yourengod

Results 4 comments of yourengod

请教热词的问题

> 1~100, but 20 is recommended 好的，多谢。

chat.load( device="cuda", use_flash_attn=True ): # 设置compile=True会卡死。

> I don't speak chinese however I can tell you that I encountered the same issue, it doesn't actually hang. It just takes literally 10-15 minutes on the first generation...

请问，说话者如何训练？

> 有写训练的文档 请问多个说话者如何训练？

Excuse me, how does this utilize FastAPI and other technologies to solve the Python Global Interpreter Lock (GIL) issue, and how can it achieve concurrency?

> How is this an issue and it achieves concurrency because every request is run in a different thread ( this is default fast-api behaviour). As far as I know...