architectures": [
|
4 |
+
"T5ForConditionalGeneration"
|
5 |
+
],
|
6 |
+
"d_ff": 2816,
|
7 |
+
"d_kv": 64,
|
8 |
+
"d_model": 1024,
|
9 |
+
"decoder_start_token_id": 0,
|
10 |
+
"dropout_rate": 0.1,
|
11 |
+
"eos_token_id": 1,
|
12 |
+
"feed_forward_proj": "gated-gelu",
|
13 |
+
"gradient_checkpointing": true,
|
14 |
+
"initializer_factor": 1.0,
|
15 |
+
"is_encoder_decoder": true,
|
16 |
+
"layer_norm_epsilon": 1e-06,
|
17 |
+
"max_length": 512,
|
18 |
+
"model_type": "t5",
|
19 |
+
"num_decoder_layers": 24,
|
20 |
+
"num_heads": 16,
|
21 |
+
"num_layers": 24,
|
22 |
+
"output_past": true,
|
23 |
+
"pad_token_id": 0,
|
24 |
+
"relative_attention_num_buckets": 32,
|
25 |
+
"tie_word_embeddings": false,
|
26 |
+
"torch_dtype": "float32",
|
27 |
+
"transformers_version": "4.9.1",
|
28 |
+
"use_cache": false,
|
29 |
+
"vocab_size": 32128
|
30 |
+
}
|
latest
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
global_step2000
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e37ea3374e90847d2f52fbeec66e4189afb8961c582f597ed237861a1d68406b
|
3 |
+
size 3132793282
|
rng_state_0.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:33d4e867339985a01d3551db4e58b19992103c762b3ebb145698ca499afe8886
|
3 |
+
size 14649
|
rng_state_1.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:07b9b533e43a86fd8d75da6e6dc2516c5a274ae1f30b985b8c6510bfe0cd7638
|
3 |
+
size 14654
|
rng_state_2.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e404ce3b064e6bb412f0226169d0deaf72d60d8d1935994e0e97d8bbfdbaebfd
|
3 |
+
size 14654
|
rng_state_3.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:8b9d1476a94159f17356260ee7fb2b4f98d4b6ec20ead888948f164e364a97eb
|
3 |
+
size 14654
|
special_tokens_map.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "additional_special_tokens": ["<extra_id_0>", "<extra_id_1>", "<extra_id_2>", "<extra_id_3>", "<extra_id_4>", "<extra_id_5>", "<extra_id_6>", "<extra_id_7>", "<extra_id_8>", "<extra_id_9>", "<extra_id_10>", "<extra_id_11>", "<extra_id_12>", "<extra_id_13>", "<extra_id_14>", "<extra_id_15>", "<extra_id_16>", "<extra_id_17>", "<extra_id_18>", "<extra_id_19>", "<extra_id_20>", "<extra_id_21>", "<extra_id_22>", "<extra_id_23>", "<extra_id_24>", "<extra_id_25>", "<extra_id_26>", "<extra_id_27>", "<extra_id_28>", "<extra_id_29>", "<extra_id_30>", "<extra_id_31>", "<extra_id_32>", "<extra_id_33>", "<extra_id_34>", "<extra_id_35>", "<extra_id_36>", "<extra_id_37>", "<extra_id_38>", "<extra_id_39>", "<extra_id_40>", "<extra_id_41>", "<extra_id_42>", "<extra_id_43>", "<extra_id_44>", "<extra_id_45>", "<extra_id_46>", "<extra_id_47>", "<extra_id_48>", "<extra_id_49>", "<extra_id_50>", "<extra_id_51>", "<extra_id_52>", "<extra_id_53>", "<extra_id_54>", "<extra_id_55>", "<extra_id_56>", "<extra_id_57>", "<extra_id_58>", "<extra_id_59>", "<extra_id_60>", "<extra_id_61>", "<extra_id_62>", "<extra_id_63>", "<extra_id_64>", "<extra_id_65>", "<extra_id_66>", "<extra_id_67>", "<extra_id_68>", "<extra_id_69>", "<extra_id_70>", "<extra_id_71>", "<extra_id_72>", "<extra_id_73>", "<extra_id_74>", "<extra_id_75>", "<extra_id_76>", "<extra_id_77>", "<extra_id_78>", "<extra_id_79>", "<extra_id_80>", "<extra_id_81>", "<extra_id_82>", "<extra_id_83>", "<extra_id_84>", "<extra_id_85>", "<extra_id_86>", "<extra_id_87>", "<extra_id_88>", "<extra_id_89>", "<extra_id_90>", "<extra_id_91>", "<extra_id_92>", "<extra_id_93>", "<extra_id_94>", "<extra_id_95>", "<extra_id_96>", "<extra_id_97>", "<extra_id_98>", "<extra_id_99>"]}
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
tokenizer_config.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>", "extra_ids": 100, "additional_special_tokens": ["<extra_id_0>", "<extra_id_1>", "<extra_id_2>", "<extra_id_3>", "<extra_id_4>", "<extra_id_5>", "<extra_id_6>", "<extra_id_7>", "<extra_id_8>", "<extra_id_9>", "<extra_id_10>", "<extra_id_11>", "<extra_id_12>", "<extra_id_13>", "<extra_id_14>", "<extra_id_15>", "<extra_id_16>", "<extra_id_17>", "<extra_id_18>", "<extra_id_19>", "<extra_id_20>", "<extra_id_21>", "<extra_id_22>", "<extra_id_23>", "<extra_id_24>", "<extra_id_25>", "<extra_id_26>", "<extra_id_27>", "<extra_id_28>", "<extra_id_29>", "<extra_id_30>", "<extra_id_31>", "<extra_id_32>", "<extra_id_33>", "<extra_id_34>", "<extra_id_35>", "<extra_id_36>", "<extra_id_37>", "<extra_id_38>", "<extra_id_39>", "<extra_id_40>", "<extra_id_41>", "<extra_id_42>", "<extra_id_43>", "<extra_id_44>", "<extra_id_45>", "<extra_id_46>", "<extra_id_47>", "<extra_id_48>", "<extra_id_49>", "<extra_id_50>", "<extra_id_51>", "<extra_id_52>", "<extra_id_53>", "<extra_id_54>", "<extra_id_55>", "<extra_id_56>", "<extra_id_57>", "<extra_id_58>", "<extra_id_59>", "<extra_id_60>", "<extra_id_61>", "<extra_id_62>", "<extra_id_63>", "<extra_id_64>", "<extra_id_65>", "<extra_id_66>", "<extra_id_67>", "<extra_id_68>", "<extra_id_69>", "<extra_id_70>", "<extra_id_71>", "<extra_id_72>", "<extra_id_73>", "<extra_id_74>", "<extra_id_75>", "<extra_id_76>", "<extra_id_77>", "<extra_id_78>", "<extra_id_79>", "<extra_id_80>", "<extra_id_81>", "<extra_id_82>", "<extra_id_83>", "<extra_id_84>", "<extra_id_85>", "<extra_id_86>", "<extra_id_87>", "<extra_id_88>", "<extra_id_89>", "<extra_id_90>", "<extra_id_91>", "<extra_id_92>", "<extra_id_93>", "<extra_id_94>", "<extra_id_95>", "<extra_id_96>", "<extra_id_97>", "<extra_id_98>", "<extra_id_99>"], "sp_model_kwargs": {}, "model_max_length": 512, "name_or_path": "google/t5-large-lm-adapt", "special_tokens_map_file": "/home/patrick/.cache/huggingface/transformers/28fa6bb11d5fd637c4b67b299f5616169510dac4cd181efc8f70674c7872c874.c94798918c92ded6aeef2d2f0e666d2cc4145eca1aa6e1336fde07f2e13e2f46", "tokenizer_class": "T5Tokenizer"}
|
trainer_state.json
ADDED
@@ -0,0 +1,1248 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 32.78688524590164,
|
5 |
+
"global_step": 2000,
|
6 |
+
"is_hyper_param_search": false,
|
7 |
+
"is_local_process_zero": true,
|
8 |
+
"is_world_process_zero": true,
|
9 |
+
"log_history": [
|
10 |
+
{
|
11 |
+
"epoch": 0.16,
|
12 |
+
"learning_rate": 2.4999999999999998e-05,
|
13 |
+
"loss": 4.5271,
|
14 |
+
"step": 10
|
15 |
+
},
|
16 |
+
{
|
17 |
+
"epoch": 0.33,
|
18 |
+
"learning_rate": 3.2525749891599525e-05,
|
19 |
+
"loss": 3.9027,
|
20 |
+
"step": 20
|
21 |
+
},
|
22 |
+
{
|
23 |
+
"epoch": 0.49,
