Benchmark-v0 / instruments_train.sh
JayceAnova's picture
Update instruments_train.sh
d3f1650 verified
export WANDB_MODE=disabled
export CUDA_LAUNCH_BLOCKING=1
ROOT=/home/jovyan/workspace
DATASET=Instruments
BASE_MODEL=$ROOT/Llama-2-7b-hf
DATA_PATH=$ROOT
INDEX=$ROOT/$DATASET/Instruments.index.json
OUTPUT_DIR=./Ins/Llama-2-7b
mkdir -p $OUTPUT_DIR
torchrun --nproc_per_node=8 finetune.py \
--base_model $BASE_MODEL \
--output_dir $OUTPUT_DIR \
--dataset $DATASET \
--data_path $DATA_PATH \
--per_device_batch_size 6 \
--gradient_accumulation_steps 2 \
--learning_rate 5e-5 \
--epochs 4 \
--weight_decay 0.01 \
--save_and_eval_strategy epoch \
--bf16 \
--deepspeed ./config/ds_z2_bf16.json \
--dataloader_num_workers 4 \
--only_train_response \
--tasks seqrec,item2index,index2item,fusionseqrec,itemsearch,preferenceobtain \
--train_prompt_sample_num 1,1,1,1,1,1 \
--train_data_sample_num 0,0,0,0,0,0 \
--index_file $INDEX
cd convert
nohup ./convert.sh $OUTPUT_DIR >convert.log 2>&1 &
cd ..