# Perplexity Can Miss SAE Feature Damage Under Quantization

arXiv: https://arxiv.org/abs/2606.03002 · HTML: https://mlsanity.com/paper/2606.03002/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability.
- Builds on 23 of its 25 references that are in the map.
- First posted on arXiv 2 Jun 2026.

## Placed in

- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 23 of 23)

- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [Learning both Weights and Connections for Efficient Neural Networks](https://mlsanity.com/paper/1506.02626/) (2015)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling](https://mlsanity.com/paper/2304.01373/) (2023)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models](https://mlsanity.com/paper/2211.10438/) (2022)
- [AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration](https://mlsanity.com/paper/2306.00978/) (2023)
- [SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot](https://mlsanity.com/paper/2301.00774/) (2023)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [Gemma: Open Models Based on Gemini Research and Technology](https://mlsanity.com/paper/2403.08295/) (2024)
- [A Simple and Effective Pruning Approach for Large Language Models](https://mlsanity.com/paper/2306.11695/) (2023)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://mlsanity.com/paper/2605.29358/) (2026)
- [Steering Language Model Refusal with Sparse Autoencoders](https://mlsanity.com/paper/2411.11296/) (2024)
- [Sparse Autoencoders Trained on the Same Data Learn Different Features](https://mlsanity.com/paper/2501.16615/) (2025)
- [Improving Steering Vectors by Targeting Sparse Autoencoder Features](https://mlsanity.com/paper/2411.02193/) (2024)
- [Evaluating Quantized Large Language Models](https://mlsanity.com/paper/2402.18158/) (2024)
- [Steering Large Language Model Activations in Sparse Spaces](https://mlsanity.com/paper/2503.00177/) (2025)
- [Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders](https://mlsanity.com/paper/2411.01220/) (2024)
- [Sparse but Wrong: Incorrect L0 Leads to Incorrect Features in Sparse Autoencoders](https://mlsanity.com/paper/2508.16560/) (2025)
- [Stable and Steerable Sparse Autoencoders with Weight Regularization](https://mlsanity.com/paper/2603.04198/) (2026)
- [Attribution-Guided Distillation of Matryoshka Sparse Autoencoders](https://mlsanity.com/paper/2512.24975/) (2025)
- [How Pruning Reshapes Features: Sparse Autoencoder Analysis of Weight-Pruned Language Models](https://mlsanity.com/paper/2603.25325/) (2026)

## Built on by (most cited first, 5 of 5)

- [Do Activation Monitors Survive Model Updates? Benchmarking, Predicting, and Repairing Activation-Monitor Staleness](https://mlsanity.com/paper/2606.15980/) (2026)
- [Perplexity Cost Understates What Activation Quantisation Breaks](https://mlsanity.com/paper/2609.23125/) (2026)
- [Cosine Similarity Is Not Evidence: Measuring the Noise Floor of Interpretability Transfer Under Quantization](https://mlsanity.com/paper/2609.30275/) (2026)
- [Linger and Lose: Knowledge Collapse in Low-Bit Language Models](https://mlsanity.com/paper/2609.32902/) (2026)
- [Saying, Not Knowing: Aggressively GGUF-Quantized Small Language Models Still Write Rare Words They Can No Longer Define](https://mlsanity.com/paper/2610.04403/) (2026)

## Abstract (arXiv)

Quantization is a standard path to deploying large language models, and quantized models are typically judged acceptable when perplexity or downstream accuracy remains close to the full-precision original. But behavioral parity need not imply feature fidelity: the sparse-autoencoder (SAE) features used to interpret a full-precision model may change after weight rounding. We test this directly by using a frozen SAE as a fixed measurement basis, encoding full-precision and round-to-nearest (RTN) quantized activations on identical tokens, and measuring per-feature survival by Pearson correlation across bit-widths from INT8 to INT4 on Pythia-70M and Gemma-2-2B. Our central finding is that perplexity can miss feature damage: on Gemma-2-2B, INT7 improves perplexity while degrading 18.7% of active SAE features, and under sliding-window evaluation INT6 also improves perplexity while only 51.3% of active features survive. Feature survival is graded rather than cliff-like, with 62.4% of active Pythia features and 51.3% of active Gemma features surviving at INT6; most non-surviving features are blurred rather than fully damaged. Survival is also predictable from full-precision feature statistics alone, with cross-validated AUC 0.92--0.97 and peak activation as the strongest marginal predictor. Finally, RTN quantization and matched-perplexity magnitude pruning damage strongly overlapping feature sets, with Jaccard overlap 0.79--0.86 and damage-score Spearman correlation 0.98. These results show that behavioral metrics alone are insufficient evidence that full-precision interpretability findings transfer to quantized models, motivating feature-level audits of compression.