28 days ago
-
Upload build/audit_contamination.py with huggingface_hub
-
Upload build/audit_contamination.py with huggingface_hub
-
Upload build/audit_contamination.py with huggingface_hub
-
Upload build/audit_contamination.py with huggingface_hub
-
audit: vectorised rolling-hash grams + sorted-array lookup; the Python-set version could not finish 1.2B tokens
-
audit: composite AUDIT_PASSED verdict; missing shards are unmeasured, not clean
-
smoke PASSED (4/4 sources per shard, 40/40 boundaries); add contamination audit; raw target 1.23B = 23% headroom