cutoff_len: 1024
dataset: Reseacrhed_data
dataset_dir: data
do_train: true
finetuning_type: lora
flash_attn: auto
fp16: true
gradient_accumulation_steps: 8
learning_rate: 2.0e-05
logging_steps: 5
lora_alpha: 16
lora_dropout: 0
lora_rank: 8
lora_target: q_proj,v_proj
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 10000
model_name_or_path: TinyLlama/TinyLlama-1.1B-Chat-v1.0
num_train_epochs: 2.0
optim: adamw_torch
output_dir: saves/LLaMA-tiny/lora/train_2024-05-05-10-57-50
packing: true
per_device_train_batch_size: 2
quantization_bit: 8
report_to: none
save_steps: 100
stage: pt
template: default
warmup_steps: 0