Upload train_multi_0725.sh

train_multi_0725.sh

@@ -0,0 +1,170 @@
+export WANDB_MODE=online
+
+cd /workspace/medvicuna
+
+torchrun --nproc_per_node=7 --master_port=20001 fastchat/train/train_eval_mem.py \
+    --model_name_or_path lmsys/vicuna-33b-v1.3 \
+    --data_path /workspace/medvicuna/33b/medqa_opt4.json \
+    --eval_data_path /workspace/medvicuna/33b/opt4_medQA_usmle_dev_plus_1272_vicunatplt.json \
+    --cache_dir /workspace/.cache \
+    --push_to_hub False \
+    --bf16 True \
+    --output_dir output_vicuna_33b_medqa_opt4 \
+    --num_train_epochs 6 \
+    --per_device_train_batch_size 8 \
+    --per_device_eval_batch_size 1 \
+    --gradient_accumulation_steps 2 \
+    --evaluation_strategy "steps" \
+    --eval_steps 150 \
+    --save_strategy "steps" \
+    --save_steps 300 \
+    --save_total_limit 100 \
+    --learning_rate 2e-5 \
+    --weight_decay 0. \
+    --warmup_ratio 0.04 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 1 \
+    --fsdp "full_shard auto_wrap offload" \
+    --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
+    --tf32 True \
+    --model_max_length 1024 \
+    --gradient_checkpointing True \
+    --lazy_preprocess True &>> /workspace/medvicuna/logs/output_vicuna_33b_medqa_opt4.log
+
+sleep 120s
+
+bash /workspace/medvicuna/push_to_hub.sh https://huggingface.co/s1ghhh/vicuna_33b_medqa_opt4_0725 /workspace/medvicuna/output_vicuna_33b_medqa_opt4 &
+
+torchrun --nproc_per_node=8 --master_port=20001 fastchat/train/train_eval_mem.py \
+    --model_name_or_path lmsys/vicuna-13b-v1.3 \
+    --data_path /workspace/medvicuna/33b/medqa_opt4.json \
+    --eval_data_path /workspace/medvicuna/33b/opt4_medQA_usmle_dev_plus_1272_vicunatplt.json \
+    --cache_dir /workspace/.cache \
+    --push_to_hub False \
+    --bf16 True \
+    --output_dir output_vicuna_13b_medqa_opt4 \
+    --num_train_epochs 6 \
+    --per_device_train_batch_size 16 \
+    --per_device_eval_batch_size 1 \
+    --gradient_accumulation_steps 1 \
+    --evaluation_strategy "steps" \
+    --eval_steps 150 \
+    --save_strategy "steps" \
+    --save_steps 300 \
+    --save_total_limit 100 \
+    --learning_rate 2e-5 \
+    --weight_decay 0. \
+    --warmup_ratio 0.04 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 1 \
+    --fsdp "full_shard auto_wrap" \
+    --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
+    --tf32 True \
+    --model_max_length 1024 \
+    --gradient_checkpointing True \
+    --lazy_preprocess True &>> /workspace/medvicuna/logs/output_vicuna_13b_medqa_opt4.log
+
+sleep 120s
+
+bash /workspace/medvicuna/push_to_hub.sh https://huggingface.co/s1ghhh/vicuna_13b_medqa_opt4_0725 /workspace/medvicuna/output_vicuna_13b_medqa_opt4 &
+
+torchrun --nproc_per_node=7 --master_port=20001 fastchat/train/train_eval_mem.py \
+    --model_name_or_path lmsys/vicuna-33b-v1.3 \
+    --data_path /workspace/medvicuna/33b/medqa_opt4.json \
+    --eval_data_path /workspace/medvicuna/33b/opt4_medQA_usmle_dev_plus_1272_vicunatplt.json \
+    --cache_dir /workspace/.cache \
+    --push_to_hub False \
+    --bf16 True \
+    --output_dir output_vicuna_33b_medqa_opt4 \
+    --num_train_epochs 6 \
+    --per_device_train_batch_size 4 \
+    --per_device_eval_batch_size 1 \
+    --gradient_accumulation_steps 2 \
+    --evaluation_strategy "steps" \
+    --eval_steps 150 \
