---
license: apache-2.0
base_model: Qwen/Qwen3-4B
pipeline_tag: text-generation
library_name: rkllm
tags:
- rkllm
- rockchip
- rk3576
- rk3588
- qwen
- qwen3
---

# Qwen3-4B-RKLLM

RKLLM-converted Qwen3-4B language-model artifacts for Rockchip RK3576 and RK3588 NPUs.

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

## Base model

- Upstream model: [Qwen/Qwen3-4B](https://huggingface.co/Qwen/Qwen3-4B)
- License: Apache-2.0
- Model type: LLM (text only)

## Conversion and variants

### Toolkit version

**RKLLM Toolkit: v1.2.3**

Use a file built for the exact target SoC.

| Target | Quantization | File | SHA256 |
| --- | --- | --- | --- |
| RK3576 | W4A16 | [Qwen3-4B_RK3576_w4a16.rkllm](RK3576/Qwen3-4B_RK3576_w4a16.rkllm) | `07b25e01a16b50c0f76c777826aea602e485395a77df4c62428cf9fa532f509e` |
| RK3576 | W4A16 (g128) | [Qwen3-4B_RK3576_w4a16_g128.rkllm](RK3576/Qwen3-4B_RK3576_w4a16_g128.rkllm) | `66b8dbc941a7fc9864583082013a2d765be9839f19e98763abfd5ec2bca166a2` |
| RK3576 | W8A8 | [Qwen3-4B_RK3576_w8a8.rkllm](RK3576/Qwen3-4B_RK3576_w8a8.rkllm) | `7f40a9c638360b1ce51dc03d2fe5005c4ded9804d13304de3c52244634e7ed26` |
| RK3588 | W8A8 | [Qwen3-4B_RK3588_w8a8.rkllm](RK3588/Qwen3-4B_RK3588_w8a8.rkllm) | `6f53a7b5ec8430f95a9fb6e6279e8a4744c9952d8ec59c4a7303b92ebc8b50d5` |

The repository also includes `Qwen3-4B_data_quant.json`, used as calibration data during conversion.

## Usage

```bash
hf download HanzoHuang/Qwen3-4B-RKLLM \
  RK3576/Qwen3-4B_RK3576_w4a16_g128.rkllm \
  --local-dir Qwen3-4B-RKLLM
```

Use the Qwen3 chat template with the RKLLM runtime. For Docker deployment, see [Hanzo-Huang/rkllm-docker](https://github.com/Hanzo-Huang/rkllm-docker).

## Limitations

These are target-specific converted artifacts. Validate quality, memory use, and runtime compatibility on your own Rockchip device.

## Acknowledgements

Thanks to the Qwen Team, Rockchip, and the RKLLM community.