bf16: true
cutoff_len: 1024
dataset: mathinstruct
dataset_dir: data
ddp_timeout: 180000000
do_train: true
finetuning_type: lora
flash_attn: auto
gradient_accumulation_steps: 1
learning_rate: 0.0002
logging_steps: 5
lora_alpha: 16
lora_dropout: 0
lora_rank: 8
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 10000
model_name_or_path: meta-llama/Llama-3.2-1B
num_train_epochs: 3.0
optim: adamw_torch
output_dir: saves/Llama-3.2-1B/lora/llama3.2-1b
packing: false
per_device_train_batch_size: 16
plot_loss: true
preprocessing_num_workers: 16
report_to: none
rope_scaling: linear
save_steps: 100
stage: sft
template: default
trust_remote_code: true
warmup_steps: 0