VibeManGeo/ zen-5-coder:q4_K_M

58 Downloads Updated 3 months ago

Zen 5 Coder is an 80B Qwen3-based coding model optimized for code generation, debugging, refactoring, and repository-scale software engineering. Supports up to 256K context length. Quantized and maintained by VibeManGeo.

ollama run VibeManGeo/zen-5-coder:q4_K_M

curl http://localhost:11434/api/chat \
  -d '{
    "model": "VibeManGeo/zen-5-coder:q4_K_M",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='VibeManGeo/zen-5-coder:q4_K_M',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'VibeManGeo/zen-5-coder:q4_K_M',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 3 months ago

3 months ago

c3fb39344aa1 · 48GB

model

archqwen3next

·

parameters79.7B

·

quantizationQ4_K_M

48GB

params

{ "num_ctx": 262144, "temperature": 0.7 }

37B

template

{{ .Prompt }}

13B

Readme

Zen-5-Coder GGUF

GGUF Q4_K_M quantization of Zen-5-Coder 80B for Ollama and compatible runtimes.

The original model was released by Zen LM in Hugging Face Transformers format.

This repository provides converted and quantized version optimized for local inference.

Full documentation: https://huggingface.co/VibeManGeo/Zen-5-Coder-GGUF

Overview

Property	Value
Model	Zen-5-Coder
Architecture	Mixture of Experts (MoE)
Parameters	80B
Original Format	Hugging Face Transformers
GGUF Conversion	llama.cpp
Repository Maintainer	VibeManGeo

Available Quantizations

Quantization	Description
Q4_K_M	Recommended default

Conversion Pipeline

All files were generated locally using the standard llama.cpp workflow:

Hugging Face Transformers
        ↓
GGUF FP16
        ↓
GGUF Quantization
        ↓
Ollama create

Tools Used

llama.cpp
convert_hf_to_gguf.py
llama-quantize
ollama

Hardware Used For Conversion

The quantization in this repository war generated and tested on:

GPU 0 NVIDIA RTX 3060 12 GB Headless
GPU 1 NVIDIA Tesla P40 24 GB Headless
AMD Ryzen 7 5700G
64 GB DDR-4 3200Mhz System RAM
Debian Linux 13.2

Actual performance will depend on context size, quantization level, GPU offloading, and runtime configuration.

Credits

Original Model

Zen LM — creators of Zen-5-Coder.

GGUF Conversion & Quantization

VibeManGeo

Fun fact: these 80B quantizations were produced before the author passed CompTIA A+ Core 1.

Acknowledgements

Special thanks to the llama.cpp developers for providing the tools that make efficient local inference and GGUF quantization possible.

Disclaimer

This repository contains converted and quantized derivatives of the original model. All credit for model architecture, training, datasets, and original weights belongs to the original authors.

Notes

The goal of this repository is to make Zen-5-Coder immediately accessible to the local inference community without requiring users to perform the conversion process themselves.

# Zen-5-Coder GGUF

GGUF Q4_K_M quantization of **Zen-5-Coder 80B** for **Ollama** and compatible runtimes.

The original model was released by **Zen LM** in Hugging Face Transformers format.

This repository provides converted and quantized version optimized for local inference.

**Full documentation:**
https://huggingface.co/VibeManGeo/Zen-5-Coder-GGUF

---

## Overview

| Property              | Value                     |
| --------------------- | ------------------------- |
| Model                 | Zen-5-Coder               |
| Architecture          | Mixture of Experts (MoE)  |
| Parameters            | 80B                       |
| Original Format       | Hugging Face Transformers |
| GGUF Conversion       | llama.cpp                 |
| Repository Maintainer | VibeManGeo                |

---

## Available Quantizations

| Quantization | Description                 |
| ------------ | --------------------------- |
| **Q4_K_M**   | Recommended default         |

---

## Conversion Pipeline

All files were generated locally using the standard llama.cpp workflow:

```text
Hugging Face Transformers
        ↓
GGUF FP16
        ↓
GGUF Quantization
        ↓
Ollama create
```

### Tools Used

* llama.cpp
* convert_hf_to_gguf.py
* llama-quantize
* ollama

---

## Hardware Used For Conversion

The quantization in this repository war generated and tested on:

* GPU 0 NVIDIA RTX 3060 12 GB Headless
* GPU 1 NVIDIA Tesla P40 24 GB Headless
* AMD Ryzen 7 5700G
* 64 GB DDR-4 3200Mhz System RAM
* Debian Linux 13.2

Actual performance will depend on context size, quantization level, GPU offloading, and runtime configuration.

---

## Credits

### Original Model

**Zen LM** — creators of Zen-5-Coder.

### GGUF Conversion & Quantization

**VibeManGeo**

**Fun fact:** these 80B quantizations were produced before the author passed CompTIA A+ Core 1.

---

## Acknowledgements

Special thanks to the **llama.cpp** developers for providing the tools that make efficient local inference and GGUF quantization possible.

---

## Disclaimer

This repository contains converted and quantized derivatives of the original model. All credit for model architecture, training, datasets, and original weights belongs to the original authors.

---

## Notes

The goal of this repository is to make Zen-5-Coder immediately accessible to the local inference community without requiring users to perform the conversion process themselves.

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)