AI Models Configuration

This repository contains configuration files and presets for various AI models, specifically tailored for ollama and llama.cpp.

Project Structure

.
├── llama.cpp
│   └── models
│       └── preset.ini       # Sampling parameters and model paths for llama.cpp
└── ollama
    └── models
        ├── Modelfile.qwen3.6-27b-q4.coding
        ├── Modelfile.qwen3.6-27b-q4.instruct
        ├── Modelfile.qwen3.6-35b-a3b-q4.coding
        └── Modelfile.tiny    # Minimal Qwen2.5 0.5B configuration

Contents

llama.cpp Presets

The llama.cpp/models/preset.ini file defines sampling parameters (temperature, top-p, penalties, etc.) for different Qwen3.6 variants:

• Qwen3.6-27B-Q4_K_M-PreciseCoding: Optimized for coding tasks with thinking preservation.
• Qwen3.6-27B-Q4_K_M-Instruct: Standard instruction-following mode.
• Qwen3.6-27B-Q4_K_M-MTP-Instruct: Multi-Token Prediction (MTP) draft configuration.
• Qwen3.6-35B-A3B-Q4_K_M-Instruct: Configuration for the 35B MoE variant.

Ollama Modelfiles

The ollama/models/ directory contains Modelfile configurations for importing models into Ollama:

• Modelfile.tiny: A concise setup using qwen2.5:0.5b-instruct-q4_K_M.
• Qwen3.6 variants: Specific configurations for 27B and 35B Qwen3.6 models optimized for coding and instruction.