[Part 1] Benchmark for Multimodal LLM Understanding of 6-Second Clips from Korean Broadcasts [Pipeline Construction]
Test Environment:
qwen3-omni-baseline-news-stt-off-30fps-vs-0.5fps
Baseline analysis of 100 clips from the news category using Qwen3-Omni without STT (Speech-to-Text) context.
[Part 2] Inference Parameter Tuning: Analysis of a 6-Second Clip Using Qwen3-Omni (OFAT Sweep)
1. Introduction
[Part 3] Analysis of a 6-Second Clip Using the Qwen3-Omni Inference Parameter Combination (Optimal Configuration Determined)
1. Introduction