theblackcat102
commited on
Commit
•
04aabb3
1
Parent(s):
f7da93f
upload models at 2500 iterations
Browse files- config.json +32 -0
- pytorch_model.bin +3 -0
- special_tokens_map.json +8 -0
- tokenizer.json +0 -0
- tokenizer_config.json +6 -0
- trainer_state.json +1696 -0
- training_args.bin +3 -0
- zero_to_fp32.py +482 -0
config.json
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "facebook/galactica-1.3b",
|
3 |
+
"_remove_final_layer_norm": false,
|
4 |
+
"activation_dropout": 0.0,
|
5 |
+
"activation_function": "gelu",
|
6 |
+
"architectures": [
|
7 |
+
"OPTForCausalLM"
|
8 |
+
],
|
9 |
+
"attention_dropout": 0.1,
|
10 |
+
"bos_token_id": 0,
|
11 |
+
"do_layer_norm_before": true,
|
12 |
+
"dropout": 0.1,
|
13 |
+
"enable_bias": true,
|
14 |
+
"eos_token_id": 2,
|
15 |
+
"ffn_dim": 8192,
|
16 |
+
"hidden_size": 2048,
|
17 |
+
"init_std": 0.02,
|
18 |
+
"layer_norm_elementwise_affine": true,
|
19 |
+
"layerdrop": 0.0,
|
20 |
+
"learned_embeddings": true,
|
21 |
+
"max_position_embeddings": 2048,
|
22 |
+
"model_type": "opt",
|
23 |
+
"num_attention_heads": 32,
|
24 |
+
"num_hidden_layers": 24,
|
25 |
+
"pad_token_id": 1,
|
26 |
+
"scale_embeddings": false,
|
27 |
+
"torch_dtype": "float16",
|
28 |
+
"transformers_version": "4.25.1",
|
29 |
+
"use_cache": true,
|
30 |
+
"vocab_size": 50002,
|
31 |
+
"word_embed_proj_dim": 2048
|
32 |
+
}
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:37f3def097356b3be26f8a392ccd4c39aa7ea999f705035efdf31dc8a728c486
|
3 |
+
size 2630533369
|
special_tokens_map.json
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"additional_special_tokens": [
|
3 |
+
"<question>",
|
4 |
+
"<answer>"
|
5 |
+
],
|
6 |
+
"eos_token": "</s>",
|
7 |
+
"pad_token": "<pad>"
|
8 |
+
}
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
tokenizer_config.json
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_max_length": 1000000000000000019884624838656,
|
3 |
+
"name_or_path": "facebook/galactica-1.3b",
|
4 |
+
"special_tokens_map_file": "/content/tokenizer/special_tokens_map.json",
|
5 |
+
"tokenizer_class": "PreTrainedTokenizerFast"
|
6 |
+
}
|
trainer_state.json
ADDED
@@ -0,0 +1,1696 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 2.6367949393779653,
|
5 |
+
"global_step": 2500,
|
6 |
+
"is_hyper_param_search": false,
|
7 |
+
"is_local_process_zero": true,
|
8 |
+
"is_world_process_zero": true,
|
9 |
+
"log_history": [
|
10 |
+
{
|
11 |
+
"epoch": 0.01,
|
12 |
+
"learning_rate": 3.599516044570672e-07,
|
13 |
+
"loss": 2.3879,
|
14 |
+
"step": 10
|
15 |
+
},
|
16 |
+
{
|
17 |
+
"epoch": 0.02,
|
18 |
+
"learning_rate": 4.6830783438602115e-07,
|
19 |
+
"loss": 2.3525,
|
20 |
+
"step": 20
|
21 |
+
},
|
22 |
+
{
|
23 |
+
"epoch": 0.03,
|
24 |
+
"learning_rate": 5.316921656139787e-07,
|
25 |
+
"loss": 2.2352,
|
26 |
+
"step": 30
|
27 |
+
},
|
28 |
+
{
|
29 |
+
"epoch": 0.04,
|
30 |
+
"learning_rate": 5.766640643149751e-07,
|
31 |
+
"loss": 2.2506,
|
32 |
+
"step": 40
|
33 |
+
},
|
34 |
+
{
|
35 |
+
"epoch": 0.05,
|
36 |
+
"learning_rate": 6.115469789851804e-07,
|
37 |
+
"loss": 2.182,
|
38 |
+
"step": 50
|
39 |
+
},
|
40 |
+
{
|
41 |
+
"epoch": 0.06,
|
42 |
+
"learning_rate": 6.400483955429327e-07,
|
43 |
+
"loss": 2.1697,
|
44 |
+
"step": 60
|
45 |
+
},
|
46 |
+
{
|
47 |
+
"epoch": 0.07,
|
48 |
+
"learning_rate": 6.641459998837218e-07,
|
49 |
+
"loss": 2.1285,
|
50 |
+
"step": 70
|
51 |
+
},
|
52 |
+
{
|
53 |
+
"epoch": 0.08,
|
54 |
+
"learning_rate": 6.850202942439291e-07,
|
55 |
+
"loss": 2.1158,
|
56 |
+
"step": 80
|
57 |
+
},
|
58 |
+
{
|
59 |
+
"epoch": 0.09,
|
60 |
+
"learning_rate": 7.034327267708902e-07,
|
61 |
+
"loss": 2.1371,
|
62 |
+
"step": 90
|
63 |
+
},
|
64 |
+
{
|
65 |
+
"epoch": 0.11,
|
66 |
+
"learning_rate": 7.199032089141344e-07,
|
67 |
+
"loss": 2.1271,
|
68 |
+
"step": 100
|
69 |
+
},
|
70 |
+
{
|
71 |
+
"epoch": 0.12,
|
72 |
+
"learning_rate": 7.348025723496237e-07,
|
73 |
+
"loss": 2.101,
|
74 |
+
"step": 110
|
75 |
+
},
|
76 |
+
{
|
77 |
+
"epoch": 0.13,
|
78 |
+
"learning_rate": 7.484046254718867e-07,
|
79 |
+
"loss": 2.1176,
|
80 |
+
"step": 120
|
81 |
+
},
|
82 |
+
{
|
83 |
+
"epoch": 0.14,
|
84 |
+
"learning_rate": 7.60917301394197e-07,
|
85 |
+
"loss": 2.0376,
|
86 |
+
"step": 130
|
87 |
+
},
|
88 |
+
{
|
89 |
+
"epoch": 0.15,
|
90 |
+
"learning_rate": 7.725022298126757e-07,
|
91 |
+
"loss": 2.1025,
|
92 |
+
"step": 140
|
93 |
+
},
|
94 |
+
{
|
95 |
+
"epoch": 0.16,
|
96 |
+
"learning_rate": 7.832875401420919e-07,
|
97 |
+
"loss": 2.1164,
|
98 |
+
"step": 150
|
99 |
+
},
|
100 |
+
{
|
101 |
+
"epoch": 0.17,
|
102 |
+
"learning_rate": 7.933765241728831e-07,
|
103 |
+
"loss": 2.0634,
|
104 |
+
"step": 160
|
105 |
+
},
|
106 |
+
{
|
107 |
+
"epoch": 0.18,
|
108 |
+
"learning_rate": 8.028536679096446e-07,
|
109 |
+
"loss": 2.0979,
|
110 |
+
"step": 170
|
111 |
+
},
|
112 |
+
{
|
113 |
+
"epoch": 0.19,
|
114 |
+
"learning_rate": 8.117889566998443e-07,
|
115 |
+
"loss": 2.0704,
|
116 |
+
"step": 180
|
117 |
+
},
|
118 |
+
{
|
119 |
+
"epoch": 0.2,
|
120 |
+
"learning_rate": 8.202410148252902e-07,
|
121 |
+
"loss": 2.0474,
|
122 |
+
"step": 190
|
123 |
+
},
|
124 |
+
{
|
125 |
+
"epoch": 0.21,
|
126 |
+
"learning_rate": 8.282594388430883e-07,
|
127 |
+
"loss": 2.0705,
|
128 |
+
"step": 200
|
129 |
+
},
|
130 |
+
{
|
131 |
+
"epoch": 0.22,
|
132 |
+
"learning_rate": 8.358865610406332e-07,
|
133 |
+
"loss": 2.0443,
|
134 |
+
"step": 210
|
135 |
+
},
|
136 |
+
{
|
137 |
+
"epoch": 0.23,
|
138 |
+
"learning_rate": 8.431588022785777e-07,
|
139 |
+
"loss": 2.1009,
|
140 |
+
"step": 220
|
141 |
+
},
|
142 |
+
{
|
143 |
+
"epoch": 0.24,
|
144 |
+
"learning_rate": 8.501077238654499e-07,
|
145 |
+
"loss": 2.0686,
|
146 |
+
"step": 230
|
147 |
+
},
|
148 |
+
{
|
149 |
+
"epoch": 0.25,
|
150 |
+
"learning_rate": 8.567608554008407e-07,
|
151 |
+
"loss": 2.0261,
|
152 |
+
"step": 240
|
153 |
+
},
|
154 |
+
{
|
155 |
+
"epoch": 0.26,
|
156 |
+
"learning_rate": 8.631423535132934e-07,
|
157 |
+
"loss": 2.0785,
|
158 |
+
"step": 250
|
159 |
+
},
|
160 |
+
{
|
161 |
+
"epoch": 0.26,
|
162 |
+
"eval_webgpt_accuracy": 0.5190179277190983,
|
163 |
+
"eval_webgpt_loss": 2.201171875,
|
164 |
+
"eval_webgpt_runtime": 429.5981,
|
165 |
+
"eval_webgpt_samples_per_second": 9.115,
|
166 |
+
"eval_webgpt_steps_per_second": 0.761,
|
167 |
+
"step": 250
|
168 |
+
},
|
169 |
+
{
|
170 |
+
"epoch": 0.26,
|
171 |
+
"eval_prompt_dialogue_accuracy": 0.5585842097604062,
|
172 |
+
"eval_prompt_dialogue_loss": 1.857421875,
|
173 |
+
"eval_prompt_dialogue_runtime": 1052.8834,
|
174 |
+
"eval_prompt_dialogue_samples_per_second": 9.791,
|
175 |
+
"eval_prompt_dialogue_steps_per_second": 0.817,
|
176 |
+
"step": 250
|
177 |
+
},
|
178 |
+
{
|
179 |
+
"epoch": 0.27,
|
180 |
+
"learning_rate": 8.692735313231511e-07,
|
181 |
+
"loss": 2.0753,
|
182 |
+
"step": 260
|
183 |
+
},
|
184 |
+
{
|
185 |
+
"epoch": 0.28,
|
186 |
+
"learning_rate": 8.751732879278018e-07,
|
187 |
+
"loss": 2.0926,
|
188 |
+
"step": 270
|
189 |
+
},
|
190 |
+
{
|
191 |
+
"epoch": 0.3,
|
192 |
+
"learning_rate": 8.808584597416296e-07,
|
193 |
+
"loss": 2.0557,
|
194 |
+
"step": 280
|
195 |
+
},
|
196 |
+
{
|
197 |
+
"epoch": 0.31,
|
198 |
+
"learning_rate": 8.863441101555992e-07,
