DDXDB

> Thanks! I understand. Here is my deduction: Local JIT can make sure that the compiled codes are optimized for a specific GPU, which runs more efficiently, but takes a...

It has been confirmed that the problem is caused by multiple buckets with different resolutions. I adjusted the picture to 512*512 and 512*1024, and the current utilization rate is 80%+

> Hi, I can't test this since I don't have intel GPU, but I'm looking into this: > > * [intel-extension-for-pytorch](https://github.com/intel/intel-extension-for-pytorch) > > Maybe this is the way to go....

> For lossless recording, you can use CQP mode with a value of 15 or lower I learned that CQP0 is lossless, but the CQP of qsv can only be...

> CQP with values below 15 will be visually indistinguishable from lossless, while the file size will be noticeably smaller I collected lossless videos to: https://github.com/cirquit/trdrop I'm not sure visually...

> CQP with values below 15 will be visually indistinguishable from lossless, while the file size will be noticeably smaller I have confirmed that setting CQP/ICQ to 1 is not...

> @minhthuc2502, sorry for the delayed response. Thank you for providing your valuable sugestions. I observed that Intel oneDNN is enabled for Conv layer. Hence was trying with that. Yes...

> I looked into chatTTS integration a while ago. It isn't really high on my priority list because it's a very resource-consuming TTS and can hardly run in real-time unless...

> Does the original ChatTTS not work on intel GPU? Are there any significant advantages to using ChatTTS-UI over the original ChatTTS? Regarding the version, I think you can open...

> 这个具体没懂，是使命令行工具调用openai 接口的吗？ 对，而且是在分发版实现