Instructions to use 88plug/MiniCPM-o-4.5-W4A16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use 88plug/MiniCPM-o-4.5-W4A16 with Transformers:

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

pipe = pipeline("image-text-to-text", model="88plug/MiniCPM-o-4.5-W4A16", 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 AutoModel
model = AutoModel.from_pretrained("88plug/MiniCPM-o-4.5-W4A16", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use 88plug/MiniCPM-o-4.5-W4A16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "88plug/MiniCPM-o-4.5-W4A16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "88plug/MiniCPM-o-4.5-W4A16",
		"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/88plug/MiniCPM-o-4.5-W4A16

SGLang

How to use 88plug/MiniCPM-o-4.5-W4A16 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 "88plug/MiniCPM-o-4.5-W4A16" \
    --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": "88plug/MiniCPM-o-4.5-W4A16",
		"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 "88plug/MiniCPM-o-4.5-W4A16" \
        --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": "88plug/MiniCPM-o-4.5-W4A16",
		"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 88plug/MiniCPM-o-4.5-W4A16 with Docker Model Runner:
```
docker model run hf.co/88plug/MiniCPM-o-4.5-W4A16
```

mlx-vlm MiniCPM-o-4.5 (quantized, 4-bit) fails with corrupted/blank vision output despite matching converter version

by joeljbrave - opened Jul 15

Discussion

joeljbrave

Jul 15

Trying to run mlx-community/MiniCPM-o-4_5-4bit locally on Apple Silicon (16GB) with mlx-vlm 0.4.0 (matching the version the checkpoint was converted with, per its model card). Text-only generation works cleanly. Image input loads and runs without crashing but produces near-blank/garbage output regardless of the actual image content.

Traced one confirmed bug: mlx_vlm/models/minicpmo/minicpmo.py's get_vision_embedding() infers a compute dtype from self.language_model.model.embed_tokens.weight.dtype — for a quantized model this is uint32 (MLX's packed 4-bit storage format), not a real float type, and casting pixel data to it corrupted input to the vision tower's conv layer (ValueError: [conv] Invalid input array with type uint32).

Patched this locally by using the layer's scales.dtype instead when quantized.

This resolved the crash, but image output is still near-blank/garbage afterward — suggesting at least one more instance of the same class of bug (a quantized layer's raw storage dtype leaking into a float-expecting computation) elsewhere in the vision or resampler path. Has anyone hit this specifically with quantized MiniCPM-o checkpoints, or found the remaining spot(s) where this happens?

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