mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-02 03:32:50 +08:00
21 lines
628 B
Bash
21 lines
628 B
Bash
#!/bin/bash
|
|
|
|
CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch \
|
|
--config_file ../accelerate/single_config.yaml \
|
|
../../src/train_bash.py \
|
|
--stage sft \
|
|
--do_predict \
|
|
--model_name_or_path ../../saves/LLaMA2-7B/full/sft \
|
|
--dataset alpaca_gpt4_en,glaive_toolcall \
|
|
--dataset_dir ../../data \
|
|
--template default \
|
|
--finetuning_type full \
|
|
--output_dir ../../saves/LLaMA2-7B/full/predict \
|
|
--overwrite_cache \
|
|
--overwrite_output_dir \
|
|
--cutoff_len 1024 \
|
|
--preprocessing_num_workers 16 \
|
|
--per_device_eval_batch_size 1 \
|
|
--max_samples 20 \
|
|
--predict_with_generate
|