Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Add BRIGHT SPLADE v3 regressions #2920

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lilyjge merged 4 commits into from
Aug 16, 2025
Merged

Add BRIGHT SPLADE v3 regressions #2920

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7cb6c14
correct path for corpus collections
lilyjge Aug 14, 2025
2301b9f
regression yamls
lilyjge Aug 16, 2025
3adf024
regression docs
lilyjge Aug 16, 2025
2998fcc
reproduce
lilyjge Aug 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
84 changes: 84 additions & 0 deletions docs/regressions/regressions-bright-aops.splade-v3.onnx.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
# Anserini Regressions: BRIGHT — AoPS

**Model**: [SPLADE-v3](https://arxiv.org/abs/2403.06789) (using ONNX for on-the-fly query encoding)

This page documents regression experiments for [BRIGHT — AoPS](https://brightbenchmark.github.io/) using using [SPLADE-v3](https://arxiv.org/abs/2403.06789).
The model itself can be download [here](https://huggingface.co/naver/splade-v3).
See the [official SPLADE repo](https://github.com/naver/splade) and the following paper for more details:

> Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant. [SPLADE-v3: New baselines for SPLADE.](https://arxiv.org/abs/2403.06789) _arXiv:2403.06789_.

In these experiments, we are using ONNX to perform query encoding on the fly.

The exact configurations for these regressions are stored in [this YAML file](../../src/main/resources/regression/bright-aops.splade-v3.onnx.yaml).
Note that this page is automatically generated from [this template](../../src/main/resources/docgen/templates/bright-aops.splade-v3.onnx.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and build Anserini to rebuild the documentation.

From one of our Waterloo servers (e.g., `orca`), the following command will perform the complete regression, end to end:

```
python src/main/python/run_regression.py --index --verify --search --regression bright-aops.splade-v3.onnx
```

All the BRIGHT corpora, encoded byt he SPLADE-v3 mode, are available for download:

```bash
wget https://huggingface.co/datasets/castorini/collections-bright/resolve/main/bright-splade-v3.tar -P collections/
tar xvf collections/bright-splade-v3.tar -C collections/
```

The tarball is 1.5 GB and has MD5 checksum `434cd776b5c40f8112d2bf888c58a516`.
After download and unpacking the corpora, the `run_regression.py` command above should work without any issue.

## Indexing

Typical indexing command:

```
bin/run.sh io.anserini.index.IndexCollection \
-threads 16 \
-collection JsonVectorCollection \
-input /path/to/bright-aops \
-generator DefaultLuceneDocumentGenerator \
-index indexes/lucene-inverted.bright-aops.splade-v3/ \
-impact -pretokenized \
>& logs/log.bright-aops &
```

The path `/path/to/bright-aops/` should point to the corpus downloaded above.
The important indexing options to note here are `-impact -pretokenized`: the first tells Anserini not to encode BM25 doclengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the pre-encoded tokens.
For additional details, see explanation of [common indexing options](../../docs/common-indexing-options.md).

## Retrieval

Topics and qrels are stored [here](https://github.com/castorini/anserini-tools/tree/master/topics-and-qrels), which is linked to the Anserini repo as a submodule.

After indexing has completed, you should be able to perform retrieval as follows:

```
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-inverted.bright-aops.splade-v3/ \
-topics tools/topics-and-qrels/topics.bright-aops.tsv.gz \
-topicReader TsvString \
-output runs/run.bright-aops.splade-v3-onnx.topics.bright-aops.txt \
-impact -pretokenized -removeQuery -hits 1000 -encoder SpladeV3 &
```

Evaluation can be performed using `trec_eval`:

```
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.bright-aops.txt runs/run.bright-aops.splade-v3-onnx.topics.bright-aops.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.bright-aops.txt runs/run.bright-aops.splade-v3-onnx.topics.bright-aops.txt
bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.bright-aops.txt runs/run.bright-aops.splade-v3-onnx.topics.bright-aops.txt
```

