
I bought Thunderbolt 5 and got USB4 40G: re-benchmarking a two-machine llama.cpp RPC setup across three links and six models
This post is a follow-up to my previous benchmark. It tests six models across three links — 5GbE,...
Tag archive

This post is a follow-up to my previous benchmark. It tests six models across three links — 5GbE,...

In the endless pursuit of "bigger, bigger, and even bigger," I finally reached this point: two...
Is Your Phone's LLM 37x Too Slow? Check This One Compiler Flag Part 3 of a series. Part...
Weights, data, licenses: the three things people conflate when they say open source.
🚀 Local LLMs in 2026: Running 70B Models Directly on Mac M4 Studio Have you ever wondered...
Meta shut down its Llama API in 2026: migrate your app and compare host costs Summary....
How Llama 4, Mistral Large 3, Gemma 4, and DeepSeek V4 are closing the gap with proprietary models in 2026.

Meta's shift from open-source Llama champion to proprietary Mango and Avocado models marks a strategic pivot. Analysis of the $14.3B Scale AI deal, delayed
Meta passe de champion open-source Llama à modèles propriétaires Mango et Avocado. Analyse du contrat de 14,3 milliards $ avec Scale AI, du retard d'Avocad
Adding gradient information to model compression takes a 2-bit LLaMA-3-70B from complete gibberish under the dominant method to a functioning model, suggesting the standard approach has been missing half the picture since 2022.
anemll-flash-llama.cpp is useful for running large MoE models on Macs with limited RAM. Instead of...