Searcharxiv⌕ Search

arXiv subjects

Tomer Solberg

Publications and source records attributed to Tomer Solberg.

4 recordsLinked to original sources

Sable: A Small-Field Sumcheck Argument for Bit-Exact Inference of Large Language Models

We present Sable, an argument system and single-GPU prover for the statement that a committed, integer-quantized transformer produced a given token sequence from a given prompt, bit for bit. Sable fixes an integer function F_O with no floating-point step, in which every rounding, normalization, softmax and sampling decision is determined by its inputs, and proves it with a constraint system whose relations provably never wrap modulo the 31-bit KoalaBear prime. The proof batches zerochecks and LogUp-GKR lookups into one sumcheck per domain, and commits to every witness tensor and weight in one multilinear commitment built from Reed-Solomon codes, BLAKE3 Merkle trees and a mixed-length BaseFold opening. The prover streams the model layer by layer on one GPU. On one NVIDIA RTX PRO 6000 Blackwell GPU, Sable proves all 64 layers of Qwen2.5-32B in 92.5 s for 4,096 tokens and 398 s for 16,384 tokens, and covers Qwen3-32B and the mixture-of-experts Qwen3-30B-A3B (4,096 tokens in 121.8 and 84.7 s). A proof costs 41 to 50 times the serving time, or 6.4 cents per 4,096-token request at public GPU prices, and 40 to 64 times less time per parameter and token than zkLLM. On WikiText-2, F_O's greedy token matches the BF16 model's at 86.3% of positions; on Qwen3, F_O is within 1.6 points of BF16 on MMLU and 1.7 points on GSM8K. A verifier with a compiled core checks a proof on one CPU core in 19 to 56 s. In the random-oracle model, the interactive proof has round-by-round soundness error 2^-103. The proof is not zero-knowledge, so the system targets open-weight models. Code: https://github.com/moonmath-ai/Sable

cs.CR↗

HyperQuant: A Rate-Distortion-Optimal Quantization Pipeline for Large Language and Diffusion Models

We present HyperQuant (Hadamard, optimallY Packing, Entropy Rice-coding), a unified post-training quantization pipeline for the weights and the KV cache of large language and diffusion transformers. Across a suite of self-contained experiments (Table 1), HyperQuant outperforms the recent HIGGS scheme at every operating point from 3 to 5 bits per scalar (bps) on weights, and beats both TurboQuant and OCTOPUS on KV quantization down to 1.7 bps. Beyond the LLM setting, HyperQuant quantizes the 19B-parameter LTX-2 DiT video model with no observable per-frame artifacts. End-to-end on an H100 at 4 bps, HyperQuant compresses the linear weights ~3.9x and the KV cache ~3.79x at near-lossless quality. HyperQuant combines four known ideas into a single construction: (i) a per-tile Randomized Hadamard Transform that makes the per-coordinate distribution of weights and activations approximately Gaussian; (ii) quantization to a low-dimensional optimal lattice (E8, D4, A2, or Z); (iii) lossless bit-stripping and near-entropy-optimal variable-length Rice coding of the lattice indices; and (iv) bias-correction methods for the KV cache that keep the reconstruction unbiased under inner products, preserving attention semantics. We further integrate the pipeline with 8-bit and 4-bit Tensor-Core MMA paths (fp8-e4m3, int8, nvfp4, mxfp4), and find that int8 beats fp8 on the post-RHT lattice output. Project page: https://moonmath.ai/hyperquant/

cs.LG↗

Understanding Higher-Spin Gravity through Bilocal Holography for Free Fermions

We consider a simple theory of N free fermions in d dimensions with O\left(N\right) or U\left(N\right) symmetry. The singlet sector of this theory is expected from holography to be dual to the notoriously complicated Vasiliev gravity. By defining bilocal fields we find an explicit holographic mapping between the two theories, similar to what was done previously for scalars, which allows us to construct a quantum version of Vasiliev gravity, including loop (=1/N) corrections. Further more, such a mapping establishes a non-supersymmetric instance of the AdS/CFT duality.

hep-th↗

The action of long strings in supersymmetric field theories

Long strings emerge in many Quantum Field Theories, for example as vortices in Abelian Higgs theories, or flux tubes in Yang-Mills theories. The actions of such objects can be expanded in the number of derivatives, around a long straight string solution. This corresponds to the expansion of energy levels in powers of 1/L, with L the length of the string. Doing so reveals that the first few terms in the expansions are universal, and only from a certain term do they become dependent on the originating field theory. Such classifications have been made before for bosonic strings. In this work we expand upon that and classify also strings with fermionic degrees of freedom, where the string breaks D=4 N=1 SUSY completely. An example is the confining string in N=1 SYM theory. We find a general method for generating supersymmetric action terms from their bosonic counterparts, as well as new fermionic terms which do not exist in the non-supersymmetric case. These terms lead to energy corrections at a lower order in 1/L than in the bosonic case.

hep-th↗