+    --save_strategy "steps" \
+    --save_steps 300 \
+    --save_total_limit 100 \
+    --learning_rate 2e-5 \
+    --weight_decay 0. \
+    --warmup_ratio 0.04 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 1 \
+    --fsdp "full_shard auto_wrap offload" \
+    --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
+    --tf32 True \
+    --model_max_length 1024 \
+    --gradient_checkpointing True \
+    --lazy_preprocess True &>> /workspace/medvicuna/logs/output_vicuna_33b_medqa_opt4.log
+
+sleep 120s
+
+bash /workspace/medvicuna/push_to_hub.sh https://huggingface.co/s1ghhh/vicuna_33b_medqa_opt4_0725 /workspace/medvicuna/output_vicuna_33b_medqa_opt4 &
+
+
+
+torchrun --nproc_per_node=7 --master_port=20001 fastchat/train/train_eval_mem.py \
+    --model_name_or_path lmsys/vicuna-33b-v1.3 \
+    --data_path /workspace/medvicuna/33b/medqa_opt4_aug.json \
+    --eval_data_path /workspace/medvicuna/33b/opt4_medQA_usmle_dev_plus_1272_vicunatplt.json \
+    --cache_dir /workspace/.cache \
+    --push_to_hub False \
+    --bf16 True \
+    --output_dir output_vicuna_33b_medqa_aug_opt4 \
+    --num_train_epochs 6 \
+    --per_device_train_batch_size 4 \
+    --per_device_eval_batch_size 1 \
+    --gradient_accumulation_steps 2 \
+    --evaluation_strategy "steps" \
+    --eval_steps 150 \
+    --save_strategy "steps" \
+    --save_steps 300 \
+    --save_total_limit 100 \
+    --learning_rate 2e-5 \
+    --weight_decay 0. \
+    --warmup_ratio 0.04 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 1 \
+    --fsdp "full_shard auto_wrap offload" \
+    --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
+    --tf32 True \
+    --model_max_length 2048 \
+    --gradient_checkpointing True \
+    --lazy_preprocess True &>> /workspace/medvicuna/logs/output_vicuna_33b_medqa_aug_opt4.log
+
+sleep 120s
+
+nohup bash /workspace/medvicuna/push_to_hub.sh https://huggingface.co/s1ghhh/vicuna_33b_medqa_opt4_aug_0725 /workspace/medvicuna/output_vicuna_33b_medqa_aug_opt4 &
+#
+#
+#torchrun --nproc_per_node=7 --master_port=20001 fastchat/train/train_eval_mem.py \
+#    --model_name_or_path lmsys/vicuna-33b-v1.3 \
+#    --data_path /workspace/medvicuna/33b/medvicuna_720k_0722.json \
+#    --eval_data_path /workspace/medvicuna/33b/opt4_medQA_usmle_dev_plus_1272_vicunatplt.json \
+#    --cache_dir /workspace/.cache \
+#    --push_to_hub False \
+#    --bf16 True \
+#    --output_dir output_vicuna_33b_medqa_opt4 \
+#    --num_train_epochs 8 \
+#    --per_device_train_batch_size 1 \
+#    --per_device_eval_batch_size 1 \
+#    --gradient_accumulation_steps 16 \
+#    --evaluation_strategy "steps" \
+#    --eval_steps 150 \
+#    --save_strategy "no" \
+#    --save_steps 300 \
+#    --save_total_limit 100 \
+#    --learning_rate 2e-5 \
+#    --weight_decay 0. \
+#    --warmup_ratio 0.04 \
+#    --lr_scheduler_type "cosine" \
+#    --logging_steps 1 \
+#    --fsdp "full_shard auto_wrap" \
+#    --fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
+#    --tf32 True \
+#    --model_max_length 2048 \
+#    --gradient_checkpointing True \
+#    --lazy_preprocess True &>> /workspace/medvicuna/logs/output_vicuna_33b_720k.log
+#
+#
+#nohup bash /workspace/medvicuna/push_to_hub.sh https://huggingface.co/s1ghhh/medvicuna_v1.1_13b_0725 /workspace/medvicuna/output_medvicuna_v1.1_13b &