FastVLM-1.5B-Stage3

Introduction

This is FastVLM-1.5B-Stage3, a multimodal language model that can understand things visually, being agentic, understand long videos and capture events, and generate structured outputs.

This model is exported from Github apple/ml-fastvlm.

Model's weight: llava-fastvithd_1.5b_stage3.zip.

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = 'FastVLM-1.5B-Stage3'
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype='auto', trust_remote_code=True)

Export to MNN

git clone https://github.com/alibaba/MNN
cd MNN/transformers/llm/export
python llmexport.py --path /path/to/FastVLM-1.5B-Stage3 --export mnn

Citation

If you find our work helpful, feel free to give us a cite.

@InProceedings{fastvlm2025,
  author = {Pavan Kumar Anasosalu Vasu, Fartash Faghri, Chun-Liang Li, Cem Koc, Nate True, Albert Antony, Gokul Santhanam, James Gabriel, Peter Grasch, Oncel Tuzel, Hadi Pouransari},
  title = {FastVLM: Efficient Vision Encoding for Vision Language Models},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {June},
  year = {2025},
}{2023}