diff --git a/docker/rocm.Dockerfile b/docker/rocm.Dockerfile
index cc2858862..d0a2c42cc 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="v0.1.6.post1"
+ENV AITER_COMMIT="v0.1.6.post3"
 ENV NO_DEPS_FLAG=""
 
 # ===============================
@@ -42,7 +42,7 @@ ENV BUILD_TRITON="0"
 ENV BUILD_LLVM="0"
 ENV BUILD_AITER_ALL="1"
 ENV BUILD_MOONCAKE="1"
-ENV AITER_COMMIT="v0.1.6.post1"
+ENV AITER_COMMIT="v0.1.6.post3"
 ENV NO_DEPS_FLAG=""
 
 # ===============================
diff --git a/python/sglang/srt/layers/communicator.py b/python/sglang/srt/layers/communicator.py
index 423bd2801..2f276b319 100644
--- a/python/sglang/srt/layers/communicator.py
+++ b/python/sglang/srt/layers/communicator.py
@@ -268,7 +268,7 @@ class LayerCommunicator:
                     residual = hidden_states
 
                     if _use_aiter and _is_gfx95_supported and ("mxfp4" in qaunt_format):
-                        hidden_states = fused_rms_mxfp4_quant(
+                        hidden_states, *_, _ = fused_rms_mxfp4_quant(
                             hidden_states,
                             self.input_layernorm.weight,
                             self.input_layernorm.variance_epsilon,
@@ -281,7 +281,7 @@ class LayerCommunicator:
                         hidden_states = self.input_layernorm(hidden_states)
                 else:
                     if _use_aiter and _is_gfx95_supported and ("mxfp4" in qaunt_format):
-                        hidden_states, residual = fused_rms_mxfp4_quant(
+                        hidden_states, *_, residual = fused_rms_mxfp4_quant(
                             hidden_states,
                             self.input_layernorm.weight,
                             self.input_layernorm.variance_epsilon,
diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py
index b190bf5d4..895b69105 100644
--- a/python/sglang/srt/models/deepseek_v2.py
+++ b/python/sglang/srt/models/deepseek_v2.py
@@ -1608,7 +1608,7 @@ 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,
@@ -1937,7 +1937,7 @@ 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,