Instructions to use icedmoca/kcode-oss-20b-mxfp4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use icedmoca/kcode-oss-20b-mxfp4 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="icedmoca/kcode-oss-20b-mxfp4",
	filename="kcode-oss-20b-mxfp4.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use icedmoca/kcode-oss-20b-mxfp4 with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf icedmoca/kcode-oss-20b-mxfp4
# Run inference directly in the terminal:
llama-cli -hf icedmoca/kcode-oss-20b-mxfp4

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf icedmoca/kcode-oss-20b-mxfp4
# Run inference directly in the terminal:
llama-cli -hf icedmoca/kcode-oss-20b-mxfp4

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf icedmoca/kcode-oss-20b-mxfp4
# Run inference directly in the terminal:
./llama-cli -hf icedmoca/kcode-oss-20b-mxfp4

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf icedmoca/kcode-oss-20b-mxfp4
# Run inference directly in the terminal:
./build/bin/llama-cli -hf icedmoca/kcode-oss-20b-mxfp4

Use Docker

docker model run hf.co/icedmoca/kcode-oss-20b-mxfp4

LM Studio
Jan
Ollama
How to use icedmoca/kcode-oss-20b-mxfp4 with Ollama:
```
ollama run hf.co/icedmoca/kcode-oss-20b-mxfp4
```

Unsloth Studio new

How to use icedmoca/kcode-oss-20b-mxfp4 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for icedmoca/kcode-oss-20b-mxfp4 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for icedmoca/kcode-oss-20b-mxfp4 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for icedmoca/kcode-oss-20b-mxfp4 to start chatting

Pi new

How to use icedmoca/kcode-oss-20b-mxfp4 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf icedmoca/kcode-oss-20b-mxfp4

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "kcode-oss-20b-mxfp4"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use icedmoca/kcode-oss-20b-mxfp4 with Docker Model Runner:
```
docker model run hf.co/icedmoca/kcode-oss-20b-mxfp4
```

Lemonade

How to use icedmoca/kcode-oss-20b-mxfp4 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull icedmoca/kcode-oss-20b-mxfp4

Run and chat with the model

lemonade run user.kcode-oss-20b-mxfp4-{{QUANT_TAG}}

List all available models

lemonade list

kcode-oss-20b-mxfp4

kcode-oss-20b-mxfp4 is a GGUF MXFP4 coding-agent model built on top of GPT-OSS 20B and optimized for terminal-native software engineering workflows, structured tool use, retrieval-grounded reasoning, and long-session coding tasks.

The model is designed primarily for:

repository navigation code editing and patch generation shell-oriented workflows structured tool calling retrieval-backed context restoration long-running agent sessions Architecture

Base architecture:

GPT-OSS 20B Mixture-of-Experts (MoE) MXFP4 quantization 131k context length GGUF runtime format

Model metadata:

24 transformer blocks 32 experts 4 active experts per token GPT-4o tokenizer format YaRN rope scaling Intended Usage

This model is intended to be paired with the Kcode runtime and orchestration layer:

exact-context replay context vault references dynamic tool schema expansion persistent memory systems multi-tool agent execution

It performs best in iterative:

edit → test → repair

coding workflows.

Prompting

Example system prompt:

You are Kcode, a terminal-native coding agent.

Repository state:

Task:

Fix the websocket reconnect logic without breaking auth refresh behavior. Runtime Compatibility

Optimized for:

llama.cpp Ollama OpenAI-compatible local servers terminal coding agents structured tool runtimes Notes

kcode-oss-20b-mxfp4 is optimized more heavily for:

coding workflows orchestration stability structured reasoning retrieval-grounded operation long-session memory behavior

than for:

roleplay creative writing unrestricted conversational chat Runtime

GitHub:

https://github.com/icedmoca/kcode

Downloads last month: 306

GGUF

Model size

21B params

Architecture

gpt-oss

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for icedmoca/kcode-oss-20b-mxfp4

Base model

openai/gpt-oss-20b

Quantized

(203)

this model