## Effectiveness

With the above commands, you should be able to reproduce the following results:

| **nDCG@10** | **SPLADE-v3**|
|:-------------------------------------------------------------------------------------------------------------|-----------|
| BRIGHT: AoPS | 0.0692 |
| **R@100** | **SPLADE-v3**|
| BRIGHT: AoPS | 0.2602 |
| **R@1000** | **SPLADE-v3**|
| BRIGHT: AoPS | 0.4644 |
84 changes: 84 additions & 0 deletions docs/regressions/regressions-bright-biology.splade-v3.onnx.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
# Anserini Regressions: BRIGHT — Biology

**Model**: [SPLADE-v3](https://arxiv.org/abs/2403.06789) (using ONNX for on-the-fly query encoding)

This page documents regression experiments for [BRIGHT — Biology](https://brightbenchmark.github.io/) using using [SPLADE-v3](https://arxiv.org/abs/2403.06789).
The model itself can be download [here](https://huggingface.co/naver/splade-v3).
See the [official SPLADE repo](https://github.com/naver/splade) and the following paper for more details:

> Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant. [SPLADE-v3: New baselines for SPLADE.](https://arxiv.org/abs/2403.06789) _arXiv:2403.06789_.

In these experiments, we are using ONNX to perform query encoding on the fly.

The exact configurations for these regressions are stored in [this YAML file](../../src/main/resources/regression/bright-biology.splade-v3.onnx.yaml).
Note that this page is automatically generated from [this template](../../src/main/resources/docgen/templates/bright-biology.splade-v3.onnx.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and build Anserini to rebuild the documentation.

From one of our Waterloo servers (e.g., `orca`), the following command will perform the complete regression, end to end:

```
python src/main/python/run_regression.py --index --verify --search --regression bright-biology.splade-v3.onnx
```

All the BRIGHT corpora, encoded byt he SPLADE-v3 mode, are available for download:

```bash
wget https://huggingface.co/datasets/castorini/collections-bright/resolve/main/bright-splade-v3.tar -P collections/
tar xvf collections/bright-splade-v3.tar -C collections/
```

The tarball is 1.5 GB and has MD5 checksum `434cd776b5c40f8112d2bf888c58a516`.
After download and unpacking the corpora, the `run_regression.py` command above should work without any issue.

## Indexing

Typical indexing command:

```
bin/run.sh io.anserini.index.IndexCollection \
-threads 16 \
-collection JsonVectorCollection \
-input /path/to/bright-biology \
-generator DefaultLuceneDocumentGenerator \
-index indexes/lucene-inverted.bright-biology.splade-v3/ \
-impact -pretokenized \
>& logs/log.bright-biology &
```

The path `/path/to/bright-biology/` should point to the corpus downloaded above.
The important indexing options to note here are `-impact -pretokenized`: the first tells Anserini not to encode BM25 doclengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the pre-encoded tokens.
For additional details, see explanation of [common indexing options](../../docs/common-indexing-options.md).

## Retrieval

Topics and qrels are stored [here](https://github.com/castorini/anserini-tools/tree/master/topics-and-qrels), which is linked to the Anserini repo as a submodule.

After indexing has completed, you should be able to perform retrieval as follows:

```
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-inverted.bright-biology.splade-v3/ \
-topics tools/topics-and-qrels/topics.bright-biology.tsv.gz \
-topicReader TsvString \
-output runs/run.bright-biology.splade-v3-onnx.topics.bright-biology.txt \
-impact -pretokenized -removeQuery -hits 1000 -encoder SpladeV3 &
```

Evaluation can be performed using `trec_eval`:

```
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.bright-biology.txt runs/run.bright-biology.splade-v3-onnx.topics.bright-biology.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.bright-biology.txt runs/run.bright-biology.splade-v3-onnx.topics.bright-biology.txt
bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.bright-biology.txt runs/run.bright-biology.splade-v3-onnx.topics.bright-biology.txt
```

