Hugging Face Trending Papers

From Attention Sensitivity to Layer Role: Revisiting Mixed-Precision Quantization of Transformers

The paper investigates post‑training quantization of transformer attention blocks by optimizing a joint loss over the Q, K, V projections rather than individual weight matrices. Using this joint attention‑based objective (JAB), the authors achieve significant compression on Mistral‑7B, recovering 77‑90% of the performance gap at 3 bits, but the method fails when MLP layers are included. A role‑aware offset rule that ignores sensitivity estimates outperforms JAB on GPT‑2 and full Mistral‑7B, demonstrating that the matrix a weight belongs to is more critical than sensitivity metrics.

arXiv Computation and Language
Sep 16

Where Post-Training Quantization Breaks Text Embedders: A Measured Map Across Four Embedder Families

arXiv:2609. 16391v1 Announce Type: cross Abstract: Weight-only post-training quantization is the cheapest way to shrink a retrieval embedder, and the received advice for applying it -- protect the embedding table, allocate bits by module sensitivity, prefer a ranking-aware objective over weight reconstruction -- was carried into LLM quantization largely intact.

By Hyojung Han
arXiv Machine Learning
Sep 2

The Structure of Quantization Damage in LLMs: Why the Next Bit Should Be Spent Globally

The paper investigates where post‑training quantization (PTQ) harms large language models (LLMs) and how to best allocate a limited precision budget. By causally raising each layer to 8‑bit precision across nine open‑weight models, the authors find that quantization damage is diffuse rather than concentrated in specific task circuits or weight statistics, and that globally refining quantization granularity outperforms selectively protecting the most recoverable layers. They also observe that the residual accuracy loss is budget‑limited and that peak recovery locations correlate with architecture within families but not across families.

By Jundong Hu, Shekar Ramachandran
arXiv Machine Learning
Oct 8

Few Bits, One Law: Toward W2A4KV2

The paper introduces CanonQ, a unified quantization‑aware training framework that separates source canonicalization from task‑aware adaptation to enable extreme low‑bit LLM compression. By applying fixed rotations and energy normalization, CanonQ maps heterogeneous tensors to canonical coordinates, allowing frozen Gaussian‑reference codebooks to be reused across layers and models. Joint training then adapts the network to the coupled errors of weight, activation, and KV cache quantization, achieving significant performance gains—up to 14.28× lower perplexity and 57.9% higher zero‑shot accuracy on LLaMA models, and substantial improvements on instruction‑tuned MobileLLM‑Pro‑1B. "whyItMatters":"The framework demonstrates that carefully designed canonicalization and joint training can dramatically improve the efficiency and accuracy of low‑bit quantized large language models, offering a practical path to deploy powerful LLMs with reduced computational and memory footprints."

By Kai Yi, Tarek Elgamal, Sruthikesh Surineni, Vignesh Vivekraja, Soumyadeep Ghosh, Steven Li
arXiv Machine Learning
Oct 28

Error Diffusion Replaces Latent Weights: Training Binary-Weight Language Models on a Few Bits of Dithered State

The paper introduces an error‑diffusion (ED) optimizer that replaces the full‑precision latent copy of binary‑weight language models with a small 2‑8‑bit residual buffer, reducing training state per weight from 97 bits to 3‑9 bits—up to 19× less. ED accumulates unrealized weight updates in the buffer and flips the binary weight when the accumulated value crosses a threshold aligned with the quantization grid, enabling efficient training on large transformers with minimal loss compared to latent‑weight baselines. The method scales to models with up to 19 M parameters on enwik8 and 70 M on wikitext‑103, maintaining performance within a few percent of reference while allowing on‑device adaptation within a 275 MB memory budget.

By Pierre-Olivier Bonin
arXiv AI
Jul 22

A Controlled Study of Attention-Only Transformers

arXiv:2607. 18363v1 Announce Type: cross Abstract: Feed-forward networks hold two thirds of a transformer's non-embedding parameters, yet the architecture has not received a necessity test that controls parameters, compute, and depth at once.

By Henry Ndubuaku, Karen Mosoyan, Jakub Mroz, Noah Cylich, Satyajit Kumar, Parkirat Sandhu, Roman Shemet, Justin H Lee
arXiv Computation and Language
Oct 1

Learning Functional Subspaces for Neural Network Compression

arXiv:2609.40127v1 Announce Type: cross Abstract: Modern transformers pair impressive capabilities with substantial memory and compute demands. Low-rank weight factorization reduces both while keepin...

By Massimo Bini, Anders Christensen, Stephan Alaniz, Judah Goldfeder, Ole Winther, Yann LeCun, Ravid Shwartz-Ziv, Zeynep Akata
arXiv AI
Sep 11

Scaling Post-Training Ternarisation to Qwen3-8B Capability Retention, Reproduction, Lossless Packing, and Packed Execution

The paper reports a large‑scale post‑training ternarisation of the Qwen3 language model, extending a conversion pipeline from the 4B to the 8B variant. Using KOTMS rotation, E2M‑ATQ adaptive ternarisation, and GPTQ‑style error compensation, the authors achieve a 1.361× perplexity ratio across three corpora and retain 78.5% of the FP16 accuracy on zero‑shot tasks, with the 8B model outperforming the 4B by 8.9 percentage points. The study also demonstrates lossless lattice‑aware packing, producing an 8.24 GiB checkpoint that preserves perplexity, and shows that direct packed execution can reach 15.52 tokens/s in 7.35 GiB, though packed GEMV remains slower than FP16 cuBLAS.

By Anirudh Malik, M Sparsh Mehra, Poojith Devan