Void

feat: allreduce and fusion kernel development

1

Currently, it is still in the draft stage. The completed parts are: - Fixed the sync error in the twoshot sync kernel. - Removed the poorly performing oneshot sync kernel....

[https://nvbugs/5655885][fix] fix invalid instruction error in 2shot ar kernel on Ampere

10

https://github.com/NVIDIA/TensorRT-LLM/issues/9086 **.acquire and .release qualifiers for fence instruction require sm_90 or higher** ## Summary by CodeRabbit * **Performance** * Enhanced barrier synchronization efficiency for newer GPU architectures * Maintained backward...

[None][fix] Enabled simultaneous support for low-precision combine and MTP.

41

## Summary by CodeRabbit * **Bug Fixes** * Enhanced low-precision combine support detection to optimize performance across different precision formats and configurations. ## Description ## Test Coverage ## PR Checklist...