[BugFix] Fix DP/EP hang (vllm-project#25906)

Signed-off-by: Lucas Wilkinson <[email protected]>

Author: LucasWilkinson

vllm/v1/worker/gpu_model_runner.py

@@ -3075,13 +3075,19 @@ def _dummy_run(
         # We currently only microbatch if the number of tokens is
         # over a certain threshold.
         if self.parallel_config.enable_dbo and allow_microbatching:
-            ubatch_slices, num_tokens_after_padding = ubatch_split(
+            ubatch_slices, ubatch_num_tokens_after_padding = ubatch_split(
                 num_scheduled_tokens,
                 total_num_scheduled_tokens,
                 total_num_scheduled_tokens,
                 uniform_decode=uniform_decode,
                 vllm_config=self.vllm_config,
             )
+            # Currently when DBO is enabled `ubatch_split` returns
+            # the num_tokens_after_padding for a single ubatch, but we have 2
+            # TODO(sage,lucas): this is cruft that should be addressed in the
+            # padding refactor.
+            if ubatch_num_tokens_after_padding is not None:
+                num_tokens_after_padding = ubatch_num_tokens_after_padding * 2
 
         # If we failed to microbatch, currently need to resynchronize
         # TODO(lucas,sage): we should be able to avoid this second sync by
@@ -3198,7 +3204,7 @@ def _dummy_run(
 
             # filter out the valid batch descriptor
             _cg_mode, batch_descriptor = self.cudagraph_dispatcher.dispatch(
-                BatchDescriptor(num_tokens=num_tokens,
+                BatchDescriptor(num_tokens=num_tokens_after_padding,
                                 uniform_decode=uniform_decode)) \
                 if not is_profile else (CUDAGraphMode.NONE, None)
             if cudagraph_runtime_mode is not None:
@@ -3212,7 +3218,13 @@ def _dummy_run(
                 cudagraph_runtime_mode = _cg_mode
 
             if ubatch_slices is not None:
-                num_tokens = num_tokens // 2
+                # Adjust values to reflect a single ubatch.
+                # TODO(sage,lucas): this is cruft that should be addressed in
+                #  the padding refactor.
+                num_tokens_after_padding = ubatch_slices[0].num_tokens
+                if num_tokens_across_dp is not None:
+                    num_tokens_across_dp[:] = num_tokens_after_padding
+
             with self.maybe_randomize_inputs(input_ids), set_forward_context(
                     attn_metadata,
                     self.vllm_config,