[Fix] Update to v0.1.5.post4 and refine HIP attention backend selection (#11161)

docker/Dockerfile.rocm

@@ -42,7 +42,7 @@ ENV BUILD_TRITON="0"
 ENV BUILD_LLVM="0"
 ENV BUILD_AITER_ALL="1"
 ENV BUILD_MOONCAKE="1"
-ENV AITER_COMMIT="v0.1.5.post3"
+ENV AITER_COMMIT="v0.1.5.post4"
 ENV NO_DEPS_FLAG="--no-deps"
 
 # ===============================

python/sglang/srt/model_executor/model_runner.py

@@ -536,9 +536,7 @@ class ModelRunner:
                 elif _is_hip:
                     head_num = self.model_config.get_num_kv_heads(self.tp_size)
                     # TODO current aiter only support head number 16 or 128 head number
-                    if (
-                        head_num == 128 or head_num == 16
-                    ) and self.spec_algorithm.is_none():
+                    if head_num == 128 or head_num == 16:
                         server_args.attention_backend = "aiter"
                     else:
                         server_args.attention_backend = "triton"