Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...

From ‘Optical Synapses’ to a ‘Photonic Brain’ — Integrated Photonic Neural Networks Toward Low-Power General-Purpose

SHANNON, CLARE, IRELAND, February 5, 2026 /EINPresswire.com/ -- A new publication from Opto-Electronic Technology; DOI ...
IEEE

SIMAX: a SIMD-Based Many-Core Accelerator for Matrix-Vector Multiplication for Transformers

Matrix-vector multiplication (MVM) is a computational bottleneck for transformer inference workloads at resource-restricted edge applications. Efficient MVM accelerator design is crucial to optimizing ...
IEEE

GAS: General-Purpose In-Memory-Computing Accelerator for Sparse Matrix Multiplication

Abstract: Sparse matrix multiplication is widely used in various practical applications. Different accelerators have been proposed to speed up sparse matrix-dense vector multiplication (SpMV), sparse ...