selfhost_qwen36_05_CD45_20260423_222309_7307347

selfhost summary report

Run dir: [v2_runs]/selfhost_qwen36_05_CD45_20260423_222309_7307347

targetprogram/profilegenn_successfulrank_keybest_i_paebest_plddtdescription
05_CD454ed5ce55c01010Switch stage 0 to best-of-N with 32 replicas for broad exploration, append stage 1 with sequence hallucination refinement gated on promising candidates.
05_CD4560fa7d6426ef000.246973570.9661878seed (balanced baseline)
05_CD458ee7f9316dbf10Switch from beam-search to diverse-beam-search with beam_width=12 and diversity_weight=0.5 for broader exploration, while doubling i_pae reward weight to -2.0 to push optimizati...

Generated 2026-04-25