fix

LucasWilkinson · LucasWilkinson · commit d5cb8a5eb85c · 2025-12-02T20:25:07.000-08:00
Signed-off-by: Lucas Wilkinson &lt;lwilkins@redhat.com&gt;
diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py
@@ -2802,7 +2802,7 @@ def _determine_batch_execution_and_padding(
             )
 
             should_ubatch, num_tokens_across_dp = coordinate_batch_across_dp(
-                num_tokens_unpadded=num_tokens_padded,
+                num_tokens_unpadded=num_tokens,
                 parallel_config=self.parallel_config,
                 allow_microbatching=allow_microbatching,
                 allow_dp_padding=allow_dp_padding,

Original file line number	Diff line number	Diff line change
`@@ -2802,7 +2802,7 @@ def _determine_batch_execution_and_padding(`
`2802`	`2802`	`)`
`2803`	`2803`
`2804`	`2804`	`should_ubatch, num_tokens_across_dp = coordinate_batch_across_dp(`
`2805`		`- num_tokens_unpadded=num_tokens_padded,`
	`2805`	`+ num_tokens_unpadded=num_tokens,`
`2806`	`2806`	`parallel_config=self.parallel_config,`
`2807`	`2807`	`allow_microbatching=allow_microbatching,`
`2808`	`2808`	`allow_dp_padding=allow_dp_padding,`