QwenLM / Qwen2-VL

Qwen2-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud.
Apache License 2.0
3.23k stars 202 forks source link

Preprocessor performance issue #514

Open SinanAkkoyun opened 2 weeks ago

SinanAkkoyun commented 2 weeks ago

I noticed that the Preprocessor doesn't have good performances on images larger than 1280x28x28, as it takes at least 3s for them (but only 400ms for 1280x28x28)

We rely on low latency and would love to know how to optimize the preprocessor to achieve minimal latency