# cvlab-columbia/viper

Code for the paper "ViperGPT: Visual Inference via Python Execution for Reasoning"

Repository: https://github.com/cvlab-columbia/viper
Canonical: https://ross.abutalabs.com/products/cvlab-columbia-viper
Language: Jupyter Notebook
License: NOASSERTION
License Family: other
Last push: 2024-01-29T21:29:37+00:00

## Health v2 (maintenance only)
Score: 30/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 90
- inputs: {"age_days": 1268, "days_push": 947, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1716, forks 131 (observed 2026-08-28T04:05:26.298671+00:00)

## What it is
ViperGPT is a research codebase that composes vision-and-language models with code generated by large language models (GPT-3.5/GPT-4) to perform visual inference and reasoning over images and videos. It provides a Python API and notebooks for generating and executing visual reasoning programs.

## Use cases
- answer questions about images using LLM-generated code
- perform visual reasoning over videos
- compose pretrained vision models into pipelines via LLM prompts
- run visual question answering on image datasets
- reproduce ViperGPT paper experiments
- prototype multimodal reasoning agents

## When to choose
- you want to reproduce or extend the ViperGPT paper's approach
- you need LLM-driven composition of vision models for image/video reasoning
- you're researching program synthesis for visual inference

## When to avoid
- you need a production-ready, actively maintained product
- you can't run CUDA/GPU vision models locally
- you want a safe sandboxed agent framework without executing LLM-generated code
- you need extensive support for newer chat models beyond GPT-3.5/GPT-4

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-inference, computer-vision, agent-framework, prompt-engineering, machine-learning
- domain: artificial-intelligence, computer-vision, large-language-models
- platform: python
- tags: viper-gpt, visual-reasoning, code-generation, research-code, paper-code, jupyter-notebook, openai, gpt-4, glip, image-captioning, blip, depth-estimation, visual-question-answering, ai-agents, research, linux, gpu

## Member repositories
- cvlab-columbia/viper (main) score 30

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:26.298671+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:34:15.163180+00:00, confidence not recorded.
  - readme: https://github.com/cvlab-columbia/viper (fetched 2026-08-28T04:05:26.298671+00:00, sha d2195098ee98)
- Data as of 2026-08-30T08:39:29.467469+00:00.