|
24 |
+
"learning_rate": 3.6928031367991554e-05,
|
25 |
+
"loss": 3.4367,
|
26 |
+
"step": 30
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"epoch": 0.66,
|
30 |
+
"learning_rate": 4.005149978319905e-05,
|
31 |
+
"loss": 3.3343,
|
32 |
+
"step": 40
|
33 |
+
},
|
34 |
+
{
|
35 |
+
"epoch": 0.82,
|
36 |
+
"learning_rate": 4.247425010840046e-05,
|
37 |
+
"loss": 3.2186,
|
38 |
+
"step": 50
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"epoch": 0.98,
|
42 |
+
"learning_rate": 4.445378125959108e-05,
|
43 |
+
"loss": 3.1259,
|
44 |
+
"step": 60
|
45 |
+
},
|
46 |
+
{
|
47 |
+
"epoch": 1.15,
|
48 |
+
"learning_rate": 4.612745100035642e-05,
|
49 |
+
"loss": 2.9406,
|
50 |
+
"step": 70
|
51 |
+
},
|
52 |
+
{
|
53 |
+
"epoch": 1.31,
|
54 |
+
"learning_rate": 4.757724967479858e-05,
|
55 |
+
"loss": 2.8567,
|
56 |
+
"step": 80
|
57 |
+
},
|
58 |
+
{
|
59 |
+
"epoch": 1.48,
|
60 |
+
"learning_rate": 4.885606273598312e-05,
|
61 |
+
"loss": 2.8893,
|
62 |
+
"step": 90
|
63 |
+
},
|
64 |
+
{
|
65 |
+
"epoch": 1.64,
|
66 |
+
"learning_rate": 4.9999999999999996e-05,
|
67 |
+
"loss": 2.9067,
|
68 |
+
"step": 100
|
69 |
+
},
|
70 |
+
{
|
71 |
+
"epoch": 1.8,
|
72 |
+
"learning_rate": 5.1034817128955624e-05,
|
73 |
+
"loss": 2.813,
|
74 |
+
"step": 110
|
75 |
+
},
|
76 |
+
{
|
77 |
+
"epoch": 1.97,
|
78 |
+
"learning_rate": 5.197953115119061e-05,
|
79 |
+
"loss": 2.8364,
|
80 |
+
"step": 120
|
81 |
+
},
|
82 |
+
{
|
83 |
+
"epoch": 2.13,
|
84 |
+
"learning_rate": 5.2848583807670914e-05,
|
85 |
+
"loss": 2.7439,
|
86 |
+
"step": 130
|
87 |
+
},
|
88 |
+
{
|
89 |
+
"epoch": 2.3,
|
90 |
+
"learning_rate": 5.365320089195593e-05,
|
91 |
+
"loss": 2.7048,
|
92 |
+
"step": 140
|
93 |
+
},
|
94 |
+
{
|
95 |
+
"epoch": 2.46,
|
96 |
+
"learning_rate": 5.4402281476392025e-05,
|
97 |
+
"loss": 2.6631,
|
98 |
+
"step": 150
|
99 |
+
},
|
100 |
+
{
|
101 |
+
"epoch": 2.62,
|
102 |
+
"learning_rate": 5.5102999566398106e-05,
|
103 |
+
"loss": 2.6804,
|
104 |
+
"step": 160
|
105 |
+
},
|
106 |
+
{
|
107 |
+
"epoch": 2.79,
|
108 |
+
"learning_rate": 5.5761223034456847e-05,
|
109 |
+
"loss": 2.6094,
|
110 |
+
"step": 170
|
111 |
+
},
|
112 |
+
{
|
113 |
+
"epoch": 2.95,
|
114 |
+
"learning_rate": 5.6381812627582644e-05,
|
115 |
+
"loss": 2.6055,
|
116 |
+
"step": 180
|
117 |
+
},
|
118 |
+
{
|
119 |
+
"epoch": 3.11,
|
120 |
+
"learning_rate": 5.696884002382071e-05,
|
121 |
+
"loss": 2.4945,
|
122 |
+
"step": 190
|
123 |
+
},
|
124 |
+
{
|
125 |
+
"epoch": 3.28,
|
126 |
+
"learning_rate": 5.752574989159952e-05,
|
127 |
+
"loss": 2.5523,
|
128 |
+
"step": 200
|
129 |
+
},
|
130 |
+
{
|
131 |
+
"epoch": 3.44,
|
132 |
+
"learning_rate": 5.805548236834797e-05,
|
133 |
+
"loss": 2.5406,
|
134 |
+
"step": 210
|
135 |
+
},
|
136 |
+
{
|
137 |
+
"epoch": 3.61,
|
138 |
+
"learning_rate": 5.856056702055516e-05,
|
139 |
+
"loss": 2.5444,
|
140 |
+
"step": 220
|
141 |
+
},
|
142 |
+
{
|
143 |
+
"epoch": 3.77,
|
144 |
+
"learning_rate": 5.9043195900439815e-05,
|
145 |
+
"loss": 2.568,
|
146 |
+
"step": 230
|
147 |
+
},
|
148 |
+
{
|
149 |
+
"epoch": 3.93,
|
150 |
+
"learning_rate": 5.950528104279014e-05,
|
151 |
+
"loss": 2.5305,
|
152 |
+
"step": 240
|
153 |
+
},
|
154 |
+
{
|
155 |
+
"epoch": 4.1,
|
156 |
+
"learning_rate": 5.9948500216800926e-05,
|
157 |
+
"loss": 2.4653,
|
158 |
+
"step": 250
|
159 |
+
},
|
160 |
+
{
|
161 |
+
"epoch": 4.26,
|
162 |
+
"learning_rate": 6.037433369927045e-05,
|
163 |
+
"loss": 2.4644,
|
164 |
+
"step": 260
|
165 |
+
},
|
166 |
+
{
|
167 |
+
"epoch": 4.43,
|
168 |
+
"learning_rate": 6.078409410397467e-05,
|
169 |
+
"loss": 2.4302,
|
170 |
+
"step": 270
|
171 |
+
},
|
172 |
+
{
|
173 |
+
"epoch": 4.59,
|
174 |
+
"learning_rate": 6.117895078355547e-05,
|
175 |
+
"loss": 2.4025,
|
176 |
+
"step": 280
|
177 |
+
},
|
178 |
+
{
|
179 |
+
"epoch": 4.75,
|
180 |
+
"learning_rate": 6.15599499474739e-05,
|
181 |
+
"loss": 2.4185,
|
182 |
+
"step": 290
|
183 |
+
},
|
184 |
+
{
|
185 |
+
"epoch": 4.92,
|
186 |
+
"learning_rate": 6.192803136799156e-05,
|
187 |
+
"loss": 2.4651,
|
188 |
+
"step": 300
|
189 |
+
},
|
190 |
+
{
|
191 |
+
"epoch": 5.08,
|
192 |
+
"learning_rate": 6.22840423458568e-05,
|
193 |
+
"loss": 2.4095,
|
194 |
+
"step": 310
|
195 |
+
},
|
196 |
+
{
|
197 |
+
"epoch": 5.25,
|
198 |
+
"learning_rate": 6.262874945799764e-05,
|
199 |
+
"loss": 2.3613,
|
200 |
+
"step": 320
|
201 |
+
},
|
202 |
+
{
|
203 |
+
"epoch": 5.41,
|
204 |
+
"learning_rate": 6.296284849694718e-05,
|
205 |
+
"loss": 2.381,
|
206 |
+
"step": 330
|
207 |
+
},
|
208 |
+
{
|
209 |
+
"epoch": 5.57,
|
210 |
+
"learning_rate": 6.328697292605637e-05,
|
211 |
+
"loss": 2.3387,
|
212 |
+
"step": 340
|
213 |
+
},
|
214 |
+
{
|
215 |
+
"epoch": 5.74,
|
216 |
+
"learning_rate": 6.360170110875688e-05,
|
217 |
+
"loss": 2.3331,
|
218 |
+
"step": 350
|
219 |
+
},
|
220 |
+
{
|
221 |
+
"epoch": 5.9,
|
222 |
+
"learning_rate": 6.390756251918218e-05,
|
223 |
+
"loss": 2.3831,
|
224 |
+
"step": 360
|
225 |
+
},
|
226 |
+
{
|
227 |
+
"epoch": 6.07,
|
228 |
+
"learning_rate": 6.420504310167487e-05,
|
229 |
+
"loss": 2.2977,
|
230 |
+
"step": 370
|
231 |
+
},
|
232 |
+
{
|
233 |
+
"epoch": 6.23,
|
234 |
+
"learning_rate": 6.449458991542025e-05,
|
235 |
+
"loss": 2.2816,
|
236 |
+
"step": 380
|
237 |
+
},
|
238 |
+
{
|
239 |
+
"epoch": 6.39,
|
240 |
+
"learning_rate": 6.477661517566246e-05,
|
241 |
+
"loss": 2.2826,
|
242 |
+
"step": 390
|
243 |
+
},
|
244 |
+
{
|
245 |
+
"epoch": 6.56,
|
246 |
+
"learning_rate": 6.505149978319905e-05,
|
247 |
+
"loss": 2.3021,
|
248 |
+
"step": 400
|
249 |
+
},
|
250 |
+
{
|
251 |
+
"epoch": 6.72,
|
252 |
+
"learning_rate": 6.531959641799339e-05,
|
253 |
+
"loss": 2.3166,
|
254 |
+
"step": 410
|
255 |
+
},
|
256 |
+
{
|
257 |
+
"epoch": 6.89,
|
258 |
+
"learning_rate": 6.55812322599475e-05,
|
259 |
+
"loss": 2.3421,
|
260 |
+
"step": 420
|
261 |
+
},
|
262 |
+
{
|
263 |
+
"epoch": 7.05,
|
264 |
+
"learning_rate": 6.583671138948966e-05,
|
265 |
+
"loss": 2.3094,
|
266 |
+
"step": 430
|
267 |
+
},
|
268 |
+
{
|
269 |
+
"epoch": 7.21,
|
270 |
+
"learning_rate": 6.608631691215467e-05,
|
271 |
+
"loss": 2.2268,
|
272 |
+
"step": 440
|
273 |
+
},
|
274 |
+
{
|
275 |
+
"epoch": 7.38,
|
276 |
+
"learning_rate": 6.633031284438359e-05,
|
277 |
+
"loss": 2.2722,
|
278 |
+
"step": 450
|
279 |
+
},
|
