模型类与 API 参考（1/6）

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer.

--> This model was published in HF papers on 2025-03-26 and contributed to Hugging Face Transformers on 2025-04-14.

Qwen2.5-Omni

Overview

The Qwen2.5-Omni model is a unified multiple modalities model proposed in Qwen2.5-Omni Technical Report from Qwen team, Alibaba Group.

The abstract from the technical report is the following:

We present Qwen2.5-Omni, an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. To enable the streaming of multimodal information inputs, both audio and visual encoders utilize a block-wise processing approach. This strategy effectively decouples the handling of long sequences of multimodal data, assigning the perceptual responsibilities to the multimodal encoder and entrusting the modeling of extended sequences to a large language model. Such a division of labor enhances the fusion of different modalities via the shared attention mechanism. To synchronize the timestamps of video inputs with audio, we organized the audio and video sequentially in an interleaved manner and propose a novel position embedding approach, named TMRoPE (Time-aligned Multimodal RoPE). To concurrently generate text and speech while avoiding interference between the two modalities, we propose Thinker-Talker architecture. In this framework, Thinker functions as a large language model tasked with text generation, while Talker is a dual-track autoregressive model that directly utilizes the hidden representations from the Thinker to produce audio tokens as output. Both the Thinker and Talker models are designed to be trained and inferred in an end-to-end manner. For decoding audio tokens in a streaming manner, we introduce a sliding-window DiT that restricts the receptive field, aiming to reduce the initial package delay. Qwen2.5-Omni outperforms the similarly sized Qwen2-VL and Qwen2-Audio in both image and audio capabilities. Furthermore, Qwen2.5-Omni achieves state-of-the-art performance on multimodal benchmarks like Omni-Bench. Notably, Qwen2.5-Omni is the first open-source model to achieve a level of performance in end-to-end speech instruction following that is comparable to its capabilities with text inputs, as evidenced by benchmarks such as MMLU and GSM8K. As for speech generation, Qwen2.5-Omni's streaming Talker outperform most existing streaming and non-streaming alternatives in robustness and naturalness.

[!TIP] Set use_kernels=True in [~PreTrainedModel.from_pretrained] to replace supported layers with optimized kernels from the Hub. Refer to Loading kernels to learn more.

Notes

Use [Qwen2_5OmniForConditionalGeneration] to generate audio and text output. To generate only one output type, use [Qwen2_5OmniThinkerForConditionalGeneration] for text-only and [Qwen2_5OmniTalkersForConditionalGeneration] for audio-only outputs.
In case out-of-memory errors when working with video input, decrease 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.
The processor has its own [~ProcessorMixin.apply_chat_template] method to convert chat messages to model inputs.

Usage example

Qwen2.5-Omni can be found on the Huggingface Hub.

Single Media inference

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 Qwen2_5OmniForConditionalGeneration, Qwen2_5OmniProcessor

model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2.5-Omni-7B",
    device_map="auto",
)
processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")

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 `Qwen2-5-OmniProcessor`
    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)

Text-only generation

To generate only text output and save compute by not loading the audio generation model, we can use Qwen2_5OmniThinkerForConditionalGeneration model.

from transformers import Qwen2_5OmniProcessor, Qwen2_5OmniThinkerForConditionalGeneration

model = Qwen2_5OmniThinkerForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2.5-Omni-7B",
    device_map="auto",
)
processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")

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 `Qwen2-5-OmniProcessor`
    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)

Batch Mixed Media Inference

The model can batch inputs composed of mixed samples of various types such as text, images, audio and videos as input when using Qwen2_5OmniThinkerForConditionalGeneration model. Here is an example.

from transformers import Qwen2_5OmniForConditionalGeneration, Qwen2_5OmniProcessor

model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2.5-Omni-7B",
    device_map="auto"
)
processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")

# 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 `Qwen2-5-OmniProcessor`
    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)

