hailin
/
jd_train
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

.

This commit is contained in:
hailin 2025-08-26 20:02:49 +08:00
parent ef116d1bc8
commit 619bebcf57
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
train_sft_ds.py
View File

@ -634,7 +634,9 @@ def main():
save_steps=args.save_steps, save_steps=args.save_steps,
save_total_limit=2, save_total_limit=2,
deepspeed=(args.deepspeed if args.deepspeed and os.path.isfile(args.deepspeed) else None), deepspeed=(args.deepspeed if args.deepspeed and os.path.isfile(args.deepspeed) else None),
dataloader_drop_last=False, # 关键：别丢尾，避免空 batch # dataloader_drop_last=False, # 关键：别丢尾，避免空 batch
dataloader_drop_last=True, # 关键：别丢尾，避免空 batch
dispatch_batches=False,
dataloader_num_workers=0, dataloader_num_workers=0,
dataloader_prefetch_factor=None, dataloader_prefetch_factor=None,
dataloader_pin_memory=False, dataloader_pin_memory=False,