docs/source/en/model_doc/qwen3_omni_moe.md
This model was published in HF papers on 2025-09-22 and contributed to Hugging Face Transformers on 2025-09-21.
The Qwen3-Omni-MOE model is a unified multiple modalities model proposed in Qwen3-Omni Technical Report from Qwen team, Alibaba Group.
The abstract from the technical report is the following:
*We present Qwen3-Omni, a single multimodal model that, for the first time, maintains state-of-the-art performance across text, image, audio, and video without any degradation relative to single-modal counterparts. Qwen3-Omni matches the performance of same-sized single-modal models within the Qwen series and excels particularly on audio tasks. Across 36 audio and audio-visual benchmarks, Qwen3-Omni achieves open-source SOTA on 32 benchmarks and overall SOTA on 22, outperforming strong closed-source models such as Gemini-2.5-Pro, Seed-ASR, and GPT-4o-Transcribe. Qwen3-Omni adopts a Thinker-Talker MoE architecture that unifies perception and generation across text, images, audio, and video, yielding fluent text and natural real-time speech. It supports text interaction in 119 languages, speech understanding in 19 languages, and speech generation in 10 languages. To reduce first-packet latency in streaming synthesis, Talker autoregressively predicts discrete speech codecs using a multi-codebook scheme. Leveraging the representational capacity of these codebooks, we replace computationally intensive block-wise diffusion with a lightweight causal ConvNet, enabling streaming from the first codec frame. In cold-start settings, Qwen3-Omni achieves a theoretical end-to-end first-packet latency of 234 ms. To further strengthen multimodal reasoning, we introduce a Thinking model that explicitly reasons over inputs from any modality. Since the research community currently lacks a general-purpose audio captioning model, we fine-tuned Qwen3-Omni-30B-A3B to obtain Qwen3-Omni-30B-A3B-Captioner, which produces detailed, low-hallucination captions for arbitrary audio inputs. Qwen3-Omni-30B-A3B, Qwen3-Omni-30B-A3B-Thinking, and Qwen3-Omni-30B-A3B-Captioner are publicly released under the Apache 2.0 license.
Qwen3OmniMoeForConditionalGeneration] to generate audio and text output. To generate only one output type, use [Qwen3OmniMoeThinkerForConditionalGeneration] for text-only and [Qwen3OmniMoeTalkerForConditionalGeneration] for audio-only outputs.processor.max_pixels. By default the maximum is set to a very large value and high resolution visuals will not be resized, unless resolution exceeds processor.max_pixels.~ProcessorMixin.apply_chat_template] method to convert chat messages to model inputs.Qwen3-Omni can be found on the Huggingface Hub.
The model can accept text, images, audio and videos as input. Here's an example code for inference.
import soundfile as sf
from transformers import Qwen3OmniMoeForConditionalGeneration, Qwen3OmniMoeProcessor
model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
device_map="auto"
)
processor = Qwen3OmniMoeProcessor.from_pretrained("Qwen/Qwen3-Omni-30B-A3B-Instruct")
conversations = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "video", "video": "/path/to/video.mp4"},
{"type": "text", "text": "What can't you hear and see in this video?"},
],
},
]
inputs = processor.apply_chat_template(
conversations,
load_audio_from_video=True,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
fps=1,
# kwargs to be passed to `Qwen3OmniMoeProcessor`
padding=True,
use_audio_in_video=True,
).to(model.device)
# Generation params for audio or text can be different and have to be prefixed with `thinker_` or `talker_`
text_ids, audio = model.generate(**inputs, use_audio_in_video=True, thinker_do_sample=False, talker_do_sample=True)
text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
sf.write(
"output.wav",
audio.reshape(-1).detach().cpu().numpy(),
samplerate=24000,
)
print(text)
To generate only text output and save compute by not loading the audio generation model, we can use Qwen3OmniMoeThinkerForConditionalGeneration model.
from transformers import Qwen3OmniMoeProcessor, Qwen3OmniMoeThinkerForConditionalGeneration
model = Qwen3OmniMoeThinkerForConditionalGeneration.from_pretrained(
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
device_map="auto",
)
processor = Qwen3OmniMoeProcessor.from_pretrained("Qwen/Qwen3-Omni-30B-A3B-Instruct")
conversations = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "video", "video": "/path/to/video.mp4"},
{"type": "text", "text": "What can't you hear and see in this video?"},
],
},
]
inputs = processor.apply_chat_template(
conversations,
load_audio_from_video=True,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
fps=1,
# kwargs to be passed to `Qwen3OmniMoeProcessor`
padding=True,
use_audio_in_video=True,
).to(model.device)
text_ids = model.generate(**inputs, use_audio_in_video=True)
text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(text)
The model can batch inputs composed of mixed samples of various types such as text, images, audio and videos as input when using Qwen3OmniMoeThinkerForConditionalGeneration model. Here is an example.
import soundfile as sf
from transformers import Qwen3OmniMoeForConditionalGeneration, Qwen3OmniMoeProcessor
model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
device_map="auto"
)
processor = Qwen3OmniMoeProcessor.from_pretrained("Qwen/Qwen3-Omni-30B-A3B-Instruct")
# Conversation with video only
conversation1 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "video", "path": "/path/to/video.mp4"},
]
}
]
# Conversation with audio only
conversation2 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "audio", "path": "/path/to/audio.wav"},
]
}
]
# Conversation with pure text
conversation3 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [{"type": "text", "text": "who are you?"}],
}
]
# Conversation with mixed media
conversation4 = [
{
"role": "system",
"content": [
{"type": "text", "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech."}
],
},
{
"role": "user",
"content": [
{"type": "image", "path": "/path/to/image.jpg"},
{"type": "video", "path": "/path/to/video.mp4"},
{"type": "audio", "path": "/path/to/audio.wav"},
{"type": "text", "text": "What are the elements can you see and hear in these media?"},
],
}
]
conversations = [conversation1, conversation2, conversation3, conversation4]
inputs = processor.apply_chat_template(
conversations,
load_audio_from_video=True,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
fps=1,
# kwargs to be passed to `Qwen3OmniMoeProcessor`
padding=True,
use_audio_in_video=True,
).to(model.thinker.device)
text_ids = model.generate(**inputs, use_audio_in_video=True)
text = processor.batch_decode(text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
print(text)
[Qwen3OmniMoeForConditionalGeneration] supports batched audio output generation. For example, below for text-to-speech batch generation. With a batch of more than one sample, the generated audio is returned as a list with one waveform per sample, each already trimmed to its own length, so shorter samples are not padded with spurious audio. A single-sample call still returns one tensor.
import soundfile as sf
from transformers import AutoModelForTextToWaveform, AutoProcessor
import torch
model_id = "Qwen/Qwen3-Omni-30B-A3B-Instruct"
model = AutoModelForTextToWaveform.from_pretrained(model_id, device_map="auto")
processor = AutoProcessor.from_pretrained(model_id)
sampling_rate = 24000 # output sampling rate
max_new_tokens = 128 # maximum number of tokens to generate
system_text = (
"You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of "
"perceiving auditory and visual inputs, as well as generating text and speech."
)
texts = [
"Hello, I'm Qwen. How can I help you today? Just let me know!",
"The weather is nice today.",
]
inputs = processor.apply_chat_template(
[
[
{"role": "system", "content": [{"type": "text", "text": system_text}]},
{
"role": "user",
"content": [
{
"type": "text",
"text": (
"Please read the following text aloud exactly as written, with no "
f"additional commentary:\n\n{text}"
),
}
],
},
]
for text in texts
],
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt",
processor_kwargs={"padding": True},
).to(model.device, dtype=model.dtype)
gen_kwargs = {
"talker_do_sample": True,
"speaker": "Ethan", # Ethan, Chelsie
"talker_max_new_tokens": max_new_tokens,
}
_, pred_waveform = model.generate(**inputs, **gen_kwargs)
for sample_id, audio in enumerate(pred_waveform):
sf.write(f"qwen3_omni_output_{sample_id}.wav", audio.detach().to(torch.float32).cpu().numpy(), sampling_rate)
print(f"Saved audio to qwen3_omni_output_{sample_id}.wav")
The model supports a wide range of resolution inputs. By default, it uses the native resolution for input, but higher resolutions can enhance performance at the cost of more computation. Users can set the minimum and maximum number of pixels to achieve an optimal configuration for their needs.
min_pixels = 128*28*28
max_pixels = 768*28*28
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-Omni-30B-A3B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels)
If users need audio output, the system prompt must be set as "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.", otherwise the audio output may not work as expected.
