jamescallander/MiniCPM3-4B_w8a8_g128_rk3588.rkllm

MiniCPM3-4B — RKLLM build for RK3588 boards

Author: @jamescallander
Source model: openbmb/MiniCPM4-0.5B · Hugging Face

Target: Rockchip RK3588 NPU via RKNN-LLM Runtime

This repository hosts a conversion of MiniCPM3-4B for use on Rockchip RK3588 single-board computers (Orange Pi 5 plus, Radxa Rock 5b+, Banana Pi M7, etc.). Conversion was performed using the RKNN-LLM toolkit

Conversion details

• RKLLM-Toolkit version: v1.2.1
• NPU driver: v0.9.8
• Python: 3.12
• Quantization: w8a8_g128
• Output: single-file .rkllm artifact
• Modifications: quantization (w8a8_g128), export to .rkllm format for RK3588 SBCs
• Tokenizer: not required at runtime (UI handles prompt I/O)

Intended use

• On-device lightweight inference on RK3588 SBCs.
• MiniCPM3-4B is a compact general-purpose model designed for efficiency, testing, and resource-constrained scenarios. Ideal for experimentation where low memory usage and fast response matter more than deep reasoning.

Limitations

• Requires 8GB free memory
• Tested on a Radxa Rock 5B+, other devices may require different drivers/toolkit versions.
• Quantization (w8a8_g128) may further reduce output fidelity.

Quick start (RK3588)

1) Install runtime

The RKNN-LLM toolkit and instructions can be found on the specific development board's manufacturer website or from airockchip's github page.

Download and install the required packages as per the toolkit's instructions.

2) Simple Flask server deployment

The simplest way the deploy the .rkllm converted model is using an example script provided in the toolkit in this directory: rknn-llm/examples/rkllm_server_demo

python3 <TOOLKIT_PATH>/rknn-llm/examples/rkllm_server_demo/flask_server.py \
  --rkllm_model_path <MODEL_PATH>/MiniCPM3-4B_w8a8_g128_rk3588.rkllm \
  --target_platform rk3588

3) Sending a request

A basic format for message request is:

{
    "model":"MiniCPM3-4B",
    "messages":[{
        "role":"user",
        "content":"<YOUR_PROMPT_HERE>"}],
    "stream":false
}

Example request using curl:

curl -s -X POST <SERVER_IP_ADDRESS>:8080/rkllm_chat \
    -H 'Content-Type: application/json' \
    -d '{"model":"MiniCPM3-4B","messages":[{"role":"user","content":"In 2 or 3 sentences, who was Napoleon Bonaparte?"}],"stream":false}'

The response is formated in the following way:

{
    "choices":[{
        "finish_reason":"stop",
        "index":0,
        "logprobs":null,
        "message":{
            "content":"<MODEL_REPLY_HERE">,
            "role":"assistant"}}],
        "created":null,
        "id":"rkllm_chat",
        "object":"rkllm_chat",
        "usage":{
            "completion_tokens":null,
            "prompt_tokens":null,
            "total_tokens":null}
}

Example response:

{"choices":[{"finish_reason":"stop","index":0,"logprobs":null,"message":{"content":"Napoleon Bonaparte (1769-1821) was a French military leader and emperor who rose to prominence during the French Revolution and went on to conquer much of Europe. His strategic brilliance and charisma earned him many followers, but his ambitions ultimately led to his downfall. After being exiled to the island of Saint Helena in the South Atlantic, he died in 1821 at the age of 51.","role":"assistant"}}],"created":null,"id":"rkllm_chat","object":"rkllm_chat","usage":{"completion_tokens":null,"prompt_tokens":null,"total_tokens":null}}

4) UI compatibility

This server exposes an OpenAI-compatible Chat Completions API.

You can connect it to any OpenAI-compatible client or UI (for example: Open WebUI)

• Configure your client with the API base: http://<SERVER_IP_ADDRESS>:8080 and use the endpoint: /rkllm_chat
• Make sure the model field matches the converted model’s name, for example:

{
 "model": "MiniCPM3-4B",
 "messages": [{"role":"user","content":"Hello!"}],
 "stream": false
}

License

This conversion follows the license of the source model: apache-2.0

• Attribution: Built with MiniCPM3-4B (OpenBMB)
• Required notice: see NOTICE
• Modifications: quantization (w8a8_g128), export to .rkllm format for RK3588 SBCs