run
training
python generate_train_scripts.py
bash train_commands.sh
eval
python generate_evl_scripts.py
bash eval_commands.sh
data gen (ft)
python sample_pipe.py --base {base} --n_digits {2} --n_samples {500}
eval
python eval_pipe.py --base {base} --cot {True} --model_name {unsloth/Phi-4 or ftpath like outputs/test_run_8_cot_True_n_digits_2/checkpoint-60} --size {200}
train sft
python ft_lora.py --base {base} --model_path {unsloth/Phi-4} --cot {include or exclude} --data_file {ft_data/data_ft_{base}_{n_digits}.txt}