From fb7e49d4eb56b5356e940f39b4a2cbeebaca67e2 Mon Sep 17 00:00:00 2001
From: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>
Date: Tue, 19 May 2026 09:18:00 +0800
Subject: [PATCH] Expose can-run-cuda-graph as a regular property on the
 embedding result (#25711)

---
 python/sglang/srt/disaggregation/prefill.py                   | 2 +-
 python/sglang/srt/dllm/mixin/scheduler.py                     | 2 +-
 .../managers/scheduler_components/batch_result_processor.py   | 2 +-
 python/sglang/srt/managers/utils.py                           | 4 ++++
 4 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/python/sglang/srt/disaggregation/prefill.py b/python/sglang/srt/disaggregation/prefill.py
index 84e476214..a64344844 100644
--- a/python/sglang/srt/disaggregation/prefill.py
+++ b/python/sglang/srt/disaggregation/prefill.py
@@ -586,7 +586,7 @@ class SchedulerDisaggregationPrefillMixin:
                     self.send_kv_chunk(req, last_chunk=False, end_idx=req.tmp_end_idx)
                 req.time_stats.set_last_chunked_prefill_finish_time()
 
-        can_run_cuda_graph = getattr(result, "can_run_cuda_graph", False)
+        can_run_cuda_graph = result.can_run_cuda_graph
         self.metrics_reporter.report_prefill_stats(
             batch=batch,
             prefill_stats=batch.prefill_stats,
diff --git a/python/sglang/srt/dllm/mixin/scheduler.py b/python/sglang/srt/dllm/mixin/scheduler.py
index 6d35864ab..ee1b8d076 100644
--- a/python/sglang/srt/dllm/mixin/scheduler.py
+++ b/python/sglang/srt/dllm/mixin/scheduler.py
@@ -92,7 +92,7 @@ class SchedulerDllmMixin:
             self.output_streamer.stream_output(batch.reqs, batch.return_logprob)
             self.token_to_kv_pool_allocator.free_group_end()
 
-        can_run_cuda_graph = getattr(result, "can_run_cuda_graph", False)
+        can_run_cuda_graph = result.can_run_cuda_graph
         self.metrics_reporter.report_prefill_stats(
             batch=batch,
             prefill_stats=batch.prefill_stats,
diff --git a/python/sglang/srt/managers/scheduler_components/batch_result_processor.py b/python/sglang/srt/managers/scheduler_components/batch_result_processor.py
index baf09cfb7..6a96fb4ef 100644
--- a/python/sglang/srt/managers/scheduler_components/batch_result_processor.py
+++ b/python/sglang/srt/managers/scheduler_components/batch_result_processor.py
@@ -324,7 +324,7 @@ class SchedulerBatchResultProcessor:
             batch.reqs, batch.return_logprob, skip_stream_req
         )
 
-        can_run_cuda_graph = getattr(result, "can_run_cuda_graph", False)
+        can_run_cuda_graph = result.can_run_cuda_graph
         self.metrics_reporter.report_prefill_stats(
             batch=batch,
             prefill_stats=batch.prefill_stats,
diff --git a/python/sglang/srt/managers/utils.py b/python/sglang/srt/managers/utils.py
index 31c5b375c..33db3942b 100644
--- a/python/sglang/srt/managers/utils.py
+++ b/python/sglang/srt/managers/utils.py
@@ -260,6 +260,10 @@ class EmbeddingBatchResult:
     pooled_hidden_states: Optional[torch.Tensor] = None
     copy_done: Optional[torch.cuda.Event] = None
 
+    @property
+    def can_run_cuda_graph(self) -> bool:
+        return False
+
     def copy_to_cpu(self):
         """Copy embeddings and pooled hidden states to CPU for overlap scheduling."""
         if isinstance(self.embeddings, torch.Tensor):