Skip to content

NVLM 1.0

A multimodal model family from NVIDIA ADLR

Visit official site
Category AI models

What NVLM 1.0 is

The model is described as a multimodal large language model family from NVIDIA ADLR, released with open weights and training code.

What you can do with it

  • Generate descriptions of images
  • Follow instructions across multimodal tasks
  • Perform OCR on text labels
  • Reason about humour in an image
  • Provide stepwise math reasoning

Model notes

  • Text-only accuracy improved after multimodal training
  • OCRBench and VQAv2 are named as benchmarks
  • Megatron-Core code is provided for training

Getting started

  • Download weights and training code from the site
  • Read the architecture documentation
  • Fine-tune for a chosen vision-language task
  • Train with the Megatron-Core code

Pros & cons

✓ What we like

  • Text-only task accuracy rose after multimodal training
  • The model can reason about humour in images

! What to watch out for

  • Fine-tuning is needed for specific vision-language tasks
  • Training relies on the Megatron-Core code base

FAQ

Who releases it?

The material names NVIDIA ADLR.

What is open sourced?

Model weights and Megatron-Core training code are listed.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review