How DeepSeek’s DSpark Makes LLMs 85% Faster Without Retraining

DeepSeek has introduced DSpark, an open-source speculative decoding method that accelerates LLM inference by up to 85% ...
techtimes

NVIDIA’s New LLM Decodes 6x More Tokens Without an Auxiliary Draft Model

Any engineer currently running Eagle3 or a multi-token-prediction setup to speed up their LLM inference now has a compelling reason to look at the alternative NVIDIA's research team published Tuesday.
XDA Developers on MSN

I tested Google's new Gemma 4 12B on my 8GB GPU, and now I don't want to go back to smaller models

Not bad for limited hardware ...
VentureBeat

Context compression finally works in production: new research cuts LLM input 16x without the accuracy hit

Context windows are becoming a computational bottleneck. The longer an agent runs, the more tokens accumulate from retrieved documents, reasoning traces and conversation history, and the more memory ...
EDN

MLPerf and the rise of latency-aware LLM benchmarking

Any discussion of modern AI system performance must include MLCommons and its MLPerf benchmark suite, which has become the industry’s de facto standard for measuring machine learning performance.
note

Multimodal LLMs That Ditched the Encoder — Reading the "Direct Projection" Architecture of Gemma 4 12B

On June 3, 2026, Google DeepMind released Gemma 4 12B (where 12B = 12G = 12 billion parameters). It is a model capable of handling images, audio, and video, making it an AI that can process multiple ...
note

Gemma 4 12B In-Depth: A New Model Bringing Full-Scale Multimodality to Laptops with an Encoder-Free Design

Gemma 4 12B is a new model in the Gemma 4 family announced by Google on June 3, 2026. It is positioned as an "encoder-free unified multimodal model optimized for laptops." The official blog (Google ...
IEEE

Towards Effective and Efficient Non-Autoregressive Decoders for Conformer and LLM-Based ASR Using Block-Based Attention Mask

Abstract: Automatic speech recognition (ASR) systems often rely on autoregressive (AR) Transformer decoder architectures, which limit efficient inference parallelization due to their sequential nature ...
GSM Arena

MediaTek unveils Dimensity 8550 with LLM Booster and support for Gemini Nano V3

Earlier this year, MediaTek unveiled the Dimensity 8500, a 4nm (TSMC N4P) chipset with all-big CPU cores (eight Cortex-A725) and a Mali-G720 MC8 GPU. Now the company is refreshing that chip as the ...
GitHub

OneCAT: Decoder-Only Auto-Regressive Model for Unified Understanding and Generation

We introduce OneCAT, a unified multimodal model that seamlessly integrates understanding, generation, and editing within a novel, pure decoder-only transformer architecture. Our framework uniquely ...