## Effectiveness

With the above commands, you should be able to reproduce the following results:

| **nDCG@10** | **SPLADE-v3**|
|:-------------------------------------------------------------------------------------------------------------|-----------|
| BRIGHT: Biology | 0.2095 |
| **R@100** | **SPLADE-v3**|
| BRIGHT: Biology | 0.5602 |
| **R@1000** | **SPLADE-v3**|
| BRIGHT: Biology | 0.8883 |
84 changes: 84 additions & 0 deletions docs/regressions/regressions-bright-earth-science.splade-v3.onnx.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
# Anserini Regressions: BRIGHT — Earth Science

**Model**: [SPLADE-v3](https://arxiv.org/abs/2403.06789) (using ONNX for on-the-fly query encoding)

This page documents regression experiments for [BRIGHT — Earth Science](https://brightbenchmark.github.io/) using using [SPLADE-v3](https://arxiv.org/abs/2403.06789).
The model itself can be download [here](https://huggingface.co/naver/splade-v3).
See the [official SPLADE repo](https://github.com/naver/splade) and the following paper for more details:

> Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant. [SPLADE-v3: New baselines for SPLADE.](https://arxiv.org/abs/2403.06789) _arXiv:2403.06789_.

In these experiments, we are using ONNX to perform query encoding on the fly.

The exact configurations for these regressions are stored in [this YAML file](../../src/main/resources/regression/bright-earth-science.splade-v3.onnx.yaml).
Note that this page is automatically generated from [this template](../../src/main/resources/docgen/templates/bright-earth-science.splade-v3.onnx.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and build Anserini to rebuild the documentation.

From one of our Waterloo servers (e.g., `orca`), the following command will perform the complete regression, end to end:

```
python src/main/python/run_regression.py --index --verify --search --regression bright-earth-science.splade-v3.onnx
```

All the BRIGHT corpora, encoded byt he SPLADE-v3 mode, are available for download:

```bash
wget https://huggingface.co/datasets/castorini/collections-bright/resolve/main/bright-splade-v3.tar -P collections/
tar xvf collections/bright-splade-v3.tar -C collections/
```

The tarball is 1.5 GB and has MD5 checksum `434cd776b5c40f8112d2bf888c58a516`.
After download and unpacking the corpora, the `run_regression.py` command above should work without any issue.

## Indexing

Typical indexing command:

```
bin/run.sh io.anserini.index.IndexCollection \
-threads 16 \
-collection JsonVectorCollection \
-input /path/to/bright-earth-science \
-generator DefaultLuceneDocumentGenerator \
-index indexes/lucene-inverted.bright-earth-science.splade-v3/ \
-impact -pretokenized \
>& logs/log.bright-earth-science &
```

The path `/path/to/bright-earth-science/` should point to the corpus downloaded above.
The important indexing options to note here are `-impact -pretokenized`: the first tells Anserini not to encode BM25 doclengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the pre-encoded tokens.
For additional details, see explanation of [common indexing options](../../docs/common-indexing-options.md).

## Retrieval

Topics and qrels are stored [here](https://github.com/castorini/anserini-tools/tree/master/topics-and-qrels), which is linked to the Anserini repo as a submodule.

After indexing has completed, you should be able to perform retrieval as follows:

```
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-inverted.bright-earth-science.splade-v3/ \
-topics tools/topics-and-qrels/topics.bright-earth-science.tsv.gz \
-topicReader TsvString \
-output runs/run.bright-earth-science.splade-v3-onnx.topics.bright-earth-science.txt \
-impact -pretokenized -removeQuery -hits 1000 -encoder SpladeV3 &
```

Evaluation can be performed using `trec_eval`:

```
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.bright-earth-science.txt runs/run.bright-earth-science.splade-v3-onnx.topics.bright-earth-science.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.bright-earth-science.txt runs/run.bright-earth-science.splade-v3-onnx.topics.bright-earth-science.txt
bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.bright-earth-science.txt runs/run.bright-earth-science.splade-v3-onnx.topics.bright-earth-science.txt
```

## Effectiveness

With the above commands, you should be able to reproduce the following results:

| **nDCG@10** | **SPLADE-v3**|
|:-------------------------------------------------------------------------------------------------------------|-----------|
| BRIGHT: Earth Science | 0.2670 |
| **R@100** | **SPLADE-v3**|
| BRIGHT: Earth Science | 0.5776 |
| **R@1000** | **SPLADE-v3**|
| BRIGHT: Earth Science | 0.8127 |
84 changes: 84 additions & 0 deletions docs/regressions/regressions-bright-economics.splade-v3.onnx.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
# Anserini Regressions: BRIGHT — Economics

