Pith. sign in

REVIEW 4 cited by

Matryoshka Quantization

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2502.06786 v3 pith:LGCTBA34 submitted 2025-02-10 cs.LG cs.AI

Matryoshka Quantization

classification cs.LG cs.AI
keywords modelint2quantizationmodelsmatquantmatryoshkaprecisionbase
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Quantizing model weights is critical for reducing the communication and inference costs of large models. However, quantizing models -- especially to low precisions like int4 or int2 -- requires a trade-off in model quality; int2, in particular, is known to severely degrade model quality. Consequently, practitioners are often forced to maintain multiple models with different quantization levels or serve a single model that best satisfies the quality-latency trade-off. On the other hand, integer data types, such as int8, inherently possess a nested (Matryoshka) structure where smaller bit-width integers, like int4 or int2, are nested within the most significant bits. Leveraging this insight, in this paper, we propose Matryoshka Quantization (MatQuant), a novel multi-scale quantization technique that alleviates the aforementioned challenge. This technique allows us to train and maintain a single quantized model but serve it with the precision demanded by the deployment. Furthermore, leveraging MatQuant's co-training and co-distillation regularization, int2 precision models extracted by MatQuant outperform standard int2 quantization by up to to 4% and 7% with OmniQuant and QAT as base algorithms respectively. Finally, we demonstrate that by using an extra bit to represent outliers, a model with an effective precision of 2.05-bit gives an additional 6% improvement with OmniQuant as the base algorithm.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. BitNet Text Embeddings

    cs.CL 2026-06 unverdicted novelty 6.0

    BITEMBED converts LLM backbones to ternary BitNet-style encoders, adapts them with contrastive pre-training and teacher distillation, and produces text embeddings at multiple precisions that perform comparably to full...

  2. BitNet Text Embeddings

    cs.CL 2026-06 conditional novelty 6.0

    BITEMBED trains 1.58-bit ternary-weight LLM embedders with contrastive pre-training, supervised distillation, and multi-precision output training, matching FP16 teachers within ~0.6 MMTEB points at ~2x CPU speed.

  3. GRINQH: Graded Input-based Quantization Hierarchy for Efficient LLM Generation

    cs.LG 2026-06 unverdicted novelty 6.0

    GRINQH introduces a graded input-based quantization hierarchy that dynamically assigns multi-precision weights using activation magnitudes as importance proxy, unifying quantization with sparsification to improve LLM ...

  4. Multi-Bitwidth Quantization for LLMs Using Additive Codebooks

    cs.LG 2026-06 unverdicted novelty 5.0

    Drop-by-Drop uses additive codebooks and Matryoshka-style training to produce one LLM model whose ordered codebook subsets give accurate reconstructions at successively higher bitwidths under a weighted MSE distortion.