- experiment=unlearn/tofu/default.yaml
- trainer=SimNPO
- task_name=tofu_Llama-3.2-1B-Instruct_forget10_SimNPO_qat-int4
- model=Llama-3.2-1B-Instruct
- forget_split=forget10
- retain_split=retain90
- holdout_split=holdout10
- model.model_args.pretrained_model_name_or_path=open-unlearning/tofu_Llama-3.2-1B-Instruct_full
- model.model_args.attn_implementation=sdpa
- trainer.args.per_device_train_batch_size=4
- trainer.args.gradient_accumulation_steps=4
- trainer.args.ddp_find_unused_parameters=true
- trainer.args.do_eval=false
- trainer.args.save_strategy=no
- +trainer.args.push_to_hub=true
- +trainer.args.hub_model_id=Jeesup/tofu_Llama-3.2-1B-Instruct_forget10_SimNPO_qat-int4
- +trainer.args.hub_strategy=end
- +trainer.args.hub_private_repo=false
- paths.output_dir=saves/unlearn/qat-baseline/tofu_Llama-3.2-1B-Instruct_forget10_SimNPO_qat-int4
- +quant=qat_w4