tiny-random/minicpm-v-4.7

This is a randomly initialized tiny model based on openbmb/MiniCPM-V-4.7-1B for testing and debugging.

Feature	Source	This repo
Checkpoint tensor elements	1.55B BF16	11.54M BF16
Safetensors weight files	3.11 GB (model.safetensors)	23.11 MB (model.safetensors)
Architecture / input	MiniCPM-V 4.7 dense; text, image, video	Same
Vocabulary rows / tied embeddings	248,094 / false	Same
Backbone layers / layer cadence	24 / six 3-linear + 1-full cycles	8 / two complete cycles
Hidden width / MLP width	1,024 / 3,584	16 / 32
Q heads / KV heads / head dimension	8 / 2 / 256	Same
Linear-attention key/value heads and dimensions	16/16, 128/128	Same
Vision layers / width / insert merger	27 / 1,152 / after layer 6	4 / 128 / after layer 2
M-RoPE	canvas; sections 11/11/10; partial 0.25	canvas; sections 11/11/10; partial 0.25
MTP	config declares one layer; no MTP tensors	Same
Checkpoint precision	BF16	Same

Two complete linear/linear/linear/full cycles are retained. The Qwen3.5 linear-attention path requires CUDA and Triton. The exact public snippet was verified on an NVIDIA A800-SXM4-80GB.

File path	Size
model.safetensors	23.1MB

Example usage

Transformers (✅ verified)

The example sends one image and one predecoded video. The Qwen3.5 linear-attention path requires CUDA and Triton. Install transformers[torch]>=5.18.0 and Pillow.

def run_transformers(model_id: str):
    import numpy as np
    import torch
    from PIL import Image, ImageDraw
    from transformers import AutoModelForImageTextToText, AutoProcessor
    from transformers.video_utils import VideoMetadata

    model_id = str(model_id)
    processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
    model = AutoModelForImageTextToText.from_pretrained(
        model_id,
        dtype=torch.bfloat16,
        device_map="cuda",
        attn_implementation="sdpa",
    )
    image = Image.new("RGB", (448, 448), "white")
    ImageDraw.Draw(image).text((20, 20), "Hello", fill="black")
    frames = np.stack(
        [
            np.asarray(Image.new("RGB", (64, 64), color))
            for color in ((220, 30, 30), (30, 220, 30), (30, 30, 220), (220, 220, 30))
        ]
    )
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "image": image},
                {"type": "video", "video": frames},
                {"type": "text", "text": "Describe the image and the video."},
            ],
        }
    ]
    inputs = processor.apply_chat_template(
        messages,
        tokenize=True,
        add_generation_prompt=True,
        return_dict=True,
        return_tensors="pt",
        downsample_mode="16x",
        max_num_frames=4,
        stack_frames=1,
        max_slice_nums=1,
        use_image_id=False,
        enable_thinking=False,
        do_sample_frames=False,
        video_metadata=[VideoMetadata(total_num_frames=4, fps=2, frames_indices=[0, 1, 2, 3])],
    ).to(model.device)
    with torch.no_grad():
        logits = model(**inputs).logits
        assert torch.isfinite(logits).all()
        output = model.generate(**inputs, downsample_mode="16x", do_sample=False, max_new_tokens=10)
    print(processor.decode(output[0, inputs["input_ids"].shape[1] :], skip_special_tokens=True))


if __name__ == "__main__":
    run_transformers(model_id="tiny-random/minicpm-v-4.7")

Creating this repo

• Code: .meta/create_model.py
• Model structure: .meta/model.log
• Environment: .meta/env.json

Change log

• 2026-11-01: Initial tiny MiniCPM-V 4.7 checkpoint.

This repo is generated by DeepSeek-V4.1-Flash.