From 36bf2131816eb6d75fcf22adb3734850e90cfb98 Mon Sep 17 00:00:00 2001 From: Nathan Price <125999937+TheCodeWrangler@users.noreply.github.com> Date: Wed, 4 Mar 2026 02:29:01 -0600 Subject: [PATCH] [Bugfix] Add missing dynamic_arg_dims for Qwen3-ASR torch.compile (#35869) Signed-off-by: Nathan Price Signed-off-by: Isotr0py Co-authored-by: Isotr0py --- vllm/model_executor/models/qwen3_asr_realtime.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/vllm/model_executor/models/qwen3_asr_realtime.py b/vllm/model_executor/models/qwen3_asr_realtime.py index a149350d18..4fb6ef5d9f 100644 --- a/vllm/model_executor/models/qwen3_asr_realtime.py +++ b/vllm/model_executor/models/qwen3_asr_realtime.py @@ -22,7 +22,6 @@ from collections.abc import AsyncGenerator, Mapping import numpy as np import torch -from vllm.compilation.decorators import support_torch_compile from vllm.config import ModelConfig, SpeechToTextConfig, VllmConfig from vllm.inputs.data import PromptType, TokensPrompt from vllm.logger import init_logger @@ -177,7 +176,6 @@ class Qwen3ASRRealtimeMultiModalProcessor(Qwen3ASRMultiModalProcessor): info=Qwen3ASRProcessingInfo, dummy_inputs=Qwen3ASRDummyInputsBuilder, ) -@support_torch_compile class Qwen3ASRRealtimeGeneration(Qwen3ASRForConditionalGeneration, SupportsRealtime): realtime_max_tokens = 64