Embarrassingly Simple Self-Distillation Improves Code Generation

TL;DR

This paper demonstrates that simple self-distillation, involving sampling and fine-tuning, significantly enhances large language models' code generation capabilities without external verification or reinforcement learning.

Contribution

It introduces a straightforward self-distillation method that improves code generation across multiple models and scales, revealing insights into decoding dynamics.

Findings

SSD improves pass@1 from 42.4% to 55.3% on LiveCodeBench v6.

Gains are concentrated on harder problems and generalize across models and scales.

SSD reshapes token distributions, balancing exploration and precision.

Abstract

Can a large language model (LLM) improve at code generation using only its own raw outputs, without a verifier, a teacher model, or reinforcement learning? We answer in the affirmative with simple self-distillation (SSD): sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning. SSD improves Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6, with gains concentrating on harder problems, and it generalizes across Qwen and Llama models at 4B, 8B, and 30B scale, including both instruct and thinking variants. To understand why such a simple method can work, we trace these gains to a precision-exploration conflict in LLM decoding and show that SSD reshapes token distributions in a context-dependent way, suppressing distractor tails where precision matters while preserving useful diversity where exploration matters. Taken together, SSD offers a complementary post-training direction for improving LLM code generation.