280 |
+
{
|
281 |
+
"epoch": 7.54,
|
282 |
+
"learning_rate": 6.656894579203935e-05,
|
283 |
+
"loss": 2.1933,
|
284 |
+
"step": 460
|
285 |
+
},
|
286 |
+
{
|
287 |
+
"epoch": 7.7,
|
288 |
+
"learning_rate": 6.680244644839293e-05,
|
289 |
+
"loss": 2.1991,
|
290 |
+
"step": 470
|
291 |
+
},
|
292 |
+
{
|
293 |
+
"epoch": 7.87,
|
294 |
+
"learning_rate": 6.703103093438967e-05,
|
295 |
+
"loss": 2.2637,
|
296 |
+
"step": 480
|
297 |
+
},
|
298 |
+
{
|
299 |
+
"epoch": 8.03,
|
300 |
+
"learning_rate": 6.725490200071283e-05,
|
301 |
+
"loss": 2.2277,
|
302 |
+
"step": 490
|
303 |
+
},
|
304 |
+
{
|
305 |
+
"epoch": 8.2,
|
306 |
+
"learning_rate": 6.747425010840046e-05,
|
307 |
+
"loss": 2.1819,
|
308 |
+
"step": 500
|
309 |
+
},
|
310 |
+
{
|
311 |
+
"epoch": 8.2,
|
312 |
+
"eval_loss": 2.786494255065918,
|
313 |
+
"eval_runtime": 13.6759,
|
314 |
+
"eval_samples_per_second": 47.602,
|
315 |
+
"eval_steps_per_second": 0.585,
|
316 |
+
"step": 500
|
317 |
+
},
|
318 |
+
{
|
319 |
+
"epoch": 8.36,
|
320 |
+
"learning_rate": 6.76892544024484e-05,
|
321 |
+
"loss": 2.1427,
|
322 |
+
"step": 510
|
323 |
+
},
|
324 |
+
{
|
325 |
+
"epoch": 8.52,
|
326 |
+
"learning_rate": 6.790008359086997e-05,
|
327 |
+
"loss": 2.173,
|
328 |
+
"step": 520
|
329 |
+
},
|
330 |
+
{
|
331 |
+
"epoch": 8.69,
|
332 |
+
"learning_rate": 6.810689674001973e-05,
|
333 |
+
"loss": 2.1895,
|
334 |
+
"step": 530
|
335 |
+
},
|
336 |
+
{
|
337 |
+
"epoch": 8.85,
|
338 |
+
"learning_rate": 6.830984399557421e-05,
|
339 |
+
"loss": 2.2101,
|
340 |
+
"step": 540
|
341 |
+
},
|
342 |
+
{
|
343 |
+
"epoch": 9.02,
|
344 |
+
"learning_rate": 6.850906723735608e-05,
|
345 |
+
"loss": 2.1926,
|
346 |
+
"step": 550
|
347 |
+
},
|
348 |
+
{
|
349 |
+
"epoch": 9.18,
|
350 |
+
"learning_rate": 6.870470067515499e-05,
|
351 |
+
"loss": 2.0861,
|
352 |
+
"step": 560
|
353 |
+
},
|
354 |
+
{
|
355 |
+
"epoch": 9.34,
|
356 |
+
"learning_rate": 6.889687139181228e-05,
|
357 |
+
"loss": 2.1092,
|
358 |
+
"step": 570
|
359 |
+
},
|
360 |
+
{
|
361 |
+
"epoch": 9.51,
|
362 |
+
"learning_rate": 6.908569983907343e-05,
|
363 |
+
"loss": 2.129,
|
364 |
+
"step": 580
|
365 |
+
},
|
366 |
+
{
|
367 |
+
"epoch": 9.67,
|
368 |
+
"learning_rate": 6.92713002910536e-05,
|
369 |
+
"loss": 2.113,
|
370 |
+
"step": 590
|
371 |
+
},
|
372 |
+
{
|
373 |
+
"epoch": 9.84,
|
374 |
+
"learning_rate": 6.945378125959108e-05,
|
375 |
+
"loss": 2.1234,
|
376 |
+
"step": 600
|
377 |
+
},
|
378 |
+
{
|
379 |
+
"epoch": 10.0,
|
380 |
+
"learning_rate": 6.963324587526918e-05,
|
381 |
+
"loss": 2.1028,
|
382 |
+
"step": 610
|
383 |
+
},
|
384 |
+
{
|
385 |
+
"epoch": 10.16,
|
386 |
+
"learning_rate": 6.980979223745634e-05,
|
387 |
+
"loss": 2.0476,
|
388 |
+
"step": 620
|
389 |
+
},
|
390 |
+
{
|
391 |
+
"epoch": 10.33,
|
392 |
+
"learning_rate": 6.998351373633953e-05,
|
393 |
+
"loss": 2.0879,
|
394 |
+
"step": 630
|
395 |
+
},
|
396 |
+
{
|
397 |
+
"epoch": 10.49,
|
398 |
+
"learning_rate": 7.015449934959717e-05,
|
399 |
+
"loss": 2.0547,
|
400 |
+
"step": 640
|
401 |
+
},
|
402 |
+
{
|
403 |
+
"epoch": 10.66,
|
404 |
+
"learning_rate": 7.032283391607138e-05,
|
405 |
+
"loss": 2.0791,
|
406 |
+
"step": 650
|
407 |
+
},
|
408 |
+
{
|
409 |
+
"epoch": 10.82,
|
410 |
+
"learning_rate": 7.048859838854671e-05,
|
411 |
+
"loss": 2.1454,
|
412 |
+
"step": 660
|
413 |
+
},
|
414 |
+
{
|
415 |
+
"epoch": 10.98,
|
416 |
+
"learning_rate": 7.065187006752065e-05,
|
417 |
+
"loss": 2.0957,
|
418 |
+
"step": 670
|
419 |
+
},
|
420 |
+
{
|
421 |
+
"epoch": 11.15,
|
422 |
+
"learning_rate": 7.08127228176559e-05,
|
423 |
+
"loss": 2.0561,
|
424 |
+
"step": 680
|
425 |
+
},
|
426 |
+
{
|
427 |
+
"epoch": 11.31,
|
428 |
+
"learning_rate": 7.097122726843138e-05,
|
429 |
+
"loss": 2.0563,
|
430 |
+
"step": 690
|
431 |
+
},
|
432 |
+
{
|
433 |
+
"epoch": 11.48,
|
434 |
+
"learning_rate": 7.112745100035642e-05,
|
435 |
+
"loss": 2.027,
|
436 |
+
"step": 700
|
437 |
+
},
|
438 |
+
{
|
439 |
+
"epoch": 11.64,
|
440 |
+
"learning_rate": 7.128145871797688e-05,
|
441 |
+
"loss": 2.0495,
|
442 |
+
"step": 710
|
443 |
+
},
|
444 |
+
{
|
445 |
+
"epoch": 11.8,
|
446 |
+
"learning_rate": 7.143331241078171e-05,
|
447 |
+
"loss": 2.019,
|
448 |
+
"step": 720
|
449 |
+
},
|
450 |
+
{
|
451 |
+
"epoch": 11.97,
|
452 |
+
"learning_rate": 7.158307150301139e-05,
|
453 |
+
"loss": 2.0242,
|
454 |
+
"step": 730
|
455 |
+
},
|
456 |
+
{
|
457 |
+
"epoch": 12.13,
|
458 |
+
"learning_rate": 7.17307929932744e-05,
|
459 |
+
"loss": 1.9962,
|
460 |
+
"step": 740
|
461 |
+
},
|
462 |
+
{
|
463 |
+
"epoch": 12.3,
|
464 |
+
"learning_rate": 7.187653158479249e-05,
|
465 |
+
"loss": 1.9971,
|
466 |
+
"step": 750
|
467 |
+
},
|
468 |
+
{
|
469 |
+
"epoch": 12.46,
|
470 |
+
"learning_rate": 7.202033980701978e-05,
|
471 |
+
"loss": 2.0236,
|
472 |
+
"step": 760
|
473 |
+
},
|
474 |
+
{
|
475 |
+
"epoch": 12.62,
|
476 |
+
"learning_rate": 7.216226812931204e-05,
|
477 |
+
"loss": 1.9923,
|
478 |
+
"step": 770
|
479 |
+
},
|
480 |
+
{
|
481 |
+
"epoch": 12.79,
|
482 |
+
"learning_rate": 7.2302365067262e-05,
|
483 |
+
"loss": 2.0244,
|
484 |
+
"step": 780
|
485 |
+
},
|
486 |
+
{
|
487 |
+
"epoch": 12.95,
|
488 |
+
"learning_rate": 7.244067728226103e-05,
|
489 |
+
"loss": 1.9846,
|
490 |
+
"step": 790
|
491 |
+
},
|
492 |
+
{
|
493 |
+
"epoch": 13.11,
|
494 |
+
"learning_rate": 7.257724967479857e-05,
|
495 |
+
"loss": 1.9811,
|
496 |
+
"step": 800
|
497 |
+
},
|
498 |
+
{
|
499 |
+
"epoch": 13.28,
|
500 |
+
"learning_rate": 7.271212547196624e-05,
|
501 |
+
"loss": 1.9709,
|
502 |
+
"step": 810
|
503 |
+
},
|
504 |
+
{
|
505 |
+
"epoch": 13.44,
|
506 |
+
"learning_rate": 7.284534630959291e-05,
|
507 |
+
"loss": 1.9652,
|
508 |
+
"step": 820
|
509 |
+
},
|
510 |
+
{
|
511 |
+
"epoch": 13.61,
|
512 |
+
"learning_rate": 7.297695230940184e-05,
|
513 |
+
"loss": 1.9605,
|
514 |
+
"step": 830
|
515 |
+
},
|
516 |
+
{
|
517 |
+
"epoch": 13.77,
|
518 |
+
"learning_rate": 7.310698215154704e-05,
|
519 |
+
"loss": 1.9692,
|
520 |
+
"step": 840
|
521 |
+
},
|
522 |
+
{
|
523 |
+
"epoch": 13.93,
|
524 |
+
"learning_rate": 7.323547314285732e-05,
|
525 |
+
"loss": 1.9945,
|
526 |
+
"step": 850
|
527 |
+
},
|
528 |
+
{
|
529 |
+
"epoch": 14.1,
|
530 |
+
"learning_rate": 7.336246128108918e-05,
|
531 |
+
"loss": 1.9222,
|
532 |
+