|
199 |
+
"loss": 1.969,
|
200 |
+
"step": 290
|
201 |
+
},
|
202 |
+
{
|
203 |
+
"epoch": 0.32,
|
204 |
+
"learning_rate": 8.916437700710459e-07,
|
205 |
+
"loss": 2.049,
|
206 |
+
"step": 300
|
207 |
+
},
|
208 |
+
{
|
209 |
+
"epoch": 0.33,
|
210 |
+
"learning_rate": 8.967696389785231e-07,
|
211 |
+
"loss": 2.063,
|
212 |
+
"step": 310
|
213 |
+
},
|
214 |
+
{
|
215 |
+
"epoch": 0.34,
|
216 |
+
"learning_rate": 9.017327541018372e-07,
|
217 |
+
"loss": 2.0055,
|
218 |
+
"step": 320
|
219 |
+
},
|
220 |
+
{
|
221 |
+
"epoch": 0.35,
|
222 |
+
"learning_rate": 9.065431335065352e-07,
|
223 |
+
"loss": 2.0109,
|
224 |
+
"step": 330
|
225 |
+
},
|
226 |
+
{
|
227 |
+
"epoch": 0.36,
|
228 |
+
"learning_rate": 9.112098978385987e-07,
|
229 |
+
"loss": 2.0601,
|
230 |
+
"step": 340
|
231 |
+
},
|
232 |
+
{
|
233 |
+
"epoch": 0.37,
|
234 |
+
"learning_rate": 9.157413744118348e-07,
|
235 |
+
"loss": 2.0343,
|
236 |
+
"step": 350
|
237 |
+
},
|
238 |
+
{
|
239 |
+
"epoch": 0.38,
|
240 |
+
"learning_rate": 9.201451866287983e-07,
|
241 |
+
"loss": 2.0123,
|
242 |
+
"step": 360
|
243 |
+
},
|
244 |
+
{
|
245 |
+
"epoch": 0.39,
|
246 |
+
"learning_rate": 9.244283311473209e-07,
|
247 |
+
"loss": 2.0567,
|
248 |
+
"step": 370
|
249 |
+
},
|
250 |
+
{
|
251 |
+
"epoch": 0.4,
|
252 |
+
"learning_rate": 9.285972447542441e-07,
|
253 |
+
"loss": 2.0171,
|
254 |
+
"step": 380
|
255 |
+
},
|
256 |
+
{
|
257 |
+
"epoch": 0.41,
|
258 |
+
"learning_rate": 9.326578625511087e-07,
|
259 |
+
"loss": 2.0287,
|
260 |
+
"step": 390
|
261 |
+
},
|
262 |
+
{
|
263 |
+
"epoch": 0.42,
|
264 |
+
"learning_rate": 9.366156687720423e-07,
|
265 |
+
"loss": 2.0357,
|
266 |
+
"step": 400
|
267 |
+
},
|
268 |
+
{
|
269 |
+
"epoch": 0.43,
|
270 |
+
"learning_rate": 9.404757413257927e-07,
|
271 |
+
"loss": 2.0556,
|
272 |
+
"step": 410
|
273 |
+
},
|
274 |
+
{
|
275 |
+
"epoch": 0.44,
|
276 |
+
"learning_rate": 9.442427909695872e-07,
|
277 |
+
"loss": 2.0166,
|
278 |
+
"step": 420
|
279 |
+
},
|
280 |
+
{
|
281 |
+
"epoch": 0.45,
|
282 |
+
"learning_rate": 9.479211958729468e-07,
|
283 |
+
"loss": 2.028,
|
284 |
+
"step": 430
|
285 |
+
},
|
286 |
+
{
|
287 |
+
"epoch": 0.46,
|
288 |
+
"learning_rate": 9.515150322075316e-07,
|
289 |
+
"loss": 2.0044,
|
290 |
+
"step": 440
|
291 |
+
},
|
292 |
+
{
|
293 |
+
"epoch": 0.47,
|
294 |
+
"learning_rate": 9.550281012990034e-07,
|
295 |
+
"loss": 2.0301,
|
296 |
+
"step": 450
|
297 |
+
},
|
298 |
+
{
|
299 |
+
"epoch": 0.48,
|
300 |
+
"learning_rate": 9.584639537944038e-07,
|
301 |
+
"loss": 2.0321,
|
302 |
+
"step": 460
|
303 |
+
},
|
304 |
+
{
|
305 |
+
"epoch": 0.5,
|
306 |
+
"learning_rate": 9.61825911230254e-07,
|
307 |
+
"loss": 1.9889,
|
308 |
+
"step": 470
|
309 |
+
},
|
310 |
+
{
|
311 |
+
"epoch": 0.51,
|
312 |
+
"learning_rate": 9.651170853297947e-07,
|
313 |
+
"loss": 2.0272,
|
314 |
+
"step": 480
|
315 |
+
},
|
316 |
+
{
|
317 |
+
"epoch": 0.52,
|
318 |
+
"learning_rate": 9.683403953103762e-07,
|
319 |
+
"loss": 2.0319,
|
320 |
+
"step": 490
|
321 |
+
},
|
322 |
+
{
|
323 |
+
"epoch": 0.53,
|
324 |
+
"learning_rate": 9.714985834422475e-07,
|
325 |
+
"loss": 2.0251,
|
326 |
+
"step": 500
|
327 |
+
},
|
328 |
+
{
|
329 |
+
"epoch": 0.53,
|
330 |
+
"eval_webgpt_accuracy": 0.522718880616189,
|
331 |
+
"eval_webgpt_loss": 2.181640625,
|
332 |
+
"eval_webgpt_runtime": 433.6644,
|
333 |
+
"eval_webgpt_samples_per_second": 9.03,
|
334 |
+
"eval_webgpt_steps_per_second": 0.754,
|
335 |
+
"step": 500
|
336 |
+
},
|
337 |
+
{
|
338 |
+
"epoch": 0.53,
|
339 |
+
"eval_prompt_dialogue_accuracy": 0.5654158849986294,
|
340 |
+
"eval_prompt_dialogue_loss": 1.810546875,
|
341 |
+
"eval_prompt_dialogue_runtime": 1051.7495,
|
342 |
+
"eval_prompt_dialogue_samples_per_second": 9.802,
|
343 |
+
"eval_prompt_dialogue_steps_per_second": 0.818,
|
344 |
+
"step": 500
|
345 |
+
},
|
346 |
+
{
|
347 |
+
"epoch": 0.54,
|
348 |
+
"learning_rate": 9.745942290665561e-07,
|
349 |
+
"loss": 2.0089,
|
350 |
+
"step": 510
|
351 |
+
},
|
352 |
+
{
|
353 |
+
"epoch": 0.55,
|
354 |
+
"learning_rate": 9.776297612521052e-07,
|
355 |
+
"loss": 2.0385,
|
356 |
+
"step": 520
|
357 |
+
},
|
358 |
+
{
|
359 |
+
"epoch": 0.56,
|
360 |
+
"learning_rate": 9.80607470246476e-07,
|
361 |
+
"loss": 1.9865,
|
362 |
+
"step": 530
|
363 |
+
},
|
364 |
+
{
|
365 |
+
"epoch": 0.57,
|
366 |
+
"learning_rate": 9.835295178567558e-07,
|
367 |
+
"loss": 1.9859,
|
368 |
+
"step": 540
|
369 |
+
},
|
370 |
+
{
|
371 |
+
"epoch": 0.58,
|
372 |
+
"learning_rate": 9.863979468777368e-07,
|
373 |
+
"loss": 2.0115,
|
374 |
+
"step": 550
|
375 |
+
},
|
376 |
+
{
|
377 |
+
"epoch": 0.59,
|
378 |
+
"learning_rate": 9.892146896705837e-07,
|
379 |
+
"loss": 2.004,
|
380 |
+
"step": 560
|
381 |
+
},
|
382 |
+
{
|
383 |
+
"epoch": 0.6,
|
384 |
+
"learning_rate": 9.919815759822018e-07,
|
385 |
+
"loss": 2.0179,
|
386 |
+
"step": 570
|
387 |
+
},
|
388 |
+
{
|
389 |
+
"epoch": 0.61,
|
390 |
+
"learning_rate": 9.94700340084553e-07,
|
391 |
+
"loss": 2.0083,
|
392 |
+
"step": 580
|
393 |
+
},
|
394 |
+
{
|
395 |
+
"epoch": 0.62,
|
396 |
+
"learning_rate": 9.97372627303682e-07,
|
397 |
+
"loss": 2.0227,
|
398 |
+
"step": 590
|
399 |
+
},
|
400 |
+
{
|
401 |
+
"epoch": 0.63,
|
402 |
+
"learning_rate": 9.999999999999997e-07,
|
403 |
+
"loss": 1.9993,
|
404 |
+
"step": 600
|
405 |
+
},
|
406 |
+
{
|
407 |
+
"epoch": 0.64,
|
408 |
+
"learning_rate": 9.959893048128342e-07,
|
409 |
+
"loss": 2.0316,
|
410 |
+
"step": 610
|
411 |
+
},
|
412 |
+
{
|
413 |
+
"epoch": 0.65,
|
414 |
+
"learning_rate": 9.915329768270946e-07,
|
415 |
+
"loss": 1.9877,
|
416 |
+
"step": 620
|
417 |
+
},
|
418 |
+
{
|
419 |
+
"epoch": 0.66,
|
420 |
+
"learning_rate": 9.870766488413547e-07,
|
421 |
+
"loss": 1.9862,
|
422 |
+
"step": 630
|
423 |
+
},
|
424 |
+
{
|
425 |
+
"epoch": 0.67,
|
426 |
+
"learning_rate": 9.82620320855615e-07,
|
427 |
+
"loss": 2.0064,
|
428 |
+
"step": 640
|
429 |
+
},
|
430 |
+
{
|
431 |
+
"epoch": 0.69,
|
432 |
+
"learning_rate": 9.781639928698751e-07,
|
433 |
+
"loss": 2.0837,
|
434 |
+
"step": 650
|
435 |
+
},
|
436 |
+
{
|
437 |
+
"epoch": 0.7,
|
438 |
+
"learning_rate": 9.737076648841355e-07,
|
439 |
+
"loss": 1.9865,
|
440 |
+
"step": 660
|
441 |
+
},
|
442 |
+
{
|
443 |
+
"epoch": 0.71,
|
444 |
+
"learning_rate": 9.692513368983956e-07,
|
445 |
+
"loss": 2.0137,
|
446 |
+
"step": 670
|
447 |
+
},
|
448 |
+
{
|
449 |
+
"epoch": 0.72,
|
450 |
+
"learning_rate": 9.64795008912656e-07,
|
451 |
+
"loss": 2.0423,
|
452 |
+
"step": 680
|
453 |
+
},
|
454 |
+
{
|
455 |
+
"epoch": 0.73,
|
456 |
+
"learning_rate": 9.603386809269163e-07,
|
457 |
+
"loss": 1.992,
|
458 |
+
"step": 690
|
459 |
+
},
|
460 |
+
{
|
461 |
+
"epoch": 0.74,
|
462 |
+
"learning_rate": 9.558823529411764e-07,
|
463 |
+
"loss": 1.9865,
|
464 |
+
"step": 700
|
465 |
+
},
|
466 |
+
{
|
467 |
+
"epoch": 0.75,
|
468 |
+
"learning_rate": 9.514260249554367e-07,
|
469 |
+
"loss": 2.0152,
|
470 |
+
"step": 710
|
471 |
+
},
|
472 |
+
{
|
473 |
+
"epoch": 0.76,
|
474 |
+
"learning_rate": 9.46969696969697e-07,
|
475 |
+
"loss": 2.0183,
|
476 |
+
"step": 720
|
477 |
+
},
|
478 |
+
{
|
479 |
+
"epoch": 0.77,
|
480 |
+
"learning_rate": 9.425133689839572e-07,
|
481 |
+
