Overview

Neelectric/Llama-3.1-8B-Instruct_SDFT_mathv00.02 is an 8 billion parameter instruction-tuned model, fine-tuned by Neelectric from the base meta-llama/Llama-3.1-8B-Instruct model. Its primary differentiation lies in its specialized training for mathematical reasoning. The model leverages the SDFT (Self-Training with On-Policy Self-Distillation) method, a technique detailed in the paper "Self-Training with On-Policy Self-Distillation for Language Model Alignment," to enhance its capabilities.

Key Capabilities

• Enhanced Mathematical Reasoning: Specifically fine-tuned on the Neelectric/OpenR1-Math-220k_all_Llama3_4096toks_SDFT dataset, which focuses on mathematical problems.
• Instruction Following: Inherits strong instruction-following abilities from its Llama-3.1-8B-Instruct base.
• SDFT Training: Utilizes a self-training and self-distillation approach for improved alignment and performance in its target domain.

Good for

• Mathematical Problem Solving: Ideal for applications requiring accurate solutions to mathematical questions and complex calculations.
• Educational Tools: Can be integrated into platforms for tutoring or generating math exercises.
• Research in Mathematical AI: Useful for exploring and benchmarking models focused on quantitative reasoning.