when inference with a video using 32 frames, the inference speed is too slow for mplug_owl3

[dongdk]

hi @LukeForeverYoung
when inference with a video using 32 frames, using params = {'num_beams': 3, 'repetition_penalty': 1.2,}, the inference time is about 30s~120s. any idea to solve it?