{"repo": "ab-rin/ecpmg", "files": {"configs/cliffwalking/ecmpg_ppo_blinded.yaml": {"088483bc38df7bef35774d8c4f7ce2b95fc9970d": {"t": 1786068218.2651966, "n": 1035, "h": "2f3a50abc80271f8", "c": "task: ecmpg_ppo_blinded\ngym_env_name: CliffWalking-v1\nlogdir: logs/ecmpg_ppo_blinded_cliffwalking\n\n# Training schedule\nnum_updates: 400\nepisodes_per_update: 32\nmax_steps: 200\ngamma: 0.99\n\n# PPO / GAE\ngae_lambda: 0.95\nclip_epsilon: 0.2\nppo_epochs: 4\nminibatch_size: 64\nvf_coef: 0.5\n\n# Entropy annealing: 0.02 → 0.005 over training\nentropy_coef: 0.02\nentropy_coef_final: 0.005\n\n# Aux scale annealing: 0.1 → 0.01 over training\naux_scale: 0.1\naux_scale_final: 0.01\n\n# Exploration\nintrinsic_beta: 0.1          # count-based intrinsic: beta / sqrt(count+1)\nepsilon_rollout: 0.05        # epsilon-greedy during rollout\nepsilon_rollout_final: 0.01  # anneals linearly to this\n\n# LLM module weighting\nllm_interval: 20\nllm_model_name: deepseek-chat   # set DEEPSEEK_API_KEY in .env\nlambda_mode: llm                # llm | fixed | random | heuristic\nmin_exploration_lambda: 0.3    # floor on exploration lambda\nuse_llm: true\nfallback_on_llm_error: true\n\n# Policy network\nlr: 0.0003\nhidden_dim: 64\n\n# Blinder\nblinder_seed: 42\nreward_binning: true\n"}}}}