skip aiter

vllm/model_executor/layers/fused_moe/layer.py

@@ -1943,7 +1943,6 @@ class FusedMoE(CustomOp):
                 topk_weights, topk_ids = self.router.select_experts(
                     hidden_states=x_orig,
                     router_logits=router_logits,
-                    # use_fused_gate=use_fused_gate,
                 )
 
                 if self.capture is not None:

vllm/v1/spec_decode/eagle.py

@@ -167,36 +167,36 @@ class SpecDecodeBaseProposer:
 
         # Determine allowed attention backends once during initialization.
         self.allowed_attn_types: tuple | None = None
-        if current_platform.is_rocm():
-            from vllm.v1.attention.backends.rocm_attn import RocmAttentionMetadata
-
-            rocm_types = [
-                TritonAttentionMetadata,
-                RocmAttentionMetadata,
-            ]
-            # ROCM_AITER_FA is an optional backend
-            if find_spec(
-                AttentionBackendEnum.ROCM_AITER_FA.get_path(include_classname=False)
-            ):
-                from vllm.v1.attention.backends.rocm_aiter_fa import (
-                    AiterFlashAttentionMetadata,
-                )
+        # if current_platform.is_rocm():
+        #     from vllm.v1.attention.backends.rocm_attn import RocmAttentionMetadata
 
-                rocm_types.append(AiterFlashAttentionMetadata)
+        #     rocm_types = [
+        #         TritonAttentionMetadata,
+        #         RocmAttentionMetadata,
+        #     ]
+        #     # ROCM_AITER_FA is an optional backend
+        #     if find_spec(
+        #         AttentionBackendEnum.ROCM_AITER_FA.get_path(include_classname=False)
+        #     ):
+        #         from vllm.v1.attention.backends.rocm_aiter_fa import (
+        #             AiterFlashAttentionMetadata,
+        #         )
 
-            # TRITON_MLA backend support for MLA models (e.g., DeepSeek)
-            from vllm.model_executor.layers.attention.mla_attention import (
-                MLACommonMetadata,
-            )
+        #         rocm_types.append(AiterFlashAttentionMetadata)
+
+        #     # TRITON_MLA backend support for MLA models (e.g., DeepSeek)
+        #     from vllm.model_executor.layers.attention.mla_attention import (
+        #         MLACommonMetadata,
+        #     )
 
-            rocm_types.append(MLACommonMetadata)
+        #     rocm_types.append(MLACommonMetadata)
 
-            # FlexAttention backend support
-            from vllm.v1.attention.backends.flex_attention import FlexAttentionMetadata
+        #     # FlexAttention backend support
+        #     from vllm.v1.attention.backends.flex_attention import FlexAttentionMetadata
 
-            rocm_types.append(FlexAttentionMetadata)
+        #     rocm_types.append(FlexAttentionMetadata)
 
-            self.allowed_attn_types = tuple(rocm_types)
+        #     self.allowed_attn_types = tuple(rocm_types)
 
         # Parse the speculative token tree.
         spec_token_tree = self.speculative_config.speculative_token_tree