ntonellotto/msmarco-passage.ance.flex

Flex Index of MSMarco Passage using ANCE

Description

This artifact represents the retrieval index of the MSMarco Passage collection produced by the ANCE model (more info here).

Usage

The following code snippete shows how to use this artifact to perform a dense retrieval experiment:

import pyterrier as pt
import pyterrier_dr
from pyterrier.measures import *

# Load the artifact
index = pt.Artifact.from_hf('ntonellotto/msmarco-passage.ance.flex')
# Load the query encoder (must be the same used to create the artifact)
model = pyterrier_dr.Ance()
# Load the dataset
dataset = pt.get_dataset('msmarco_passage')
# Run the PyTerrier experiment
pt.Experiment(
    [model >> index],
    dataset.get_topics('test-2019'),
    dataset.get_qrels('test-2019'),
    eval_metrics=[RR@10, Recall(rel=2)@100, Recall@100, nDCG@10, "mrt"],
    names=["Dense Retrieval"]
)

You should get something like:

name	RR@10	R(rel=2)@100	R@100	nDCG@10	mrt
Dense Retrieval	0.976744	0.607269	0.514744	0.718238	280.014161

Reproduction

Generating the data

The data has been generated using the following code snippet:

import torch
import pyterrier as pt
import pyterrier_dr

# Select the available Torch's backend
if torch.backends.mps.is_available():
    device = torch.device("mps")
elif torch.cuda.is_available():
    device = torch.device("cuda")
else:
    device = torch.device("cpu")
print(f"Using device: {device}")
# Create the encoder model
tct = pyterrier_dr.Ance(device=device, batch_size=256)
# Create the destination index folder
index = pyterrier_dr.FlexIndex("./msmarco-passage.ance.flex")
# Indexing
(tct >> index).index(pt.get_dataset("msmarco_passage").get_corpus_iter())

print(f"Indexed {len(index)} documents")

Uploading the data

To upload the index to Hugging Face, use the following code snippet (the dataset sheet will be created automatically):

import pyterrier as pt

# Load the artifact from disk
artifact = pt.Artifact.load("./msmarco-passage.ance.flex")
# Upload the artifact to HF
artifact.to_hf('ntonellotto/msmarco-passage.ance.flex')

Metadata

{
  "type": "dense_index",
  "format": "flex",
  "vec_size": 768,
  "doc_count": 8841823
}