microsoft/mscclpp
1
0
Fork 0
mirror of https://github.com/microsoft/mscclpp.git synced 2026-06-29 19:07:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP

Browse Source
This commit is contained in:
Binyang Li
2026-06-16 02:58:26 +00:00
parent 74dce951bf
commit dcd16d433d
1 changed files with 0 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python/mscclpp/ext/ep/README.md
View File

@@ -700,12 +700,3 @@ expert_output = fp8_grouped_mlp(
output = moe_comm.combine(expert_output, handle)
```
## Open questions
- Whether `weights` should be required for dispatch or allowed to be `None`
for unweighted combine.
- Whether `DispatchHandle` should copy `topk_ids` and `weights` for safety or
hold references for lower overhead.
- How to represent backend-specific quantization formats beyond FP8, NVFP4,
and MXFP8 without growing the dispatch argument list.