OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Search & Research AI

Parade

PARADE is research code and data for an end-to-end document reranking model that aggregates passage representations using pretrained BERT or ELECTRA models.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Parade — OpenViST project profile in Search & Research AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use PARADE for research or development work that needs a passage-based reranker and can supply prepared retrieval, relevance and corpus data.

Who it is for

  • Information-retrieval researchers examining passage-based document reranking.
  • Developers adapting BERT- or ELECTRA-based ranking experiments.
  • Researchers reproducing the README’s Robust04 example or exploring its TREC-COVID branch.

Documented uses

  • Rerank retrieved documents using passage representation aggregation.
  • Compare the PARADE-Avg, PARADE-Attn, PARADE-Max and PARADE model variants.
  • Prepare five-fold Robust04 data and train and evaluate a reranking model.
  • Explore the TREC-COVID challenge experiment referenced by the repository.
Before you choose

Requirements & limits

  • The implementation supports BERT and ELECTRA, with four named aggregation variants.
  • The documented Robust04 workflow requires qrels, queries, a retrieval run, corpus text, passage splitting and TFRecord preparation.
  • The README recommends fine-tuning pretrained models on the MS MARCO passage collection and also lists fine-tuned checkpoints as an alternative initialization.
  • TPU use requires uploading training and testing data to Google Cloud Storage.
  • The repository describes a trec_eval configuration for significance testing.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare Robust04 data for a five-fold experiment

Convert the documented Robust04 inputs into five train/test fold directories for the subsequent PARADE workflow.

Before you start

  • Robust04 qrels, queries and retrieval-run files; the README points to Anserini for these standard inputs.
  • Corpus text in docno-and-content pairs, with documents split into passages and written to TFRecord files.
  • An output_dir configured for the conversion script.

Steps

  1. Obtain the standard qrels, query and trec_run files and corpus text; the README refers readers to Anserini for these inputs.

  2. Split documents into passages and write the passages to TFRecord files.

  3. Run scripts/run.convert.data.sh.

  4. Check the output_dir for five generated subfolders, each containing a train file and a test file.

Expected outcome

Five prepared fold directories for the model-training and evaluation stage.

Guide limitations

  • The guide covers data preparation only; it does not establish a completed training or evaluation run.
  • The README says that when running on TPU, training and testing data must be uploaded to Google Cloud Storage.
  • The README does not specify the commands or settings for obtaining every input file.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
GitHub - canjiali/PARADE: code and data to faciliate BERT/ELECTRA for document ranking. Details refer to the paper - PARADE: Passage Representation Aggregation for Document Reranking. · GitHub · Official homepage title · retained in its source language
How we review sourcesSuggest a correction