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

Deco

Deco is an official Python and PyTorch research implementation of Dynamic Correction Decoding, a method that adjusts multimodal large language model output logits using selected preceding layers to mitigate hallucinations.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Deco — 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 Deco when you have a multimodal language model codebase and want to experiment with a decoding-time correction method or reproduce the repository's evaluation workflows.

Who it is for

  • Researchers evaluating decoding approaches for multimodal language models.
  • Developers modifying an existing multimodal model generation pipeline.

Documented uses

  • Add DeCo options to a multimodal model's generation call.
  • Adjust correction strength, token thresholds, and selected early-exit layers.
  • Run documented hallucination-related evaluation workflows using MSCOCO 2014 data.
Before you choose

Requirements & limits

The public GitHub repository specifies an MIT license.

  • The README's example calls model.generate with image input, hidden-state output, and DeCo-specific parameters.
  • The implementation is identified as residing in transformers/generation/utils.py.
  • The documented environment uses Conda with Python 3.9 and installs dependencies from requirements.txt.
  • Evaluation guidance requires MSCOCO 2014 data; layer intervals may be adjusted for the model.

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 the Deco Python environment

Create the documented Conda environment and install the repository dependencies before proceeding to model use.

Before you start

  • A Conda installation
  • The Deco repository and its requirements.txt file

Steps

  1. Create an environment with Python 3.9 using conda create -n deco python==3.9.

  2. Activate it with conda activate deco.

  3. Install the listed dependencies with the command below.

    Documented command
    pip install -r requirements.txt

Expected outcome

A Conda environment named deco with the repository requirements installed.

Guide limitations

  • These setup steps do not themselves run a model or verify compatibility with a particular multimodal model.
  • The README does not specify deployment targets.
Documentation 1 ↗
Evidence and transparency

Sources & review

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