"step": 860
|
533 |
+
},
|
534 |
+
{
|
535 |
+
"epoch": 14.26,
|
536 |
+
"learning_rate": 7.348798131546546e-05,
|
537 |
+
"loss": 1.9283,
|
538 |
+
"step": 870
|
539 |
+
},
|
540 |
+
{
|
541 |
+
"epoch": 14.43,
|
542 |
+
"learning_rate": 7.36120668037542e-05,
|
543 |
+
"loss": 1.9376,
|
544 |
+
"step": 880
|
545 |
+
},
|
546 |
+
{
|
547 |
+
"epoch": 14.59,
|
548 |
+
"learning_rate": 7.37347501661228e-05,
|
549 |
+
"loss": 1.9247,
|
550 |
+
"step": 890
|
551 |
+
},
|
552 |
+
{
|
553 |
+
"epoch": 14.75,
|
554 |
+
"learning_rate": 7.385606273598311e-05,
|
555 |
+
"loss": 1.9218,
|
556 |
+
"step": 900
|
557 |
+
},
|
558 |
+
{
|
559 |
+
"epoch": 14.92,
|
560 |
+
"learning_rate": 7.397603480802732e-05,
|
561 |
+
"loss": 1.9492,
|
562 |
+
"step": 910
|
563 |
+
},
|
564 |
+
{
|
565 |
+
"epoch": 15.08,
|
566 |
+
"learning_rate": 7.409469568363888e-05,
|
567 |
+
"loss": 1.9235,
|
568 |
+
"step": 920
|
569 |
+
},
|
570 |
+
{
|
571 |
+
"epoch": 15.25,
|
572 |
+
"learning_rate": 7.421207371384837e-05,
|
573 |
+
"loss": 1.8671,
|
574 |
+
"step": 930
|
575 |
+
},
|
576 |
+
{
|
577 |
+
"epoch": 15.41,
|
578 |
+
"learning_rate": 7.432819633999247e-05,
|
579 |
+
"loss": 1.909,
|
580 |
+
"step": 940
|
581 |
+
},
|
582 |
+
{
|
583 |
+
"epoch": 15.57,
|
584 |
+
"learning_rate": 7.444309013222118e-05,
|
585 |
+
"loss": 1.8568,
|
586 |
+
"step": 950
|
587 |
+
},
|
588 |
+
{
|
589 |
+
"epoch": 15.74,
|
590 |
+
"learning_rate": 7.45567808259892e-05,
|
591 |
+
"loss": 1.9199,
|
592 |
+
"step": 960
|
593 |
+
},
|
594 |
+
{
|
595 |
+
"epoch": 15.9,
|
596 |
+
"learning_rate": 7.46692933566561e-05,
|
597 |
+
"loss": 1.9247,
|
598 |
+
"step": 970
|
599 |
+
},
|
600 |
+
{
|
601 |
+
"epoch": 16.07,
|
602 |
+
"learning_rate": 7.478065189231236e-05,
|
603 |
+
"loss": 1.895,
|
604 |
+
"step": 980
|
605 |
+
},
|
606 |
+
{
|
607 |
+
"epoch": 16.23,
|
608 |
+
"learning_rate": 7.489087986493874e-05,
|
609 |
+
"loss": 1.8821,
|
610 |
+
"step": 990
|
611 |
+
},
|
612 |
+
{
|
613 |
+
"epoch": 16.39,
|
614 |
+
"learning_rate": 7.5e-05,
|
615 |
+
"loss": 1.8423,
|
616 |
+
"step": 1000
|
617 |
+
},
|
618 |
+
{
|
619 |
+
"epoch": 16.39,
|
620 |
+
"eval_loss": 2.925347328186035,
|
621 |
+
"eval_runtime": 13.9926,
|
622 |
+
"eval_samples_per_second": 46.524,
|
623 |
+
"eval_steps_per_second": 0.572,
|
624 |
+
"step": 1000
|
625 |
+
},
|
626 |
+
{
|
627 |
+
"epoch": 16.56,
|
628 |
+
"learning_rate": 7.510803434456605e-05,
|
629 |
+
"loss": 1.8519,
|
630 |
+
"step": 1010
|
631 |
+
},
|
632 |
+
{
|
633 |
+
"epoch": 16.72,
|
634 |
+
"learning_rate": 7.521500429404794e-05,
|
635 |
+
"loss": 1.8578,
|
636 |
+
"step": 1020
|
637 |
+
},
|
638 |
+
{
|
639 |
+
"epoch": 16.89,
|
640 |
+
"learning_rate": 7.532093061762931e-05,
|
641 |
+
"loss": 1.8676,
|
642 |
+
"step": 1030
|
643 |
+
},
|
644 |
+
{
|
645 |
+
"epoch": 17.05,
|
646 |
+
"learning_rate": 7.54258334824695e-05,
|
647 |
+
"loss": 1.8492,
|
648 |
+
"step": 1040
|
649 |
+
},
|
650 |
+
{
|
651 |
+
"epoch": 17.21,
|
652 |
+
"learning_rate": 7.552973247674843e-05,
|
653 |
+
"loss": 1.8542,
|
654 |
+
"step": 1050
|
655 |
+
},
|
656 |
+
{
|
657 |
+
"epoch": 17.38,
|
658 |
+
"learning_rate": 7.563264663161926e-05,
|
659 |
+
"loss": 1.8312,
|
660 |
+
"step": 1060
|
661 |
+
},
|
662 |
+
{
|
663 |
+
"epoch": 17.54,
|
664 |
+
"learning_rate": 7.573459444213023e-05,
|
665 |
+
"loss": 1.8554,
|
666 |
+
"step": 1070
|
667 |
+
},
|
668 |
+
{
|
669 |
+
"epoch": 17.7,
|
670 |
+
"learning_rate": 7.583559388717374e-05,
|
671 |
+
"loss": 1.8484,
|
672 |
+
"step": 1080
|
673 |
+
},
|
674 |
+
{
|
675 |
+
"epoch": 17.87,
|
676 |
+
"learning_rate": 7.593566244851558e-05,
|
677 |
+
"loss": 1.8485,
|
678 |
+
"step": 1090
|
679 |
+
},
|
680 |
+
{
|
681 |
+
"epoch": 18.03,
|
682 |
+
"learning_rate": 7.603481712895562e-05,
|
683 |
+
"loss": 1.8505,
|
684 |
+
"step": 1100
|
685 |
+
},
|
686 |
+
{
|
687 |
+
"epoch": 18.2,
|
688 |
+
"learning_rate": 7.613307446966643e-05,
|
689 |
+
"loss": 1.8163,
|
690 |
+
"step": 1110
|
691 |
+
},
|
692 |
+
{
|
693 |
+
"epoch": 18.36,
|
694 |
+
"learning_rate": 7.623045056675453e-05,
|
695 |
+
"loss": 1.8382,
|
696 |
+
"step": 1120
|
697 |
+
},
|
698 |
+
{
|
699 |
+
"epoch": 18.52,
|
700 |
+
"learning_rate": 7.632696108708549e-05,
|
701 |
+
"loss": 1.8251,
|
702 |
+
"step": 1130
|
703 |
+
},
|
704 |
+
{
|
705 |
+
"epoch": 18.69,
|
706 |
+
"learning_rate": 7.642262128341181e-05,
|
707 |
+
"loss": 1.8252,
|
708 |
+
"step": 1140
|
709 |
+
},
|
710 |
+
{
|
711 |
+
"epoch": 18.85,
|
712 |
+
"learning_rate": 7.651744600884029e-05,
|
713 |
+
"loss": 1.849,
|
714 |
+
"step": 1150
|
715 |
+
},
|
716 |
+
{
|
717 |
+
"epoch": 19.02,
|
718 |
+
"learning_rate": 7.661144973067295e-05,
|
719 |
+
"loss": 1.8202,
|
720 |
+
"step": 1160
|
721 |
+
},
|
722 |
+
{
|
723 |
+
"epoch": 19.18,
|
724 |
+
"learning_rate": 7.670464654365404e-05,
|
725 |
+
"loss": 1.8013,
|
726 |
+
"step": 1170
|
727 |
+
},
|
728 |
+
{
|
729 |
+
"epoch": 19.34,
|
730 |
+
"learning_rate": 7.679705018265312e-05,
|
731 |
+
"loss": 1.8149,
|
732 |
+
"step": 1180
|
733 |
+
},
|
734 |
+
{
|
735 |
+
"epoch": 19.51,
|
736 |
+
"learning_rate": 7.688867403481326e-05,
|
737 |
+
"loss": 1.7919,
|
738 |
+
"step": 1190
|
739 |
+
},
|
740 |
+
{
|
741 |
+
"epoch": 19.67,
|
742 |
+
"learning_rate": 7.697953115119061e-05,
|
743 |
+
"loss": 1.801,
|
744 |
+
"step": 1200
|
745 |
+
},
|
746 |
+
{
|
747 |
+
"epoch": 19.84,
|
748 |
+
"learning_rate": 7.706963425791124e-05,
|
749 |
+
"loss": 1.8286,
|
750 |
+
"step": 1210
|
751 |
+
},
|
752 |
+
{
|
753 |
+
"epoch": 20.0,
|
754 |
+
"learning_rate": 7.71589957668687e-05,
|
755 |
+
"loss": 1.7945,
|
756 |
+
"step": 1220
|
757 |
+
},
|
758 |
+
{
|
759 |
+
"epoch": 20.16,
|
760 |
+
"learning_rate": 7.724762778598493e-05,
|
761 |
+
"loss": 1.7619,
|
762 |
+
"step": 1230
|
763 |
+
},
|
764 |
+
{
|
765 |
+
"epoch": 20.33,
|
766 |
+
"learning_rate": 7.733554212905587e-05,
|
767 |
+
"loss": 1.7693,
|
768 |
+
"step": 1240
|
769 |
+
},
|
770 |
+
{
|
771 |
+
"epoch": 20.49,
|
772 |
+
"learning_rate": 7.74227503252014e-05,
|
773 |
+
"loss": 1.7689,
|
774 |
+
"step": 1250
|
775 |
+
},
|
776 |
+
{
|
777 |
+
"epoch": 20.66,
|
778 |
+
"learning_rate": 7.750926362793907e-05,
|
779 |
+
"loss": 1.77,
|
780 |
+
"step": 1260
|
781 |
+
},
|
782 |
+
{
|
783 |
+
"epoch": 20.82,
|
784 |
+
"learning_rate": 7.759509302389892e-05,
