AdarshSingh7647/Eklav-8B-Math-CotGen

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026Architecture:Transformer Featherless Exclusive Cold

AdarshSingh7647/Eklav-8B-Math-CotGen is an 8 billion parameter language model based on Qwen/Qwen3-8B, fine-tuned for mathematical reasoning tasks. It utilizes a standard full trace CoT (Chain-of-Thought) SFT (Supervised Fine-Tuning) method, serving as a baseline for the Eklav project. This model excels in math reasoning benchmarks like GSM8K and MATH-500, making it suitable for complex arithmetic and logical problem-solving.

Loading preview...

Eklav-8B-Math-CotGen Overview

Eklav-8B-Math-CotGen is an 8 billion parameter model built upon the Qwen/Qwen3-8B base architecture, specifically designed for advanced mathematical reasoning. This model represents a standard full trace CoT SFT baseline within the broader Eklav project, which aims to develop models that can learn to continue a teacher's reasoning process rather than merely imitating it end-to-end.

Key Capabilities & Training

  • Mathematical Reasoning: The model is specialized in solving complex math problems, as evidenced by its strong performance across various benchmarks.
  • Chain-of-Thought (CoT) Distillation: It is trained using a standard full trace CoT Supervised Fine-Tuning method, where the model learns from complete reasoning traces.
  • Baseline Model: This particular iteration serves as a comparative baseline to measure improvements from the Eklav training methodology, which focuses on conditioning the model's reasoning on partial teacher traces.

Performance Highlights

Evaluated on a single run per benchmark, Eklav-8B-Math-CotGen demonstrates robust performance in mathematical and general reasoning tasks:

  • GSM8K: Achieved 95.3% Pass@1.
  • MATH-500: Achieved 89.4% Pass@1.
  • AIME (1983-2024): Scored 54.0% Pass@1.
  • MMLU: Achieved 82.2% Pass@1, indicating strong general knowledge and reasoning capabilities alongside its math specialization.

Good for

  • Mathematical Problem Solving: Ideal for applications requiring accurate and detailed step-by-step mathematical reasoning.
  • Research in CoT Methods: Useful as a strong baseline for researchers exploring advanced Chain-of-Thought distillation techniques and reasoning continuation.

Finetunes
1 models