---
license: mit
---

# Frames2LoRA SmolVLM Checkpoints

## Frames2LoRA: Parametric Video Internalization for Vision-Language Models

Official implementation of **Frames2LoRA**

[**Manan Suri**](https://manansuri.com/)  ·  [**Sarvesh Baskar**](https://sarvesh-369.github.io/)  ·  [**Dinesh Manocha**](https://www.cs.umd.edu/people/dmanocha)

*University of Maryland, College Park*

[![project page](https://img.shields.io/badge/🌐_project-page-3b82f6?style=flat-square)](https://frames2lora.github.io/)
[![arxiv paper](https://img.shields.io/badge/📄_arxiv_paper-2606.04351-b31b1b?style=flat-square)](https://arxiv.org/abs/2606.04351)
[![code](https://img.shields.io/badge/GitHub-Frames2LoRA-181717?style=flat-square)](https://github.com/frames2lora/Frames2LoRA)

This repository contains two Frames2LoRA Stage 1 checkpoint files:

- `frames2lora-smolvlm2-500m-best-ce.pt` for `HuggingFaceTB/SmolVLM2-500M-Video-Instruct`
- `frames2lora-smolvlm2-2.2b-best-ce.pt` for `HuggingFaceTB/SmolVLM2-2.2B-Instruct`

## Cite us

```bibtex
@misc{suri2026frames2loraparametricvideointernalization,
      title={Frames2LoRA: Parametric Video Internalization for Vision-Language Models},
      author={Manan Suri and Sarvesh Baskar and Dinesh Manocha},
      year={2026},
      eprint={2606.04351},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2606.04351},
}
```