[feature] extend DBO to XBO (#30120)

Signed-off-by: jiangkuaixue123 <jiangxiaozhou111@163.com>
Co-authored-by: root <root@hk01dgx028.cm.cluster>

vllm/config/parallel.py

@@ -156,6 +156,8 @@ class ParallelConfig:
 
     enable_dbo: bool = False
     """Enable dual batch overlap for the model executor."""
+    ubatch_size: int = 0
+    """Number of ubatch size."""
 
     dbo_decode_token_threshold: int = 32
     """The threshold for dual batch overlap for batches only containing decodes.
@@ -325,6 +327,14 @@ class ParallelConfig:
         including data parallelism."""
         return self.world_size * self.data_parallel_size
 
+    @property
+    def use_ubatching(self) -> bool:
+        return self.enable_dbo or self.ubatch_size > 1
+
+    @property
+    def num_ubatches(self) -> int:
+        return 2 if self.enable_dbo else self.ubatch_size
+
     def get_next_dp_init_port(self) -> int:
         """
         We might need to initialize process groups in multiple

vllm/config/vllm.py

@@ -870,9 +870,12 @@ class VllmConfig:
                     f"cudagraph_mode={self.compilation_config.cudagraph_mode}"
                 )
 
-        if self.parallel_config.enable_dbo:
+        if self.parallel_config.use_ubatching:
             a2a_backend = self.parallel_config.all2all_backend
-            assert a2a_backend in ["deepep_low_latency", "deepep_high_throughput"], (
+            assert a2a_backend in [
+                "deepep_low_latency",
+                "deepep_high_throughput",
+            ], (
                 "Microbatching currently only supports the deepep_low_latency and "
                 f"deepep_high_throughput all2all backend. {a2a_backend} is not "
                 "supported. To fix use --all2all-backend=deepep_low_latency or "