Topic: quantization techniques
-
Google's TurboQuant AI Memory Compression Shakes Chip Stocks
Google's new TurboQuant AI algorithm compresses a key memory component in AI models by at least sixfold, reducing it to 3 bits per value while maintaining accuracy. The breakthrough targets the costly key-value cache, a bottleneck for AI inference, and triggered a sharp sell-off in memory stocks ...
Read More » -
Google's TurboQuant AI Cuts LLM Memory Use by 6x
The high memory demands of large language models (LLMs) are a key factor in current high memory prices, driven by the substantial memory consumption of their key-value caches. Google's new TurboQuant compression technique dramatically shrinks an LLM's memory footprint and accelerates performance ...
Read More » -
Google's Gemma 4 AI models now use Apache 2.0 license
Google has released the new Gemma 4 family of open-weight AI models, now under a permissive Apache 2.0 license, making them more accessible for commercial and research use. The lineup includes four sizes designed for local hardware, with the largest models (26B Mixture of Experts and 31B Dense) o...
Read More » -
DeepSeek R1: Quantum Breakthrough Shrinks AI Model
Researchers tested an uncensored AI model's ability to answer sensitive questions, using GPT-5 as a judge, and found it provided factual responses comparable to Western models. Multiverse is developing technology to compress AI models for greater efficiency, aiming to reduce energy use and costs ...
Read More » -
Nvidia AI Chief: Why Open Models Matter in AI
Organizations in developing nations sometimes prefer older, well-tested AI models over newer ones for predictable output, as software upgrades can change behavior and break systems built around specific prompts. Nvidia emphasizes that companies need skills to quickly evaluate and adopt new models...
Read More »