# Diagnostic-Driven Layer-Wise Compensation for Post-Training Quantization of Encoder-Decoder ASR Models

arXiv: https://arxiv.org/abs/2601.02455 · HTML: https://mlsanity.com/paper/2601.02455/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 31 references that are in the map.
- First posted on arXiv 5 Jan 2026.

## Builds on (most cited first, 21 of 21)

- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration](https://mlsanity.com/paper/2306.00978/) (2023)
- [ZeroQuant: Efficient and Affordable Post-Training Quantization for Large-Scale Transformers](https://mlsanity.com/paper/2206.01861/) (2022)
- [QuIP: 2-Bit Quantization of Large Language Models With Guarantees](https://mlsanity.com/paper/2307.13304/) (2023)
- [QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks](https://mlsanity.com/paper/2402.04396/) (2024)
- [SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression](https://mlsanity.com/paper/2306.03078/) (2023)
- [Atom: Low-bit Quantization for Efficient and Accurate LLM Serving](https://mlsanity.com/paper/2310.19102/) (2023)
- [Extreme Compression of Large Language Models via Additive Quantization](https://mlsanity.com/paper/2401.06118/) (2024)
- [TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation](https://mlsanity.com/paper/1805.04699/) (2018)
- [SPGISpeech: 5,000 hours of transcribed financial audio for fully formatted end-to-end speech recognition](https://mlsanity.com/paper/2104.02014/) (2021)
- [Outlier Suppression+: Accurate quantization of large language models by equivalent and optimal shifting and scaling](https://mlsanity.com/paper/2304.09145/) (2023)
- [Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization](https://mlsanity.com/paper/2504.09629/) (2025)
- [4-bit Conformer with Native Quantization Aware Training for Speech Recognition](https://mlsanity.com/paper/2203.15952/) (2022)
- [Shrinking Bigfoot: Reducing wav2vec 2.0 footprint](https://mlsanity.com/paper/2103.15760/) (2021)
- [Integer-only Zero-shot Quantization for Efficient Speech Recognition](https://mlsanity.com/paper/2103.16827/) (2021)
- [Moonshine: Speech Recognition for Live Transcription and Voice Commands](https://mlsanity.com/paper/2410.15608/) (2024)
- [A Model for Every User and Budget: Label-Free and Personalized Mixed-Precision Quantization](https://mlsanity.com/paper/2307.12659/) (2023)
- [Extremely Low Footprint End-to-End ASR System for Smart Device](https://mlsanity.com/paper/2104.05784/) (2021)
- [Effective and Efficient Mixed Precision Quantization of Speech Foundation Models](https://mlsanity.com/paper/2501.03643/) (2025)
- [One-pass Multiple Conformer and Foundation Speech Systems Compression and Quantization Using An All-in-one Neural Model](https://mlsanity.com/paper/2406.10160/) (2024)
- [StableQuant: Layer Adaptive Post-Training Quantization for Speech Foundation Models](https://mlsanity.com/paper/2504.14915/) (2025)

## Abstract (arXiv)

Layer-wise post-training quantization reconstructs each layer from inputs already altered by the quantized prefix. QEP compensates for this drift with one model-wide coefficient, conflating the model-level operating point with residual variation across layers. We present FADE, which constructs layer-specific coefficients from normalized round-to-nearest distortion and a heuristic calibrated-solver response. It requires no training or per-model coefficient search and adds no inference-time operation. We evaluate seven Whisper, Moonshine, and Qwen3-ASR models at 3 and 4 bits on four English ASR benchmarks. Across 38 settings, FADE lowers mean word error rate relative to fixed QEP-0.5 in 31, although a development-tuned global coefficient recovers much of this gap. The largest absolute reductions occur in 3-bit settings whose final error remains too high for practical use; these results measure collapse mitigation rather than deployable accuracy. In two lower-WER 3-bit cases, FADE reduces the tuned-global result from 3.67 to 3.10 and from 13.03 to 11.63. Most 4-bit differences from the tuned control are within a descriptive tolerance, and one reverses. Paired reruns and within-seed permutations on an outcome-informed subset support assignment sensitivity in selected cases, but do not estimate a matrix-wide success rate. FADE is therefore a layer-wise alternative when per-model coefficient search is unavailable, not a universal replacement for tuned global compensation.