Batch audio generation

[Qwen2_5OmniForConditionalGeneration] 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/Qwen2.5-Omni-7B"
model = AutoModelForTextToWaveform.from_pretrained(model_id, device_map="auto")
processor = AutoProcessor.from_pretrained(model_id)
sampling_rate = 24000  # output sampling rate
max_new_tokens = 200  # 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?",
    "The weather is nice today. Let's go for a walk.",
]
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"qwen2_5_omni_output_{sample_id}.wav",
        audio.detach().to(torch.float32).cpu().numpy(),
        sampling_rate,
    )
    print(f"Saved audio to qwen2_5_omni_output_{sample_id}.wav")

Usage Tips

Image Resolution trade-off

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/Qwen2.5-Omni-7B", min_pixels=min_pixels, max_pixels=max_pixels)

Prompt for audio output

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.",
}

Use audio output or not

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 = Qwen2_5OmniForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2.5-Omni-7B",
    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 = Qwen2_5OmniForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2.5-Omni-7B",
    device_map="auto",
    enable_audio_output=True,
)
...
text_ids = model.generate(**inputs, return_audio=False)

Change voice type of output audio

Qwen2.5-Omni 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/Qwen2.5-Omni-7B" 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")

Flash-Attention 2 to speed up generation

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 Qwen2_5OmniForConditionalGeneration

model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
    "Qwen/Qwen2.5-Omni-7B",
    device_map="auto",
    attn_implementation="flash_attention_2",
)

Qwen2_5OmniConfig

[[autodoc]] Qwen2_5OmniConfig

Qwen2_5OmniVisionEncoderConfig

[[autodoc]] Qwen2_5OmniVisionEncoderConfig

Qwen2_5OmniTextConfig

[[autodoc]] Qwen2_5OmniTextConfig

Qwen2_5OmniDiTConfig

[[autodoc]] Qwen2_5OmniDiTConfig

Qwen2_5OmniBigVGANConfig

[[autodoc]] Qwen2_5OmniBigVGANConfig

Qwen2_5OmniAudioEncoderConfig

[[autodoc]] Qwen2_5OmniAudioEncoderConfig

Qwen2_5OmniProcessor

[[autodoc]] Qwen2_5OmniProcessor - call

Qwen2_5OmniForConditionalGeneration

[[autodoc]] Qwen2_5OmniForConditionalGeneration - forward

Qwen2_5OmniPreTrainedModelForConditionalGeneration

[[autodoc]] Qwen2_5OmniPreTrainedModelForConditionalGeneration

Qwen2_5OmniThinkerConfig

[[autodoc]] Qwen2_5OmniThinkerConfig

Qwen2_5OmniThinkerForConditionalGeneration

[[autodoc]] Qwen2_5OmniThinkerForConditionalGeneration

Qwen2_5OmniThinkerTextModel

[[autodoc]] Qwen2_5OmniThinkerTextModel

Qwen2_5OmniTalkerConfig

[[autodoc]] Qwen2_5OmniTalkerConfig

Qwen2_5OmniTalkerForConditionalGeneration

[[autodoc]] Qwen2_5OmniTalkerForConditionalGeneration

Qwen2_5OmniTalkerModel

[[autodoc]] Qwen2_5OmniTalkerModel

Qwen2_5OmniToken2WavConfig

[[autodoc]] Qwen2_5OmniToken2WavConfig

Qwen2_5OmniToken2WavModel

[[autodoc]] Qwen2_5OmniToken2WavModel

Qwen2_5OmniToken2WavDiTModel

[[autodoc]] Qwen2_5OmniToken2WavDiTModel

Qwen2_5OmniToken2WavBigVGANModel

[[autodoc]] Qwen2_5OmniToken2WavBigVGANModel

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer.

-->

Pipelines

The pipelines are a great and easy way to use models for inference. These pipelines are objects that abstract most of the complex code from the library, offering a simple API dedicated to several tasks, including Named Entity Recognition, Masked Language Modeling, Sentiment Analysis, Feature Extraction and Question Answering. See the task list for examples of use.

