### model
model_name_or_path: shenzhi-wang/Llama3.1-8B-Chinese-Chat

### method
stage: sft
do_train: true
finetuning_type: lora
lora_target: all
# quantization_bit: 4                   
loraplus_lr_ratio: 16.0                 # use LoRA+ with lambda=16.0
upcast_layernorm: true

### dataset
dataset: alpaca_mgtv_p2
template: llama3
cutoff_len: 4096
max_samples: 25000
overwrite_cache: true
preprocessing_num_workers: 16

### output
output_dir: saves/llama3.1_8b/lora/sft_p2
logging_steps: 175
save_steps: 35
plot_loss: true
resume_from_checkpoint: True

### train
per_device_train_batch_size: 16
gradient_accumulation_steps: 8
learning_rate: 1.0e-4
num_train_epochs: 1.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

### eval
val_size: 0.1
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 35

report_to: none
run_name: llama3.1_8b_p2 # optional