Support for apple/FastVLM-0.5B

[rasgaard]

https://huggingface.co/apple/FastVLM-0.5B

Multimodal models such as Apple's new FastVLM-0.5B have lots of mobile/edge use-cases. It would be super nice to see support in the ExecuTorch ecosystem! :)

[mergennachin]

cc @kimishpatel @jackzhxng @larryliu0820 -- this seems like a good one to validate as part of the vlm models (e.g., llava, gemma3, smolvlm)

[jackzhxng]

Is the modeling code on transformers yet? I couldn't find it for some reason

[rasgaard]

@jackzhxng It looks to me like the modeling code comes from llava_qwen.py in the Hugging Face model repo. It registers LlavaQwen2ForCausalLM into transformers.

[jackzhxng]

Llava, I see. This is in the works, will let you know!