Skip to content

rag24-doc-segmented-test-umbrela.splade-v3.onnx.md

Latest commit

 

History

History
112 lines (78 loc) · 7.05 KB

rag24-doc-segmented-test-umbrela.splade-v3.onnx.md

File metadata and controls

112 lines (78 loc) · 7.05 KB

Anserini Regressions: TREC 2024 RAG Track Test Topics

Model: SPLADE-v3 (using ONNX for on-the-fly query encoding)

This page describes regression experiments for ranking on the segmented version of the MS MARCO V2.1 document corpus using the test topics (= queries in TREC parlance), which is integrated into Anserini's regression testing framework. This corpus was derived from the MS MARCO V2 segmented document corpus and prepared for the TREC 2024 RAG Track.

The model itself can be download here. See the official SPLADE repo and the following paper for more details:

Carlos Lassance, Hervé Déjean, Thibault Formal, and Stéphane Clinchant. SPLADE-v3: New baselines for SPLADE. arXiv:2403.06789.

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

Evaluation uses (automatically generated) UMBRELA qrels over all 301 topics from the TREC 2024 RAG Track test set. More details can be found in the following two papers:

Shivani Upadhyay, Ronak Pradeep, Nandan Thakur, Daniel Campos, Nick Craswell, Ian Soboroff, and Jimmy Lin. A Large-Scale Study of Relevance Assessments with Large Language Models Using UMBRELA. Proceedings of the 2025 International ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval (ICTIR 2025), pages 358-368, July 2025, Padua, Italy.

Shivani Upadhyay, Ronak Pradeep, Nandan Thakur, Daniel Campos, Nick Craswell, Ian Soboroff, Hoa Trang Dang, and Jimmy Lin. A Large-Scale Study of Relevance Assessments with Large Language Models: An Initial Look. arXiv:2411.08275, November 2024.

The exact configurations for these regressions are stored in this YAML file. Note that this page is automatically generated from this template as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and then run bin/build.sh to rebuild the documentation.

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

bin/run.sh io.anserini.reproduce.ReproduceFromDocumentCollection --index --verify --search --config rag24-doc-segmented-test-umbrela.splade-v3.onnx

We make available a version of the MS MARCO V2.1 segmented document corpus that has already been encoded with SPLADE-v3.

From any machine, the following command will download the corpus and perform the complete regression, end to end:

bin/run.sh io.anserini.reproduce.ReproduceFromDocumentCollection --download --index --verify --search --config rag24-doc-segmented-test-umbrela.splade-v3.onnx

The run_regression.py script automates the following steps, but if you want to perform each step manually, simply copy/paste from the commands below and you'll obtain the same regression results.

Corpus Download

Download the corpus and unpack into collections/:

wget https://rgw.cs.uwaterloo.ca/pyserini/data/msmarco_v2.1_doc_segmented_splade-v3.tar -P collections/
tar xvf collections/msmarco_v2.1_doc_segmented_splade-v3.tar -C collections/

To confirm, msmarco-v2.1-doc-segmented-splade-v3.tar is 125 GB and has MD5 checksum c62490569364a1eb0101da1ca4a894d9. With the corpus downloaded, the following command will perform the remaining steps below:

bin/run.sh io.anserini.reproduce.ReproduceFromDocumentCollection --index --verify --search --config rag24-doc-segmented-test-umbrela.splade-v3.onnx \
  --corpus-path collections/msmarco_v2.1_doc_segmented_splade-v3

Indexing

Typical indexing command:

bin/run.sh io.anserini.index.IndexCollection \
  -threads 24 \
  -collection JsonVectorCollection \
  -input /path/to/msmarco-v2.1-doc-segmented-splade-v3 \
  -generator DefaultLuceneDocumentGenerator \
  -index indexes/lucene-inverted.msmarco-v2.1-doc-segmented.splade-v3/ \
  -impact -pretokenized \
  >& logs/log.msmarco-v2.1-doc-segmented-splade-v3 &

The setting of -input should be a directory containing the compressed jsonl files that comprise the corpus.

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.

Retrieval

Here, we are using 89 test topics from the TREC 2024 RAG Track with manual relevance judgments from NIST assessors. Topics and qrels are stored here, 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.msmarco-v2.1-doc-segmented.splade-v3/ \
  -topics tools/topics-and-qrels/topics.rag24.test.txt \
  -topicReader TsvString \
  -output runs/run.msmarco-v2.1-doc-segmented-splade-v3.splade-v3-onnx.topics.rag24.test.txt \
  -impact -pretokenized -removeQuery -hits 1000 -encoder SpladeV3 &

Evaluation can be performed using trec_eval:

bin/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.rag24.test-umbrela-all.txt runs/run.msmarco-v2.1-doc-segmented-splade-v3.splade-v3-onnx.topics.rag24.test.txt
bin/trec_eval -c -m ndcg_cut.100 tools/topics-and-qrels/qrels.rag24.test-umbrela-all.txt runs/run.msmarco-v2.1-doc-segmented-splade-v3.splade-v3-onnx.topics.rag24.test.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.rag24.test-umbrela-all.txt runs/run.msmarco-v2.1-doc-segmented-splade-v3.splade-v3-onnx.topics.rag24.test.txt

Effectiveness

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

nDCG@20 SPLADE-v3
RAG 24: Test queries 0.5167
nDCG@100 SPLADE-v3
RAG 24: Test queries 0.4587
R@100 SPLADE-v3
RAG 24: Test queries 0.2437