ByteDance-Seed / ByteDance-Seed/Triton-distributed
Tried to generate ptx automatically with xla？
Open
question
- Dominant language
- Python
- Stars
- 1.5k
- Forks
- 172
- PR merge metrics
- No merged PRs in 30d
Description
Hello, I see that in the architecture diagram of tilelink, compilation and execution phases are separated, so in principle, runtime should be able to access any ptx diagram.
Due to the high cost of writing each kernel by hand, have you considered using xla for automatic subgraph compilation to generate ptx?
Contributor guide
Assessment
This issue has not been assessed yet.