|
785 |
+
"loss": 1.7765,
|
786 |
+
"step": 1270
|
787 |
+
},
|
788 |
+
{
|
789 |
+
"epoch": 20.98,
|
790 |
+
"learning_rate": 7.768024924119671e-05,
|
791 |
+
"loss": 1.7791,
|
792 |
+
"step": 1280
|
793 |
+
},
|
794 |
+
{
|
795 |
+
"epoch": 21.15,
|
796 |
+
"learning_rate": 7.776474275748121e-05,
|
797 |
+
"loss": 1.7514,
|
798 |
+
"step": 1290
|
799 |
+
},
|
800 |
+
{
|
801 |
+
"epoch": 21.31,
|
802 |
+
"learning_rate": 7.784858380767091e-05,
|
803 |
+
"loss": 1.7564,
|
804 |
+
"step": 1300
|
805 |
+
},
|
806 |
+
{
|
807 |
+
"epoch": 21.48,
|
808 |
+
"learning_rate": 7.793178239139409e-05,
|
809 |
+
"loss": 1.7541,
|
810 |
+
"step": 1310
|
811 |
+
},
|
812 |
+
{
|
813 |
+
"epoch": 21.64,
|
814 |
+
"learning_rate": 7.801434828014625e-05,
|
815 |
+
"loss": 1.7519,
|
816 |
+
"step": 1320
|
817 |
+
},
|
818 |
+
{
|
819 |
+
"epoch": 21.8,
|
820 |
+
"learning_rate": 7.809629102417713e-05,
|
821 |
+
"loss": 1.7862,
|
822 |
+
"step": 1330
|
823 |
+
},
|
824 |
+
{
|
825 |
+
"epoch": 21.97,
|
826 |
+
"learning_rate": 7.817761995912018e-05,
|
827 |
+
"loss": 1.7724,
|
828 |
+
"step": 1340
|
829 |
+
},
|
830 |
+
{
|
831 |
+
"epoch": 22.13,
|
832 |
+
"learning_rate": 7.825834421237515e-05,
|
833 |
+
"loss": 1.7565,
|
834 |
+
"step": 1350
|
835 |
+
},
|
836 |
+
{
|
837 |
+
"epoch": 22.3,
|
838 |
+
"learning_rate": 7.833847270925543e-05,
|
839 |
+
"loss": 1.7346,
|
840 |
+
"step": 1360
|
841 |
+
},
|
842 |
+
{
|
843 |
+
"epoch": 22.46,
|
844 |
+
"learning_rate": 7.841801417891016e-05,
|
845 |
+
"loss": 1.7238,
|
846 |
+
"step": 1370
|
847 |
+
},
|
848 |
+
{
|
849 |
+
"epoch": 22.62,
|
850 |
+
"learning_rate": 7.84969771600309e-05,
|
851 |
+
"loss": 1.738,
|
852 |
+
"step": 1380
|
853 |
+
},
|
854 |
+
{
|
855 |
+
"epoch": 22.79,
|
856 |
+
"learning_rate": 7.857537000635237e-05,
|
857 |
+
"loss": 1.7446,
|
858 |
+
"step": 1390
|
859 |
+
},
|
860 |
+
{
|
861 |
+
"epoch": 22.95,
|
862 |
+
"learning_rate": 7.865320089195594e-05,
|
863 |
+
"loss": 1.7395,
|
864 |
+
"step": 1400
|
865 |
+
},
|
866 |
+
{
|
867 |
+
"epoch": 23.11,
|
868 |
+
"learning_rate": 7.87304778163845e-05,
|
869 |
+
"loss": 1.7533,
|
870 |
+
"step": 1410
|
871 |
+
},
|
872 |
+
{
|
873 |
+
"epoch": 23.28,
|
874 |
+
"learning_rate": 7.880720860957641e-05,
|
875 |
+
"loss": 1.7101,
|
876 |
+
"step": 1420
|
877 |
+
},
|
878 |
+
{
|
879 |
+
"epoch": 23.44,
|
880 |
+
"learning_rate": 7.888340093662653e-05,
|
881 |
+
"loss": 1.7145,
|
882 |
+
"step": 1430
|
883 |
+
},
|
884 |
+
{
|
885 |
+
"epoch": 23.61,
|
886 |
+
"learning_rate": 7.895906230238123e-05,
|
887 |
+
"loss": 1.7496,
|
888 |
+
"step": 1440
|
889 |
+
},
|
890 |
+
{
|
891 |
+
"epoch": 23.77,
|
892 |
+
"learning_rate": 7.903420005587436e-05,
|
893 |
+
"loss": 1.7416,
|
894 |
+
"step": 1450
|
895 |
+
},
|
896 |
+
{
|
897 |
+
"epoch": 23.93,
|
898 |
+
"learning_rate": 7.910882139461093e-05,
|
899 |
+
"loss": 1.7315,
|
900 |
+
"step": 1460
|
901 |
+
},
|
902 |
+
{
|
903 |
+
"epoch": 24.1,
|
904 |
+
"learning_rate": 7.918293336870439e-05,
|
905 |
+
"loss": 1.7224,
|
906 |
+
"step": 1470
|
907 |
+
},
|
908 |
+
{
|
909 |
+
"epoch": 24.26,
|
910 |
+
"learning_rate": 7.925654288487392e-05,
|
911 |
+
"loss": 1.716,
|
912 |
+
"step": 1480
|
913 |
+
},
|
914 |
+
{
|
915 |
+
"epoch": 24.43,
|
916 |
+
"learning_rate": 7.932965671030685e-05,
|
917 |
+
"loss": 1.704,
|
918 |
+
"step": 1490
|
919 |
+
},
|
920 |
+
{
|
921 |
+
"epoch": 24.59,
|
922 |
+
"learning_rate": 7.940228147639202e-05,
|
923 |
+
"loss": 1.6873,
|
924 |
+
"step": 1500
|
925 |
+
},
|
926 |
+
{
|
927 |
+
"epoch": 24.59,
|
928 |
+
"eval_loss": 3.106438159942627,
|
929 |
+
"eval_runtime": 13.9459,
|
930 |
+
"eval_samples_per_second": 46.681,
|
931 |
+
"eval_steps_per_second": 0.574,
|
932 |
+
"step": 1500
|
933 |
+
},
|
934 |
+
{
|
935 |
+
"epoch": 24.75,
|
936 |
+
"learning_rate": 7.947442368232923e-05,
|
937 |
+
"loss": 1.7098,
|
938 |
+
"step": 1510
|
939 |
+
},
|
940 |
+
{
|
941 |
+
"epoch": 24.92,
|
942 |
+
"learning_rate": 7.954608969861931e-05,
|
943 |
+
"loss": 1.7058,
|
944 |
+
"step": 1520
|
945 |
+
},
|
946 |
+
{
|
947 |
+
"epoch": 25.08,
|
948 |
+
"learning_rate": 7.961728577043997e-05,
|
949 |
+
"loss": 1.7189,
|
950 |
+
"step": 1530
|
951 |
+
},
|
952 |
+
{
|
953 |
+
"epoch": 25.25,
|
954 |
+
"learning_rate": 7.968801802091157e-05,
|
955 |
+
"loss": 1.6689,
|
956 |
+
"step": 1540
|
957 |
+
},
|
958 |
+
{
|
959 |
+
"epoch": 25.41,
|
960 |
+
"learning_rate": 7.975829245425728e-05,
|
961 |
+
"loss": 1.709,
|
962 |
+
"step": 1550
|
963 |
+
},
|
964 |
+
{
|
965 |
+
"epoch": 25.57,
|
966 |
+
"learning_rate": 7.982811495886153e-05,
|
967 |
+
"loss": 1.6881,
|
968 |
+
"step": 1560
|
969 |
+
},
|
970 |
+
{
|
971 |
+
"epoch": 25.74,
|
972 |
+
"learning_rate": 7.989749131023083e-05,
|
973 |
+
"loss": 1.7032,
|
974 |
+
"step": 1570
|
975 |
+
},
|
976 |
+
{
|
977 |
+
"epoch": 25.9,
|
978 |
+
"learning_rate": 7.996642717386056e-05,
|
979 |
+
"loss": 1.6887,
|
980 |
+
"step": 1580
|
981 |
+
},
|
982 |
+
{
|
983 |
+
"epoch": 26.07,
|
984 |
+
"learning_rate": 8.003492810801127e-05,
|
985 |
+
"loss": 1.6961,
|
986 |
+
"step": 1590
|
987 |
+
},
|
988 |
+
{
|
989 |
+
"epoch": 26.23,
|
990 |
+
"learning_rate": 8.01029995663981e-05,
|
991 |
+
"loss": 1.6701,
|
992 |
+
"step": 1600
|
993 |
+
},
|
994 |
+
{
|
995 |
+
"epoch": 26.39,
|
996 |
+
"learning_rate": 8.017064690079624e-05,
|
997 |
+
"loss": 1.69,
|
998 |
+
"step": 1610
|
999 |
+
},
|
1000 |
+
{
|
1001 |
+
"epoch": 26.56,
|
1002 |
+
"learning_rate": 8.023787536356576e-05,
|
1003 |
+
"loss": 1.7125,
|
1004 |
+
"step": 1620
|
1005 |
+
},
|
1006 |
+
{
|
1007 |
+
"epoch": 26.72,
|
1008 |
+
"learning_rate": 8.030469011009893e-05,
|
1009 |
+
"loss": 1.6606,
|
1010 |
+
"step": 1630
|
1011 |
+
},
|
1012 |
+
{
|
1013 |
+
"epoch": 26.89,
|
1014 |
+
"learning_rate": 8.037109620119243e-05,
|
1015 |
+
"loss": 1.6649,
|
1016 |
+
"step": 1640
|
1017 |
+
},
|
1018 |
+
{
|
1019 |
+
"epoch": 27.05,
|
1020 |
+
"learning_rate": 8.043709860534764e-05,
|
1021 |
+
"loss": 1.6699,
|
1022 |
+
"step": 1650
|
1023 |
+
},
|
1024 |
+
{
|
1025 |
+
"epoch": 27.21,
|
1026 |
+
"learning_rate": 8.050270220100136e-05,
|
1027 |
+
"loss": 1.645,
|
1028 |
+
"step": 1660
|
1029 |
+
},
|
1030 |
+
{
|
1031 |
+
"epoch": 27.38,
|
1032 |
+
"learning_rate": 8.056791177868957e-05,
|
1033 |
+
