Environment

At least 2 * A800 is required. Preferable 8 * A800.

pip install requirements.txt

Training

torchrun --nproc_per_node=8 --master_port=1234 finetune.py \
    --base_model garage-bAInd/Platypus2-70B-instruct \
    --data_path Beauty \
    --task_type sequential \
    --cache_dir cache_dir/ \
    --output_dir output_dir/ \
    --batch_size 16 \
    --micro_batch_size 1 \
    --num_epochs 3 \
    --learning_rate 0.0003 \
    --cutoff_len 4096 \
    --val_set_size 0 \
    --lora_r 16 \
    --lora_alpha 16 \
    --lora_dropout 0.05 \
    --lora_target_modules '[gate_proj, down_proj, up_proj]' \
    --train_on_inputs False \
    --add_eos_token False \
    --group_by_length False \
    --prompt_template_name alpaca \
    --lr_scheduler 'cosine' \
    --warmup_steps 100

Inference

torchrun --nproc_per_node=8 --master_port=1234 inference.py \
    --base_model garage-bAInd/Platypus2-70B-instruct \
    --data_path Beauty \
    --task_type sequential \
    --checkpoint_dir checkpoint_dir \
    --cache_dir cache_dir/ \
    --output_dir output_dir/ \
    --batch_size 16 \
    --micro_batch_size 1