"loss": 2.0077,
|
482 |
+
"step": 730
|
483 |
+
},
|
484 |
+
{
|
485 |
+
"epoch": 0.78,
|
486 |
+
"learning_rate": 9.380570409982175e-07,
|
487 |
+
"loss": 1.9948,
|
488 |
+
"step": 740
|
489 |
+
},
|
490 |
+
{
|
491 |
+
"epoch": 0.79,
|
492 |
+
"learning_rate": 9.336007130124776e-07,
|
493 |
+
"loss": 2.0004,
|
494 |
+
"step": 750
|
495 |
+
},
|
496 |
+
{
|
497 |
+
"epoch": 0.79,
|
498 |
+
"eval_webgpt_accuracy": 0.5246072651558654,
|
499 |
+
"eval_webgpt_loss": 2.171875,
|
500 |
+
"eval_webgpt_runtime": 427.6682,
|
501 |
+
"eval_webgpt_samples_per_second": 9.157,
|
502 |
+
"eval_webgpt_steps_per_second": 0.765,
|
503 |
+
"step": 750
|
504 |
+
},
|
505 |
+
{
|
506 |
+
"epoch": 0.79,
|
507 |
+
"eval_prompt_dialogue_accuracy": 0.5691780443314355,
|
508 |
+
"eval_prompt_dialogue_loss": 1.78515625,
|
509 |
+
"eval_prompt_dialogue_runtime": 1055.9887,
|
510 |
+
"eval_prompt_dialogue_samples_per_second": 9.762,
|
511 |
+
"eval_prompt_dialogue_steps_per_second": 0.814,
|
512 |
+
"step": 750
|
513 |
+
},
|
514 |
+
{
|
515 |
+
"epoch": 0.8,
|
516 |
+
"learning_rate": 9.29144385026738e-07,
|
517 |
+
"loss": 2.0092,
|
518 |
+
"step": 760
|
519 |
+
},
|
520 |
+
{
|
521 |
+
"epoch": 0.81,
|
522 |
+
"learning_rate": 9.246880570409982e-07,
|
523 |
+
"loss": 1.9648,
|
524 |
+
"step": 770
|
525 |
+
},
|
526 |
+
{
|
527 |
+
"epoch": 0.82,
|
528 |
+
"learning_rate": 9.202317290552585e-07,
|
529 |
+
"loss": 2.0095,
|
530 |
+
"step": 780
|
531 |
+
},
|
532 |
+
{
|
533 |
+
"epoch": 0.83,
|
534 |
+
"learning_rate": 9.157754010695186e-07,
|
535 |
+
"loss": 1.9805,
|
536 |
+
"step": 790
|
537 |
+
},
|
538 |
+
{
|
539 |
+
"epoch": 0.84,
|
540 |
+
"learning_rate": 9.11319073083779e-07,
|
541 |
+
"loss": 1.999,
|
542 |
+
"step": 800
|
543 |
+
},
|
544 |
+
{
|
545 |
+
"epoch": 0.85,
|
546 |
+
"learning_rate": 9.068627450980392e-07,
|
547 |
+
"loss": 2.0067,
|
548 |
+
"step": 810
|
549 |
+
},
|
550 |
+
{
|
551 |
+
"epoch": 0.86,
|
552 |
+
"learning_rate": 9.024064171122994e-07,
|
553 |
+
"loss": 1.9908,
|
554 |
+
"step": 820
|
555 |
+
},
|
556 |
+
{
|
557 |
+
"epoch": 0.88,
|
558 |
+
"learning_rate": 8.979500891265597e-07,
|
559 |
+
"loss": 1.9952,
|
560 |
+
"step": 830
|
561 |
+
},
|
562 |
+
{
|
563 |
+
"epoch": 0.89,
|
564 |
+
"learning_rate": 8.9349376114082e-07,
|
565 |
+
"loss": 1.9924,
|
566 |
+
"step": 840
|
567 |
+
},
|
568 |
+
{
|
569 |
+
"epoch": 0.9,
|
570 |
+
"learning_rate": 8.890374331550802e-07,
|
571 |
+
"loss": 2.022,
|
572 |
+
"step": 850
|
573 |
+
},
|
574 |
+
{
|
575 |
+
"epoch": 0.91,
|
576 |
+
"learning_rate": 8.845811051693403e-07,
|
577 |
+
"loss": 2.0003,
|
578 |
+
"step": 860
|
579 |
+
},
|
580 |
+
{
|
581 |
+
"epoch": 0.92,
|
582 |
+
"learning_rate": 8.801247771836007e-07,
|
583 |
+
"loss": 2.0315,
|
584 |
+
"step": 870
|
585 |
+
},
|
586 |
+
{
|
587 |
+
"epoch": 0.93,
|
588 |
+
"learning_rate": 8.756684491978609e-07,
|
589 |
+
"loss": 1.9881,
|
590 |
+
"step": 880
|
591 |
+
},
|
592 |
+
{
|
593 |
+
"epoch": 0.94,
|
594 |
+
"learning_rate": 8.712121212121211e-07,
|
595 |
+
"loss": 1.946,
|
596 |
+
"step": 890
|
597 |
+
},
|
598 |
+
{
|
599 |
+
"epoch": 0.95,
|
600 |
+
"learning_rate": 8.667557932263814e-07,
|
601 |
+
"loss": 2.0094,
|
602 |
+
"step": 900
|
603 |
+
},
|
604 |
+
{
|
605 |
+
"epoch": 0.96,
|
606 |
+
"learning_rate": 8.622994652406417e-07,
|
607 |
+
"loss": 1.9853,
|
608 |
+
"step": 910
|
609 |
+
},
|
610 |
+
{
|
611 |
+
"epoch": 0.97,
|
612 |
+
"learning_rate": 8.578431372549019e-07,
|
613 |
+
"loss": 1.9821,
|
614 |
+
"step": 920
|
615 |
+
},
|
616 |
+
{
|
617 |
+
"epoch": 0.98,
|
618 |
+
"learning_rate": 8.533868092691621e-07,
|
619 |
+
"loss": 1.9795,
|
620 |
+
"step": 930
|
621 |
+
},
|
622 |
+
{
|
623 |
+
"epoch": 0.99,
|
624 |
+
"learning_rate": 8.489304812834224e-07,
|
625 |
+
"loss": 1.9971,
|
626 |
+
"step": 940
|
627 |
+
},
|
628 |
+
{
|
629 |
+
"epoch": 1.0,
|
630 |
+
"learning_rate": 8.444741532976827e-07,
|
631 |
+
"loss": 2.0708,
|
632 |
+
"step": 950
|
633 |
+
},
|
634 |
+
{
|
635 |
+
"epoch": 1.01,
|
636 |
+
"learning_rate": 8.400178253119428e-07,
|
637 |
+
"loss": 1.9623,
|
638 |
+
"step": 960
|
639 |
+
},
|
640 |
+
{
|
641 |
+
"epoch": 1.02,
|
642 |
+
"learning_rate": 8.355614973262032e-07,
|
643 |
+
"loss": 1.9546,
|
644 |
+
"step": 970
|
645 |
+
},
|
646 |
+
{
|
647 |
+
"epoch": 1.03,
|
648 |
+
"learning_rate": 8.311051693404634e-07,
|
649 |
+
"loss": 1.9878,
|
650 |
+
"step": 980
|
651 |
+
},
|
652 |
+
{
|
653 |
+
"epoch": 1.04,
|
654 |
+
"learning_rate": 8.266488413547237e-07,
|
655 |
+
"loss": 1.9843,
|
656 |
+
"step": 990
|
657 |
+
},
|
658 |
+
{
|
659 |
+
"epoch": 1.05,
|
660 |
+
"learning_rate": 8.221925133689838e-07,
|
661 |
+
"loss": 1.9807,
|
662 |
+
"step": 1000
|
663 |
+
},
|
664 |
+
{
|
665 |
+
"epoch": 1.05,
|
666 |
+
"eval_webgpt_accuracy": 0.5253906990392405,
|
667 |
+
"eval_webgpt_loss": 2.16796875,
|
668 |
+
"eval_webgpt_runtime": 428.1397,
|
669 |
+
"eval_webgpt_samples_per_second": 9.147,
|
670 |
+
"eval_webgpt_steps_per_second": 0.764,
|
671 |
+
"step": 1000
|
672 |
+
},
|
673 |
+
{
|
674 |
+
"epoch": 1.05,
|
675 |
+
"eval_prompt_dialogue_accuracy": 0.5718087356550049,
|
676 |
+
"eval_prompt_dialogue_loss": 1.7685546875,
|
677 |
+
"eval_prompt_dialogue_runtime": 1050.2103,
|
678 |
+
"eval_prompt_dialogue_samples_per_second": 9.816,
|
679 |
+
"eval_prompt_dialogue_steps_per_second": 0.819,
|
680 |
+
"step": 1000
|
681 |
+
},
|
682 |
+
{
|
683 |
+
"epoch": 1.07,
|
684 |
+
"learning_rate": 8.177361853832442e-07,
|
685 |
+
"loss": 1.9773,
|
686 |
+
"step": 1010
|
687 |
+
},
|
688 |
+
{
|
689 |
+
"epoch": 1.08,
|
690 |
+
"learning_rate": 8.132798573975044e-07,
|
691 |
+
"loss": 2.0443,
|
692 |
+
"step": 1020
|
693 |
+
},
|
694 |
+
{
|
695 |
+
"epoch": 1.09,
|
696 |
+
"learning_rate": 8.088235294117646e-07,
|
697 |
+
"loss": 1.9528,
|
698 |
+
"step": 1030
|
699 |
+
},
|
700 |
+
{
|
701 |
+
"epoch": 1.1,
|
702 |
+
"learning_rate": 8.043672014260249e-07,
|
703 |
+
"loss": 1.9365,
|
704 |
+
"step": 1040
|
705 |
+
},
|
706 |
+
{
|
707 |
+
"epoch": 1.11,
|
708 |
+
"learning_rate": 7.999108734402852e-07,
|
709 |
+
"loss": 1.9579,
|
710 |
+
"step": 1050
|
711 |
+
},
|
712 |
+
{
|
713 |
+
"epoch": 1.12,
|
714 |
+
"learning_rate": 7.954545454545454e-07,
|
715 |
+
"loss": 1.974,
|
716 |
+
"step": 1060
|
717 |
+
},
|
718 |
+
{
|
719 |
+
"epoch": 1.13,
|
720 |
+
"learning_rate": 7.909982174688057e-07,
|
721 |
+
"loss": 1.9537,
|
722 |
+
"step": 1070
|
723 |
+
},
|
724 |
+
{
|
725 |
+
"epoch": 1.14,
|
726 |
+
"learning_rate": 7.865418894830659e-07,
|
727 |
+
"loss": 1.9373,
|
728 |
+
"step": 1080
|
729 |
+
},
|
730 |
+
{
|
731 |
+
"epoch": 1.15,
|
732 |
+
"learning_rate": 7.820855614973262e-07,
|
733 |
+
"loss": 1.9555,
|
734 |
+
"step": 1090
|
735 |
+
},
|
736 |
+
{
|
737 |
+
"epoch": 1.16,
|
738 |
+
"learning_rate": 7.776292335115863e-07,
|
739 |
+
"loss": 1.973,
|
740 |
+
"step": 1100
|
741 |
+
},
|
742 |
+
{
|
743 |
+
"epoch": 1.17,
|
744 |
+
"learning_rate": 7.731729055258467e-07,
|
745 |
+
"loss": 1.9401,
|
746 |
+
"step": 1110
|
747 |
+
},
|
748 |
+
{
|
749 |
+
"epoch": 1.18,
|
750 |
+
"learning_rate": 7.687165775401069e-07,
|
751 |
+
"loss": 1.9508,
|
752 |
+
"step": 1120
|
753 |
+
},
|
754 |
+
{
|
755 |
+
