OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Parco

PARCO is an open-source Python research project implementing parallel autoregressive models for multi-agent combinatorial optimization, with examples and scripts for vehicle-routing, pickup-and-delivery, and scheduling problems.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Parco — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use PARCO when you want a research codebase and problem-specific examples for experimenting with learned combinatorial optimization, rather than a hosted optimization interface.

Who it is for

  • Researchers studying neural and combinatorial optimization.
  • Developers exploring learned approaches to vehicle routing, pickup and delivery, or flow-shop scheduling.

Documented uses

  • Explore parallel autoregressive model code for combinatorial optimization.
  • Train an experiment for HCVRP, OMDCPDP, or FFSP.
  • Evaluate models using greedy decoding or sampling.
  • Regenerate problem data or run the included OR-Tools solving script.
Before you choose

Requirements & limits

The repository is public and licensed under MIT. Its documented workflow uses Python and uv.

  • The documented quickstarts cover three specific problem families, not a general-purpose optimization interface.
  • The README describes both greedy and sampling evaluation modes.
  • The initial FFSP implementation is kept in a legacy folder because it was not integrated into RL4CO.
  • The README recommends uv and provides a script for downloading Hugging Face data and checkpoints.

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

Set up PARCO and retrieve its data and checkpoints

Create the project environment and use the documented script to download data and checkpoints.

Before you start

  • A Python development environment with uv installed.
  • A local copy of the PARCO repository.

Steps

  1. Create and activate a virtual environment with the command below and source .venv/bin/activate.

    Documented command
    uv venv

  2. Install the project dependencies with the command below.

    Documented command
    uv sync --all-extras

  3. Run the command below to download data and checkpoints from Hugging Face.

    Documented command
    python scripts/download_hf.py

Expected outcome

The documented environment setup is complete and the repository’s download script has been invoked for data and checkpoints.

Guide limitations

  • The README does not specify hardware requirements or describe how to troubleshoot download or installation problems.
  • This is a setup and retrieval guide; it does not verify that the environment or downloaded files work.
Documentation 1 ↗
Source-based practical guide

Train a PARCO experiment

Use the documented training script to select an experiment for HCVRP, OMDCPDP, or FFSP.

Before you start

  • The PARCO repository and its dependencies installed as described in the README.

Steps

  1. Run the command below to select HCVRP.

    Documented command
    python train.py experiment=hcvrp

  2. For the other documented experiments, replace hcvrp with omdcpdp or ffsp.

Expected outcome

The training script is invoked with the selected problem experiment.

Guide limitations

  • The README does not provide a full account of training configuration or hardware needs.
  • The initial FFSP implementation is retained in a legacy folder because it was not integrated in RL4CO.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
ai4co/parco · Official homepage title · retained in its source language
How we review sourcesSuggest a correction