Instructions to use tencent/Youtu-VL-4B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tencent/Youtu-VL-4B-Instruct with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="tencent/Youtu-VL-4B-Instruct", trust_remote_code=True)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("tencent/Youtu-VL-4B-Instruct", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tencent/Youtu-VL-4B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tencent/Youtu-VL-4B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tencent/Youtu-VL-4B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/tencent/Youtu-VL-4B-Instruct

SGLang

How to use tencent/Youtu-VL-4B-Instruct with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "tencent/Youtu-VL-4B-Instruct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tencent/Youtu-VL-4B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "tencent/Youtu-VL-4B-Instruct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tencent/Youtu-VL-4B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use tencent/Youtu-VL-4B-Instruct with Docker Model Runner:
```
docker model run hf.co/tencent/Youtu-VL-4B-Instruct
```

The latest version of transformers breaks this model

#12

by jrp2024 - opened Aug 30

Discussion

jrp2024

Aug 30

Crash: tencent/Youtu-VL-4B-Instruct

Maintainer evidence

The model repo code imports DefaultFastImageProcessorKwargs from transformers.image_processing_utils_fast, which transformers 5.16 removed — a model-repo-code compatibility break.

Root exception and chain

builtins.ImportError: cannot import name 'DefaultFastImageProcessorKwargs' from 'transformers.image_processing_utils_fast' (unknown location)
builtins.ValueError: Model loading failed: cannot import name 'DefaultFastImageProcessorKwargs' from 'transformers.image_processing_utils_fast' (unknown location)

Execution and provenance

Complete traceback

Traceback (most recent call last):
  File "~/Documents/AI/mlx/check_models/src/check_models.py", line 13593, in _run_model_generation
    model, processor, config = _load_model(params)
                               ~~~~~~~~~~~^^^^^^^^
  File "~/Documents/AI/mlx/check_models/src/check_models.py", line 12726, in _load_model
    model, processor = load(
                       ~~~~^
        path_or_hf_repo=params.model_identifier,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<5 lines>...
        quantize_activations=params.quantize_activations,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/check_models/src/check_models.py", line 818, in _typed_mlx_vlm_load
    loaded: tuple[nn.Module, ProcessorMixin] = _mlx_vlm_load(
                                               ~~~~~~~~~~~~~^
        path_or_hf_repo=path_or_hf_repo,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<5 lines>...
        **kwargs,
        ^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/utils.py", line 1217, in load
    processor = load_processor(model_path, True, eos_token_ids=eos_token_id, **kwargs)
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/utils.py", line 1362, in load_processor
    processor = AutoProcessor.from_pretrained(model_path, **kwargs)
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/base.py", line 644, in _patched_auto_processor_from_pretrained
    return previous_from_pretrained.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/base.py", line 644, in _patched_auto_processor_from_pretrained
    return previous_from_pretrained.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/base.py", line 644, in _patched_auto_processor_from_pretrained
    return previous_from_pretrained.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  [Previous line repeated 9 more times]
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/molmo2/processing.py", line 768, in _patched_auto_processor_from_pretrained_molmo2
    return _original_auto_processor_from_pretrained_molmo2.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/base.py", line 644, in _patched_auto_processor_from_pretrained
    return previous_from_pretrained.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/base.py", line 644, in _patched_auto_processor_from_pretrained
    return previous_from_pretrained.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/mlx-vlm/mlx_vlm/models/base.py", line 644, in _patched_auto_processor_from_pretrained
    return previous_from_pretrained.__func__(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        cls, pretrained_model_name_or_path, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  [Previous line repeated 11 more times]
  File "~/miniconda3/envs/mlx-vlm/lib/python3.14/site-packages/transformers/models/auto/processing_auto.py", line 326, in from_pretrained
    return processor_class.from_pretrained(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/miniconda3/envs/mlx-vlm/lib/python3.14/site-packages/transformers/processing_utils.py", line 1735, in from_pretrained
    args = cls._get_arguments_from_pretrained(pretrained_model_name_or_path, processor_dict, **kwargs)
  File "~/miniconda3/envs/mlx-vlm/lib/python3.14/site-packages/transformers/processing_utils.py", line 1875, in _get_arguments_from_pretrained
    sub_processor = auto_processor_class.from_pretrained(
        pretrained_model_name_or_path, subfolder=subfolder, **kwargs
    )
  File "~/miniconda3/envs/mlx-vlm/lib/python3.14/site-packages/transformers/models/auto/image_processing_auto.py", line 672, in from_pretrained
    image_processor_class = get_class_from_dynamic_module(class_ref, pretrained_model_name_or_path, **kwargs)
  File "~/miniconda3/envs/mlx-vlm/lib/python3.14/site-packages/transformers/dynamic_module_utils.py", line 623, in get_class_from_dynamic_module
    return get_class_in_module(class_name, final_module, force_reload=force_download)
  File "~/miniconda3/envs/mlx-vlm/lib/python3.14/site-packages/transformers/dynamic_module_utils.py", line 309, in get_class_in_module
    module_spec.loader.exec_module(module)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
  File "<frozen importlib._bootstrap_external>", line 759, in exec_module
  File "<frozen importlib._bootstrap>", line 491, in _call_with_frames_removed
  File "~/.cache/huggingface/modules/transformers_modules/_8d30a0e49662a1d628a472b12df264dbcd768753/71940f1d5c3bbe2f/image_processing_siglip2_fast.py", line 7, in <module>
    from transformers.image_processing_utils_fast import (
    ...<3 lines>...
    )
ImportError: cannot import name 'DefaultFastImageProcessorKwargs' from 'transformers.image_processing_utils_fast' (unknown location)

