Run dir: [v2_runs]/selfhost_qwen36_23_BetV1_20260422_091648_7241305
| target | program/profile | gen | n_successful | rank_key | best_i_pae | best_plddt | description |
|---|---|---|---|---|---|---|---|
| 23_BetV1 | 3b991ed25262 | 0 | 0 | 0.33225274 | 0.92051405 | seed (default_deep heuristic baseline) | |
| 23_BetV1 | 65d030db4a34 | 1 | 14 | 0.19169782 | 0.92294097 | Two-stage budgeting: stage 0 cheap exploration (batch=4, nsamples=8, 200 steps), stage 1 clone for quality refinement (batch=8, nsamples=4, 400 steps). | |
| 23_BetV1 | 8df6a6e42769 | 2 | Increase beam width to 12 in both stages and strengthen iPAE reward weight to -1.5 to improve top-end structural quality. | ||||
| 23_BetV1 | ef3ab39d6fbc | 2 | 8 | 0.19340725 | 0.92051405 | Strengthen i_pae and plddt reward weights to -3.0 and 1.5, raise filter threshold to -1.5 to retain top-25%, and append stage 3 as deep refinement gated on stage 2 passing i_pae... | |
| 23_BetV1 | f55d68a576ff | 1 | 0 | Switch to diverse-beam-search with doubled beam width and higher diversity weight, and add contact rewards to guide interface packing and lower iPAE. |
Generated 2026-04-25