cutoff_len: 1024
dataset: truth_dev_0716
dataset_dir: data
do_predict: true
finetuning_type: full
flash_attn: auto
max_new_tokens: 512
max_samples: 100000
model_name_or_path: saves/LLaMA3-8B-Chat/full/train_2024-07-23-06-00-05_llama3
output_dir: saves/LLaMA3-8B-Chat/full/eval_2024-07-23-06-52-5_llama3
per_device_eval_batch_size: 2
predict_with_generate: true
preprocessing_num_workers: 16
quantization_method: bitsandbytes
stage: sft
temperature: 0.95
template: llama3
top_p: 0.7