Beyond Temperature: Hyperfitting as a Late-Stage Geometric Expansion

2026-05-21

AI Summary

This study investigates the intrinsic mechanism by which "overfitting" during large language model fine-tuning enhances generation quality and mitigates repetition. Through entropy-matching controlled experiments, inter-layer feature analysis, and dynamic ranking ablation, the work refutes the prevailing hypothesis that overfitting is equivalent to temperature scaling. Instead, it reveals that overfitting fundamentally arises from a context-dependent dynamic token re-ranking mechanism, driven by geometric expansion in the feature space of the Transformer’s final layer—approximately +80.8 dimensions. Building on this insight, the authors propose Late-Stage LoRA, a parameter-efficient strategy that fine-tunes only the last five layers, achieving significant improvements in both generation diversity and quality with minimal parameter updates.

Abstract

Recent work has identified a counterintuitive phenomenon termed "Hyperfitting", where fine-tuning Large Language Models (LLMs) to near-zero training loss on small datasets surprisingly enhances open-ended generation quality and mitigates repetition in greedy decoding. While effective, the underlying mechanism remains poorly understood, with the extremely low-entropy output distributions suggesting a potential equivalence to simple temperature scaling. In this work, we demonstrate that this phenomenon is fundamentally distinct from distribution sharpening; entropy-matched control experiments reveal that temperature scaling fails to replicate the diversity gains of hyperfitting. Furthermore, we falsify the hypothesis of static vocabulary reweighting, showing through ablation studies that hyperfitting relies on a dynamic, context-dependent rank reordering mechanism. Layer-wise analysis localizes this effect to a "Terminal Expansion" in the final transformer block, where a substantial geometric expansion of the feature space (Delta Dim approx +80.8) facilitates the promotion of deep-tail tokens. Additionally, we introduce Late-Stage LoRA, a targeted fine-tuning strategy that updates only the final 5 layers, yielding robust generation with minimal parameter updates

Problem

• Hyperfitting
• Large Language Models
• Temperature Scaling
• Geometric Expansion
• Open-ended Generation

Innovation

• Hyperfitting
• Terminal Expansion
• Geometric Expansion
• Late-Stage LoRA
• Context-Dependent Rank Reordering