| --- |
| license: apache-2.0 |
| task_categories: |
| - text-retrieval |
| tags: |
| - sparse-retrieval |
| - learned-sparse-retrieval |
| - anserini |
| - beir |
| - msmarco |
| size_categories: |
| - 10M<n<100M |
| configs: |
| - config_name: beir-v1.0.0-arguana |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-arguana/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-arguana/queries/test/*.parquet |
| - config_name: beir-v1.0.0-bioasq |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-bioasq/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-bioasq/queries/test/*.parquet |
| - config_name: beir-v1.0.0-climate-fever |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-climate-fever/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-climate-fever/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-android |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-android/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-android/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-english |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-english/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-english/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-gaming |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-gaming/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-gaming/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-gis |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-gis/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-gis/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-mathematica |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-mathematica/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-mathematica/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-physics |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-physics/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-physics/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-programmers |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-programmers/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-programmers/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-stats |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-stats/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-stats/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-tex |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-tex/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-tex/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-unix |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-unix/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-unix/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-webmasters |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-webmasters/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-webmasters/queries/test/*.parquet |
| - config_name: beir-v1.0.0-cqadupstack-wordpress |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-cqadupstack-wordpress/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-cqadupstack-wordpress/queries/test/*.parquet |
| - config_name: beir-v1.0.0-dbpedia-entity |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-dbpedia-entity/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-dbpedia-entity/queries/test/*.parquet |
| - config_name: beir-v1.0.0-fever |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-fever/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-fever/queries/test/*.parquet |
| - config_name: beir-v1.0.0-fiqa |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-fiqa/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-fiqa/queries/test/*.parquet |
| - config_name: beir-v1.0.0-hotpotqa |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-hotpotqa/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-hotpotqa/queries/test/*.parquet |
| - config_name: beir-v1.0.0-nfcorpus |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-nfcorpus/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-nfcorpus/queries/test/*.parquet |
| - config_name: beir-v1.0.0-nq |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-nq/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-nq/queries/test/*.parquet |
| - config_name: beir-v1.0.0-quora |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-quora/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-quora/queries/test/*.parquet |
| - config_name: beir-v1.0.0-robust04 |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-robust04/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-robust04/queries/test/*.parquet |
| - config_name: beir-v1.0.0-scidocs |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-scidocs/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-scidocs/queries/test/*.parquet |
| - config_name: beir-v1.0.0-scifact |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-scifact/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-scifact/queries/test/*.parquet |
| - config_name: beir-v1.0.0-signal1m |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-signal1m/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-signal1m/queries/test/*.parquet |
| - config_name: beir-v1.0.0-trec-covid |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-trec-covid/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-trec-covid/queries/test/*.parquet |
| - config_name: beir-v1.0.0-trec-news |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-trec-news/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-trec-news/queries/test/*.parquet |
| - config_name: beir-v1.0.0-webis-touche2020 |
| data_files: |
| - split: corpus |
| path: beir-v1.0.0-webis-touche2020/corpus/*.parquet |
| - split: queries |
| path: beir-v1.0.0-webis-touche2020/queries/test/*.parquet |
| - config_name: msmarco-passage |
| data_files: |
| - split: corpus |
| path: msmarco-passage/corpus/*.parquet |
| - split: dl19_passage |
| path: msmarco-passage/queries/dl19-passage/*.parquet |
| - split: dl20 |
| path: msmarco-passage/queries/dl20/*.parquet |
| - split: dev_subset |
| path: msmarco-passage/queries/msmarco-passage.dev-subset/*.parquet |
| --- |
| |
| # SlideOn sparse vectors for MS MARCO passage and BEIR |
|
|
| Pre-computed learned sparse vectors from [lightonai/SlideOn](https://huggingface.co/lightonai/SlideOn) |
| (revision `370c600`) for the MS MARCO v1 passage corpus and the 29 BEIR v1.0.0 corpora |
| distributed with [Anserini](https://github.com/castorini/anserini), plus their evaluation queries. |
| Document ids, query ids and texts follow Anserini exactly (BEIR documents are encoded as `title + " " + text`), |
| so the vectors drop into Anserini/Pyserini impact indexes. See the |
| [SlideOn & mSlideOn blog post](https://huggingface.co/blog/lightonai/slideon-mslideon) for the models. |
|
|
| ## Layout |
|
|
| One config per dataset; each row is one document or query: |
|
|
| | column | type | content | |
| | --- | --- | --- | |
| | `id` | string | Anserini docid / qid | |
| | `indices` | list<uint16> | active latents (0–32767), **sorted by decreasing weight** | |
| | `weights` | list<float16> | the model's weights, exactly as it outputs them (FP16, no quantization) | |
|
|
| Scores are raw dot products between query and document vectors (no log, no normalization). |
|
|
| ## Choosing the sparsity (Matryoshka budgets) |
|
|
| SlideOn was trained with Matryoshka budgets of 512 or 1024 document latents and 32, 64 or 128 query latents. |
| Because each row is sorted by weight, keeping the first `k` entries is exactly the model's `max_active_dims=k` output: |
|
|
| - **documents** store the top **1024** latents (the largest trained document budget; docs with fewer active latents keep all of them). |
| Truncating further (e.g. to 512) reduces index size and latency. |
| - **queries** are stored unpruned; truncate to the trained query budgets for faster search. |
|
|
| ```python |
| from datasets import load_dataset |
| |
| docs = load_dataset("lightonai/SlideOn-sparse-vectors", "beir-v1.0.0-scifact", split="corpus") |
| k = 512 |
| docs = docs.map(lambda r: {"indices": r["indices"][:k], "weights": r["weights"][:k]}) |
| ``` |
|
|
| ## Anserini |
|
|
| `to_anserini.py` (in this repo) truncates to top-k, quantizes weights as `round(w * 100)` (the SPLADE convention; |
| costs ≤ 0.001 nDCG@10 in our checks) and writes `JsonVectorCollection` documents and `TsvString` topics: |
|
|
| ```bash |
| python to_anserini.py beir-v1.0.0-scifact/corpus collections/scifact/part0.jsonl.gz --k 512 |
| python to_anserini.py beir-v1.0.0-scifact/queries/test topics.scifact.tsv.gz --query |
| |
| bin/run.sh io.anserini.index.IndexCollection -collection JsonVectorCollection -input collections/scifact \ |
| -index indexes/scifact.slideon -generator DefaultLuceneDocumentGenerator -threads 16 -impact -pretokenized |
| bin/run.sh io.anserini.search.SearchCollection -index indexes/scifact.slideon -topics topics.scifact.tsv.gz \ |
| -topicReader TsvString -output run.scifact.txt -impact -pretokenized -removeQuery -hits 1000 |
| ``` |
|
|
| ## Citation |
|
|
| ```bibtex |
| @misc{moura2026slideon, |
| author = {Moura Junior, Paulo Roberto and Sourty, Raphaël}, |
| title = {{SlideOn} \& {mSlideOn}: Sparsified late-interaction and dense retrievers}, |
| year = {2026}, |
| howpublished = {Hugging Face Blog}, |
| url = {https://huggingface.co/blog/lightonai/slideon-mslideon} |
| } |
| ``` |
|
|
