Hello,
I am currently evaluating Qwen3-Omni-30B-A3B-Thinking on the OmniVideoBench benchmark. My current test results show a score of 34.4, whereas the Instruct model achieved 37.7. Max_Frames set to 256.
Could you please clarify the specific settings or inference parameters used for the Thinking mode. I am looking to reproduce the 38.8 score reported in the paper. Any guidance on the sampling configurations and other settings would be greatly appreciated.
Thank you!
Hello,
I am currently evaluating Qwen3-Omni-30B-A3B-Thinking on the OmniVideoBench benchmark. My current test results show a score of 34.4, whereas the Instruct model achieved 37.7. Max_Frames set to 256.
Could you please clarify the specific settings or inference parameters used for the Thinking mode. I am looking to reproduce the 38.8 score reported in the paper. Any guidance on the sampling configurations and other settings would be greatly appreciated.
Thank you!