---
library_name: pytorch
---

![smolvlm_logo](resource/SmolVLM.png)

SmolVLM is a family of compact vision–language models designed for efficient multimodal understanding by integrating lightweight visual encoders with small language models, with a focus on edge deployment and low-latency multimodal AI.

Original paper: [SmolVLM: Redefining small and efficient multimodal models](https://arxiv.org/abs/2504.05299)

# SmolVLM2-500M-Video-Instruct

SmolVLM2-500M-Video-Instruct is a highly efficient ~500M-parameter variant optimized for low-memory footprint and fast multimodal inference. It is well suited for applications such as visual question answering, image captioning, document understanding, and real-time multimodal assistants on edge devices or resource-constrained environments.

Model Configuration:
- Reference implementation: [smollm](https://github.com/huggingface/smollm)
- Original Weight: [SmolVLM2-500M-Video-Instruct](https://huggingface.co/HuggingFaceTB/SmolVLM2-500M-Video-Instruct)
- Resolution: 3x512x512
- Support Cooper version:
    - Cooper SDK: [2.5.4]
    - Cooper Foundry: [2.3]

| Model | Device | Model Link |
| :-----: | :-----: | :-----: |
| SmolVLM2-500M-Video-Instruct | CV7 | [Model_Link](https://huggingface.co/Ambarella/SmolVLM2/blob/main/cv7_smolvlm2_video_instruct_500M.tar) |
| SmolVLM2-500M-Video-Instruct | CV72 | [Model_Link](https://huggingface.co/Ambarella/SmolVLM2/blob/main/cv72_smolvlm2_video_instruct_500M.tar) |
| SmolVLM2-500M-Video-Instruct | CV75 | [Model_Link](https://huggingface.co/Ambarella/SmolVLM2/blob/main/cv75_smolvlm2_video_instruct_500M.tar) |