# NVlabs/prismer

The implementation of "Prismer: A Vision-Language Model with Multi-Task Experts".

Repository: https://github.com/NVlabs/prismer
Canonical: https://ross.abutalabs.com/products/prismer
Homepage: https://shikun.io/projects/prismer
Language: Python
License: NOASSERTION
License Family: other
Topics: image-captioning, language-model, multi-modal-learning, multi-task-learning, vision-language-model, vision-and-language, vqa
Last push: 2024-01-17T21:55:00+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 91
- inputs: {"age_days": 1281, "days_push": 959, "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 1309, forks 76 (observed 2026-08-28T04:04:19.467901+00:00)

## What it is
Official PyTorch implementation of Prismer, a data- and parameter-efficient vision-language model that ensembles pre-trained task-specific experts for multi-modal reasoning. It supports pre-training and fine-tuning on image captioning and VQA benchmarks using Huggingface Accelerate.

## Use cases
- generate captions for images
- answer visual questions about images
- pretrain a vision-language model on image-text datasets
- fine-tune a multimodal model with fewer training samples
- reproduce the Prismer paper results
- run a vision-language model demo

## When to choose
- you need a parameter-efficient vision-language model for captioning or VQA
- you want to experiment with multi-task expert ensembles
- you are reproducing or extending the Prismer research

## When to avoid
- you need a production-ready multimodal API or chatbot
- you want a maintained general-purpose VLM with broad ecosystem support
- you need text-only language modeling

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, nlp, llm-training
- domain: artificial-intelligence, machine-learning, computer-vision
- platform: python
- tags: vision-language-model, multi-task-learning, image-captioning, vqa, pytorch, research-code, multimodal, natural-language-processing, gpu, linux

## Member repositories
- NVlabs/prismer (main) score 30

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:19.467901+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 2026-08-30T04:49:57.960418+00:00, confidence not recorded.
  - readme: https://github.com/NVlabs/prismer (fetched 2026-08-28T04:04:19.467901+00:00, sha 6b5015b5f397)
  - homepage: https://shikun.io/projects/prismer (fetched 2026-08-29T12:08:11.139181+00:00, sha c3c292466124)
- Data as of 2026-08-30T08:39:29.467469+00:00.