99 lines
2.8 KiB
JSON
99 lines
2.8 KiB
JSON
|
{
|
||
|
|
"model": "Qwen/Qwen2.5-7B-Instruct",
|
||
|
|
"output_dir": "qwen-apostate",
|
||
|
|
"profile": "balanced",
|
||
|
|
"device": "cuda",
|
||
|
|
"load_in_4bit": true,
|
||
|
|
"compute_dtype": "bfloat16",
|
||
|
|
"seed": 0,
|
||
|
|
"resume": false,
|
||
|
|
"cache_activations": true,
|
||
|
|
"activation_cache_dir": null,
|
||
|
|
"harmful_path": "mlabonne/harmful_behaviors:train:text|/host/data/harmful.txt",
|
||
|
|
"harmless_path": "mlabonne/harmless_alpaca:train:text|/host/data/harmless.txt",
|
||
|
|
"harmful_test": "mlabonne/harmful_behaviors:test:text|JailbreakBench/JBB-Behaviors@behaviors:harmful:Goal|/host/data/refusal_calibration.txt",
|
||
|
|
"harmless_test": "mlabonne/harmless_alpaca:test:text",
|
||
|
|
"preserve_path": null,
|
||
|
|
"n_harmful": 600,
|
||
|
|
"n_harmless": 600,
|
||
|
|
"n_eval": 300,
|
||
|
|
"max_new_tokens": 32,
|
||
|
|
"batch_size": 24,
|
||
|
|
"baseline_eval_n": 24,
|
||
|
|
"head_sweep": true,
|
||
|
|
"head_sweep_min": 3.5,
|
||
|
|
"head_sweep_max": 5.5,
|
||
|
|
"head_sweep_step": 0.5,
|
||
|
|
"head_sweep_top_k": 6,
|
||
|
|
"head_sweep_probe_n": 8,
|
||
|
|
"head_sweep_eval_n": 48,
|
||
|
|
"head_sweep_probe_classifier": false,
|
||
|
|
"fit_response_activations": false,
|
||
|
|
"fit_response_n": 160,
|
||
|
|
"fit_response_tokens": 32,
|
||
|
|
"refusal_rank": 1,
|
||
|
|
"variance_threshold": 0.9,
|
||
|
|
"max_rank": 1,
|
||
|
|
"direction_layer_frac": 0.6,
|
||
|
|
"direction_scope": "global",
|
||
|
|
"causal_targeting": true,
|
||
|
|
"causal_floor": 0.1,
|
||
|
|
"causal_temperature": 1.0,
|
||
|
|
"preserve_rank": 8,
|
||
|
|
"refine_refusal": true,
|
||
|
|
"refine_max_scale": 2.0,
|
||
|
|
"refine_steps": 6,
|
||
|
|
"refine_deescalate": true,
|
||
|
|
"refine_kl_steps": 10,
|
||
|
|
"refine_scale_rerank_k": 2,
|
||
|
|
"refine_kl_layer_steps": 10,
|
||
|
|
"refine_kl_layer_candidates": 8,
|
||
|
|
"repair_steps": 10,
|
||
|
|
"repair_candidates": 10,
|
||
|
|
"repair_rerank_k": 4,
|
||
|
|
"repair_probe_candidates": 24,
|
||
|
|
"repair_probe_ref_n": 12,
|
||
|
|
"repair_probe_kl_n": 16,
|
||
|
|
"repair_probe_positions": 8,
|
||
|
|
"repair_refusal_regress_slack": 0.01,
|
||
|
|
"repair_stop_kl_frac": 0.8,
|
||
|
|
"repair_min_alpha": 0.001,
|
||
|
|
"repair_min_kl_gain": 0.003,
|
||
|
|
"repair_min_refusal_gain": 0.005,
|
||
|
|
"repair_min_score_gain": 0.01,
|
||
|
|
"repair_eval_n": 48,
|
||
|
|
"repair_kl_n": 64,
|
||
|
|
"refine_refusal_slack": 0.01,
|
||
|
|
"guard_max_iters": 2,
|
||
|
|
"guard_leakage_eps": 0.15,
|
||
|
|
"guard_alpha_step": 0.25,
|
||
|
|
"optimize": true,
|
||
|
|
"n_trials": 16,
|
||
|
|
"adaptive_trials": true,
|
||
|
|
"kl_weight": 3.0,
|
||
|
|
"kl_target": 0.06,
|
||
|
|
"kl_target_weight": 10.0,
|
||
|
|
"kl_quad_weight": 14.0,
|
||
|
|
"kl_over_budget_weight": 36.0,
|
||
|
|
"refusal_target_weight": 4.0,
|
||
|
|
"refusal_quad_weight": 8.0,
|
||
|
|
"kl_positions": 32,
|
||
|
|
"opt_capability": true,
|
||
|
|
"opt_capability_weight": 1.0,
|
||
|
|
"opt_capability_code_n": 4,
|
||
|
|
"opt_capability_math_n": 4,
|
||
|
|
"opt_eval_n": 32,
|
||
|
|
"opt_gen_tokens": 32,
|
||
|
|
"opt_objective": "generation",
|
||
|
|
"opt_rerank_k": 3,
|
||
|
|
"opt_guard": true,
|
||
|
|
"opt_early_stop": true,
|
||
|
|
"opt_early_stop_margin": 0.02,
|
||
|
|
"prune": false,
|
||
|
|
"prune_max_frac": 0.25,
|
||
|
|
"prune_kl": 0.04,
|
||
|
|
"max_kl": 0.16,
|
||
|
|
"target_refusal": 0.03,
|
||
|
|
"save_dtype": "bfloat16",
|
||
|
|
"bake": true
|
||
|
|
}
|