cutoff_len: 1024
dataset: alpaca_en
dataset_dir: data
do_train: true
finetuning_type: lora
flash_attn: auto
fp16: true
gradient_accumulation_steps: 8
learning_rate: 0.0002
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: 500
model_name_or_path: microsoft/phi-1_5
num_train_epochs: 3.0
optim: adamw_torch
output_dir: saves/Phi-1.5-1.3B/lora/train_2024-05-09-19-57-19
packing: false
per_device_train_batch_size: 2
report_to: none
save_steps: 100
stage: sft
template: default
warmup_steps: 0