"epoch": 1.19,
|
756 |
+
"learning_rate": 7.642602495543672e-07,
|
757 |
+
"loss": 1.934,
|
758 |
+
"step": 1130
|
759 |
+
},
|
760 |
+
{
|
761 |
+
"epoch": 1.2,
|
762 |
+
"learning_rate": 7.598039215686273e-07,
|
763 |
+
"loss": 1.9325,
|
764 |
+
"step": 1140
|
765 |
+
},
|
766 |
+
{
|
767 |
+
"epoch": 1.21,
|
768 |
+
"learning_rate": 7.553475935828877e-07,
|
769 |
+
"loss": 1.9914,
|
770 |
+
"step": 1150
|
771 |
+
},
|
772 |
+
{
|
773 |
+
"epoch": 1.22,
|
774 |
+
"learning_rate": 7.508912655971479e-07,
|
775 |
+
"loss": 1.978,
|
776 |
+
"step": 1160
|
777 |
+
},
|
778 |
+
{
|
779 |
+
"epoch": 1.23,
|
780 |
+
"learning_rate": 7.464349376114081e-07,
|
781 |
+
"loss": 1.9442,
|
782 |
+
"step": 1170
|
783 |
+
},
|
784 |
+
{
|
785 |
+
"epoch": 1.24,
|
786 |
+
"learning_rate": 7.419786096256684e-07,
|
787 |
+
"loss": 1.9598,
|
788 |
+
"step": 1180
|
789 |
+
},
|
790 |
+
{
|
791 |
+
"epoch": 1.26,
|
792 |
+
"learning_rate": 7.375222816399287e-07,
|
793 |
+
"loss": 1.9529,
|
794 |
+
"step": 1190
|
795 |
+
},
|
796 |
+
{
|
797 |
+
"epoch": 1.27,
|
798 |
+
"learning_rate": 7.330659536541889e-07,
|
799 |
+
"loss": 1.9675,
|
800 |
+
"step": 1200
|
801 |
+
},
|
802 |
+
{
|
803 |
+
"epoch": 1.28,
|
804 |
+
"learning_rate": 7.286096256684492e-07,
|
805 |
+
"loss": 1.9433,
|
806 |
+
"step": 1210
|
807 |
+
},
|
808 |
+
{
|
809 |
+
"epoch": 1.29,
|
810 |
+
"learning_rate": 7.241532976827094e-07,
|
811 |
+
"loss": 1.9002,
|
812 |
+
"step": 1220
|
813 |
+
},
|
814 |
+
{
|
815 |
+
"epoch": 1.3,
|
816 |
+
"learning_rate": 7.196969696969697e-07,
|
817 |
+
"loss": 1.9349,
|
818 |
+
"step": 1230
|
819 |
+
},
|
820 |
+
{
|
821 |
+
"epoch": 1.31,
|
822 |
+
"learning_rate": 7.152406417112298e-07,
|
823 |
+
"loss": 1.9414,
|
824 |
+
"step": 1240
|
825 |
+
},
|
826 |
+
{
|
827 |
+
"epoch": 1.32,
|
828 |
+
"learning_rate": 7.107843137254902e-07,
|
829 |
+
"loss": 1.9555,
|
830 |
+
"step": 1250
|
831 |
+
},
|
832 |
+
{
|
833 |
+
"epoch": 1.32,
|
834 |
+
"eval_webgpt_accuracy": 0.526057600642472,
|
835 |
+
"eval_webgpt_loss": 2.1640625,
|
836 |
+
"eval_webgpt_runtime": 428.0509,
|
837 |
+
"eval_webgpt_samples_per_second": 9.148,
|
838 |
+
"eval_webgpt_steps_per_second": 0.764,
|
839 |
+
"step": 1250
|
840 |
+
},
|
841 |
+
{
|
842 |
+
"epoch": 1.32,
|
843 |
+
"eval_prompt_dialogue_accuracy": 0.5739159961148795,
|
844 |
+
"eval_prompt_dialogue_loss": 1.755859375,
|
845 |
+
"eval_prompt_dialogue_runtime": 1048.4647,
|
846 |
+
"eval_prompt_dialogue_samples_per_second": 9.832,
|
847 |
+
"eval_prompt_dialogue_steps_per_second": 0.82,
|
848 |
+
"step": 1250
|
849 |
+
},
|
850 |
+
{
|
851 |
+
"epoch": 1.33,
|
852 |
+
"learning_rate": 7.063279857397504e-07,
|
853 |
+
"loss": 1.983,
|
854 |
+
"step": 1260
|
855 |
+
},
|
856 |
+
{
|
857 |
+
"epoch": 1.34,
|
858 |
+
"learning_rate": 7.018716577540107e-07,
|
859 |
+
"loss": 1.983,
|
860 |
+
"step": 1270
|
861 |
+
},
|
862 |
+
{
|
863 |
+
"epoch": 1.35,
|
864 |
+
"learning_rate": 6.974153297682709e-07,
|
865 |
+
"loss": 1.9449,
|
866 |
+
"step": 1280
|
867 |
+
},
|
868 |
+
{
|
869 |
+
"epoch": 1.36,
|
870 |
+
"learning_rate": 6.929590017825312e-07,
|
871 |
+
"loss": 1.9886,
|
872 |
+
"step": 1290
|
873 |
+
},
|
874 |
+
{
|
875 |
+
"epoch": 1.37,
|
876 |
+
"learning_rate": 6.885026737967914e-07,
|
877 |
+
"loss": 1.9334,
|
878 |
+
"step": 1300
|
879 |
+
},
|
880 |
+
{
|
881 |
+
"epoch": 1.38,
|
882 |
+
"learning_rate": 6.840463458110517e-07,
|
883 |
+
"loss": 1.9732,
|
884 |
+
"step": 1310
|
885 |
+
},
|
886 |
+
{
|
887 |
+
"epoch": 1.39,
|
888 |
+
"learning_rate": 6.795900178253119e-07,
|
889 |
+
"loss": 1.9457,
|
890 |
+
"step": 1320
|
891 |
+
},
|
892 |
+
{
|
893 |
+
"epoch": 1.4,
|
894 |
+
"learning_rate": 6.751336898395722e-07,
|
895 |
+
"loss": 1.9691,
|
896 |
+
"step": 1330
|
897 |
+
},
|
898 |
+
{
|
899 |
+
"epoch": 1.41,
|
900 |
+
"learning_rate": 6.706773618538324e-07,
|
901 |
+
"loss": 2.0199,
|
902 |
+
"step": 1340
|
903 |
+
},
|
904 |
+
{
|
905 |
+
"epoch": 1.42,
|
906 |
+
"learning_rate": 6.662210338680927e-07,
|
907 |
+
"loss": 1.9492,
|
908 |
+
"step": 1350
|
909 |
+
},
|
910 |
+
{
|
911 |
+
"epoch": 1.43,
|
912 |
+
"learning_rate": 6.617647058823529e-07,
|
913 |
+
"loss": 1.9553,
|
914 |
+
"step": 1360
|
915 |
+
},
|
916 |
+
{
|
917 |
+
"epoch": 1.44,
|
918 |
+
"learning_rate": 6.573083778966132e-07,
|
919 |
+
"loss": 1.9766,
|
920 |
+
"step": 1370
|
921 |
+
},
|
922 |
+
{
|
923 |
+
"epoch": 1.46,
|
924 |
+
"learning_rate": 6.528520499108734e-07,
|
925 |
+
"loss": 1.9496,
|
926 |
+
"step": 1380
|
927 |
+
},
|
928 |
+
{
|
929 |
+
"epoch": 1.47,
|
930 |
+
"learning_rate": 6.483957219251337e-07,
|
931 |
+
"loss": 1.9677,
|
932 |
+
"step": 1390
|
933 |
+
},
|
934 |
+
{
|
935 |
+
"epoch": 1.48,
|
936 |
+
"learning_rate": 6.439393939393939e-07,
|
937 |
+
"loss": 1.9043,
|
938 |
+
"step": 1400
|
939 |
+
},
|
940 |
+
{
|
941 |
+
"epoch": 1.49,
|
942 |
+
"learning_rate": 6.394830659536542e-07,
|
943 |
+
"loss": 1.9571,
|
944 |
+
"step": 1410
|
945 |
+
},
|
946 |
+
{
|
947 |
+
"epoch": 1.5,
|
948 |
+
"learning_rate": 6.350267379679144e-07,
|
949 |
+
"loss": 1.9631,
|
950 |
+
"step": 1420
|
951 |
+
},
|
952 |
+
{
|
953 |
+
"epoch": 1.51,
|
954 |
+
"learning_rate": 6.305704099821747e-07,
|
955 |
+
"loss": 1.9587,
|
956 |
+
"step": 1430
|
957 |
+
},
|
958 |
+
{
|
959 |
+
"epoch": 1.52,
|
960 |
+
"learning_rate": 6.261140819964349e-07,
|
961 |
+
"loss": 1.9621,
|
962 |
+
"step": 1440
|
963 |
+
},
|
964 |
+
{
|
965 |
+
"epoch": 1.53,
|
966 |
+
"learning_rate": 6.216577540106952e-07,
|
967 |
+
"loss": 1.932,
|
968 |
+
"step": 1450
|
969 |
+
},
|
970 |
+
{
|
971 |
+
"epoch": 1.54,
|
972 |
+
"learning_rate": 6.172014260249554e-07,
|
973 |
+
"loss": 1.9682,
|
974 |
+
"step": 1460
|
975 |
+
},
|
976 |
+
{
|
977 |
+
"epoch": 1.55,
|
978 |
+
"learning_rate": 6.127450980392157e-07,
|
979 |
+
"loss": 1.9636,
|
980 |
+
"step": 1470
|
981 |
+
},
|
982 |
+
{
|
983 |
+
"epoch": 1.56,
|
984 |
+
"learning_rate": 6.082887700534759e-07,
|
985 |
+
"loss": 1.9538,
|
986 |
+
"step": 1480
|
987 |
+
},
|
988 |
+
{
|
989 |
+
"epoch": 1.57,
|
990 |
+
"learning_rate": 6.038324420677362e-07,
|
991 |
+
"loss": 1.9676,
|
992 |
+
"step": 1490
|
993 |
+
},
|
994 |
+
{
|
995 |
+
"epoch": 1.58,
|
996 |
+
"learning_rate": 5.998217468805704e-07,
|
997 |
+
"loss": 1.9459,
|
998 |
+
"step": 1500
|
999 |
+
},
|
1000 |
+
{
|
1001 |
+
"epoch": 1.58,
|
1002 |
+
"eval_webgpt_accuracy": 0.526490033682041,
|
1003 |
+
"eval_webgpt_loss": 2.162109375,
|
1004 |
+
"eval_webgpt_runtime": 435.1199,
|
1005 |
+
"eval_webgpt_samples_per_second": 9.0,
|
1006 |
+
"eval_webgpt_steps_per_second": 0.752,
|
1007 |
+
"step": 1500
|
1008 |
+
},
|
1009 |
+
{
|
1010 |
+
"epoch": 1.58,
|
1011 |
+
"eval_prompt_dialogue_accuracy": 0.5751027402282565,
|
1012 |
+
"eval_prompt_dialogue_loss": 1.748046875,
|
1013 |
+
"eval_prompt_dialogue_runtime": 1051.1439,
|
1014 |
+
"eval_prompt_dialogue_samples_per_second": 9.807,
|
1015 |
+
"eval_prompt_dialogue_steps_per_second": 0.818,
|
1016 |
+
"step": 1500
|
1017 |
+
},
|
1018 |
+
{
|
1019 |
+
"epoch": 1.59,
|
1020 |
+
"learning_rate": 5.953654188948306e-07,
|
1021 |
+
"loss": 1.9592,
|
1022 |
+
"step": 1510
|
1023 |
+
},
|
1024 |
+
{
|
1025 |
+
"epoch": 1.6,
|
1026 |
+
