[Spyre-Next] Pytorch Native Attention on Spyre: 4D Attention Kernel#914

Closed

jvlunteren wants to merge 54 commits into

torch-spyre:mainfrom

jvlunteren:pytorch_native_attention_v2

Commits on Mar 23, 2026

Integrated custom attention backend

bohnstingl
authored and
jvlunteren
committed
Formatting issues

bohnstingl
authored and
jvlunteren
committed
Changed the name of the attention operation

bohnstingl
authored and
jvlunteren
committed
Changed filename

bohnstingl
authored and
jvlunteren
committed
Implemented gather to avoid using full KV cache

bohnstingl
authored and
jvlunteren
committed
Removed .item() calls

bohnstingl
authored and
jvlunteren
committed
Cleanup and adding of example

bohnstingl
authored and
jvlunteren
committed
Lint

bohnstingl
authored and
jvlunteren
committed
Added testcase for attention backend

bohnstingl
authored and
jvlunteren
committed
Added missing utils file

bohnstingl
authored and
jvlunteren
committed
Reformat

bohnstingl
authored and
jvlunteren
committed
Functional update

bohnstingl
authored and
jvlunteren
committed
Lint issues

bohnstingl
authored and
jvlunteren
committed
:art: linting, vllm compatibility, test integration

joerunde
authored and
jvlunteren
committed
refactored attention backend to support compilation and execution on Spyre
jvlunteren
committed
formatting
jvlunteren
committed
add unit test
jvlunteren
committed
formatting
jvlunteren
committed

Commits on Mar 24, 2026

Commits on Mar 25, 2026

Commits on Mar 30, 2026

Commits on Apr 2, 2026

Commits on Apr 3, 2026

Commits on Apr 9, 2026

Removed enforce_eager argument
bohnstingl
committed

Commits on Apr 13, 2026

Merge branch 'main' into pytorch_native_attention
jvlunteren
authored

Commits on Apr 14, 2026

Commits on Apr 15, 2026

clarify shape comments
jvlunteren
committed