{
"role": "system",
"content": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, capable of perceiving auditory and visual inputs, as well as generating text and speech.",
}
The model supports both text and audio outputs, if users do not need audio outputs, they can set enable_audio_output in the from_pretrained function. This option will save about ~2GB of GPU memory but the return_audio option for generate function will only allow to be set at False.
model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
device_map="auto",
enable_audio_output=False,
)
In order to obtain a flexible experience, we recommend that users set enable_audio_output at True when initializing the model through from_pretrained function, and then decide whether to return audio when generate function is called. When return_audio is set to False, the model will only return text outputs to get text responses faster.
model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
device_map="auto",
enable_audio_output=True,
)
...
text_ids = model.generate(**inputs, return_audio=False)
Qwen3-Omni-MOE supports the ability to change the voice of the output audio. Users can use the speaker parameter of generate function to specify the voice type. The "Qwen/Qwen3-Omni-30B-A3B-Instruct" checkpoint support two voice types: Chelsie and Ethan, while Chelsie is a female voice and Ethan is a male voice. By default, if speaker is not specified, the default voice type is Chelsie.
text_ids, audio = model.generate(**inputs, speaker="Chelsie")
text_ids, audio = model.generate(**inputs, speaker="Ethan")
First, make sure to install the latest version of Flash Attention 2:
pip install -U flash-attn --no-build-isolation
Also, you should have hardware that is compatible with FlashAttention 2. Read more about it in the official documentation of the flash attention repository. FlashAttention-2 can only be used when a model is loaded in torch.float16 or torch.bfloat16.
To load and run a model using FlashAttention-2, add attn_implementation="flash_attention_2" when loading the model:
from transformers import Qwen3OmniMoeForConditionalGeneration
model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
"Qwen/Qwen3-Omni-30B-A3B-Instruct",
device_map="auto",
attn_implementation="flash_attention_2",
)
[[autodoc]] Qwen3OmniMoeConfig
[[autodoc]] Qwen3OmniMoeThinkerConfig
[[autodoc]] Qwen3OmniMoeTalkerConfig
[[autodoc]] Qwen3OmniMoeTextConfig
[[autodoc]] Qwen3OmniMoeVisionEncoderConfig
[[autodoc]] Qwen3OmniMoeAudioEncoderConfig
[[autodoc]] Qwen3OmniMoeTalkerTextConfig
[[autodoc]] Qwen3OmniMoeTalkerCodePredictorConfig
[[autodoc]] Qwen3OmniMoeForConditionalGeneration
[[autodoc]] Qwen3OmniMoeThinkerTextModel
[[autodoc]] Qwen3OmniMoeThinkerForConditionalGeneration
[[autodoc]] Qwen3OmniMoeTalkerForConditionalGeneration
[[autodoc]] Qwen3OmniMoePreTrainedModel
[[autodoc]] Qwen3OmniMoePreTrainedModelForConditionalGeneration
[[autodoc]] Qwen3OmniMoeTalkerModel
[[autodoc]] Qwen3OmniMoeThinkerTextPreTrainedModel
[[autodoc]] Qwen3OmniMoeProcessor - call
[[autodoc]] Qwen3OmniMoeCode2Wav
[[autodoc]] Qwen3OmniMoeCode2WavDecoderBlock
[[autodoc]] Qwen3OmniMoeCode2WavTransformerModel
[[autodoc]] Qwen3OmniMoeTalkerCodePredictorModel
[[autodoc]] Qwen3OmniMoeTalkerCodePredictorModelForConditionalGeneration