"learning_rate": 5.909090909090909e-07,
|
1027 |
+
"loss": 1.9291,
|
1028 |
+
"step": 1520
|
1029 |
+
},
|
1030 |
+
{
|
1031 |
+
"epoch": 1.61,
|
1032 |
+
"learning_rate": 5.864527629233511e-07,
|
1033 |
+
"loss": 1.9082,
|
1034 |
+
"step": 1530
|
1035 |
+
},
|
1036 |
+
{
|
1037 |
+
"epoch": 1.62,
|
1038 |
+
"learning_rate": 5.819964349376115e-07,
|
1039 |
+
"loss": 1.9357,
|
1040 |
+
"step": 1540
|
1041 |
+
},
|
1042 |
+
{
|
1043 |
+
"epoch": 1.63,
|
1044 |
+
"learning_rate": 5.775401069518716e-07,
|
1045 |
+
"loss": 1.9502,
|
1046 |
+
"step": 1550
|
1047 |
+
},
|
1048 |
+
{
|
1049 |
+
"epoch": 1.65,
|
1050 |
+
"learning_rate": 5.730837789661319e-07,
|
1051 |
+
"loss": 1.9892,
|
1052 |
+
"step": 1560
|
1053 |
+
},
|
1054 |
+
{
|
1055 |
+
"epoch": 1.66,
|
1056 |
+
"learning_rate": 5.686274509803921e-07,
|
1057 |
+
"loss": 1.983,
|
1058 |
+
"step": 1570
|
1059 |
+
},
|
1060 |
+
{
|
1061 |
+
"epoch": 1.67,
|
1062 |
+
"learning_rate": 5.641711229946524e-07,
|
1063 |
+
"loss": 1.9745,
|
1064 |
+
"step": 1580
|
1065 |
+
},
|
1066 |
+
{
|
1067 |
+
"epoch": 1.68,
|
1068 |
+
"learning_rate": 5.597147950089126e-07,
|
1069 |
+
"loss": 1.9541,
|
1070 |
+
"step": 1590
|
1071 |
+
},
|
1072 |
+
{
|
1073 |
+
"epoch": 1.69,
|
1074 |
+
"learning_rate": 5.552584670231729e-07,
|
1075 |
+
"loss": 1.9482,
|
1076 |
+
"step": 1600
|
1077 |
+
},
|
1078 |
+
{
|
1079 |
+
"epoch": 1.7,
|
1080 |
+
"learning_rate": 5.508021390374332e-07,
|
1081 |
+
"loss": 1.9341,
|
1082 |
+
"step": 1610
|
1083 |
+
},
|
1084 |
+
{
|
1085 |
+
"epoch": 1.71,
|
1086 |
+
"learning_rate": 5.463458110516934e-07,
|
1087 |
+
"loss": 1.971,
|
1088 |
+
"step": 1620
|
1089 |
+
},
|
1090 |
+
{
|
1091 |
+
"epoch": 1.72,
|
1092 |
+
"learning_rate": 5.418894830659536e-07,
|
1093 |
+
"loss": 1.9179,
|
1094 |
+
"step": 1630
|
1095 |
+
},
|
1096 |
+
{
|
1097 |
+
"epoch": 1.73,
|
1098 |
+
"learning_rate": 5.374331550802139e-07,
|
1099 |
+
"loss": 1.9049,
|
1100 |
+
"step": 1640
|
1101 |
+
},
|
1102 |
+
{
|
1103 |
+
"epoch": 1.74,
|
1104 |
+
"learning_rate": 5.329768270944741e-07,
|
1105 |
+
"loss": 1.9407,
|
1106 |
+
"step": 1650
|
1107 |
+
},
|
1108 |
+
{
|
1109 |
+
"epoch": 1.75,
|
1110 |
+
"learning_rate": 5.285204991087344e-07,
|
1111 |
+
"loss": 1.9338,
|
1112 |
+
"step": 1660
|
1113 |
+
},
|
1114 |
+
{
|
1115 |
+
"epoch": 1.76,
|
1116 |
+
"learning_rate": 5.240641711229946e-07,
|
1117 |
+
"loss": 1.9874,
|
1118 |
+
"step": 1670
|
1119 |
+
},
|
1120 |
+
{
|
1121 |
+
"epoch": 1.77,
|
1122 |
+
"learning_rate": 5.19607843137255e-07,
|
1123 |
+
"loss": 1.9699,
|
1124 |
+
"step": 1680
|
1125 |
+
},
|
1126 |
+
{
|
1127 |
+
"epoch": 1.78,
|
1128 |
+
"learning_rate": 5.151515151515151e-07,
|
1129 |
+
"loss": 1.9128,
|
1130 |
+
"step": 1690
|
1131 |
+
},
|
1132 |
+
{
|
1133 |
+
"epoch": 1.79,
|
1134 |
+
"learning_rate": 5.106951871657754e-07,
|
1135 |
+
"loss": 1.9382,
|
1136 |
+
"step": 1700
|
1137 |
+
},
|
1138 |
+
{
|
1139 |
+
"epoch": 1.8,
|
1140 |
+
"learning_rate": 5.062388591800356e-07,
|
1141 |
+
"loss": 1.9514,
|
1142 |
+
"step": 1710
|
1143 |
+
},
|
1144 |
+
{
|
1145 |
+
"epoch": 1.81,
|
1146 |
+
"learning_rate": 5.01782531194296e-07,
|
1147 |
+
"loss": 1.9499,
|
1148 |
+
"step": 1720
|
1149 |
+
},
|
1150 |
+
{
|
1151 |
+
"epoch": 1.82,
|
1152 |
+
"learning_rate": 4.973262032085561e-07,
|
1153 |
+
"loss": 1.9458,
|
1154 |
+
"step": 1730
|
1155 |
+
},
|
1156 |
+
{
|
1157 |
+
"epoch": 1.84,
|
1158 |
+
"learning_rate": 4.928698752228163e-07,
|
1159 |
+
"loss": 1.9095,
|
1160 |
+
"step": 1740
|
1161 |
+
},
|
1162 |
+
{
|
1163 |
+
"epoch": 1.85,
|
1164 |
+
"learning_rate": 4.884135472370767e-07,
|
1165 |
+
"loss": 1.9237,
|
1166 |
+
"step": 1750
|
1167 |
+
},
|
1168 |
+
{
|
1169 |
+
"epoch": 1.85,
|
1170 |
+
"eval_webgpt_accuracy": 0.5268101464515921,
|
1171 |
+
"eval_webgpt_loss": 2.16015625,
|
1172 |
+
"eval_webgpt_runtime": 427.8942,
|
1173 |
+
"eval_webgpt_samples_per_second": 9.152,
|
1174 |
+
"eval_webgpt_steps_per_second": 0.764,
|
1175 |
+
"step": 1750
|
1176 |
+
},
|
1177 |
+
{
|
1178 |
+
"epoch": 1.85,
|
1179 |
+
"eval_prompt_dialogue_accuracy": 0.5762533856613786,
|
1180 |
+
"eval_prompt_dialogue_loss": 1.740234375,
|
1181 |
+
"eval_prompt_dialogue_runtime": 1049.6135,
|
1182 |
+
"eval_prompt_dialogue_samples_per_second": 9.822,
|
1183 |
+
"eval_prompt_dialogue_steps_per_second": 0.819,
|
1184 |
+
"step": 1750
|
1185 |
+
},
|
1186 |
+
{
|
1187 |
+
"epoch": 1.86,
|
1188 |
+
"learning_rate": 4.839572192513369e-07,
|
1189 |
+
"loss": 1.9181,
|
1190 |
+
"step": 1760
|
1191 |
+
},
|
1192 |
+
{
|
1193 |
+
"epoch": 1.87,
|
1194 |
+
"learning_rate": 4.795008912655971e-07,
|
1195 |
+
"loss": 1.9393,
|
1196 |
+
"step": 1770
|
1197 |
+
},
|
1198 |
+
{
|
1199 |
+
"epoch": 1.88,
|
1200 |
+
"learning_rate": 4.750445632798574e-07,
|
1201 |
+
"loss": 1.9278,
|
1202 |
+
"step": 1780
|
1203 |
+
},
|
1204 |
+
{
|
1205 |
+
"epoch": 1.89,
|
1206 |
+
"learning_rate": 4.705882352941176e-07,
|
1207 |
+
"loss": 1.972,
|
1208 |
+
"step": 1790
|
1209 |
+
},
|
1210 |
+
{
|
1211 |
+
"epoch": 1.9,
|
1212 |
+
"learning_rate": 4.6613190730837784e-07,
|
1213 |
+
"loss": 1.9545,
|
1214 |
+
"step": 1800
|
1215 |
+
},
|
1216 |
+
{
|
1217 |
+
"epoch": 1.91,
|
1218 |
+
"learning_rate": 4.616755793226381e-07,
|
1219 |
+
"loss": 1.9324,
|
1220 |
+
"step": 1810
|
1221 |
+
},
|
1222 |
+
{
|
1223 |
+
"epoch": 1.92,
|
1224 |
+
"learning_rate": 4.5721925133689835e-07,
|
1225 |
+
"loss": 1.9856,
|
1226 |
+
"step": 1820
|
1227 |
+
},
|
1228 |
+
{
|
1229 |
+
"epoch": 1.93,
|
1230 |
+
"learning_rate": 4.5276292335115863e-07,
|
1231 |
+
"loss": 1.9013,
|
1232 |
+
"step": 1830
|
1233 |
+
},
|
1234 |
+
{
|
1235 |
+
"epoch": 1.94,
|
1236 |
+
"learning_rate": 4.4830659536541886e-07,
|
1237 |
+
"loss": 1.9001,
|
1238 |
+
"step": 1840
|
1239 |
+
},
|
1240 |
+
{
|
1241 |
+
"epoch": 1.95,
|
1242 |
+
"learning_rate": 4.4385026737967914e-07,
|
1243 |
+
"loss": 1.9326,
|
1244 |
+
"step": 1850
|
1245 |
+
},
|
1246 |
+
{
|
1247 |
+
"epoch": 1.96,
|
1248 |
+
"learning_rate": 4.3939393939393937e-07,
|
1249 |
+
"loss": 1.9358,
|
1250 |
+
"step": 1860
|
1251 |
+
},
|
1252 |
+
{
|
1253 |
+
"epoch": 1.97,
|
1254 |
+
"learning_rate": 4.3493761140819965e-07,
|
1255 |
+
"loss": 2.0019,
|
1256 |
+
"step": 1870
|
1257 |
+
},
|
1258 |
+
{
|
1259 |
+
"epoch": 1.98,
|
1260 |
+
"learning_rate": 4.304812834224599e-07,
|
1261 |
+
"loss": 1.941,
|
1262 |
+
"step": 1880
|
1263 |
+
},
|
1264 |
+
{
|
1265 |
+
"epoch": 1.99,
|
1266 |
+
"learning_rate": 4.260249554367201e-07,
|
1267 |
+
"loss": 1.9325,
|
1268 |
+
"step": 1890
|
1269 |
+
},
|
1270 |
+
{
|
1271 |
+
"epoch": 2.0,
|
1272 |
+
"learning_rate": 4.2112299465240644e-07,
|
1273 |
+
"loss": 2.0644,
|
1274 |
+
"step": 1900
|
1275 |
+
},
|
1276 |
+
{
|
1277 |
+
"epoch": 2.01,
|
1278 |
+
"learning_rate": 4.1666666666666667e-07,
|
1279 |
+
"loss": 1.9158,
|
1280 |
+
"step": 1910
|
1281 |
+
},
|
1282 |
+
{
|
1283 |
+
"epoch": 2.03,
|
1284 |
+
"learning_rate": 4.122103386809269e-07,
|
1285 |
+
"loss": 1.9409,
|
1286 |
+
"step": 1920
|
1287 |
+
},
|
1288 |
+
{
|
1289 |
+
"epoch": 2.04,
|
1290 |
+
"learning_rate": 4.0775401069518717e-07,
|
1291 |
+
"loss": 1.926,
|
1292 |
+
"step": 1930
|
1293 |
+
},
|
1294 |
+
{
|
1295 |
+
"epoch": 2.05,
|
