gen_ai_apps / vlm_chat app -> Qwen3-VL-2B-Instruct model doesn't work

[Michael]

Qwen3-VL-2B-Instruct.hef fails with HAILO_INVALID_OPERATION(6) on HailoRT 5.3.0 — generate_impl fails despite successful model load

Hi guys I just discovered this thread…had the same problem too, thanks for sharing the solution. I amended convert_resize_image() in backend.py to 512x288 in the signature instead of 336x336 and it now works with Qwen3 . Will have to rem to change it back if using Qwen2 tho! Would be great if the Qwen2 and Qwen3 models were compiled to accept a range of image sizes, or at least, the same image size.