From 12a8f5793490d99bf359b80f3e43ec309a4000b9 Mon Sep 17 00:00:00 2001 From: LL <56384135+OrangePure@users.noreply.github.com> Date: Tue, 31 Mar 2026 12:42:21 +0800 Subject: [PATCH] Fix Qwen3-TTS Base on NPU running failed (#2353) Signed-off-by: LL Signed-off-by: Canlin Guo Co-authored-by: Canlin Guo --- vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_talker.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_talker.py b/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_talker.py index ca9c8ca41..87fadae8c 100644 --- a/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_talker.py +++ b/vllm_omni/model_executor/models/qwen3_tts/qwen3_tts_talker.py @@ -1122,7 +1122,7 @@ class Qwen3TTSTalkerForConditionalGeneration(nn.Module): ) # Prefer GPU for encoder if available; otherwise keep CPU. dev = next(self.parameters()).device - if getattr(dev, "type", None) == "cuda": + if dev.type != "cpu": try: tok.model.to(dev) tok.device = dev