Racktic/alchemy-eval-logs

Eval logs + exact oracle for zero-shot Symbolic Alchemy (no-summary / summary-replace / summary-augment). Data: <model>/<condition>/{results.jsonl,summary.json,normalized.json,traj/}; oracle_all_episodes.json (exact oracle, ALL 1000); difficulty_index.json; hard_set_20.json. Analysis: QWEN_FAILURE_MODES.md, QWEN_SUMMARY_DRIFT.md. Full method/results below. Symbolic Alchemy — Memory-RL: training + eval trajectories This repo now holds BOTH (1) the frontier-model eval baseline… See the full description on the dataset page: https://huggingface.co/datasets/Racktic/alchemy-eval-logs.

<iframe src="https://silytics.si/embed/item/9c2a943e-733d-409b-89b0-50fb553dd683" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.