Research map / Architectures and efficiency / Quantization
FPTQuant: Function-Preserving Transforms for LLM Quantization
Boris van Breugel, Yelysei Bondarenko, Paul N. Whatmough, Markus Nagel
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add per-token residual scaling Sn to the other transforms (ablation) | FPTQuant, Llama-3.2 3B it W4A4KV4 · WikiText-2 | perplexity: 12.78 -> 12.57 | Improved |
| add pre-RoPE transform Tk (ablation) | FPTQuant, Llama-3.2 3B it W4A4KV4 · 5-shot MMLU | accuracy: 35.46 -> 38.33 | Improved |
| add multihead value transform Tv (ablation) | FPTQuant, Llama-3.2 3B it W4A4KV4 · WikiText-2 | perplexity: 12.78 -> 11.84 | Improved |
| use all three novel transforms together instead of baseline transforms only (ablation) | FPTQuant, Llama-3.2 3B it W4A4KV4 · 5-shot MMLU | accuracy: 35.46 -> 44.64 | Improved |
| apply dynamic per-token scaling Sn inside FFN (ablation) | Llama-3.2 3B it first blocks · FFN intermediate activation kurtosis | kurtosis: 2278 -> 544 (-76.1%) | Improved |
| FPTQuant vs SpinQuant and FlatQuant in the hardest static setting (all attention/MLP activations quantized) | Llama-3.2 3B it W4A4KV4 · WikiText-2 setting (iii) | perplexity: FlatQuant 18.60 -> FPTQuant 17.17 | Improved |
| quantize queries and keys to 4 bit with the mergeable Pre-RoPE transform instead of non-mergeable transforms | FPTQuant vs FlatQuant · WikiText-2 setting (ii), W4A4KV4 | perplexity: FlatQuant 12.30 vs FPTQuant 13.99 | Hurt |
| FPTQuant instead of FlatQuant on very challenging W4A4KV4 static quantization | Llama-3 8B · zero-shot CSR avg | accuracy: FlatQuant 61.43 vs FPTQuant 52.96 | Hurt |
| use FPTQuant mergeable transforms instead of FlatQuant at W4A8KV8 | Llama-2 7B static · WikiText-2 | perplexity: FlatQuant 6.46 -> FPTQuant 5.85 | Improved |
| use student-teacher JSD loss instead of next-token prediction loss for end-to-end training (ablation) | FPTQuant and baselines · generalization | generalization | Improved |
| add locally optimized (Lp norm) transform initialization before end-to-end training (ablation) | FPTQuant · end-to-end training | quantization error | Improved |
| compare FPTQuant with FlatQuant speed on static INT4 prefill | FPTQuant · single transformer block INT4 prefill | speedup: 2.8-3.9x over FP16; 15-29% faster than FlatQuant | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FlatQuant — FlatQuant: Flatness Matters for LLM Quantization (ICML 2025) · better on 10 of 13 shared results
- SpinQuant — SpinQuant: LLM Quantization with Learned Rotations (ICLR 2025) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OSTQuant: Refining Large Language Model Quantization with Orthogonal and Scaling Transformations for Better Distribution Fitting (ICLR 2025 · similarity 0.92)
- DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs (NeurIPS 2024 · similarity 0.91)
- AffineQuant: Affine Transformation Quantization for Large Language Models (ICLR 2024 · similarity 0.91)
- Structured Transforms for Low-Overhead Quantization of Language Models (NeurIPS 2026 · similarity 0.90)
- FlatQuant: Flatness Matters for LLM Quantization (ICML 2025 · similarity 0.90)
- Norm Tweaking: High-Performance Low-Bit Quantization of Large Language Models (AAAI 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
