tofu_Llama-3.1-8B-Instruct_forget05_NPO/NPO.log

[2026-08-13 20:32:37,656][model][INFO] - Setting pad_token as eos token: <|eot_id|>
[2026-08-13 20:32:37,656][__main__][INFO] - Enabling input_require_grads on LlamaForCausalLM for gradient checkpointing
[2026-08-13 20:32:41,236][evaluator][INFO] - Evaluations stored in the experiment directory: saves/unlearn/baselines/tofu/NPO_lr2e-05_beta0.1_alpha2_epoch10/Llama-3.1-8B-Instruct/forget05
[2026-08-13 20:34:24,404][trainer][INFO] - NPO Trainer loaded, output_dir: saves/unlearn/baselines/tofu/NPO_lr2e-05_beta0.1_alpha2_epoch10/Llama-3.1-8B-Instruct/forget05
[2026-08-13 20:34:24,405][trainer][INFO] - Method args: {'gamma': 1.0, 'alpha': 2, 'retain_loss_type': 'NLL', 'beta': 0.1}
[2026-08-13 20:34:24,405][__main__][INFO] - Training setup: epochs=10, batch_size=4, learning_rate=2e-05, gradient_accumulation_steps=4, weight_decay=0.01
[2026-08-24 14:06:12,102][model][INFO] - Setting pad_token as eos token: <|eot_id|>
[2026-08-24 14:06:12,103][__main__][INFO] - Enabling input_require_grads on LlamaForCausalLM for gradient checkpointing
[2026-08-24 14:06:15,484][evaluator][INFO] - Evaluations stored in the experiment directory: saves/unlearn/baselines/tofu/NPO/Llama-3.1-8B-Instruct/forget05
[2026-08-24 14:07:43,097][trainer][INFO] - NPO Trainer loaded, output_dir: saves/unlearn/baselines/tofu/NPO/Llama-3.1-8B-Instruct/forget05
[2026-08-24 14:07:43,099][trainer][INFO] - Method args: {'gamma': 1.0, 'alpha': 2, 'retain_loss_type': 'NLL', 'beta': 0.1}
[2026-08-24 14:07:43,099][__main__][INFO] - Training setup: epochs=10, batch_size=4, learning_rate=2e-05, gradient_accumulation_steps=4, weight_decay=0.01
[2026-08-24 15:02:45,283][model][INFO] - Setting pad_token as eos token: <|eot_id|>
[2026-08-24 15:02:45,284][__main__][INFO] - Enabling input_require_grads on LlamaForCausalLM for gradient checkpointing
[2026-08-24 15:02:48,348][evaluator][INFO] - Evaluations stored in the experiment directory: saves/unlearn/baselines/tofu/NPO/Llama-3.1-8B-Instruct/forget05
[2026-08-24 15:04:15,208][trainer][INFO] - NPO Trainer loaded, output_dir: saves/unlearn/baselines/tofu/NPO/Llama-3.1-8B-Instruct/forget05
[2026-08-24 15:04:15,209][trainer][INFO] - Method args: {'gamma': 1.0, 'alpha': 2, 'retain_loss_type': 'NLL', 'beta': 0.1}
[2026-08-24 15:04:15,209][__main__][INFO] - Training setup: epochs=10, batch_size=4, learning_rate=2e-05, gradient_accumulation_steps=4, weight_decay=0.01