Author: @jamescallander
Source model: deepseek-ai/deepseek-coder-1.3b-instruct · Hugging Face
Target: Rockchip RK3588 NPU via RKNN-LLM Runtime

This repository hosts a conversion of deepseek-coder-1.3b-instruct 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
Tokenizer: not required at runtime (UI handles prompt I/O)

⚠️ Code generation disclaimer

🛑 This model may produce incorrect or insecure code.

It is intended for research, educational, and experimental purposes only.
Always review, test, and validate code outputs before using them in real projects.
Do not rely on outputs for production, security-sensitive, or safety-critical systems.
Use responsibly and in compliance with the source model’s license and restrictions.

Intended use

On-device deployment of a code-specialized LLM on RK3588 SBCs.
deepseek-coder-1.3b-instruct is tuned for programming tasks, code completion, and instruction-following in developer workflows, optimized for smaller edge hardware.

Limitations

Requires 2.5GB free memory
Quantized build (w8a8_g128) may show small quality differences vs. full-precision upstream.
Tested on Radxa Rock 5B+; other devices may require different drivers/toolkit versions.
Generated code should always be reviewed before use in production systems.

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

bash

python3 <TOOLKIT_PATH>/rknn-llm/examples/rkllm_server_demo/flask_server.py \
  --rkllm_model_path <MODEL_PATH>/deepseek-coder-1.3b-instruct_w8a8_g128_rk3588.rkllm \
  --target_platform rk3588

3) Sending a request

A basic format for message request is:

json

{
    "model":"deepseek-coder-1.3b-instruct",
    "messages":[{
        "role":"user",
        "content":"<YOUR_PROMPT_HERE>"}],
    "stream":false
}

Example request using curl:

bash

curl -s -X POST <SERVER_IP_ADDRESS>:8080/rkllm_chat \
    -H 'Content-Type: application/json' \
    -d '{"model":"deepseek-coder-1.3b-instruct","messages":[{"role":"user","content":"Create a python function to calculate factorials using recursive method."}],"stream":false}'

The response is formated in the following way:

json

{
    "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":"Sure! Here is the Python code for calculating factorial of an number (n) by implementing it in Recursion Method : ```python def Factorial(num): # Define Function with parameter num if num == 1 or num==0:# Base Case to stop recursive call when we reach one. It's the definition for factorial of any number n, where fact(n) = n * (n-1)! return 1 else: # Recursion Call -> Factorial function is called again with decreasing value until it reaches base case return num*Factorial(num-1) ``` You can call this Function as follows : `print (fact_of_5())`, where 5 will be the number for which you want to find factorial. This function works only with non negative integers and zero! It doesn't work properly if called without arguments or it is given a floating point argument because of integer division in Python2 when using `num*Factorial(num-1)`, this would result into an infinite recursion loop as the base case will never be reached.","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:

json

{
 "model": "deepseek-coder-1.3b-instruct",
 "messages": [{"role":"user","content":"Hello!"}],
 "stream": false
}

License

This conversion follows the DeepSeek License Agreement

Required notice: see NOTICE
Modifications: quantization (w8a8_g128), export to .rkllm format for RK3588 SBCs.
Use Restrictions: You may not use this model or its derivatives for prohibited purposes listed in Attachment A of the DeepSeek License Agreement (including military use, harming minors, generating PII without authorization, harassment, or discrimination)

jamescallander/deepseek-coder-1.3b-instruct_w8a8_g128_rk3588.rkllm

deepseek-coder-1.3b-instruct — RKLLM build for RK3588 boards

Built with DeepSeek (DeepSeek License Agreement)

Conversion details

⚠️ Code generation disclaimer

Intended use

Limitations

Quick start (RK3588)

1) Install runtime

2) Simple Flask server deployment

3) Sending a request

4) UI compatibility

License

jamescallander/
deepseek-coder-1.3b-instruct_w8a8_g128_rk3588.rkllm