Sashank-810/IDC_Global_Merged

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Nov 26, 2025License:llama3.1Architecture:Transformer Cold

Sashank-810/IDC_Global_Merged is an 8 billion parameter instruction-tuned causal language model, based on Meta's Llama-3.1-8B-Instruct, fine-tuned for math tutoring and doubt clarification. This model integrates an IDC critic adapter, enabling it to provide step-by-step math help and critique student answers. It offers an 11.12 percentage point accuracy gain over its base model in structured evaluations, making it suitable for educational applications requiring detailed mathematical assistance.

Loading preview...

Overview

Sashank-810/IDC_Global_Merged is an 8 billion parameter language model built upon the meta-llama/Llama-3.1-8B-Instruct base. It has been specifically fine-tuned to function as a math tutor, incorporating an Integrated Doubt Clarification (IDC) critic adapter. This integration allows the model to not only provide solutions but also to critique and help fix student answers, offering a comprehensive educational support tool.

Key Capabilities

  • Math Tutoring: Provides step-by-step explanations and solutions for mathematical problems.
  • Critique and Fix: Evaluates student responses and offers corrective feedback.
  • Enhanced Accuracy: Demonstrates a significant improvement in structured evaluation accuracy, achieving a +11.12 percentage point gain over the base Llama-3.1-8B-Instruct model.
  • Text Generation Quality: Shows improved BLEU, ROUGE, and METEOR scores compared to the base model, indicating higher quality and relevance in generated text.

Intended Use Cases

  • Educational Tutoring: Ideal for assisting students with math homework and concepts.
  • Doubt Clarification: Helps users understand complex mathematical topics.
  • Interactive Learning: Supports scenarios where students need feedback on their problem-solving approaches.

Performance Metrics

The fine-tuned model achieved 35.00% accuracy in structured evaluations, compared to the base model's 23.88%. Text generation metrics also show notable improvements:

  • BLEU: 59.31 (Fine-Tuned) vs. 38.24 (Base)
  • ROUGE-1: 0.4423 (Fine-Tuned) vs. 0.2947 (Base)
  • METEOR: 0.2478 (Fine-Tuned) vs. 0.1633 (Base)