DeepSeek/DeepSeek-Coder
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepSeek-Coder synced 2025-06-26 18:25:53 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Update finetune_deepseekcoder.py

Browse Source 
Using torch.float16 or torch.cuda.amp can significantly reduce memory usage and speed up training by performing computations with lower precision.
This commit is contained in:
Rahul Dubey
2025-01-29 12:38:12 +05:30
committed by GitHub
parent b7ba565956
commit 78d0fd332a
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
finetune/finetune_deepseekcoder.py
View File

@@ -143,7 +143,7 @@ def train():
model = transformers.AutoModelForCausalLM.from_pretrained( model = transformers.AutoModelForCausalLM.from_pretrained(
model_args.model_name_or_path, model_args.model_name_or_path,
torch_dtype=torch.bfloat16 torch_dtype=torch.float16
) )
if training_args.local_rank == 0: if training_args.local_rank == 0: