Datasetpublic

jpq-repro/msmarco-passage-train__tct_colbertfaiss2opqM96_nbits8ps159744neg200ibnlr__

msmarco-passage-train__tct_colbert__faiss2opq__M96_nbits8__ps159744__neg200__ibn__lr__ Description This is the PyTerrier JPQIndex for MSMARCO v1 passage corpus, which corresponds to an result from the SIGIR 2026 reproducibility paper. Usage # Load the artifact import pyterrier as pt import pyterrier_dr index = pt.Artifact.from_hf('jpq-repro/msmarco-passage-train__tct_colbert__faiss2opq__M96_nbits8__ps159744__neg200__ibn__lr__') model =… See the full description on the dataset page: https://huggingface.co/datasets/jpq-repro/msmarco-passage-train__tct_colbert__faiss2opq__M96_nbits8__ps159744__neg200__ibn__lr__.

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes42downloads

Dataset Card

msmarco-passage-train_tctcolbert_faiss2opqM96nbits8_ps159744neg200ibnlr_

Description

This is the PyTerrier JPQIndex for MSMARCO v1 passage corpus, which corresponds to an result from the SIGIR 2026 reproducibility paper.

Usage

python

# Load the artifact
import pyterrier as pt
import pyterrier_dr
index = pt.Artifact.from_hf('jpq-repro/msmarco-passage-train__tct_colbert__faiss2opq__M96_nbits8__ps159744__neg200__ibn__lr__')
model = pyterrier_dr.TctColBert.hnp()
model.model = model.model.from_pretrained(index.path)
retr_pipe = model >> index.retriever_pq()
retr_pipe.search("what are chemical reactions?")

Benchmarks

TREC-DL 2019: nDCG@10 0.669942

Metadata

{
  "type": "dense_index",
  "format": "jpq",
  "M": 96,
  "Ks": 256,
  "dsub": 8,
  "doc_count": 8841823,
  "opq": true
}