Hugging Face Overview
Vex Amber Mini 1.2: A Compact Model for Code and Math
Vex Amber Mini 1.2, developed by Arioron, is a 0.6 billion parameter decoder-only transformer model. It is an advancement from Vex Amber Mini 1.0, built on the Qwen3-0.6B base, and is particularly distinguished by its capabilities in mathematical reasoning and code generation.
Key Capabilities
- Exceptional Code Generation: Achieves a Pass@1 score of 21.34% on HumanEval and 38.7% on MBPP, indicating strong performance in generating efficient algorithms and solving programming challenges.
- Robust Mathematical Reasoning: Demonstrates 65.2% accuracy on GSM8K and 45.8% on MATH, showcasing its ability to solve complex mathematical problems and explain steps.
- Optimized Training: Trained on a specialized dataset comprising 45% code, 30% mathematical content, 15% general reasoning, and 10% conversational data, enhancing its focus areas.
- Efficient Architecture: Features a Transformer-based decoder with Rotary Positional Embeddings (RoPE) and a context length of 8,192 tokens, designed for efficient processing.
Good For
- Code Completion and Generation: Ideal for integrating into IDEs, assisting with programming tasks, and generating code snippets.
- Mathematical Problem Solving: Useful for educational assistance, solving equations, and explaining mathematical concepts.
- Research and Prototyping: Suitable for technical documentation, research in mathematics and computer science, and developing technical chatbots.
While powerful for its size, users should note its 0.6B parameter count may limit performance on extremely complex, multi-step reasoning tasks compared to much larger models (7B+).