AdarshSingh7647/Eklav-9B-Math

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026Architecture:Transformer Featherless Exclusive Cold

AdarshSingh7647/Eklav-9B-Math 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 unique "Eklav" training method, which involves hint-conditioned supervised fine-tuning, allowing the model to learn to continue reasoning from partial traces. This approach enables the model to excel in math problem-solving by developing its own reasoning capabilities rather than merely imitating full reasoning chains.

Loading preview...

Eklav-9B-Math: A Specialized Model for Mathematical Reasoning

Eklav-9B-Math is a 9 billion parameter model developed by AdarshSingh7647, built upon the zai-org/GLM-Z1-9B-0414 base architecture. Its core innovation lies in its unique "Eklav" training methodology, which focuses on teaching the model to continue reasoning from partial hints rather than simply reproducing complete reasoning traces. This hint-conditioned supervised fine-tuning (SFT) approach aims to foster independent reasoning capabilities in the model.

Key Capabilities & Training

  • Hint-Conditioned Reasoning: The model is trained to pick up a teacher's reasoning mid-thought, learning to continue a partial reasoning trace and produce the final answer. This differs from standard full-trace Chain-of-Thought (CoT) distillation.
  • Mathematical Task Focus: Eklav-9B-Math is specifically designed and optimized for mathematical reasoning tasks.
  • Performance: It demonstrates competitive performance on various math benchmarks. Compared to standard full-trace CoT SFT, it shows only a -0.7% average pass@1 across 6 math benchmarks (excluding MMLU).

Benchmarks

Performance metrics (Pass@1 %) on key benchmarks include:

  • AIME 1983-2024: 60.7%
  • GSM8K: 95.3%
  • MATH-500: 92.5%
  • Omni-MATH: 41.6%

When to Use This Model

This model is particularly well-suited for applications requiring robust mathematical problem-solving and reasoning. Its unique training paradigm makes it an interesting choice for tasks where understanding and extending logical thought processes in mathematics are crucial.

Finetunes
2 models