FP8 Enablement for TE layers + HSTU attn#197

Open

esoba wants to merge 13 commits intoNVIDIA:mainfrom

Commits on Sep 18, 2025

Commits on Sep 19, 2025

Added flags for HSTU FP8 computation + wrapped forward method of pipeline in te amp cm
esoba
committed

Commits on Oct 1, 2025

Commits on Oct 8, 2025

Working implementation of FP8 for ranking
esoba
committed
Added TE dependency in Dockerfiles, updated HSTU readme to reflect quant_mode argument name in python API
esoba
committed

Commits on Oct 13, 2025

propagate quant mode to fused layer
esoba
committed

Commits on Oct 20, 2025

Commits on Oct 21, 2025