There are two categories of pipeline abstractions to be aware about:

The [pipeline] which is the most powerful object encapsulating all other pipelines.
Task-specific pipelines are available for audio, computer vision, natural language processing, and multimodal tasks.

The pipeline abstraction

The pipeline abstraction is a wrapper around all the other available pipelines. It is instantiated as any other pipeline but can provide additional quality of life.

Simple call on one item:

>>> pipe = pipeline("text-classification")
>>> pipe("This restaurant is awesome")
[{'label': 'POSITIVE', 'score': 0.9998743534088135}]

If you want to use a specific model from the hub you can ignore the task if the model on the hub already defines it:

>>> pipe = pipeline(model="FacebookAI/roberta-large-mnli")
>>> pipe("This restaurant is awesome")
[{'label': 'NEUTRAL', 'score': 0.7313136458396912}]

To call a pipeline on many items, you can call it with a list.

>>> pipe = pipeline("text-classification")
>>> pipe(["This restaurant is awesome", "This restaurant is awful"])
[{'label': 'POSITIVE', 'score': 0.9998743534088135},
 {'label': 'NEGATIVE', 'score': 0.9996669292449951}]

To iterate over full datasets it is recommended to use a dataset directly. This means you don't need to allocate the whole dataset at once, nor do you need to do batching yourself. This should work just as fast as custom loops on GPU. If it doesn't don't hesitate to create an issue.

import datasets
from transformers import pipeline
from transformers.pipelines.pt_utils import KeyDataset
from tqdm.auto import tqdm

pipe = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h", device=0)
dataset = datasets.load_dataset("s3prl/superb", name="asr", split="test")

# KeyDataset (only *pt*) will simply return the item in the dict returned by the dataset item
# as we're not interested in the *target* part of the dataset. For sentence pair use KeyPairDataset
for out in tqdm(pipe(KeyDataset(dataset, "file"))):
    print(out)
    # {"text": "NUMBER TEN FRESH NELLY IS WAITING ON YOU GOOD NIGHT HUSBAND"}
    # {"text": ....}
    # ....

For ease of use, a generator is also possible:

from transformers import pipeline

pipe = pipeline("text-classification")

def data():
    while True:
        # This could come from a dataset, a database, a queue or HTTP request
        # in a server
        # Caveat: because this is iterative, you cannot use `num_workers > 1` variable
        # to use multiple threads to preprocess data. You can still have 1 thread that
        # does the preprocessing while the main runs the big inference
        yield "This is a test"

for out in pipe(data()):
    print(out)
    # {"text": "NUMBER TEN FRESH NELLY IS WAITING ON YOU GOOD NIGHT HUSBAND"}
    # {"text": ....}
    # ....

[[autodoc]] pipeline

Pipeline batching

All pipelines can use batching. This will work whenever the pipeline uses its streaming ability (so when passing lists or Dataset or generator).

from transformers import pipeline
from transformers.pipelines.pt_utils import KeyDataset
import datasets

dataset = datasets.load_dataset("stanfordnlp/imdb", name="plain_text", split="unsupervised")
pipe = pipeline("text-classification", device=0)
for out in pipe(KeyDataset(dataset, "text"), batch_size=8, truncation="only_first"):
    print(out)
    # [{'label': 'POSITIVE', 'score': 0.9998743534088135}]
    # Exactly the same output as before, but the content are passed
    # as batches to the model

However, this is not automatically a win for performance. It can be either a 10x speedup or 5x slowdown depending on hardware, data and the actual model being used.

