flypythoncom/GPT2-Chinese
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update to transformers

Browse Source
This commit is contained in:
Duzeyao
2019-10-25 23:37:34 +08:00
parent f4ab09186c
commit 9b630a53fe
8 changed files with 20 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
train_single.py
View File

@@ -1,4 +1,4 @@
import pytorch_transformers
import transformers
import torch
import os
import json
@@ -69,7 +69,7 @@ def main():
from tokenizations import tokenization_bert
os.environ["CUDA_VISIBLE_DEVICES"] = args.device # 此处设置程序使用哪些显卡
model_config = pytorch_transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
model_config = transformers.modeling_gpt2.GPT2Config.from_json_file(args.model_config)
print('config:\n' + model_config.to_json_string())
n_ctx = model_config.n_ctx
@@ -101,9 +101,9 @@ def main():
print('files built')
if not args.pretrained_model:
model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel(config=model_config)
model = transformers.modeling_gpt2.GPT2LMHeadModel(config=model_config)
else:
model = pytorch_transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
model = transformers.modeling_gpt2.GPT2LMHeadModel.from_pretrained(args.pretrained_model)
model.train()
model.to(device)
multi_gpu = False
@@ -115,8 +115,8 @@ def main():
total_steps = int(full_len / stride * epochs / batch_size / gradient_accumulation)
print('total steps = {}'.format(total_steps))
optimizer = pytorch_transformers.AdamW(model.parameters(), lr=lr, correct_bias=True)
scheduler = pytorch_transformers.WarmupLinearSchedule(optimizer, warmup_steps=warmup_steps,
optimizer = transformers.AdamW(model.parameters(), lr=lr, correct_bias=True)
scheduler = transformers.WarmupLinearSchedule(optimizer, warmup_steps=warmup_steps,
t_total=total_steps)
if fp16:
try: