AdarshSingh7647/Eklav-9B-Math-CotGen
AdarshSingh7647/Eklav-9B-Math-CotGen is a 9 billion parameter language model based on the zai-org/GLM-Z1-9B-0414 architecture, specifically fine-tuned for mathematical reasoning tasks. It utilizes a CotGen (standard full trace CoT SFT) training method, serving as a baseline for the Eklav project's approach to learning partial reasoning traces. This model excels in math problem-solving, achieving 95.5% on GSM8K and 91.2% on MATH-500, making it suitable for applications requiring strong mathematical capabilities.
Loading preview...
Overview
AdarshSingh7647/Eklav-9B-Math-CotGen is a 9 billion parameter model built upon the zai-org/GLM-Z1-9B-0414 base architecture. It is specifically designed and trained for mathematical reasoning tasks using a standard full trace CoT (Chain-of-Thought) Supervised Fine-Tuning (SFT) method, referred to as CotGen. This model serves as a baseline for the Eklav project, which aims to train models to continue a teacher's partial reasoning trace rather than imitating it end-to-end.
Key Capabilities
- Mathematical Reasoning: Optimized for solving complex math problems, demonstrating strong performance across various math benchmarks.
- CoT SFT Baseline: Represents a standard approach to CoT distillation, providing a comparative measure for more advanced reasoning training methods.
- High Accuracy on Math Benchmarks: Achieves notable results such as:
- 95.5% on GSM8K
- 91.2% on MATH-500
- 59.3% on AIME 1983-2024
- 81.0% on MMLU
Good For
- Applications requiring robust mathematical problem-solving abilities.
- Research and development in reasoning and CoT methods, particularly as a strong baseline for comparison.
- Tasks involving complex numerical and logical deduction.