WooYoungSeok/qwen2.5-math-7b-descriptive-verifier-v2-trval-halfA

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026Architecture:Transformer Featherless Exclusive Cold

WooYoungSeok/qwen2.5-math-7b-descriptive-verifier-v2-trval-halfA is a 7.6 billion parameter Qwen2.5-Math-7B-Instruct based model developed by WooYoungSeok. It is specifically fine-tuned to act as a mathematical error verifier, determining if a natural-language error description is 'aligned' or 'not_aligned' with a given incorrect student solution to a math problem. This model excels at precisely evaluating the accuracy of error descriptions in mathematical contexts, achieving 96.95% accuracy on the v2 test set.

Loading preview...

Overview

WooYoungSeok/qwen2.5-math-7b-descriptive-verifier-v2-trval-halfA is a specialized 7.6 billion parameter model, fine-tuned from Qwen/Qwen2.5-Math-7B-Instruct. Its primary function is to act as a descriptive mathematical error verifier. Given a math question, an incorrect student solution, and a natural-language error description, the model outputs aligned if the description is supported by the solution, or not_aligned otherwise. This model was trained on "half A" of a custom dataset, comprising 1511 cases, with loss applied only to the assistant label during SFT.

Key Capabilities

  • Mathematical Error Verification: Precisely determines if an error description accurately reflects an error in a student's math solution.
  • Binary Classification: Provides a clear aligned or not_aligned output, suitable for automated grading or feedback systems.
  • High Accuracy: Achieved an accuracy of 0.9695 (95% CI [0.9552, 0.9815]) on the v2 test set, with a macro-F1 of 0.9695.
  • Robust Evaluation: Evaluated using exact match after stripping whitespace, with a low invalid output rate of 0.0000.

Good for

  • Automated Feedback Systems: Integrating into educational platforms to provide instant, objective feedback on student error descriptions.
  • Quality Assurance for Educational Content: Verifying the accuracy of pre-written error explanations for math problems.
  • Research in AI for Education: Studying the alignment between natural language descriptions and mathematical errors.
  • Specialized Math Tutoring Tools: Enhancing tools that help students understand their mistakes by validating error explanations.

Finetunes
1 models