From b75e813c031683c0c26c60811d1d8ce540e01e7c Mon Sep 17 00:00:00 2001
From: hailin <hailin.zhao@gdzx.xyz>
Date: Tue, 7 Apr 2026 03:22:48 -0700
Subject: [PATCH] feat: enable flash_attention_2 for Qwen3-TTS

---
 modules/tts/qwen3_tts.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/modules/tts/qwen3_tts.py b/modules/tts/qwen3_tts.py
index 436c01f..8c6ff59 100644
--- a/modules/tts/qwen3_tts.py
+++ b/modules/tts/qwen3_tts.py
@@ -41,6 +41,7 @@ class TTSProvider(TTSProviderBase):
             model_path,
             device_map=device,
             dtype=dtype,
+            attn_implementation="flash_attention_2",
         )
         self.tokenizer = Qwen3TTSTokenizer.from_pretrained(tokenizer_path)