**Model**: [SPLADE-v3](https://arxiv.org/abs/2403.06789) (using ONNX for on-the-fly query encoding)

This page documents regression experiments for [BRIGHT — Economics](https://brightbenchmark.github.io/) using using [SPLADE-v3](https://arxiv.org/abs/2403.06789).
The model itself can be download [here](https://huggingface.co/naver/splade-v3).
See the [official SPLADE repo](https://github.com/naver/splade) and the following paper for more details:

> Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant. [SPLADE-v3: New baselines for SPLADE.](https://arxiv.org/abs/2403.06789) _arXiv:2403.06789_.

In these experiments, we are using ONNX to perform query encoding on the fly.

The exact configurations for these regressions are stored in [this YAML file](../../src/main/resources/regression/bright-economics.splade-v3.onnx.yaml).
Note that this page is automatically generated from [this template](../../src/main/resources/docgen/templates/bright-economics.splade-v3.onnx.template) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and build Anserini to rebuild the documentation.

From one of our Waterloo servers (e.g., `orca`), the following command will perform the complete regression, end to end:

```
python src/main/python/run_regression.py --index --verify --search --regression bright-economics.splade-v3.onnx
```

All the BRIGHT corpora, encoded byt he SPLADE-v3 mode, are available for download:

```bash
wget https://huggingface.co/datasets/castorini/collections-bright/resolve/main/bright-splade-v3.tar -P collections/
tar xvf collections/bright-splade-v3.tar -C collections/
```

The tarball is 1.5 GB and has MD5 checksum `434cd776b5c40f8112d2bf888c58a516`.
After download and unpacking the corpora, the `run_regression.py` command above should work without any issue.

## Indexing

Typical indexing command:

```
bin/run.sh io.anserini.index.IndexCollection \
-threads 16 \
-collection JsonVectorCollection \
-input /path/to/bright-economics \
-generator DefaultLuceneDocumentGenerator \
-index indexes/lucene-inverted.bright-economics.splade-v3/ \
-impact -pretokenized \
>& logs/log.bright-economics &
```

The path `/path/to/bright-economics/` should point to the corpus downloaded above.
The important indexing options to note here are `-impact -pretokenized`: the first tells Anserini not to encode BM25 doclengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the pre-encoded tokens.
For additional details, see explanation of [common indexing options](../../docs/common-indexing-options.md).

## Retrieval

Topics and qrels are stored [here](https://github.com/castorini/anserini-tools/tree/master/topics-and-qrels), which is linked to the Anserini repo as a submodule.

After indexing has completed, you should be able to perform retrieval as follows:

```
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-inverted.bright-economics.splade-v3/ \
-topics tools/topics-and-qrels/topics.bright-economics.tsv.gz \
-topicReader TsvString \
-output runs/run.bright-economics.splade-v3-onnx.topics.bright-economics.txt \
-impact -pretokenized -removeQuery -hits 1000 -encoder SpladeV3 &
```

Evaluation can be performed using `trec_eval`:

```
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.bright-economics.txt runs/run.bright-economics.splade-v3-onnx.topics.bright-economics.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.bright-economics.txt runs/run.bright-economics.splade-v3-onnx.topics.bright-economics.txt
bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.bright-economics.txt runs/run.bright-economics.splade-v3-onnx.topics.bright-economics.txt
```

## Effectiveness

With the above commands, you should be able to reproduce the following results:

| **nDCG@10** | **SPLADE-v3**|
|:-------------------------------------------------------------------------------------------------------------|-----------|
| BRIGHT: Economics | 0.1604 |
| **R@100** | **SPLADE-v3**|
| BRIGHT: Economics | 0.4478 |
| **R@1000** | **SPLADE-v3**|
| BRIGHT: Economics | 0.7804 |
Loading