mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-02 19:52:50 +08:00
37 lines
926 B
Bash
37 lines
926 B
Bash
# BAdam layer-wise
|
|
export CUDA_VISIBLE_DEVICES=0
|
|
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
|
|
python ../../../src/train_bash.py \
|
|
--stage sft \
|
|
--do_train \
|
|
--model_name_or_path meta-llama/Llama-2-7b-hf \
|
|
--dataset alpaca_gpt4_en,glaive_toolcall \
|
|
--dataset_dir ../../../data \
|
|
--template default \
|
|
--finetuning_type full \
|
|
--output_dir ../../../saves/LLaMA2-7B/badam \
|
|
--overwrite_cache \
|
|
--overwrite_output_dir \
|
|
--cutoff_len 1024 \
|
|
--preprocessing_num_workers 32 \
|
|
--per_device_train_batch_size 8 \
|
|
--per_device_eval_batch_size 5 \
|
|
--gradient_accumulation_steps 2 \
|
|
--lr_scheduler_type cosine \
|
|
--logging_steps 10 \
|
|
--warmup_steps 20 \
|
|
--save_steps 100 \
|
|
--eval_steps 100 \
|
|
--evaluation_strategy steps \
|
|
--load_best_model_at_end \
|
|
--learning_rate 5e-5 \
|
|
--num_train_epochs 3.0 \
|
|
--val_size 0.1 \
|
|
--plot_loss \
|
|
--use_badam \
|
|
--switch_mode descending \
|
|
--badam_verbose 2 \
|
|
--switch_block_every 50 \
|
|
--pure_bf16 \
|
|
|