[Usage]: How to set the language in Qwen3-Asr

[xyqsgdog-ctrl]

Your current environment

import requests

url = "http://0.0.0.0:9881/v1/chat/completions"
headers = {"Content-Type": "application/json"}
target_language = "Arabic"

data = {
    "messages": [
        {
            "role": "user",
            "content": [
                {"type": "text", "text": f"Transcribe the following audio into {target_language}."},
                {
                    "type": "audio_url",
                    "audio_url": {
                        "url": "file:///nas/xyq/Qwen-Asr-POC/5262126349647872/1770072158499.mp3" # "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_en.wav"
                    },
                }
            ],
        }
    ]
}

response = requests.post(url, headers=headers, json=data, timeout=300)
response.raise_for_status()
content = response.json()['choices'][0]['message']['content']
print(content)
language, text = content.split("<asr_text>")

print(language)
print(text)

(base) ➜ Qwen3-ASR git:(main) ✗ python3 infer_vllm_demo.py
language English<asr_text>I'm a geek. I'm a geek. I'm a geek. Yeah, I'm a geek. I'm a geek. I'm a geek. I'm a geek. I'm a geek.
language English
I'm a geek. I'm a geek. I'm a geek. Yeah, I'm a geek. I'm a geek. I'm a geek. I'm a geek. I'm a geek.
(base) ➜ Qwen3-ASR git:(main) ✗

I want the language to be Arabic

How would you like to use vllm

I want to run inference of a [specific model](put link here). I don't know how to integrate it with vllm.

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[lixikun]

+1

[ucanseeme-cmd]

Have you tried Transcriptions API?
https://docs.vllm.ai/en/latest/serving/openai_compatible_server/#transcriptions-api

[xyqsgdog-ctrl]

Have you tried Transcriptions API? https://docs.vllm.ai/en/latest/serving/openai_compatible_server/#transcriptions-api

`from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:9882/v1",
    api_key="empty",
)

audio_path = "/nas/xyq/Qwen-Asr-POC/temp_audio.mp3"

# Upload audio file from disk
with open(audio_path, "rb") as audio_file:
    transcription = client.audio.transcriptions.create(
        model="/nas/llm_models/Qwen3/Qwen3-ASR-0.6B",
        file=audio_file,
        language="ar",
        response_format="json",
    )

print(transcription.text)` The output is : language Chinese<asr_text>我覺得，我覺得，我覺得。 I want it to be Arabic.

[gordon0414]

This issue is caused by qwen-asr package, not vLLM itself.

In get_generation_prompt, only to_language is checked but the Whisper-compatible
language parameter (passed as language arg) is never used as fallback.

I submitted a fix: QwenLM/Qwen3-ASR#101

One-line change:

# before
if to_language is None:
# after
effective_language = to_language or language
if effective_language is None:

Tested with Qwen3-ASR-1.7B on vLLM — language=ko now correctly injects
the language prefill into the assistant prompt.

[Anthonyy232]

Would this also fix if using through transformers? I'm using it right now and im pretty sure it doesn't respect language either 😆

[xyqsgdog-ctrl]

This issue is caused by qwen-asr package, not vLLM itself.

In get_generation_prompt, only to_language is checked but the Whisper-compatible language parameter (passed as language arg) is never used as fallback.

I submitted a fix: QwenLM/Qwen3-ASR#101

One-line change:

before

if to_language is None:

after

effective_language = to_language or language
if effective_language is None:
Tested with Qwen3-ASR-1.7B on vLLM — language=ko now correctly injects the language prefill into the assistant prompt.

sloved, thx a lot