AdarshSingh7647/Eklav-14B-Math-CotGen
AdarshSingh7647/Eklav-14B-Math-CotGen is a 14 billion parameter language model based on Qwen3-14B, specifically fine-tuned for mathematical reasoning tasks. It utilizes a standard full trace CoT (Chain-of-Thought) SFT (Supervised Fine-Tuning) baseline, serving as a comparative model for the Eklav training method. This model excels in generating detailed mathematical reasoning steps and achieving high scores on various math benchmarks.
Loading preview...
Eklav-14B-Math-CotGen Overview
Eklav-14B-Math-CotGen is a 14 billion parameter model built upon the Qwen3-14B base architecture, specifically designed for advanced mathematical reasoning. This particular version represents a standard full trace Chain-of-Thought (CoT) Supervised Fine-Tuning (SFT) baseline, which is used to measure improvements achieved by the novel Eklav training method.
Key Capabilities
- Mathematical Reasoning: The model is fine-tuned to generate step-by-step reasoning for complex mathematical problems.
- CoT Distillation Baseline: It serves as a strong baseline for evaluating new CoT distillation techniques, demonstrating performance with traditional full trace SFT.
- Benchmark Performance: Achieves competitive results across various math benchmarks, including:
- GSM8K: 95.9%
- MATH-500: 90.5%
- AIME 1983-2024: 58.6%
- MMLU: 83.9%
Good For
- Mathematical Problem Solving: Ideal for applications requiring detailed, reasoned solutions to math problems.
- Research in CoT Methods: Useful as a comparative model for researchers developing and evaluating new Chain-of-Thought training or distillation techniques.
- Educational Tools: Can be integrated into systems that assist with learning or verifying mathematical solutions.