[Fix] Remove redundant allreduce fusion block and skip TP=1 (#20621)

python/sglang/srt/server_args.py

@@ -1706,13 +1706,6 @@ class ServerArgs:
 
             quant_method = get_quantization_config(hf_config)
             is_mxfp4_quant_format = quant_method == "mxfp4"
-            if is_blackwell_supported():
-                # workaround for https://github.com/flashinfer-ai/flashinfer/issues/2006
-                if not self.enable_dp_attention and self.nnodes == 1:
-                    self.enable_flashinfer_allreduce_fusion = True
-                    logger.info(
-                        "Enable FlashInfer AllReduce Fusion on sm100 for GptOssForCausalLM"
-                    )
             if not self.enable_dp_attention and self.nnodes == 1 and is_hip():
                 # TODO (Hubert): Put this back later
                 # self.enable_aiter_allreduce_fusion = True
@@ -2076,6 +2069,7 @@ class ServerArgs:
                 "Qwen3_5ForConditionalGeneration",
             ]
             and (is_sm90_supported() or is_sm100_supported())
+            and self.tp_size > 1
             and not self.enable_dp_attention
             and self.attn_cp_size <= 1
             and self.nnodes == 1
@@ -2083,6 +2077,9 @@ class ServerArgs:
             and self.moe_a2a_backend == "none"
         ):
             self.enable_flashinfer_allreduce_fusion = True
+            logger.info(
+                f"Auto-enabling FlashInfer AllReduce Fusion on SM90/SM10X for {model_arch}"
+            )
 
     def _handle_mamba_radix_cache(
         self,