Example where it's mostly a speedup:

from transformers import pipeline
from torch.utils.data import Dataset
from tqdm.auto import tqdm

pipe = pipeline("text-classification", device=0)

class MyDataset(Dataset):
    def __len__(self):
        return 5000

    def __getitem__(self, i):
        return "This is a test"

dataset = MyDataset()

for batch_size in [1, 8, 64, 256]:
    print("-" * 30)
    print(f"Streaming batch_size={batch_size}")
    for out in tqdm(pipe(dataset, batch_size=batch_size), total=len(dataset)):
        pass

# On GTX 970
------------------------------
Streaming no batching
100%|██████████████████████████████████████████████████████████████████████| 5000/5000 [00:26<00:00, 187.52it/s]
------------------------------
Streaming batch_size=8
100%|█████████████████████████████████████████████████████████████████████| 5000/5000 [00:04<00:00, 1205.95it/s]
------------------------------
Streaming batch_size=64
100%|█████████████████████████████████████████████████████████████████████| 5000/5000 [00:02<00:00, 2478.24it/s]
------------------------------
Streaming batch_size=256
100%|█████████████████████████████████████████████████████████████████████| 5000/5000 [00:01<00:00, 2554.43it/s]
(diminishing returns, saturated the GPU)

Example where it's mostly a slowdown:

class MyDataset(Dataset):
    def __len__(self):
        return 5000

    def __getitem__(self, i):
        if i % 64 == 0:
            n = 100
        else:
            n = 1
        return "This is a test" * n

This is a occasional very long sentence compared to the other. In that case, the whole batch will need to be 400 tokens long, so the whole batch will be [64, 400] instead of [64, 4], leading to the high slowdown. Even worse, on bigger batches, the program simply crashes.

------------------------------
Streaming no batching
100%|█████████████████████████████████████████████████████████████████████| 1000/1000 [00:05<00:00, 183.69it/s]
------------------------------
Streaming batch_size=8
100%|█████████████████████████████████████████████████████████████████████| 1000/1000 [00:03<00:00, 265.74it/s]
------------------------------
Streaming batch_size=64
100%|██████████████████████████████████████████████████████████████████████| 1000/1000 [00:26<00:00, 37.80it/s]
------------------------------
Streaming batch_size=256
  0%|                                                                                 | 0/1000 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "/home/nicolas/src/transformers/test.py", line 42, in <module>
    for out in tqdm(pipe(dataset, batch_size=256), total=len(dataset)):
....
    q = q / math.sqrt(dim_per_head)  # (bs, n_heads, q_length, dim_per_head)
RuntimeError: CUDA out of memory. Tried to allocate 376.00 MiB (GPU 0; 3.95 GiB total capacity; 1.72 GiB already allocated; 354.88 MiB free; 2.46 GiB reserved in total by PyTorch)

There are no good (general) solutions for this problem, and your mileage may vary depending on your use cases. Rule of thumb:

For users, a rule of thumb is:

Measure performance on your load, with your hardware. Measure, measure, and keep measuring. Real numbers are the only way to go.
If you are latency constrained (live product doing inference), don't batch.
If you are using CPU, don't batch.
If you are using throughput (you want to run your model on a bunch of static data), on GPU, then:
If you have no clue about the size of the sequence_length ("natural" data), by default don't batch, measure and try tentatively to add it, add OOM checks to recover when it will fail (and it will at some point if you don't control the sequence_length.)
If your sequence_length is super regular, then batching is more likely to be VERY interesting, measure and push it until you get OOMs.
The larger the GPU the more likely batching is going to be more interesting
As soon as you enable batching, make sure you can handle OOMs nicely.

Pipeline chunk batching

zero-shot-classification and question-answering are slightly specific in the sense, that a single input might yield multiple forward pass of a model. Under normal circumstances, this would yield issues with batch_size argument.

In order to circumvent this issue, both of these pipelines are a bit specific, they are ChunkPipeline instead of regular Pipeline. In short:

preprocessed = pipe.preprocess(inputs)
model_outputs = pipe.forward(preprocessed)
outputs = pipe.postprocess(model_outputs)

Now becomes:

all_model_outputs = []
for preprocessed in pipe.preprocess(inputs):
    model_outputs = pipe.forward(preprocessed)
    all_model_outputs.append(model_outputs)
outputs = pipe.postprocess(all_model_outputs)

This should be very transparent to your code because the pipelines are used in the same way.

