---
license: apache-2.0
base_model: openbmb/MiniCPM3-4B
pipeline_tag: text-generation
library_name: rkllm
tags:
- rkllm
- rockchip
- rk3576
- rk3588
- minicpm3
---

# MiniCPM3-4B-RKLLM

RKLLM-converted MiniCPM3-4B language-model artifacts for Rockchip RK3576 and RK3588 NPUs.

These are hardware-specific `.rkllm` binaries, not Transformers checkpoints. They require a compatible Rockchip RKLLM runtime and cannot be loaded directly with Transformers, llama.cpp, or Ollama.

## Base model

- Upstream model: [openbmb/MiniCPM3-4B](https://huggingface.co/openbmb/MiniCPM3-4B)
- License: Apache-2.0
- Model type: LLM (text only)

## Conversion and variants

### Toolkit version

**RKLLM Toolkit: v1.3.0**

Use a file built for the exact target SoC.

| Target | Quantization | File | SHA256 |
| --- | --- | --- | --- |
| RK3576 | W4A16 (g128) | [MiniCPM3-4B_RK3576_w4a16_g128.rkllm](RK3576/MiniCPM3-4B_RK3576_w4a16_g128.rkllm) | `ec05211d17c4bb05a685588ff335608233b59722f6ecc296976d61dbc88c4128` |
| RK3576 | W8A8 | [MiniCPM3-4B_RK3576_w8a8.rkllm](RK3576/MiniCPM3-4B_RK3576_w8a8.rkllm) | `3795c6a22e5f12de0c4243fc187da55f516b530de3125734874d340856160aec` |
| RK3588 | W8A8 | [MiniCPM3-4B_RK3588_w8a8.rkllm](RK3588/MiniCPM3-4B_RK3588_w8a8.rkllm) | `3815a59593df7aebda90f86844caf2fd625d9d0e996e35aa31fd74790d10b130` |

The repository also includes `MiniCPM3-4B_data_quant.json`, used as calibration data during conversion.

## Usage

```bash
hf download HanzoHuang/MiniCPM3-4B-RKLLM \
  RK3576/MiniCPM3-4B_RK3576_w4a16_g128.rkllm \
  --local-dir MiniCPM3-4B-RKLLM
```

Run the downloaded file with the RKLLM runtime and the upstream MiniCPM3 chat template. For Docker deployment, see [Hanzo-Huang/rkllm-docker](https://github.com/Hanzo-Huang/rkllm-docker).

## Limitations

These artifacts are target-specific conversions. Validate quality, memory use, and runtime compatibility on your own hardware.

## Acknowledgements

Thanks to OpenBMB, Rockchip, and the RKLLM community.