Contribute to the privateuse1 backend. #98406
Description
🚀 The feature, motivation and pitch
This issue is used to discuss how to improve the PrivateUse1 backend to facilitate third-party manufacturers to access Pytorch.
With the popularity of pytorch and the evolution of computing acceleration hardware, the strong coupling between pytorch and cuda has become a serious problem. So the completeness of PrivateUse1 is what third-party hardware manufacturers need. After all, we can't add more enumerated types to DeviceType unless we are a big company like Apple or Intel (just kidding🫡).
For each feature, I will summarize it into this issue.
Please join us, thank you.
Alternatives
No response
Additional context
No response