Oct 13 8 days ago 06:52 Results README: setup, numbers, findings, caveats, reproduction 06:49 Upload results/full/answers.json with huggingface_hub 06:49 Upload results/full/retrieval.json with huggingface_hub 06:49 Upload results/full/per_row.json with huggingface_hub 06:49 Upload results/full/metrics.json with huggingface_hub 06:49 Upload results/full/artifacts/answers.json with huggingface_hub Oct 2 19 days ago 07:14 Upload results/full/artifacts/retrieval.json with huggingface_hub 07:09 Upload results/full/artifacts/emb.npy with huggingface_hub 07:08 Upload results/full/artifacts/triples.json with huggingface_hub 06:33 fix fetch_artifact: copyfile instead of move (hub cache symlinks) 06:28 Upload results/full/artifacts/state.json with huggingface_hub 06:27 add stage-artifact push/fetch resume logic (protects long GPU run against external cancellation) Sep 29 22 days ago 21:08 ppr: max-score per chunk instead of sum (hub-entity inflation fix) 21:06 Fix 2Wiki context dict-form parsing (root cause of 400 dropped queries), fix supporting_facts diagnostic crash 07:00 fix: case-insensitive needle survival check + empty-gold diagnostics 06:30 Fix needle truncation bug (insert needles within first 200 words), rewrite PPR as personalized PageRank with teleport, needle questions name projects, add all-recall@depth and needle survival check 05:35 Fix embed stage: pre-truncate texts to 510 tokens with bge tokenizer (bge-large 512-token limit crashed llm.embed) 04:30 Fixed experiment script: body/query/question_type fix, fact-level gold, unique needle batches, needle depth=max(5,K), divergence diagnostics Sep 28 23 days ago 23:13 Upload versions.json with huggingface_hub 22:46 initial commit 
