lattice quantization
Quantization normally rounds each number individually, like snapping each coordinate to a grid. Lattice quantization rounds a whole group of numbers at once, projecting them onto a structured grid in high-dimensional space, such as the E8 lattice. Because that grid exploits correlations across dimensions, it can achieve lower error than scalar quantization at the same bit budget, effectively packing more information into the same number of bits.
The papers show this idea has broad traction for large language models. One line uses lattice methods to compress key-value caches at just 2 bits per key, matching or beating 3-bit scalar schemes. Others apply lattice quantization to model weights, either during training with special gradient estimators or arithmetically without codebook search, reporting accuracy on par with full precision and large memory and throughput gains. The common thread is that moving from per-number to group-level quantization makes extreme compression practical without sacrificing quality.
- Papers
- 4
- Released code
- 1
- First seen
- Oct 2026
- Latest
- Oct 2026
4 papers in the last two months, against 0 in the two before.
Start here
Two papers to read first, in the order they came out- 1EntroPack compresses neural weights at finely adjustable storage rates
EntroPack entropy-codes row-normalized E8 lattice quantized weights to hit arbitrary bitrates with lower error than fixed-width formats and fast GPU reconstruction.
Oct 2026 - 2E8 lattice vector quantization enables 2-bit keys for LLM KV-cache compression
E8 lattice vector quantization with 2-bit keys and 2-bit values achieves 6x KV-cache compression while preserving near-baseline LLM quality.
Oct 2026
How fast it is moving
Often studied with
vector quantizationVector quantization encodes data as indices into a fixed set of codebook vectors, reducing storage and computation at some cost in precision.3 togetherquantization-aware trainingTraining a neural network with simulated quantization in the forward pass so its weights and activations adapt to low-precision formats, improving deployment accuracy.2 togetherkv cache compressionReducing the size of the key-value cache in transformers via selective retention, low-rank updates, or adaptive compression.1 togetherThe papers
Most central to this idea first, not most recent.
- Academiccs.LG
Language models can be trained for compression nearly 13 times faster
SQuAT improves low-bit LLM compression by applying separate straight-through estimators for shape and gain in vector quantization.
University of Copenhagen, Qualcomm AI Research · Oct 2026
- Academiccs.LG
E8 lattice vector quantization enables 2-bit keys for LLM KV-cache compression
E8 lattice vector quantization with 2-bit keys and 2-bit values achieves 6x KV-cache compression while preserving near-baseline LLM quality.
Technical University of Berlin, logsight.ai · Oct 2026
- Independentcs.LGcode
EntroPack compresses neural weights at finely adjustable storage rates
EntroPack entropy-codes row-normalized E8 lattice quantized weights to hit arbitrary bitrates with lower error than fixed-width formats and fast GPU reconstruction.
Oct 2026
- Chinese Techcs.AI
E8 lattice quantization preserves model accuracy at two-bit precision
E8-QAT delivers accurate 2-bit LLM weights by projecting eight-weight groups onto an implicit E8 lattice without explicit codebook storage or search.
Peking University, City University of Hong Kong · Oct 2026
Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.