From 427f7c1d4fdd7cb82978cd805e3cea6ccca4b678 Mon Sep 17 00:00:00 2001
From: ixlmar <206748156+ixlmar@users.noreply.github.com>
Date: Wed, 27 May 2026 12:37:13 +0000
Subject: [PATCH] perf: remove sync after FlashInfer attention plan()

Signed-off-by: ixlmar <206748156+ixlmar@users.noreply.github.com>
---
 tensorrt_llm/_torch/attention_backend/flashinfer.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/tensorrt_llm/_torch/attention_backend/flashinfer.py b/tensorrt_llm/_torch/attention_backend/flashinfer.py
index 49964319a18a..3322422075dc 100644
--- a/tensorrt_llm/_torch/attention_backend/flashinfer.py
+++ b/tensorrt_llm/_torch/attention_backend/flashinfer.py
@@ -1006,7 +1006,6 @@ def _plan_with_params(self,
                         self.kv_layout,
                         backend="cudnn",
                     ))
-            torch.cuda.current_stream().synchronize()
             if self.num_contexts <= 0:
                 raise ValueError(
                     "FlashInfer ragged prefill without KV cache requires "