"loss": 1.65,
|
1034 |
+
"step": 1670
|
1035 |
+
},
|
1036 |
+
{
|
1037 |
+
"epoch": 27.54,
|
1038 |
+
"learning_rate": 8.063273204314657e-05,
|
1039 |
+
"loss": 1.6552,
|
1040 |
+
"step": 1680
|
1041 |
+
},
|
1042 |
+
{
|
1043 |
+
"epoch": 27.7,
|
1044 |
+
"learning_rate": 8.069716761534183e-05,
|
1045 |
+
"loss": 1.6772,
|
1046 |
+
"step": 1690
|
1047 |
+
},
|
1048 |
+
{
|
1049 |
+
"epoch": 27.87,
|
1050 |
+
"learning_rate": 8.076122303445684e-05,
|
1051 |
+
"loss": 1.6664,
|
1052 |
+
"step": 1700
|
1053 |
+
},
|
1054 |
+
{
|
1055 |
+
"epoch": 28.03,
|
1056 |
+
"learning_rate": 8.082490275980384e-05,
|
1057 |
+
"loss": 1.6539,
|
1058 |
+
"step": 1710
|
1059 |
+
},
|
1060 |
+
{
|
1061 |
+
"epoch": 28.2,
|
1062 |
+
"learning_rate": 8.088821117268871e-05,
|
1063 |
+
"loss": 1.6616,
|
1064 |
+
"step": 1720
|
1065 |
+
},
|
1066 |
+
{
|
1067 |
+
"epoch": 28.36,
|
1068 |
+
"learning_rate": 8.095115257821987e-05,
|
1069 |
+
"loss": 1.6379,
|
1070 |
+
"step": 1730
|
1071 |
+
},
|
1072 |
+
{
|
1073 |
+
"epoch": 28.52,
|
1074 |
+
"learning_rate": 8.1013731207065e-05,
|
1075 |
+
"loss": 1.6536,
|
1076 |
+
"step": 1740
|
1077 |
+
},
|
1078 |
+
{
|
1079 |
+
"epoch": 28.69,
|
1080 |
+
"learning_rate": 8.107595121715735e-05,
|
1081 |
+
"loss": 1.6506,
|
1082 |
+
"step": 1750
|
1083 |
+
},
|
1084 |
+
{
|
1085 |
+
"epoch": 28.85,
|
1086 |
+
"learning_rate": 8.113781669535373e-05,
|
1087 |
+
"loss": 1.66,
|
1088 |
+
"step": 1760
|
1089 |
+
},
|
1090 |
+
{
|
1091 |
+
"epoch": 29.02,
|
1092 |
+
"learning_rate": 8.119933165904515e-05,
|
1093 |
+
"loss": 1.6548,
|
1094 |
+
"step": 1770
|
1095 |
+
},
|
1096 |
+
{
|
1097 |
+
"epoch": 29.18,
|
1098 |
+
"learning_rate": 8.126050005772234e-05,
|
1099 |
+
"loss": 1.6408,
|
1100 |
+
"step": 1780
|
1101 |
+
},
|
1102 |
+
{
|
1103 |
+
"epoch": 29.34,
|
1104 |
+
"learning_rate": 8.132132577449732e-05,
|
1105 |
+
"loss": 1.6533,
|
1106 |
+
"step": 1790
|
1107 |
+
},
|
1108 |
+
{
|
1109 |
+
"epoch": 29.51,
|
1110 |
+
"learning_rate": 8.138181262758264e-05,
|
1111 |
+
"loss": 1.6508,
|
1112 |
+
"step": 1800
|
1113 |
+
},
|
1114 |
+
{
|
1115 |
+
"epoch": 29.67,
|
1116 |
+
"learning_rate": 8.144196437172959e-05,
|
1117 |
+
"loss": 1.6302,
|
1118 |
+
"step": 1810
|
1119 |
+
},
|
1120 |
+
{
|
1121 |
+
"epoch": 29.84,
|
1122 |
+
"learning_rate": 8.150178469962686e-05,
|
1123 |
+
"loss": 1.6319,
|
1124 |
+
"step": 1820
|
1125 |
+
},
|
1126 |
+
{
|
1127 |
+
"epoch": 30.0,
|
1128 |
+
"learning_rate": 8.156127724326073e-05,
|
1129 |
+
"loss": 1.623,
|
1130 |
+
"step": 1830
|
1131 |
+
},
|
1132 |
+
{
|
1133 |
+
"epoch": 30.16,
|
1134 |
+
"learning_rate": 8.16204455752384e-05,
|
1135 |
+
"loss": 1.6299,
|
1136 |
+
"step": 1840
|
1137 |
+
},
|
1138 |
+
{
|
1139 |
+
"epoch": 30.33,
|
1140 |
+
"learning_rate": 8.167929321007533e-05,
|
1141 |
+
"loss": 1.6187,
|
1142 |
+
"step": 1850
|
1143 |
+
},
|
1144 |
+
{
|
1145 |
+
"epoch": 30.49,
|
1146 |
+
"learning_rate": 8.17378236054479e-05,
|
1147 |
+
"loss": 1.6138,
|
1148 |
+
"step": 1860
|
1149 |
+
},
|
1150 |
+
{
|
1151 |
+
"epoch": 30.66,
|
1152 |
+
"learning_rate": 8.179604016341247e-05,
|
1153 |
+
"loss": 1.6418,
|
1154 |
+
"step": 1870
|
1155 |
+
},
|
1156 |
+
{
|
1157 |
+
"epoch": 30.82,
|
1158 |
+
"learning_rate": 8.1853946231592e-05,
|
1159 |
+
"loss": 1.6433,
|
1160 |
+
"step": 1880
|
1161 |
+
},
|
1162 |
+
{
|
1163 |
+
"epoch": 30.98,
|
1164 |
+
"learning_rate": 8.19115451043311e-05,
|
1165 |
+
"loss": 1.6416,
|
1166 |
+
"step": 1890
|
1167 |
+
},
|
1168 |
+
{
|
1169 |
+
"epoch": 31.15,
|
1170 |
+
"learning_rate": 8.196884002382071e-05,
|
1171 |
+
"loss": 1.6244,
|
1172 |
+
"step": 1900
|
1173 |
+
},
|
1174 |
+
{
|
1175 |
+
"epoch": 31.31,
|
1176 |
+
"learning_rate": 8.202583418119318e-05,
|
1177 |
+
"loss": 1.6141,
|
1178 |
+
"step": 1910
|
1179 |
+
},
|
1180 |
+
{
|
1181 |
+
"epoch": 31.48,
|
1182 |
+
"learning_rate": 8.208253071758874e-05,
|
1183 |
+
"loss": 1.6033,
|
1184 |
+
"step": 1920
|
1185 |
+
},
|
1186 |
+
{
|
1187 |
+
"epoch": 31.64,
|
1188 |
+
"learning_rate": 8.213893272519434e-05,
|
1189 |
+
"loss": 1.6287,
|
1190 |
+
"step": 1930
|
1191 |
+
},
|
1192 |
+
{
|
1193 |
+
"epoch": 31.8,
|
1194 |
+
"learning_rate": 8.219504324825564e-05,
|
1195 |
+
"loss": 1.6268,
|
1196 |
+
"step": 1940
|
1197 |
+
},
|
1198 |
+
{
|
1199 |
+
"epoch": 31.97,
|
1200 |
+
"learning_rate": 8.225086528406294e-05,
|
1201 |
+
"loss": 1.6297,
|
1202 |
+
"step": 1950
|
1203 |
+
},
|
1204 |
+
{
|
1205 |
+
"epoch": 32.13,
|
1206 |
+
"learning_rate": 8.23064017839119e-05,
|
1207 |
+
"loss": 1.6012,
|
1208 |
+
"step": 1960
|
1209 |
+
},
|
1210 |
+
{
|
1211 |
+
"epoch": 32.3,
|
1212 |
+
"learning_rate": 8.236165565403982e-05,
|
1213 |
+
"loss": 1.6203,
|
1214 |
+
"step": 1970
|
1215 |
+
},
|
1216 |
+
{
|
1217 |
+
"epoch": 32.46,
|
1218 |
+
"learning_rate": 8.241662975653826e-05,
|
1219 |
+
"loss": 1.6107,
|
1220 |
+
"step": 1980
|
1221 |
+
},
|
1222 |
+
{
|
1223 |
+
"epoch": 32.62,
|
1224 |
+
"learning_rate": 8.247132691024267e-05,
|
1225 |
+
"loss": 1.6107,
|
1226 |
+
"step": 1990
|
1227 |
+
},
|
1228 |
+
{
|
1229 |
+
"epoch": 32.79,
|
1230 |
+
"learning_rate": 8.252574989159953e-05,
|
1231 |
+
"loss": 1.5886,
|
1232 |
+
"step": 2000
|
1233 |
+
},
|
1234 |
+
{
|
1235 |
+
"epoch": 32.79,
|
1236 |
+
"eval_loss": 3.1871249675750732,
|
1237 |
+
"eval_runtime": 13.1185,
|
1238 |
+
"eval_samples_per_second": 49.624,
|
1239 |
+
"eval_steps_per_second": 0.61,
|
1240 |
+
"step": 2000
|
1241 |
+
}
|
1242 |
+
],
|
1243 |
+
"max_steps": 100000,
|
1244 |
+
"num_train_epochs": 1640,
|
1245 |
+
"total_flos": 394015948800.0,
|
1246 |
+
"trial_name": null,
|
1247 |
+
"trial_params": null
|
1248 |
+
}
|
training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:429c0dc53ed2ff5e8b7c901bbcf2bb07691a9d6ef79fc60cc81ec69a50afe4f7
|
3 |
+
size 4015
|
zero_to_fp32.py
ADDED
@@ -0,0 +1,348 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
|
3 |
+
# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
|
4 |
+
# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
|
5 |
+
# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
|
6 |
+
# application.
|
7 |
+
#
|
8 |
+
# example: python zero_to_fp32.py . pytorch_model.bin
|
9 |
+
|
10 |
+
import argparse
|
11 |
+
import torch
|
12 |
+
import glob
|
13 |
+
import os
|
14 |
+
from collections import OrderedDict
