ROCm: update aiter and its usage to fused moe (bloat16, fp8, fp8 block-quant) (#4053)

This commit is contained in:
HAI
2025-03-04 03:00:46 -08:00
committed by GitHub
parent e0a2c96308
commit 51d25405a7
4 changed files with 82 additions and 40 deletions

View File

@@ -18,7 +18,7 @@ ARG TRITON_COMMIT="improve_fa_decode_3.0.0"
ARG AITER_REPO="https://github.com/ROCm/aiter.git"
ARG AITER_COMMIT="dev/testx"
ARG AITER_COMMIT="testx"
RUN git clone ${SGL_REPO} \
&& cd sglang \