Demo-131 Cuda graph with optimized paged attention#990

Open

PanZezhong1725 wants to merge 66 commits intomainfrom

Commits on Dec 29, 2025

Issue/862 - Fix compilation errors (missing headers, cub namespace) to enable inference on Hygon platform
gongchensu
committed

Commits on Jan 4, 2026

Commits on Jan 22, 2026

Commits on Jan 27, 2026

issue/987 - add .cpp files to ninetoothed includes
wooway777
committed
issue/978 - metax cuda graph impl and wrappings
wooway777
committed
issue/846 - Refactor embedding to support device-side input and CUDA graph recording

gongchensu
authored and
wooway777
committed
issue/900 - support embedding on iluvatar, metax, and moore
wooway777
committed
issue/900 - adapt to graph and adjust test script
wooway777
committed
issue/900 - maintains classic embedding for devices yet to be worked on
wooway777
committed
issue/791 fix add_rmsnorm api and rmsnorm module

PanZezhong1725
authored and
wooway777
committed
issue/884 - add_rms_norm on iluvatar, metax and moore
wooway777
committed
issue/632 - adapt to iluvatar core 20
wooway777
committed
issue/791 - fix add_rmsnorm api on mtx and mth
wooway777
committed
issue/810 support more ops as graph op

PanZezhong1725
authored and
wooway777
committed
issue/985 - adjust cxflags and cxxflags for lua scripts
wooway777
committed
issue/402 - convenient ninetoothed util

voltjia
authored and
wooway777
committed
issue/925 - Speed up `scripts/build_ntops.py` and `src/infiniop/ninetoothed/build.py` with `concurrent.futures`

voltjia
authored and
wooway777
committed
issue/940 - check build result and implicitly require build.py for build ntops
wooway777
committed
issue/935 - add metax include dir for ninetoothed
wooway777
committed
issue/919 - ninetoothed flash attention
wooway777
committed
issue/931 - ninetoothed swiglu for nv, il, mtx
wooway777
committed
issue/923 - ninetoothed kv caching for nv, il, mtx
wooway777
committed
issue/979 optimize paged attention

PanZezhong1725
authored and
wooway777
committed
issue/979 - removed commented paged attn codes
wooway777
committed
issue/983 - adapted the optimized paged attention to metax
wooway777
committed
demo131 - patch lua flags and includes
wooway777
committed
issue/811 use relax graph capture mode, add compile flag for graph instantiate
PanZezhong1725
committed
Merge pull request #989 from InfiniTensor/issue/811-fix
PanZezhong1725
authored

Commits on Jan 29, 2026

issue/995 fix paged attn on iluvatar
zhangyue207
authored

Commits on Feb 11, 2026

Commits on Feb 12, 2026