[Common][PyTorch] EP dispatch with unfused MXFP8 quantization - #3270

Open

phu0ngng wants to merge 13 commits into

NVIDIA:mainfrom

phu0ngng:ep_mxfp8

Commits on Aug 6, 2026

EP Dispatch with MXFP8
phu0ngng
committed
81ba2c5
restore gitignore
phu0ngng
committed
e4a572d
EP Dispatch with MXFP8
phu0ngng
committed
df4aaff
Separate EP dispatch-forward and combine-backward quant recipes
phu0ngng
committed
01c7e1c
[PyTorch] Simplify EP quantize path and guard zero-copy pool alloc under CUDA graph capture
phu0ngng
committed
a6f54a6
[PyTorch] Add eager-mode coverage for MXFP8 EP combine backward
phu0ngng
committed
196419b
[PyTorch] Supply persistent symm-mem recv buffers for zero-copy 1F1B CUDA-graph capture
phu0ngng
committed
8934d2b
shorten quant enum
phu0ngng
committed
ea10ceb
[PyTorch] Enable MXFP8 EP combine backward under zero-copy
phu0ngng
committed
b27bf5c
[PyTorch] Release symm-mem pool in ep_finalize
phu0ngng
committed
40f1bc8
[PyTorch] Drop unused device param from release_symm_mem_pool
phu0ngng
committed
1c30b1c
Merge branch 'main' into ep_mxfp8
phu0ngng
authored
dc05a03

Commits on Aug 7, 2026

Merge branch 'main' into ep_mxfp8
phu0ngng
authored
0ab4057