SmolVLM2 500M Video Instruct - Video inference

[youchi1]

Question

Hey, is it possible to setup video inference through transformers.js (may be somehow else?) for the model SmolVLM2 500M Video Instruct? I can't make it work, but I saw, that it is possible in py transformers.

I want to create something similar to https://huggingface.co/spaces/HuggingFaceTB/SmolVLM2-HighlightGenerator/tree/main but with full local WebGPU inference.

Thanks in advance. cc: @xenova