Tencent's Youtu-Parsing-Omni is a 5B model that turns documents, images, audio and video into structured records.
Design: One shared encoder handles every modality, and one JSON schema covers all ten parsing tasks. Video frames can attend to the concurrent audio track.
Results: The paper reports 96.96 Overall on OmniDocBench v1.6 and 75.08 Avg. on OmniParsingBench, the best open-weight model on the latter. Its advantage is concentrated on audio and video.
Practical: Code and model weights are public, and it ships with a vLLM plugin, task prompts and inference examples.
Sign in to comment.
TL;DR of the paper, generated automatically.
Tencent's Youtu-Parsing-Omni is a 5B model that turns documents, images, audio and video into structured records.