This is a simplified view, since the pipeline can handle automatically the batch to you! Meaning you don't have to care about how many forward passes you inputs are actually going to trigger, you can optimize the batch_size independently of the inputs. The caveats from the previous section still apply.

Pipeline FP16 inference

Models can be run in FP16 which can be significantly faster on GPU while saving memory. Most models will not suffer noticeable performance loss from this. The larger the model, the less likely that it will.

To enable FP16 inference, you can simply pass dtype=torch.float16 or dtype='float16' to the pipeline constructor. Note that this only works for models with a PyTorch backend. Your inputs will be converted to FP16 internally.

Pipeline custom code

If you want to override a specific pipeline.

Don't hesitate to create an issue for your task at hand, the goal of the pipeline is to be easy to use and support most cases, so transformers could maybe support your use case.

If you want to try simply you can:

Subclass your pipeline of choice

class MyPipeline(TextClassificationPipeline):
    def postprocess():
        # Your code goes here
        scores = scores * 100
        # And here

my_pipeline = MyPipeline(model=model, tokenizer=tokenizer, ...)
# or if you use *pipeline* function, then:
my_pipeline = pipeline(model="xxxx", pipeline_class=MyPipeline)

That should enable you to do all the custom code you want.

Implementing a pipeline

Implementing a new pipeline

Audio

Pipelines available for audio tasks include the following.

AudioClassificationPipeline

[[autodoc]] AudioClassificationPipeline - call - all

AutomaticSpeechRecognitionPipeline

[[autodoc]] AutomaticSpeechRecognitionPipeline - call - all

TextToAudioPipeline

[[autodoc]] TextToAudioPipeline - call - all

ZeroShotAudioClassificationPipeline

[[autodoc]] ZeroShotAudioClassificationPipeline - call - all

Computer vision

Pipelines available for computer vision tasks include the following.

DepthEstimationPipeline

[[autodoc]] DepthEstimationPipeline - call - all

ImageClassificationPipeline

[[autodoc]] ImageClassificationPipeline - call - all

ImageSegmentationPipeline

[[autodoc]] ImageSegmentationPipeline - call - all

KeypointMatchingPipeline

[[autodoc]] KeypointMatchingPipeline - call - all

ObjectDetectionPipeline

[[autodoc]] ObjectDetectionPipeline - call - all

VideoClassificationPipeline

[[autodoc]] VideoClassificationPipeline - call - all

ZeroShotImageClassificationPipeline

[[autodoc]] ZeroShotImageClassificationPipeline - call - all

ZeroShotObjectDetectionPipeline

[[autodoc]] ZeroShotObjectDetectionPipeline - call - all

Natural Language Processing

Pipelines available for natural language processing tasks include the following.

FillMaskPipeline

[[autodoc]] FillMaskPipeline - call - all

TableQuestionAnsweringPipeline

[[autodoc]] TableQuestionAnsweringPipeline - call

TextClassificationPipeline

[[autodoc]] TextClassificationPipeline - call - all

TextGenerationPipeline

[[autodoc]] TextGenerationPipeline - call - all

TokenClassificationPipeline

[[autodoc]] TokenClassificationPipeline - call - all

ZeroShotClassificationPipeline

[[autodoc]] ZeroShotClassificationPipeline - call - all

Multimodal

Pipelines available for multimodal tasks include the following.

DocumentQuestionAnsweringPipeline

[[autodoc]] DocumentQuestionAnsweringPipeline - call - all

FeatureExtractionPipeline

[[autodoc]] FeatureExtractionPipeline - call - all

ImageFeatureExtractionPipeline

[[autodoc]] ImageFeatureExtractionPipeline - call - all

ImageTextToTextPipeline

[[autodoc]] ImageTextToTextPipeline - call - all

AnyToAnyPipeline

[[autodoc]] AnyToAnyPipeline - call - all

MaskGenerationPipeline

[[autodoc]] MaskGenerationPipeline - call - all

Parent class: `Pipeline`

[[autodoc]] Pipeline

模型类与 API 参考（1/3）