Upload ms-swift/sft.sh with huggingface_hub

ms-swift/sft.sh

@@ -0,0 +1,29 @@
+# 22GB
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+swift sft \
+    --model /oss/wangyujia/BIO/pretrain_output/qwen2.5-7b-instruct-bio/bio_all/save1epoch/checkpoint-1300 \
+    --train_type lora \
+    --dataset /oss/wangyujia/pretrain-bench/locate/deeplocmulti/train.jsonl \
+    --torch_dtype bfloat16 \
+    --num_train_epochs 2 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --learning_rate 1e-4 \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --gradient_accumulation_steps 16 \
+    --eval_steps 5 \
+    --save_steps 5 \
+    --save_total_limit 5 \
+    --logging_steps 5 \
+    --max_length 8192 \
+    --output_dir /oss/wangyujia/BIO/sft/qwen2.5-7b-bio-deeplocmulti \
+    --system 'You are a helpful assistant.' \
+    --warmup_ratio 0.05 \
+    --dataloader_num_workers 8 \
+    --model_author swift \
+    --model_name qwen_bio_sft_deeplocmulti \
+    --save_only_model true
+
+       # --dataset /fs-computility/ai-shen/wangyujia/pretrain-bench/data/SciKnowEval_bio_swift_new_.jsonl \