-
Notifications
You must be signed in to change notification settings - Fork 265
Pull requests: NVIDIA/cudnn-frontend
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Fix split-K heuristic for FROST LA
cat-enhancements
mod-frost
orig-nv-eng
Reported or requested by NVIDIA engineering.
fix-nightly-ci-failure
cat-bug
Reports of incorrect behavior, crashes, regressions, or unexpected results.
mod-frost
optimize-moe-grouped-mm-perf
cat-enhancements
cat-feature
Requests for new functionality, APIs, examples, or behavior improvements.
mod-frost
DSA backward SM90: fix three top-k and attention-sink boundary failures
#785
opened Aug 28, 2026 by
SuperGoodGame
Loading…
2 of 4 tasks
benchmark/e2e: per-block perf share for every published Qwen3.5/3.6/3.8 configuration
#782
opened Aug 27, 2026 by
YangXu1990uiuc
Collaborator
Loading…
4 tasks done
frost gemm: 64-tall tile family for M-starved shapes, measured cluster tie-breaks
#781
opened Aug 27, 2026 by
brandonfzhang
Collaborator
Loading…
refactor(sdpa): remove the experimental SDPA torch op
cat-cleanup
mod-frost
orig-nv-eng
Reported or requested by NVIDIA engineering.
Add Triton NVFP4 attention QAT backward
#778
opened Aug 27, 2026 by
SolitaryThinker
Loading…
2 of 4 tasks
Add Flex Attention CuTe DSL kernels
cat-enhancements
mod-cutedsl
CuTeDSL kernels, generated kernels, examples, or related integration work.
orig-nv-eng
Reported or requested by NVIDIA engineering.
Ship the SDPA repro tool as Build, packaging, tooling, dependency, release, or repository maintenance work.
mod-infra
Infrastructure, CI/CD, build systems, packaging, releases, or repo maintenance.
orig-nv-eng
Reported or requested by NVIDIA engineering.
python -m cudnn.repro
cat-infra
frost(sdpa): support attention bias in SDPA forward engines
cat-feature
Requests for new functionality, APIs, examples, or behavior improvements.
mod-frost
orig-nv-eng
Reported or requested by NVIDIA engineering.
Avoid copies in BackendDescriptor constructor
cat-enhancements
mod-frontend
cuDNN frontend APIs, operation graph construction, plans, and user-facing wrappers.
orig-external
Reported or requested by an external user, customer, or community contributor.
perf(dsa): specialize SM100 sparse-attention backward for H32/D576
#760
opened Aug 26, 2026 by
icavan
Contributor
Loading…
2 of 4 tasks
docs(sdpa): clarify O strides in fp16_fwd sample (#184)
cat-doc
Documentation changes, examples, API references, tutorials, or wording fixes.
mod-frontend
cuDNN frontend APIs, operation graph construction, plans, and user-facing wrappers.
orig-external
Reported or requested by an external user, customer, or community contributor.
Add Rubin MXFP8 MoE expert-parallel training support
#750
opened Aug 26, 2026 by
zhibinz-nv
Loading…
4 tasks
Add GDP support to FROST LA
cat-feature
Requests for new functionality, APIs, examples, or behavior improvements.
mod-frost
orig-nv-eng
Reported or requested by NVIDIA engineering.
Change RoPE computation to bf16 precision to match MCore's triton implementation
#723
opened Aug 24, 2026 by
ghadiaravi13
•
Draft
3 tasks
Add public GELU-MLP and a ModelOpt-anchored Qwen-Image NVFP4 proxy
cat-feature
Requests for new functionality, APIs, examples, or behavior improvements.
mod-frontend
cuDNN frontend APIs, operation graph construction, plans, and user-facing wrappers.
mod-infra
Infrastructure, CI/CD, build systems, packaging, releases, or repo maintenance.
orig-nv-eng
Reported or requested by NVIDIA engineering.
frost(sdpa): port #585 LPT schedulers to the SM107 fp8 sibling; defaults must obey declared knob domains
cat-bug
Reports of incorrect behavior, crashes, regressions, or unexpected results.
mod-cutedsl
CuTeDSL kernels, generated kernels, examples, or related integration work.
mod-frost
orig-nv-eng
Reported or requested by NVIDIA engineering.
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.