Issue/846 - 修改embedding算子支持图录制#859

Open

gongchensu wants to merge 3 commits intoInfiniTensor:mainfrom

gongchensu:Issue/846

Commits on Dec 26, 2025

Issue/846 - Refactor embedding to support device-side input and CUDA graph recording
gongchensu
committed
Issue/846 - Ensure embedding tensors are on the same device. Change format.
gongchensu
committed
Issue/846 - Optimize embedding kernel with vectorized memory access and __ldg
gongchensu
committed