Embedding 23000 functions with BAAI/bge-small-en-v1.5...
group 1 of 23
group 2 of 23
group 3 of 23
group 4 of 23
group 5 of 23
group 6 of 23
group 7 of 23
group 8 of 23
group 9 of 23
group 10 of 23
group 11 of 23
group 12 of 23
group 13 of 23
group 14 of 23
group 15 of 23
group 16 of 23
group 17 of 23
group 18 of 23
group 19 of 23
group 20 of 23
group 21 of 23
group 22 of 23
group 23 of 23
CodeSearchNet Python, csn_python_valid.parquet, model BAAI/bge-small-en-v1.5
Queries: 23000 in 23 groups of 1000 (1 right answer, 999 distractors), seed 2026
keyword  MRR 0.8067   hit rate at 5: 0.8824
meaning  MRR 0.9037   hit rate at 5: 0.9498
both     MRR 0.8799   hit rate at 5: 0.9405
Functions that didn't parse (docstring left in): 289
Queries with no keyword matches: 28
