Skip to content
Computer vision

RK182X-VLM-MinerU2.5-Pro-2604-1.2B

VLM Conversion for Rockchip RK182X Chips

2 min readupdated

This model is a ported version of the VLM (vision-language model) OpenDataLab/MinerU2.5-Pro-2604-1.2B tailored for the RK182X series NPU hardware accelerators (RK1820 and RK1828 chips). Featuring 1.2 billion parameters, the base AI model is designed to efficiently recognize and parse complex documents, including layout formatting, text, formulas, and tables. The new version is optimized to run on embedded devices using the RKNN Runtime v1.1.0 environment.

Key features

  • NPU Optimization: Weights are converted to run on RK1820 and RK1828 hardware platforms using w4a16 quantization.
  • Document parsing: Accurate extraction of text, tables, and formulas with support for a context window of up to 8192 tokens.
  • Flexible configurations: Variants are available with vision resolutions of 1036x1036 (for RK1828) and 448x448 (for RK1820 and RK1828).
  • Ready-to-use environments: The repository includes configuration files and manifests with SHA-256 verification support.

What you can do

  • Deploy on hardware: Install the model on devices powered by Rockchip RK182X processors for local document recognition.
  • Automate PDF processing: Integrate the model into your pipeline for processing scanned documents, reports, and research papers.
  • Recognize formulas and tables: Leverage VLM capabilities to digitize complex printed materials without sending data to the cloud.

Example

Downloading the model repository using the Hugging Face CLI utility:

hf download UGREENDEV/RK182X-VLM-MinerU2.5-Pro-2604-1.2B --revision v1.1.0 --local-dir ./RK182X-VLM-MinerU2.5-Pro-2604-1.2B
Read moreDetails
The original model is distributed under the Apache License 2.0.
A compatible RKNN Runtime version and drivers are required (verify via rknn-smi -v).
Operation is supported on RK1820 and RK1828 chips with fp16 KV cache.
Discuss Telegram 0 203

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.3 GB · ready files0.8 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.1 GB0.6 GB
  • GGUFweights 0.0 GB · ready files0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # UGREENDEV/RK182X-VLM-MinerU2.5-Pro-2604-1.2B
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download UGREENDEV/RK182X-VLM-MinerU2.5-Pro-2604-1.2B --local-dir ~/models/RK182X-VLM-MinerU2.5-Pro-2604-1.2B
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of RK182X-VLM-MinerU2.5-Pro-2604-1.2B (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

UGREENDEV/RK182X-VLM-MinerU2.5-Pro-2604-1.2B
Author
UGREENDEV
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

2030

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments