supra-nexus-o1-thinking-gguf

🔗 Model Collection

Base Models

• 🤖 supra-nexus-o1-instruct - Instruction-following model
• 💭 supra-nexus-o1-thinking - Chain-of-thought reasoning model

Available Formats

Instruction Model

• 📦 GGUF | 🍎 MLX | ⚡ MLX 4-bit

Thinking Model

• 📦 GGUF | 🍎 MLX | ⚡ MLX 4-bit

Training Data

• 📊 supra-nexus-o1-training - Complete training dataset

💡 Key Features

• Transparent Reasoning: Shows thought process using <thinking> tags
• Chain-of-Thought: Step-by-step problem solving approach
• Self-Improvement: Trained with recursive improvement examples
• Multi-Format: Available in multiple formats for different platforms

🚀 Quick Start

Using with llama.cpp

Available Quantizations

• F16 - Full 16-bit precision (largest, most accurate)
• Q8_0 - 8-bit quantization (good balance)
• Q5_K_M - 5-bit quantization (recommended)
• Q4_K_M - 4-bit quantization (smallest)

📈 Performance

• Complex reasoning tasks
• Step-by-step problem solving
• Mathematical computations
• Code generation and debugging
• Creative writing with logical structure

🏗️ Architecture

• Custom fine-tuning for reasoning
• Chain-of-thought training
• Self-improvement capabilities
• Identity preservation techniques

🔬 Training Details

• Base Model: Qwen/Qwen2.5-7B-Instruct
• Training Framework: Zoo Gym
• Dataset: supra-nexus-o1-training
• Training Duration: Multiple iterations with self-improvement
• Hardware: NVIDIA A100 GPUs

📚 Resources

• 📖 GitHub Repository - Source code and documentation
• 🏢 Supra Foundation - Organization behind O1
• 🐦 Twitter - Latest updates
• 💬 Discord - Community support

📄 Citation

📝 License