docgen.templates.rag24-doc-segmented-test.arctic-embed-l.parquet.shard03.flat.onnx.template Maven / Gradle / Ivy

Go to download
# Anserini Regressions: TREC 2024 RAG Track Test Topics

**Model**: Snowflake's [Arctic-embed-l](https://huggingface.co/Snowflake/snowflake-arctic-embed-l) with flat indexes (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.

We build on embeddings made available by Snowflake on [Hugging Face Datasets](https://huggingface.co/datasets/Snowflake/msmarco-v2.1-snowflake-arctic-embed-l), which contains vectors already encoded by the Arctic-embed-l model.
The complete dataset comprises 60 parquet files (from `00` to `59`).
Due to its large size (472 GB), we have divided the vectors into ten shards, each comprised of six files:
for example `shard00` spans `00.parquet` to `05.parquet`; `shard01` spans the next six parquet files, etc.

This page documents experiments for `shard03`; we expect the corpus to be in `msmarco_v2.1_doc_segmented.arctic-embed-l/shard00` (relative to the base collection path).
In these experiments, we are performing query inference "on-the-fly" with ONNX, using flat vector indexes.

The exact configurations for these regressions are stored in [this YAML file](${yaml}).
Note that this page is automatically generated from [this template](${template}) as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead.

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 ${test_name}
```

## Indexing

Typical indexing command:

```
${index_cmds}
```

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

For additional details, see explanation of [common indexing options](${root_path}/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:

```
${ranking_cmds}
```

Evaluation can be performed using `trec_eval`:

```
${eval_cmds}
```

## Effectiveness

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

${effectiveness}