|
15 |
+
|
16 |
+
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
17 |
+
# DeepSpeed data structures it has to be available in the current python environment.
|
18 |
+
import deepspeed
|
19 |
+
from deepspeed.utils import logger
|
20 |
+
|
21 |
+
debug = 0
|
22 |
+
|
23 |
+
# load to cpu
|
24 |
+
device = torch.device('cpu')
|
25 |
+
|
26 |
+
|
27 |
+
def get_model_state_file(checkpoint_dir, zero_stage):
|
28 |
+
if not os.path.isdir(checkpoint_dir):
|
29 |
+
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
30 |
+
|
31 |
+
# there should be only one file
|
32 |
+
if zero_stage == 2:
|
33 |
+
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
34 |
+
elif zero_stage == 3:
|
35 |
+
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
36 |
+
|
37 |
+
if not os.path.exists(file):
|
38 |
+
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
39 |
+
|
40 |
+
return file
|
41 |
+
|
42 |
+
|
43 |
+
def get_optim_files(checkpoint_dir):
|
44 |
+
# XXX: need to test that this simple glob rule works for multi-node setup too
|
45 |
+
optim_files = sorted(glob.glob(os.path.join(checkpoint_dir, "*_optim_states.pt")))
|
46 |
+
|
47 |
+
if len(optim_files) == 0:
|
48 |
+
raise FileNotFoundError(
|
49 |
+
f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
|
50 |
+
|
51 |
+
return optim_files
|
52 |
+
|
53 |
+
|
54 |
+
def parse_model_state(file):
|
55 |
+
state_dict = torch.load(file, map_location=device)
|
56 |
+
|
57 |
+
if "buffer_names" not in state_dict:
|
58 |
+
raise ValueError(f"{file} is not a model state checkpoint")
|
59 |
+
buffer_names = state_dict["buffer_names"]
|
60 |
+
if debug:
|
61 |
+
print("Found buffers:", buffer_names)
|
62 |
+
|
63 |
+
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
64 |
+
buffers = {
|
65 |
+
k: v.float()
|
66 |
+
for k,
|
67 |
+
v in state_dict["module"].items() if k in buffer_names
|
68 |
+
}
|
69 |
+
return buffers
|
70 |
+
|
71 |
+
|
72 |
+
def parse_optim_states(files, ds_checkpoint_dir):
|
73 |
+
|
74 |
+
total_files = len(files)
|
75 |
+
state_dicts = []
|
76 |
+
for f in files:
|
77 |
+
state_dicts.append(torch.load(f, map_location=device))
|
78 |
+
|
79 |
+
if not "zero_stage" in state_dicts[0]['optimizer_state_dict']:
|
80 |
+
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
81 |
+
zero_stage = state_dicts[0]['optimizer_state_dict']["zero_stage"]
|
82 |
+
world_size = state_dicts[0]['optimizer_state_dict']["partition_count"]
|
83 |
+
param_shapes = state_dicts[0]["param_shapes"]
|
84 |
+
|
85 |
+
if world_size != total_files:
|
86 |
+
raise ValueError(
|
87 |
+
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
88 |
+
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
89 |
+
)
|
90 |
+
|
91 |
+
# the groups are named differently in each stage
|
92 |
+
if zero_stage == 2:
|
93 |
+
fp32_groups_key = "single_partition_of_fp32_groups"
|
94 |
+
elif zero_stage == 3:
|
95 |
+
fp32_groups_key = "fp32_flat_groups"
|
96 |
+
else:
|
97 |
+
raise ValueError(f"unknown zero stage {zero_stage}")
|
98 |
+
|
99 |
+
# if there is more than one param group, there will be multiple flattened tensors - one
|
100 |
+
# flattened tensor per group - for simplicity merge them into a single tensor
|
101 |
+
#
|
102 |
+
# XXX: could make the script more memory efficient for when there are multiple groups - it
|
103 |
+
# will require matching the sub-lists of param_shapes for each param group flattened tensor
|
104 |
+
fp32_flat_groups = [
|
105 |
+
torch.cat(state_dicts[i]['optimizer_state_dict'][fp32_groups_key],
|
106 |
+
0) for i in range(len(state_dicts))
|
107 |
+
]
|
108 |
+
|
109 |
+
return zero_stage, world_size, param_shapes, fp32_flat_groups
|
110 |
+
|
111 |
+
|
112 |
+
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
113 |
+
remainder = unpartitioned_numel % world_size
|
114 |
+
padding_numel = (world_size - remainder) if remainder else 0
|
115 |
+
partitioned_numel = int(unpartitioned_numel / world_size)
|
116 |
+
return partitioned_numel, padding_numel
|
117 |
+
|
118 |
+
|
119 |
+
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
|
120 |
+
"""
|
121 |
+
Returns fp32 state_dict reconstructed from ds checkpoint
|
122 |
+
|
123 |
+
Args:
|
124 |
+
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
125 |
+
|
126 |
+
"""
|
127 |
+
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
128 |
+
|
129 |
+
optim_files = get_optim_files(ds_checkpoint_dir)
|
130 |
+
zero_stage, world_size, param_shapes, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
131 |
+
print(
|
132 |
+
f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
133 |
+
|
134 |
+
model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
|
135 |
+
buffers = parse_model_state(model_file)
|
136 |
+
|
137 |
+
# Reconstruction protocol:
|
138 |
+
#
|
139 |
+
# - for zero2 we just need to concat the partitions back to back and reconsolidate over one huge
|
140 |
+
# flat buffer - no need to deal with padding since if there is any it will be only in the tail
|
141 |
+
# of the last partition so there it will be just left out
|
142 |
+
#
|
143 |
+
# - for zero3 we need to zip the partitions together at boundary of each param, re-consolidating
|
144 |
+
# each param, while dealing with padding if any
|
145 |
+
|
146 |
+
if debug:
|
147 |
+
for i in range(world_size):
|
148 |
+
print(f"fp32_flat_groups[i].shape={fp32_flat_groups[i].shape}")
|
149 |
+
|
150 |
+
if zero_stage == 2:
|
151 |
+
# XXX: memory usage doubles here (zero2)
|
152 |
+
full_single_fp32_vector = torch.cat(fp32_flat_groups, 0)
|
153 |
+
avail_numel = full_single_fp32_vector.numel()
|
154 |
+
elif zero_stage == 3:
|
155 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
156 |
+
|
157 |
+
if debug:
|
158 |
+
wanted_params = len(param_shapes)
|
159 |
+
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
160 |
+
# not asserting if there is a mismatch due to possible padding
|
161 |
+
print(f"Have {avail_numel} numels to process.")
|
162 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
163 |
+
|
164 |
+
state_dict = OrderedDict()
|
165 |
+
|
166 |
+
# buffers
|
167 |
+
state_dict.update(buffers)
|
168 |
+
if debug:
|
169 |
+
print(f"added {len(buffers)} buffers")
|
170 |
+
|
171 |
+
# params
|
172 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
173 |
+
# out-of-core computing solution
|
