Fix SFT fine tune output dir and CUDA DEVICE NUM

This commit is contained in:
Xiang Long 2024-02-01 10:07:33 +08:00
parent bd80f792e4
commit 213716ff0a

View File

@ -2,9 +2,9 @@ formatted_time=$(date +"%Y%m%d%H%M%S")
echo $formatted_time
deepspeed --include localhost:1,2,4,5 finetune.py \
deepspeed --include localhost:0,1,2,3 finetune.py \
--model_name_or_path <your_model_name_or_path> \
--output_dir output/AdvertiseGenLoRA/$formatted_time/ \
--output_dir output/AdvertiseGenSFT/$formatted_time/ \
--train_data_path data/AdvertiseGenChatML/train.json \
--eval_data_path data/AdvertiseGenChatML/dev.json \
--learning_rate 1e-3 --per_device_train_batch_size 1 \