1296 |
+
"learning_rate": 4.032976827094474e-07,
|
1297 |
+
"loss": 1.9212,
|
1298 |
+
"step": 1940
|
1299 |
+
},
|
1300 |
+
{
|
1301 |
+
"epoch": 2.06,
|
1302 |
+
"learning_rate": 3.988413547237077e-07,
|
1303 |
+
"loss": 1.9745,
|
1304 |
+
"step": 1950
|
1305 |
+
},
|
1306 |
+
{
|
1307 |
+
"epoch": 2.07,
|
1308 |
+
"learning_rate": 3.943850267379679e-07,
|
1309 |
+
"loss": 1.9079,
|
1310 |
+
"step": 1960
|
1311 |
+
},
|
1312 |
+
{
|
1313 |
+
"epoch": 2.08,
|
1314 |
+
"learning_rate": 3.899286987522282e-07,
|
1315 |
+
"loss": 1.9336,
|
1316 |
+
"step": 1970
|
1317 |
+
},
|
1318 |
+
{
|
1319 |
+
"epoch": 2.09,
|
1320 |
+
"learning_rate": 3.854723707664884e-07,
|
1321 |
+
"loss": 1.9242,
|
1322 |
+
"step": 1980
|
1323 |
+
},
|
1324 |
+
{
|
1325 |
+
"epoch": 2.1,
|
1326 |
+
"learning_rate": 3.8101604278074864e-07,
|
1327 |
+
"loss": 1.9357,
|
1328 |
+
"step": 1990
|
1329 |
+
},
|
1330 |
+
{
|
1331 |
+
"epoch": 2.11,
|
1332 |
+
"learning_rate": 3.765597147950089e-07,
|
1333 |
+
"loss": 1.9036,
|
1334 |
+
"step": 2000
|
1335 |
+
},
|
1336 |
+
{
|
1337 |
+
"epoch": 2.11,
|
1338 |
+
"eval_webgpt_accuracy": 0.5268452465359726,
|
1339 |
+
"eval_webgpt_loss": 2.158203125,
|
1340 |
+
"eval_webgpt_runtime": 428.2029,
|
1341 |
+
"eval_webgpt_samples_per_second": 9.145,
|
1342 |
+
"eval_webgpt_steps_per_second": 0.764,
|
1343 |
+
"step": 2000
|
1344 |
+
},
|
1345 |
+
{
|
1346 |
+
"epoch": 2.11,
|
1347 |
+
"eval_prompt_dialogue_accuracy": 0.5769595660938633,
|
1348 |
+
"eval_prompt_dialogue_loss": 1.7373046875,
|
1349 |
+
"eval_prompt_dialogue_runtime": 1052.6454,
|
1350 |
+
"eval_prompt_dialogue_samples_per_second": 9.793,
|
1351 |
+
"eval_prompt_dialogue_steps_per_second": 0.817,
|
1352 |
+
"step": 2000
|
1353 |
+
},
|
1354 |
+
{
|
1355 |
+
"epoch": 2.12,
|
1356 |
+
"learning_rate": 3.7210338680926915e-07,
|
1357 |
+
"loss": 1.8764,
|
1358 |
+
"step": 2010
|
1359 |
+
},
|
1360 |
+
{
|
1361 |
+
"epoch": 2.13,
|
1362 |
+
"learning_rate": 3.6764705882352943e-07,
|
1363 |
+
"loss": 1.9615,
|
1364 |
+
"step": 2020
|
1365 |
+
},
|
1366 |
+
{
|
1367 |
+
"epoch": 2.14,
|
1368 |
+
"learning_rate": 3.6319073083778966e-07,
|
1369 |
+
"loss": 1.914,
|
1370 |
+
"step": 2030
|
1371 |
+
},
|
1372 |
+
{
|
1373 |
+
"epoch": 2.15,
|
1374 |
+
"learning_rate": 3.5873440285204994e-07,
|
1375 |
+
"loss": 1.8962,
|
1376 |
+
"step": 2040
|
1377 |
+
},
|
1378 |
+
{
|
1379 |
+
"epoch": 2.16,
|
1380 |
+
"learning_rate": 3.542780748663101e-07,
|
1381 |
+
"loss": 1.8999,
|
1382 |
+
"step": 2050
|
1383 |
+
},
|
1384 |
+
{
|
1385 |
+
"epoch": 2.17,
|
1386 |
+
"learning_rate": 3.4982174688057034e-07,
|
1387 |
+
"loss": 1.9456,
|
1388 |
+
"step": 2060
|
1389 |
+
},
|
1390 |
+
{
|
1391 |
+
"epoch": 2.18,
|
1392 |
+
"learning_rate": 3.453654188948306e-07,
|
1393 |
+
"loss": 1.9347,
|
1394 |
+
"step": 2070
|
1395 |
+
},
|
1396 |
+
{
|
1397 |
+
"epoch": 2.19,
|
1398 |
+
"learning_rate": 3.4090909090909085e-07,
|
1399 |
+
"loss": 1.9102,
|
1400 |
+
"step": 2080
|
1401 |
+
},
|
1402 |
+
{
|
1403 |
+
"epoch": 2.2,
|
1404 |
+
"learning_rate": 3.3645276292335113e-07,
|
1405 |
+
"loss": 1.9355,
|
1406 |
+
"step": 2090
|
1407 |
+
},
|
1408 |
+
{
|
1409 |
+
"epoch": 2.22,
|
1410 |
+
"learning_rate": 3.3199643493761136e-07,
|
1411 |
+
"loss": 1.9204,
|
1412 |
+
"step": 2100
|
1413 |
+
},
|
1414 |
+
{
|
1415 |
+
"epoch": 2.23,
|
1416 |
+
"learning_rate": 3.2754010695187164e-07,
|
1417 |
+
"loss": 1.9396,
|
1418 |
+
"step": 2110
|
1419 |
+
},
|
1420 |
+
{
|
1421 |
+
"epoch": 2.24,
|
1422 |
+
"learning_rate": 3.2308377896613187e-07,
|
1423 |
+
"loss": 1.9057,
|
1424 |
+
"step": 2120
|
1425 |
+
},
|
1426 |
+
{
|
1427 |
+
"epoch": 2.25,
|
1428 |
+
"learning_rate": 3.1862745098039215e-07,
|
1429 |
+
"loss": 1.9241,
|
1430 |
+
"step": 2130
|
1431 |
+
},
|
1432 |
+
{
|
1433 |
+
"epoch": 2.26,
|
1434 |
+
"learning_rate": 3.141711229946524e-07,
|
1435 |
+
"loss": 1.9383,
|
1436 |
+
"step": 2140
|
1437 |
+
},
|
1438 |
+
{
|
1439 |
+
"epoch": 2.27,
|
1440 |
+
"learning_rate": 3.097147950089126e-07,
|
1441 |
+
"loss": 1.9197,
|
1442 |
+
"step": 2150
|
1443 |
+
},
|
1444 |
+
{
|
1445 |
+
"epoch": 2.28,
|
1446 |
+
"learning_rate": 3.052584670231729e-07,
|
1447 |
+
"loss": 1.8983,
|
1448 |
+
"step": 2160
|
1449 |
+
},
|
1450 |
+
{
|
1451 |
+
"epoch": 2.29,
|
1452 |
+
"learning_rate": 3.008021390374331e-07,
|
1453 |
+
"loss": 1.9677,
|
1454 |
+
"step": 2170
|
1455 |
+
},
|
1456 |
+
{
|
1457 |
+
"epoch": 2.3,
|
1458 |
+
"learning_rate": 2.963458110516934e-07,
|
1459 |
+
"loss": 1.9104,
|
1460 |
+
"step": 2180
|
1461 |
+
},
|
1462 |
+
{
|
1463 |
+
"epoch": 2.31,
|
1464 |
+
"learning_rate": 2.918894830659536e-07,
|
1465 |
+
"loss": 1.8994,
|
1466 |
+
"step": 2190
|
1467 |
+
},
|
1468 |
+
{
|
1469 |
+
"epoch": 2.32,
|
1470 |
+
"learning_rate": 2.874331550802139e-07,
|
1471 |
+
"loss": 1.9157,
|
1472 |
+
"step": 2200
|
1473 |
+
},
|
1474 |
+
{
|
1475 |
+
"epoch": 2.33,
|
1476 |
+
"learning_rate": 2.829768270944741e-07,
|
1477 |
+
"loss": 1.918,
|
1478 |
+
"step": 2210
|
1479 |
+
},
|
1480 |
+
{
|
1481 |
+
"epoch": 2.34,
|
1482 |
+
"learning_rate": 2.7852049910873435e-07,
|
1483 |
+
"loss": 1.9053,
|
1484 |
+
"step": 2220
|
1485 |
+
},
|
1486 |
+
{
|
1487 |
+
"epoch": 2.35,
|
1488 |
+
"learning_rate": 2.7406417112299463e-07,
|
1489 |
+
"loss": 1.8939,
|
1490 |
+
"step": 2230
|
1491 |
+
},
|
1492 |
+
{
|
1493 |
+
"epoch": 2.36,
|
1494 |
+
"learning_rate": 2.6960784313725486e-07,
|
1495 |
+
"loss": 1.8959,
|
1496 |
+
"step": 2240
|
1497 |
+
},
|
1498 |
+
{
|
1499 |
+
"epoch": 2.37,
|
1500 |
+
"learning_rate": 2.6515151515151514e-07,
|
1501 |
+
"loss": 1.9326,
|
1502 |
+
"step": 2250
|
1503 |
+
},
|
1504 |
+
{
|
1505 |
+
"epoch": 2.37,
|
1506 |
+
"eval_webgpt_accuracy": 0.5270740990861342,
|
1507 |
+
"eval_webgpt_loss": 2.158203125,
|
1508 |
+
"eval_webgpt_runtime": 429.6721,
|
1509 |
+
"eval_webgpt_samples_per_second": 9.114,
|
1510 |
+
"eval_webgpt_steps_per_second": 0.761,
|
1511 |
+
"step": 2250
|
1512 |
+
},
|
1513 |
+
{
|
1514 |
+
"epoch": 2.37,
|
1515 |
+
"eval_prompt_dialogue_accuracy": 0.577335217985265,
|
1516 |
+
"eval_prompt_dialogue_loss": 1.7333984375,
|
1517 |
+
"eval_prompt_dialogue_runtime": 1050.2434,
|
1518 |
+
"eval_prompt_dialogue_samples_per_second": 9.816,
|
1519 |
+
"eval_prompt_dialogue_steps_per_second": 0.819,
|
1520 |
+
"step": 2250
|
1521 |
+
},
|
1522 |
+
{
|
1523 |
+
"epoch": 2.38,
|
1524 |
+
"learning_rate": 2.6069518716577537e-07,
|
1525 |
+
"loss": 1.8933,
|
1526 |
+
"step": 2260
|
1527 |
+
},
|
1528 |
+
{
|
1529 |
+
"epoch": 2.39,
|
1530 |
+
"learning_rate": 2.5623885918003565e-07,
|
1531 |
+
"loss": 1.9186,
|
1532 |
+
"step": 2270
|
1533 |
+
},
|
1534 |
+
{
|
1535 |
+
"epoch": 2.4,
|
1536 |
+
"learning_rate": 2.517825311942959e-07,
|
1537 |
+
"loss": 1.9149,
|
1538 |
+
"step": 2280
|
1539 |
+
},
|
1540 |
+
{
|
1541 |
+
"epoch": 2.42,
|
1542 |
+
"learning_rate": 2.473262032085561e-07,
|
1543 |
+
"loss": 1.9288,
|
1544 |
+
"step": 2290
|
1545 |
+
},
|
1546 |
+
{
|
1547 |
+
"epoch": 2.43,
|
1548 |
+
"learning_rate": 2.428698752228164e-07,
|
1549 |
+
"loss": 1.9082,
|
1550 |
+
"step": 2300
|
1551 |
+
},
|
1552 |
+
{
|
1553 |
+
"epoch": 2.44,
|
1554 |
+
"learning_rate": 2.3841354723707664e-07,
|
1555 |
+
"loss": 1.9195,
|
1556 |
+
"step": 2310
|
1557 |
+
},
|
1558 |
+
{
|
1559 |
+
"epoch": 2.45,
|
1560 |
+
"learning_rate": 2.339572192513369e-07,
|
1561 |
+
"loss": 1.9262,
|
1562 |
+
"step": 2320
|
1563 |
+
},
|
1564 |
+
{
|
1565 |
+
"epoch": 2.46,
|
1566 |
+
"learning_rate": 2.2950089126559712e-07,
|
1567 |
+
"loss": 1.9096,
|
1568 |
+
"step": 2330
|
1569 |
+
},
|
1570 |
+
{
|
1571 |
+
"epoch": 2.47,
|
1572 |
+
"learning_rate": 2.2504456327985737e-07,
|
1573 |
+
"loss": 1.913,
|
1574 |
+
"step": 2340
|
1575 |
+
},
|
1576 |
+
{
|
1577 |
+
"epoch": 2.48,
|
1578 |
+
"learning_rate": 2.2058823529411763e-07,
|
1579 |
+
"loss": 1.9242,
|
1580 |
+
"step": 2350
|
1581 |
+
},
|
1582 |
+
{
|
1583 |
+
"epoch": 2.49,
|
1584 |
+
"learning_rate": 2.1613190730837788e-07,
|
1585 |
+
"loss": 1.9175,
|
1586 |
+
"step": 2360
|
1587 |
+
},
|
1588 |
+
{
|
1589 |
+
"epoch": 2.5,
|
1590 |
+
"learning_rate": 2.1167557932263814e-07,
|
1591 |
+
"loss": 1.8917,
|
1592 |
+
"step": 2370
|
1593 |
+
},
|
1594 |
+
{
|
1595 |
+
"epoch": 2.51,
|
1596 |
+
"learning_rate": 2.072192513368984e-07,
|
1597 |
+
"loss": 1.9024,
|
1598 |
+
"step": 2380
|
1599 |
+
},
|
1600 |
+
{
|
1601 |
+
"epoch": 2.52,
|
1602 |
+
"learning_rate": 2.0276292335115864e-07,
|
1603 |
+
"loss": 1.9243,
|
1604 |
+
"step": 2390
|
1605 |
+
},
|
1606 |
+
{
|
1607 |
+
"epoch": 2.53,
|
1608 |
+
"learning_rate": 1.9830659536541887e-07,
|
1609 |
+
"loss": 1.9269,
|
1610 |
+
"step": 2400
|
1611 |
+
},
|
1612 |
+
{
|
1613 |
+
"epoch": 2.54,
|
1614 |
+
"learning_rate": 1.9385026737967912e-07,
|
1615 |
+
"loss": 1.961,
|
1616 |
+
"step": 2410
|
1617 |
+
},
|
1618 |
+
{
|
1619 |
+
"epoch": 2.55,
|
1620 |
+
"learning_rate": 1.8939393939393938e-07,
|
1621 |
+
"loss": 1.9414,
|
1622 |
+
"step": 2420
|
1623 |
+
},
|
1624 |
+
{
|
1625 |
+
"epoch": 2.56,
|
1626 |
+
"learning_rate": 1.8493761140819963e-07,
|
1627 |
+
"loss": 1.9306,
|
1628 |
+
"step": 2430
|
1629 |
+
},
|
1630 |
+
{
|
1631 |
+
"epoch": 2.57,
|
1632 |
+
"learning_rate": 1.804812834224599e-07,
|
1633 |
+
"loss": 1.9385,
|
1634 |
+
"step": 2440
|
1635 |
+
},
|
1636 |
+
{
|
1637 |
+
"epoch": 2.58,
|
1638 |
+
"learning_rate": 1.7602495543672014e-07,
|
1639 |
+
"loss": 1.9304,
|
1640 |
+
"step": 2450
|
1641 |
+
},
|
1642 |
+
{
|
1643 |
+
"epoch": 2.59,
|
1644 |
+
"learning_rate": 1.715686274509804e-07,
|
1645 |
+
"loss": 1.9214,
|
1646 |
+
"step": 2460
|
1647 |
+
},
|
1648 |
+
{
|
1649 |
+
"epoch": 2.61,
|
1650 |
+
"learning_rate": 1.6711229946524065e-07,
|
1651 |
+
"loss": 1.9854,
|
1652 |
+
"step": 2470
|
1653 |
+
},
|
1654 |
+
{
|
1655 |
+
"epoch": 2.62,
|
1656 |
+
"learning_rate": 1.6265597147950088e-07,
|
1657 |
+
"loss": 1.91,
|
1658 |
+
"step": 2480
|
1659 |
+
},
|
1660 |
+
{
|
1661 |
+
"epoch": 2.63,
|
1662 |
+
"learning_rate": 1.5819964349376113e-07,
|
1663 |
+
"loss": 1.9317,
|
1664 |
+
"step": 2490
|
1665 |
+
},
|
1666 |
+
{
|
1667 |
+
"epoch": 2.64,
|
1668 |
+
"learning_rate": 1.5463458110516933e-07,
|
1669 |
+
"loss": 1.9087,
|
1670 |
+
"step": 2500
|
1671 |
+
},
|
1672 |
+
{
|
1673 |
+
"epoch": 2.64,
|
1674 |
+
"eval_webgpt_accuracy": 0.5270825231063856,
|
1675 |
+
"eval_webgpt_loss": 2.15625,
|
1676 |
+
"eval_webgpt_runtime": 426.8235,
|
1677 |
+
"eval_webgpt_samples_per_second": 9.175,
|
1678 |
+
"eval_webgpt_steps_per_second": 0.766,
|
1679 |
+
"step": 2500
|
1680 |
+
},
|
1681 |
+
{
|
1682 |
+
"epoch": 2.64,
|
1683 |
+
"eval_prompt_dialogue_accuracy": 0.5776905643690232,
|
1684 |
+
"eval_prompt_dialogue_loss": 1.7314453125,
|
1685 |
+
"eval_prompt_dialogue_runtime": 1054.4289,
|
1686 |
+
"eval_prompt_dialogue_samples_per_second": 9.777,
|
1687 |
+
"eval_prompt_dialogue_steps_per_second": 0.816,
|
1688 |
+
"step": 2500
|
1689 |
+
}
|
1690 |
+
],
|
1691 |
+
"max_steps": 2844,
|
1692 |
+
"num_train_epochs": 3,
|
1693 |
+
"total_flos": 147144228831232.0,
|
1694 |
+
"trial_name": null,
|
1695 |
+
"trial_params": null
|
1696 |
+
}
|
training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:eb9607fb0053bbe67a5f7e6792f0aaef0aa879aef697c0b5c8814f6482f93625
|
3 |
+
size 4719
|
zero_to_fp32.py
ADDED
@@ -0,0 +1,482 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 math
|
14 |
+
import os
|
15 |
+
import re
|
16 |
+
from collections import OrderedDict
|
17 |
+
|
18 |
+
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
19 |
+
# DeepSpeed data structures it has to be available in the current python environment.
|
20 |
+
from deepspeed.utils import logger
|
21 |
+
from deepspeed.checkpoint.constants import (DS_VERSION,
|
22 |
+
OPTIMIZER_STATE_DICT,
|
23 |
+
SINGLE_PARTITION_OF_FP32_GROUPS,
|
24 |
+
FP32_FLAT_GROUPS,
|
25 |
+
ZERO_STAGE,
|
26 |
+
PARTITION_COUNT,
|
27 |
+
PARAM_SHAPES,
|
28 |
+
BUFFER_NAMES)
|
29 |
+
|
30 |
+
debug = 0
|
31 |
+
|
32 |
+
# load to cpu
|
33 |
+
device = torch.device('cpu')
|
34 |
+
|
35 |
+
|
36 |
+
def atoi(text):
|
37 |
+
return int(text) if text.isdigit() else text
|
38 |
+
|
39 |
+
|
40 |
+
def natural_keys(text):
|
41 |
+
'''
|
42 |
+
alist.sort(key=natural_keys) sorts in human order
|
43 |
+
http://nedbatchelder.com/blog/200712/human_sorting.html
|
44 |
+
(See Toothy's implementation in the comments)
|
45 |
+
'''
|
46 |
+
return [atoi(c) for c in re.split(r'(\d+)', text)]
|
47 |
+
|
48 |
+
|
49 |
+
def get_model_state_file(checkpoint_dir, zero_stage):
|
50 |
+
if not os.path.isdir(checkpoint_dir):
|
51 |
+
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
52 |
+
|
53 |
+
# there should be only one file
|
54 |
+
if zero_stage == 2:
|
55 |
+
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
56 |
+
elif zero_stage == 3:
|
57 |
+
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
58 |
+
|
59 |
+
if not os.path.exists(file):
|
60 |
+
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
61 |
+
|
62 |
+
return file
|
63 |
+
|
64 |
+
|
65 |
+
def get_optim_files(checkpoint_dir):
|
66 |
+
# XXX: need to test that this simple glob rule works for multi-node setup too
|
67 |
+
optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
|
68 |
+
"*_optim_states.pt")),
|
69 |
+
key=natural_keys)
|
70 |
+
|
71 |
+
if len(optim_files) == 0:
|
72 |
+
raise FileNotFoundError(
|
73 |
+
f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
|
74 |
+
|
75 |
+
return optim_files
|
76 |
+
|
77 |
+
|
78 |
+
def parse_model_state(file):
|
79 |
+
state_dict = torch.load(file, map_location=device)
|
80 |
+
|
81 |
+
if BUFFER_NAMES not in state_dict:
|
82 |
+
raise ValueError(f"{file} is not a model state checkpoint")
|
83 |
+
buffer_names = state_dict[BUFFER_NAMES]
|
84 |
+
if debug:
|
85 |
+
print("Found buffers:", buffer_names)
|
86 |
+
|
87 |
+
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
88 |
+
buffers = {
|
89 |
+
k: v.float()
|
90 |
+
for k,
|
91 |
+
v in state_dict["module"].items() if k in buffer_names
|
92 |
+
}
|
93 |
+
param_shapes = state_dict[PARAM_SHAPES]
|
94 |
+
|
95 |
+
ds_version = state_dict.get(DS_VERSION, None)
|
96 |
+
|
97 |
+
return buffers, param_shapes, ds_version
|
98 |
+
|
99 |
+
|
100 |
+
def parse_optim_states(files, ds_checkpoint_dir):
|
101 |
+
|
102 |
+
total_files = len(files)
|
103 |
+
state_dicts = []
|
104 |
+
for f in files:
|
105 |
+
state_dicts.append(torch.load(f, map_location=device))