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "~/Documents/AI/mlx/check_models/src/check_models.py", line 14606, in process_image_with_model
    output: GenerationResult | SupportsGenerationResult = _run_model_generation(
                                                          ~~~~~~~~~~~~~~~~~~~~~^
        params=params,
        ^^^^^^^^^^^^^^
        phase_callback=_update_phase,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
        phase_timer=phase_timer,
        ^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "~/Documents/AI/mlx/check_models/src/check_models.py", line 13608, in _run_model_generation
    raise _tag_exception_failure_phase(ValueError(error_details), "model_load") from load_err
ValueError: Model loading failed: cannot import name 'DefaultFastImageProcessorKwargs' from 'transformers.image_processing_utils_fast' (unknown location)

Captured stdout/stderr

=== STDERR ===
Downloading bytes:           |  0.00B
Reconstructing (incomplete total...): |          |  0.00B /  0.00B
Fetching 19 files:   0%|          | 0/19 [00:00<?, ?it/s]
Fetching 19 files: 100%|##########| 19/19 [00:00<00:00, 2785.16it/s]
Download complete: :           |  0.00B
Reconstruction complete: |          |  0.00B /  0.00B
Download complete: :           |  0.00B
Reconstruction complete: |          |  0.00B /  0.00B
[01:47:22] DEBUG    HF Cache Info for tencent/Youtu-VL-4B-Instruct: size=10229.6 MB, files=27

Reproduction inputs

Image format: JPEG
Image dimensions: 9,984 x 5,616 pixels
Image size: 33,850,802 bytes
Image SHA-256: b318746396941f675647ccf9ebdf8652161618926a84538fac85170096a7f92c

Exact prompt

Create British-English catalogue metadata from the image and supplied context.

Treat any capture date/time and GPS as authoritative facts, but do not claim they are visible. Descriptive hints may be incomplete or wrong: retain details supported by the image, correct conflicts, and add important visible details. Prefer image evidence when a hint conflicts, and omit uncertain details.

Context: Authoritative context:
- Capture date/time: 2026-08-28 16:17:42 UTC+01:00

Descriptive hints:
- Description hint: A white motor cruiser boat named 'Wavey Katey II' cruises along a calm waterway, flying a British maritime flag, past a rustic wooden riverside house and lush green trees.
- Keyword hints: Boat, Boat driver, Boat fender, Boating, Cabin cruiser, Canopy, Cottage, Cruising, Flag, Foliage, Leisure, Motorboat, Nautical, Outboard motor, Passenger, Railing, River, Riverbank, Shrubs, Trees

Write:
- a concrete 5-10-word title;
- a 1-2-sentence factual description combining relevant context with the main visible subject, setting, action, lighting, and distinctive details;
- 10-18 unique, comma-separated keywords covering relevant context and visible details.

Return exactly these three sections and nothing else:
Title:
Description:
Keywords:

The crash occurred during model load, before image decoding, so the exact
input image is not required: substitute any local image for the placeholder
path and run one native mlx-vlm process.

python -m mlx_vlm.generate --model tencent/Youtu-VL-4B-Instruct --image any-local-image.jpg --prompt x --max-tokens 8 --temperature 0.0 --revision 8d30a0e49662a1d628a472b12df264dbcd768753 --trust-remote-code

Provenance and Environment

Components

Component	Value
mlx-vlm	0.7.0rc0
mlx	0.32.3.dev20260829+052e77db9
mlx-lm	0.32.0
transformers	5.16.1
tokenizers	0.23.1
huggingface-hub	1.29.0
Pillow	12.3.0
Python Version	3.14.7
macOS Version	26.6.2
GPU/Chip	Apple M5 Max
check_models	0.16.5; revision 7299db1db9c3863984bb90b7ee4129779c30877b; clean

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment