Skip to content

fix+feat(gemma2): fix SWA correctness bugs and add FlexAttention fused softcap+SWA path#4308

Open
nvegesna-netizen wants to merge 5 commits into
NVIDIA-NeMo:mainfrom
nvegesna-netizen:nvegesna/fix-gemma2-swa-and-assertions
Open

fix+feat(gemma2): fix SWA correctness bugs and add FlexAttention fused softcap+SWA path#4308
nvegesna-netizen wants to merge 5 commits into
NVIDIA-NeMo:mainfrom
nvegesna-netizen:nvegesna/fix-gemma2-swa-and-assertions