From 763888b5a85da1927aa6a4c73ddc2930edd8fdd1 Mon Sep 17 00:00:00 2001 From: yctseng0211 Date: Tue, 9 Dec 2025 01:09:23 +0800 Subject: [PATCH] [AMD] change fused rms quant interface for aiter upgrade (#14497) --- docker/rocm.Dockerfile | 4 ++-- python/sglang/srt/models/deepseek_v2.py | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/docker/rocm.Dockerfile b/docker/rocm.Dockerfile index d196bf173..de05e97e4 100644 --- a/docker/rocm.Dockerfile +++ b/docker/rocm.Dockerfile @@ -31,7 +31,7 @@ ENV BUILD_TRITON="0" ENV BUILD_LLVM="0" ENV BUILD_AITER_ALL="1" ENV BUILD_MOONCAKE="1" -ENV AITER_COMMIT="v0.1.7.post1" +ENV AITER_COMMIT="v0.1.7.post5" ENV NO_DEPS_FLAG="" # =============================== @@ -42,7 +42,7 @@ ENV BUILD_TRITON="0" ENV BUILD_LLVM="0" ENV BUILD_AITER_ALL="0" ENV BUILD_MOONCAKE="1" -ENV AITER_COMMIT="v0.1.7.post2" +ENV AITER_COMMIT="v0.1.7.post5" ENV NO_DEPS_FLAG="" # =============================== # Chosen arch and args diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py index dc30b4f0a..9fdfbcb6a 100644 --- a/python/sglang/srt/models/deepseek_v2.py +++ b/python/sglang/srt/models/deepseek_v2.py @@ -1839,7 +1839,7 @@ class DeepseekV2AttentionMLA(nn.Module): current_stream.wait_stream(self.alt_stream) else: if _use_aiter_gfx95 and self.q_b_proj.weight.dtype == torch.uint8: - q, k_nope, *_ = fused_rms_mxfp4_quant( + q, _, k_nope, *_ = fused_rms_mxfp4_quant( q, self.q_a_layernorm.weight, self.q_a_layernorm.variance_epsilon,