Instructions to use adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy with Transformers:

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

pipe = pipeline("image-text-to-text", model="adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy")
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 AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy")
model = AutoModelForMultimodalLM.from_pretrained("adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy", device_map="auto")
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?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy",
		"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/adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy

SGLang

How to use adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy 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 "adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy" \
    --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": "adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy",
		"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 "adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy" \
        --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": "adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy",
		"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 adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy with Docker Model Runner:
```
docker model run hf.co/adityabhushannagar/MiniCPM-v-4.7-35B-A3B-Copy
```

MiniCPM-V-4.7-35B-A3B

Mirror of OpenBMB/MiniCPM-V-4.7-35B-A3B from ModelScope. This is an unofficial copy. The source README is only a ModelScope stub with no model card, so everything below was read from the config files in this repo.

Integrity: all 27 files match the ModelScope source. The 16 weight shards and tokenizer.json were compared by SHA-256, and the other files byte-for-byte. All 1,152 tensors in the index are present in the shards.

Summary


Architecture	`MiniCPMV4_7ForConditionalGeneration` (`minicpmv4_7`)
Parameters	35.2B total (BF16), about 3B active per token (MoE)
Inputs	text, image, video
Output	text
Audio	not supported (audio tokens are defined in the tokenizer, but there is no audio encoder in the weights)
Context	262,144 tokens. No YaRN or other extension is configured

Architecture

Language model: qwen3_5_moe_text, 40 layers, hidden size 2048. Linear-attention layers alternate with periodic full-attention layers. 256 experts, 8 active per token.
Vision tower: 27 layers, hidden size 1152, input 980 px, patch size 14.
Projector: model.merger, with 16x downsampling of vision tokens (downsample_mode: 16x).
Images: large images are split into up to 9 slices (max_slice_nums: 9). Normalization uses mean and std of 0.5.
Positions: interleaved M-RoPE (mrope_mode: canvas), rope_theta 1e7.
Special tokens: <|image_pad|> (id 248056), <|video_pad|> (id 248057), <image>/</image>, <slice>/</slice>.

Weight groups in the index: language model 692 tensors, vision tower 453, merger 6, lm_head 1.

Usage

Generation defaults (generation_config.json): temperature=1.0, top_k=20, top_p=0.95.

This repo has no custom .py files, but preprocessor_config.json references image_processing_minicpmv4_7.py, video_processing_minicpmv4_7.py and processing_minicpmv4_7.py through auto_map. The ModelScope source doesn't include them either. You will need a transformers release that supports minicpmv4_7 natively, or the code files from OpenBMB's GitHub.

from transformers import AutoModelForImageTextToText, AutoProcessor

repo = "adityabhushannagar/MiniCPM-V-4.7-35B-A3B"
processor = AutoProcessor.from_pretrained(repo, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
    repo, dtype="bfloat16", device_map="auto", trust_remote_code=True
)

This snippet is untested. It depends on the processor code being available as described above.

Notes

No benchmarks or license text were published with the source files, and none are claimed here. The license is marked other, so check OpenBMB's terms before any use.
Weights occupy about 70.4 GB across 16 safetensors shards.

Source

Original release: OpenBMB, via ModelScope. See the OpenBMB organization for official model cards of related MiniCPM-V releases.

Downloads last month: 79

Safetensors

Model size

35B params

Tensor type

BF16