# config for https://github.com/hiyouga/LLaMA-Factory
### model
model_name_or_path: Qwen/Qwen2-7B-Instruct
### method
stage: sft
do_train: true
finetuning_type: lora
lora_target: all

### dataset
dataset_dir: data
dataset: llm-complex-reasoning-train-qwen2-72b-instruct-correct,Infinity-Instruct-0625
template: qwen
cutoff_len: 2048
max_samples: 128000
overwrite_cache: false
preprocessing_num_workers: 8
# deepspeed: ./LLaMA-Factory/examples/deepspeed/ds_z2_config.json

### output
output_dir: output/qwen2-7b-instruct/sft-lora
logging_steps: 5
save_steps: 300
plot_loss: true
overwrite_output_dir: true

### train
per_device_train_batch_size: 8
gradient_accumulation_steps: 8
learning_rate: 5.0e-5
num_train_epochs: 2.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

### eval
val_size: 0.1
per_device_eval_batch_size: 4
eval_strategy: steps
eval_steps: 100