174 |
+
offset = 0
|
175 |
+
total_numel = 0
|
176 |
+
total_params = 0
|
177 |
+
for name, shape in param_shapes.items():
|
178 |
+
|
179 |
+
unpartitioned_numel = shape.numel()
|
180 |
+
total_numel += unpartitioned_numel
|
181 |
+
total_params += 1
|
182 |
+
|
183 |
+
if zero_stage == 2:
|
184 |
+
if debug:
|
185 |
+
print(
|
186 |
+
f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
|
187 |
+
)
|
188 |
+
state_dict[name] = full_single_fp32_vector.narrow(
|
189 |
+
0,
|
190 |
+
offset,
|
191 |
+
unpartitioned_numel).view(shape)
|
192 |
+
offset += unpartitioned_numel
|
193 |
+
|
194 |
+
elif zero_stage == 3:
|
195 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
196 |
+
|
197 |
+
if debug:
|
198 |
+
print(
|
199 |
+
f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
200 |
+
)
|
201 |
+
|
202 |
+
# XXX: memory usage doubles here (zero3)
|
203 |
+
state_dict[name] = torch.cat(
|
204 |
+
tuple(fp32_flat_groups[i].narrow(0,
|
205 |
+
offset,
|
206 |
+
partitioned_numel)
|
207 |
+
for i in range(world_size)),
|
208 |
+
0).view(shape)
|
209 |
+
offset += partitioned_numel + partitioned_padding_numel
|
210 |
+
|
211 |
+
if zero_stage == 3:
|
212 |
+
offset *= world_size
|
213 |
+
|
214 |
+
# Sanity check
|
215 |
+
if offset != avail_numel:
|
216 |
+
raise ValueError(
|
217 |
+
f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
218 |
+
|
219 |
+
print(
|
220 |
+
f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
|
221 |
+
)
|
222 |
+
|
223 |
+
return state_dict
|
224 |
+
|
225 |
+
|
226 |
+
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
|
227 |
+
"""
|
228 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
229 |
+
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
230 |
+
via a model hub.
|
231 |
+
|
232 |
+
Args:
|
233 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder
|
234 |
+
- ``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``
|
235 |
+
|
236 |
+
Returns:
|
237 |
+
- pytorch ``state_dict``
|
238 |
+
|
239 |
+
Note: this approach may not work if your application doesn't have sufficient free CPU memory and
|
240 |
+
you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
241 |
+
the checkpoint.
|
242 |
+
|
243 |
+
A typical usage might be ::
|
244 |
+
|
245 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
246 |
+
# do the training and checkpoint saving
|
247 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
248 |
+
model = model.cpu() # move to cpu
|
249 |
+
model.load_state_dict(state_dict)
|
250 |
+
# submit to model hub or save the model to share with others
|
251 |
+
|
252 |
+
In this example the ``model`` will no longer be useable in the deepspeed context of the same
|
253 |
+
application. i.e. you will need to re-initialize the deepspeed engine, since
|
254 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
255 |
+
|
256 |
+
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
257 |
+
|
258 |
+
"""
|
259 |
+
if tag is None:
|
260 |
+
latest_path = os.path.join(checkpoint_dir, 'latest')
|
261 |
+
if os.path.isfile(latest_path):
|
262 |
+
with open(latest_path, 'r') as fd:
|
263 |
+
tag = fd.read().strip()
|
264 |
+
else:
|
265 |
+
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
266 |
+
|
267 |
+
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
268 |
+
|
269 |
+
if not os.path.isdir(ds_checkpoint_dir):
|
270 |
+
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
271 |
+
|
272 |
+
return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
|
273 |
+
|
274 |
+
|
275 |
+
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
|
276 |
+
"""
|
277 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
278 |
+
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
279 |
+
|
280 |
+
Args:
|
281 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
282 |
+
- ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
|
283 |
+
- ``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``
|
284 |
+
"""
|
285 |
+
|
286 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
287 |
+
print(f"Saving fp32 state dict to {output_file}")
|
288 |
+
torch.save(state_dict, output_file)
|
289 |
+
|
290 |
+
|
291 |
+
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
292 |
+
"""
|
293 |
+
1. Put the provided model to cpu
|
294 |
+
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
295 |
+
3. Load it into the provided model
|
296 |
+
|
297 |
+
Args:
|
298 |
+
- ``model``: the model object to update
|
299 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
300 |
+
- ``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``
|
301 |
+
|
302 |
+
Returns:
|
303 |
+
- ``model`: modified model
|
304 |
+
|
305 |
+
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
306 |
+
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
307 |
+
conveniently placed for you in the checkpoint folder.
|
308 |
+
|
309 |
+
A typical usage might be ::
|
310 |
+
|
311 |
+
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
312 |
+
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
313 |
+
# submit to model hub or save the model to share with others
|
314 |
+
|
315 |
+
Note, that once this was run, the ``model`` will no longer be useable in the deepspeed context
|
316 |
+
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
317 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
318 |
+
|
319 |
+
"""
|
320 |
+
logger.info(f"Extracting fp32 weights")
|
321 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
322 |
+
|
323 |
+
logger.info(f"Overwriting model with fp32 weights")
|
324 |
+
model = model.cpu()
|
325 |
+
model.load_state_dict(state_dict, strict=False)
|
326 |
+
|
327 |
+
return model
|
328 |
+
|
329 |
+
|
330 |
+
if __name__ == "__main__":
|
331 |
+
|
332 |
+
parser = argparse.ArgumentParser()
|
333 |
+
parser.add_argument(
|
334 |
+
"checkpoint_dir",
|
335 |
+
type=str,
|
336 |
+
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
337 |
+
parser.add_argument(
|
338 |
+
"output_file",
|
339 |
+
type=str,
|
340 |
+
help=
|
341 |
+
"path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
|
342 |
+
)
|
343 |
+
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
344 |
+
args = parser.parse_args()
|
345 |
+
|
346 |
+
debug = args.debug
|
347 |
+
|
348 |
+
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)
|