stage: sft
do_train: true
max_steps: 100
do_eval: false
save_strategy: steps
save_steps: 50
logging_steps: 10
fp16: true
bf16: false
overwrite_output_dir: true
per_device_train_batch_size: 1
gradient_accumulation_steps: 1
gradient_checkpointing: true
model_name_or_path: Qwen/Qwen2.5-0.5B
finetuning_type: lora
dataset: my_custom_sft
dataset_dir: /scratch/zrs2020/LlamaFactoryHelper/LLaMA-Factory/data
template: default
cutoff_len: 8096
val_size: 0.1
lora_rank: 8
lora_alpha: 16
lora_dropout: 0.05
lora_target: all
output_dir: /scratch/zrs2020/LlamaFactoryHelper/experiments/lf_torch_test__local/checkpoints