OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
RAG & Knowledge Systems

Dyprag

Dyprag is a Python research implementation of Dynamic Parametric Retrieval Augmented Generation, a method that uses a parameter translator to incorporate document-derived knowledge into language models at test time.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Dyprag — OpenViST project profile in RAG & Knowledge Systems.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Dyprag when you need research code for experimenting with document-derived parameterized knowledge during language-model inference, and can work with the associated model and retrieval-data setup.

Who it is for

  • Researchers studying retrieval-augmented generation and test-time knowledge enhancement
  • Developers prototyping parameterized retrieval or comparing RAG methods
  • Practitioners exploring interpretability experiments for parametric retrieval

Documented uses

  • Prototype a dynamic parametric RAG pipeline for test-time knowledge enhancement.
  • Compare DyPRAG inference with a vanilla language-model baseline.
  • Evaluate outputs using the repository's RAGTruth evaluation instructions.
  • Adapt model-specific code to support another language model.
Before you choose

Requirements & limits

The public repository lists an Apache-2.0 license.

  • The documented environment uses Python 3.10.4, Conda and dependencies from requirements.txt.
  • The BM25 data-preparation route uses a Wikipedia passage dump and Elasticsearch 8.15.0.
  • The README describes dyprag and dyprag_combine inference modes and a separate RAGTruth evaluation script.
  • Adapting the implementation to another language model requires adding model files and modifying model-loading and forward-pass code.

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 the documented Python environment

Create the Conda environment and install the repository's Python dependencies.

Before you start

  • A local checkout of the Dyprag repository
  • Conda

Steps

  1. Change into the DyPRAG repository directory.

  2. Create a Conda environment named dyprag with Python 3.10.4.

  3. Activate the dyprag environment.

  4. Install dependencies from requirements.txt with pip.

Expected outcome

A Conda environment with the repository's declared Python dependencies installed.

Guide limitations

  • These setup instructions do not cover data preparation, model weights or running inference.
Documentation 1 ↗
Source-based practical guide

Prepare a Wikipedia BM25 retrieval index

Download the specified Wikipedia passage dump and build an Elasticsearch index using the repository's preparation script.

Before you start

  • The DyPRAG repository and its Python environment
  • Elasticsearch 8.15.0
  • The Wikipedia passage dump from the DPR data source

Steps

  1. Download the psgs_w100.tsv.gz Wikipedia dump into data/dpr and decompress it.

  2. Download and extract Elasticsearch 8.15.0, then start Elasticsearch in the background as shown in the README.

  3. From the repository directory, run prep_elastic.py with the dump path data/dpr/psgs_w100.tsv and index name wiki.

Expected outcome

The README's specified Wikipedia passage data is indexed in Elasticsearch under the name wiki.

Guide limitations

  • This recipe covers retrieval-index preparation only; it does not prepare benchmark datasets or run DyPRAG inference.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
Trae1ounG/DyPRAG · Official homepage title · retained in its source language
How we review sourcesSuggest a correction