|
106 |
+
|
107 |
+
if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
|
108 |
+
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
109 |
+
zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
|
110 |
+
world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
|
111 |
+
|
112 |
+
# For ZeRO-2 each param group can have different partition_count as data parallelism for expert
|
113 |
+
# parameters can be different from data parallelism for non-expert parameters. So we can just
|
114 |
+
# use the max of the partition_count to get the dp world_size.
|
115 |
+
|
116 |
+
if type(world_size) is list:
|
117 |
+
world_size = max(world_size)
|
118 |
+
|
119 |
+
if world_size != total_files:
|
120 |
+
raise ValueError(
|
121 |
+
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
122 |
+
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
123 |
+
)
|
124 |
+
|
125 |
+
# the groups are named differently in each stage
|
126 |
+
if zero_stage == 2:
|
127 |
+
fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
|
128 |
+
elif zero_stage == 3:
|
129 |
+
fp32_groups_key = FP32_FLAT_GROUPS
|
130 |
+
else:
|
131 |
+
raise ValueError(f"unknown zero stage {zero_stage}")
|
132 |
+
|
133 |
+
if zero_stage == 2:
|
134 |
+
fp32_flat_groups = [
|
135 |
+
state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
|
136 |
+
for i in range(len(state_dicts))
|
137 |
+
]
|
138 |
+
elif zero_stage == 3:
|
139 |
+
# if there is more than one param group, there will be multiple flattened tensors - one
|
140 |
+
# flattened tensor per group - for simplicity merge them into a single tensor
|
141 |
+
#
|
142 |
+
# XXX: could make the script more memory efficient for when there are multiple groups - it
|
143 |
+
# will require matching the sub-lists of param_shapes for each param group flattened tensor
|
144 |
+
|
145 |
+
fp32_flat_groups = [
|
146 |
+
torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
|
147 |
+
0) for i in range(len(state_dicts))
|
148 |
+
]
|
149 |
+
|
150 |
+
return zero_stage, world_size, fp32_flat_groups
|
151 |
+
|
152 |
+
|
153 |
+
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
|
154 |
+
"""
|
155 |
+
Returns fp32 state_dict reconstructed from ds checkpoint
|
156 |
+
|
157 |
+
Args:
|
158 |
+
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
159 |
+
|
160 |
+
"""
|
161 |
+
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
162 |
+
|
163 |
+
optim_files = get_optim_files(ds_checkpoint_dir)
|
164 |
+
zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
165 |
+
print(
|
166 |
+
f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
167 |
+
|
168 |
+
model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
|
169 |
+
buffers, param_shapes, ds_version = parse_model_state(model_file)
|
170 |
+
print(f'Parsing checkpoint created by deepspeed=={ds_version}')
|
171 |
+
|
172 |
+
if zero_stage == 2:
|
173 |
+
return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
|
174 |
+
param_shapes,
|
175 |
+
fp32_flat_groups,
|
176 |
+
buffers)
|
177 |
+
elif zero_stage == 3:
|
178 |
+
return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
|
179 |
+
param_shapes,
|
180 |
+
fp32_flat_groups,
|
181 |
+
buffers)
|
182 |
+
|
183 |
+
|
184 |
+
def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
|
185 |
+
param_shapes,
|
186 |
+
fp32_flat_groups,
|
187 |
+
buffers):
|
188 |
+
|
189 |
+
# Reconstruction protocol:
|
190 |
+
#
|
191 |
+
# XXX: document this
|
192 |
+
|
193 |
+
if debug:
|
194 |
+
for i in range(world_size):
|
195 |
+
for j in range(len(fp32_flat_groups[0])):
|
196 |
+
print(
|
197 |
+
f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
|
198 |
+
|
199 |
+
# XXX: memory usage doubles here (zero2)
|
200 |
+
num_param_groups = len(fp32_flat_groups[0])
|
201 |
+
merged_single_partition_of_fp32_groups = []
|
202 |
+
for i in range(num_param_groups):
|
203 |
+
merged_partitions = [sd[i] for sd in fp32_flat_groups]
|
204 |
+
full_single_fp32_vector = torch.cat(merged_partitions, 0)
|
205 |
+
merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
|
206 |
+
avail_numel = sum([
|
207 |
+
full_single_fp32_vector.numel()
|
208 |
+
for full_single_fp32_vector in merged_single_partition_of_fp32_groups
|
209 |
+
])
|
210 |
+
|
211 |
+
if debug:
|
212 |
+
wanted_params = sum([len(shapes) for shapes in param_shapes])
|
213 |
+
wanted_numel = sum(
|
214 |
+
[sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
|
215 |
+
# not asserting if there is a mismatch due to possible padding
|
216 |
+
print(f"Have {avail_numel} numels to process.")
|
217 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
218 |
+
|
219 |
+
state_dict = OrderedDict()
|
220 |
+
|
221 |
+
# buffers
|
222 |
+
state_dict.update(buffers)
|
223 |
+
if debug:
|
224 |
+
print(f"added {len(buffers)} buffers")
|
225 |
+
|
226 |
+
# params
|
227 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
228 |
+
# out-of-core computing solution
|
229 |
+
total_numel = 0
|
230 |
+
total_params = 0
|
231 |
+
for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
|
232 |
+
offset = 0
|
233 |
+
avail_numel = full_single_fp32_vector.numel()
|
234 |
+
for name, shape in shapes.items():
|
235 |
+
|
236 |
+
unpartitioned_numel = shape.numel()
|
237 |
+
total_numel += unpartitioned_numel
|
238 |
+
total_params += 1
|
239 |
+
|
240 |
+
if debug:
|
241 |
+
print(
|
242 |
+
f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
|
243 |
+
)
|
244 |
+
state_dict[name] = full_single_fp32_vector.narrow(
|
245 |
+
0,
|
246 |
+
offset,
|
247 |
+
unpartitioned_numel).view(shape)
|
248 |
+
offset += unpartitioned_numel
|
249 |
+
|
250 |
+
# Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
|
251 |
+
# avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
|
252 |
+
# paddings performed in the code it's almost impossible to predict the exact numbers w/o the
|
253 |
+
# live optimizer object, so we are checking that the numbers are within the right range
|
254 |
+
align_to = 2 * world_size
|
255 |
+
|
256 |
+
def zero2_align(x):
|
257 |
+
return align_to * math.ceil(x / align_to)
|
258 |
+
|
259 |
+
if debug:
|
260 |
+
print(f"original offset={offset}, avail_numel={avail_numel}")
|
261 |
+
|
262 |
+
offset = zero2_align(offset)
|
263 |
+
avail_numel = zero2_align(avail_numel)
|
264 |
+
|
265 |
+
if debug:
|
266 |
+
print(f"aligned offset={offset}, avail_numel={avail_numel}")
|
267 |
+
|
268 |
+
# Sanity check
|
269 |
+
if offset != avail_numel:
|
270 |
+
raise ValueError(
|
271 |
+
f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
272 |
+
|
273 |
+
print(
|
274 |
+
f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
|
275 |
+
)
|
276 |
+
|
277 |
+
return state_dict
|
278 |
+
|
279 |
+
|
280 |
+
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
281 |
+
remainder = unpartitioned_numel % world_size
|
282 |
+
padding_numel = (world_size - remainder) if remainder else 0
|
283 |
+
partitioned_numel = math.ceil(unpartitioned_numel / world_size)
|
284 |
+
return partitioned_numel, padding_numel
|
285 |
+
|
286 |
+
|
287 |
+
def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
|
288 |
+
param_shapes,
|
289 |
+
fp32_flat_groups,
|
290 |
+
buffers):
|
291 |
+
|
292 |
+
# Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
|
293 |
+
# param, re-consolidating each param, while dealing with padding if any
|
294 |
+
|
295 |
+
avail_numel = fp32_flat_groups[0].numel() * world_size
|
296 |
+
# merge list of dicts, preserving order
|
297 |
+
param_shapes = {k: v for d in param_shapes for k, v in d.items()}
|
298 |
+
|
299 |
+
if debug:
|
300 |
+
for i in range(world_size):
|
301 |
+
print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
|
302 |
+
|
303 |
+
wanted_params = len(param_shapes)
|
304 |
+
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
305 |
+
# not asserting if there is a mismatch due to possible padding
|
306 |
+
print(f"Have {avail_numel} numels to process.")
|
307 |
+
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
308 |
+
|
309 |
+
state_dict = OrderedDict()
|
310 |
+
|
311 |
+
# buffers
|
312 |
+
state_dict.update(buffers)
|
313 |
+
if debug:
|
314 |
+
print(f"added {len(buffers)} buffers")
|
315 |
+
|
316 |
+
# params
|
317 |
+
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
318 |
+
# out-of-core computing solution
|
319 |
+
offset = 0
|
320 |
+
total_numel = 0
|
321 |
+
total_params = 0
|
322 |
+
for name, shape in param_shapes.items():
|
323 |
+
|
324 |
+
unpartitioned_numel = shape.numel()
|
325 |
+
total_numel += unpartitioned_numel
|
326 |
+
total_params += 1
|
327 |
+
|
328 |
+
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
329 |
+
|
330 |
+
if debug:
|
331 |
+
print(
|
332 |
+
f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
333 |
+
)
|
334 |
+
|
335 |
+
# XXX: memory usage doubles here
|
336 |
+
state_dict[name] = torch.cat(
|
337 |
+
tuple(fp32_flat_groups[i].narrow(0,
|
338 |
+
offset,
|
339 |
+
partitioned_numel)
|
340 |
+
for i in range(world_size)),
|
341 |
+
0).narrow(0,
|
342 |
+
0,
|
343 |
+
unpartitioned_numel).view(shape)
|
344 |
+
offset += partitioned_numel
|
345 |
+
|
346 |
+
offset *= world_size
|
347 |
+
|
348 |
+
# Sanity check
|
349 |
+
if offset != avail_numel:
|
350 |
+
raise ValueError(
|
351 |
+
f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
352 |
+
|
353 |
+
print(
|
354 |
+
f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
|
355 |
+
)
|
356 |
+
|
357 |
+
return state_dict
|
358 |
+
|
359 |
+
|
360 |
+
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
|
361 |
+
"""
|
362 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
363 |
+
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
364 |
+
via a model hub.
|
365 |
+
|
366 |
+
Args:
|
367 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder
|
368 |
+
- ``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``
|
369 |
+
|
370 |
+
Returns:
|
371 |
+
- pytorch ``state_dict``
|
372 |
+
|
373 |
+
Note: this approach may not work if your application doesn't have sufficient free CPU memory and
|
374 |
+
you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
375 |
+
the checkpoint.
|
376 |
+
|
377 |
+
A typical usage might be ::
|
378 |
+
|
379 |
+
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
380 |
+
# do the training and checkpoint saving
|
381 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
382 |
+
model = model.cpu() # move to cpu
|
383 |
+
model.load_state_dict(state_dict)
|
384 |
+
# submit to model hub or save the model to share with others
|
385 |
+
|
386 |
+
In this example the ``model`` will no longer be usable in the deepspeed context of the same
|
387 |
+
application. i.e. you will need to re-initialize the deepspeed engine, since
|
388 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
389 |
+
|
390 |
+
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
391 |
+
|
392 |
+
"""
|
393 |
+
if tag is None:
|
394 |
+
latest_path = os.path.join(checkpoint_dir, 'latest')
|
395 |
+
if os.path.isfile(latest_path):
|
396 |
+
with open(latest_path, 'r') as fd:
|
397 |
+
tag = fd.read().strip()
|
398 |
+
else:
|
399 |
+
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
400 |
+
|
401 |
+
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
402 |
+
|
403 |
+
if not os.path.isdir(ds_checkpoint_dir):
|
404 |
+
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
405 |
+
|
406 |
+
return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
|
407 |
+
|
408 |
+
|
409 |
+
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
|
410 |
+
"""
|
411 |
+
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
412 |
+
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
413 |
+
|
414 |
+
Args:
|
415 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
416 |
+
- ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
|
417 |
+
- ``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``
|
418 |
+
"""
|
419 |
+
|
420 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
421 |
+
print(f"Saving fp32 state dict to {output_file}")
|
422 |
+
torch.save(state_dict, output_file)
|
423 |
+
|
424 |
+
|
425 |
+
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
426 |
+
"""
|
427 |
+
1. Put the provided model to cpu
|
428 |
+
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
429 |
+
3. Load it into the provided model
|
430 |
+
|
431 |
+
Args:
|
432 |
+
- ``model``: the model object to update
|
433 |
+
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
434 |
+
- ``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``
|
435 |
+
|
436 |
+
Returns:
|
437 |
+
- ``model`: modified model
|
438 |
+
|
439 |
+
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
440 |
+
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
441 |
+
conveniently placed for you in the checkpoint folder.
|
442 |
+
|
443 |
+
A typical usage might be ::
|
444 |
+
|
445 |
+
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
446 |
+
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
447 |
+
# submit to model hub or save the model to share with others
|
448 |
+
|
449 |
+
Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
|
450 |
+
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
451 |
+
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
452 |
+
|
453 |
+
"""
|
454 |
+
logger.info(f"Extracting fp32 weights")
|
455 |
+
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
456 |
+
|
457 |
+
logger.info(f"Overwriting model with fp32 weights")
|
458 |
+
model = model.cpu()
|
459 |
+
model.load_state_dict(state_dict, strict=False)
|
460 |
+
|
461 |
+
return model
|
462 |
+
|
463 |
+
|
464 |
+
if __name__ == "__main__":
|
465 |
+
|
466 |
+
parser = argparse.ArgumentParser()
|
467 |
+
parser.add_argument(
|
468 |
+
"checkpoint_dir",
|
469 |
+
type=str,
|
470 |
+
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
471 |
+
parser.add_argument(
|
472 |
+
"output_file",
|
473 |
+
type=str,
|
474 |
+
help=
|
475 |
+
"path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
|
476 |
+
)
|
477 |
+
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
478 |
+
args = parser.parse_args()
|
479 |
+
|
480 |
+
debug = args.debug
|
481 |
+
|
482 |
+
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)
|