Fix processor resize defaults for vLLM compatibility

#1
by JungleGym - opened
ARC Lab, Tencent PCG org

Remove explicit do_resize: false from the video and image processor configs so the checkpoint falls back to the upstream Qwen3-VL default behavior. This keeps Transformers examples unchanged when do_resize=False is passed explicitly, while restoring compatibility for runtimes like vLLM that rely on the saved processor defaults.

JungleGym changed pull request status to merged
JungleGym deleted the refs/pr/1 ref

Sign up or log in to comment