Yesterday
-
Remove docid_sft so OpenInstIR strictly contains retrieval (instruction, query, doc) data
-
Add DocID generator SFT training config ('docid', 1,205,874 document-docid pairs)
2 days ago
-
Add full unsampled Parquet shards for large tasks (batch 3: LePaRD)
-
Add full unsampled Parquet shards for large tasks (batch 2: Reddit)
-
Add full unsampled Parquet shards for large tasks (batch 1: TAD, TAS2, CodeSearchNet, CLERC, PubMedQA-Context, PubMedQA-Answer, AGNews, CC-News, Xsum, zsRE, Fever, ToT)
-
Upload OpenInstIR (69 tasks across 6 domains, SFT Parquet split + dataset card)
-
initial commit