#Mixture Of Experts

Models Aren't Getting Dumber — They're Getting Unbundled

Active parameters collapsed while reasoning scores soared, and your retrieval stack quietly became responsible for the facts.

Article · 1w ago0

Run Mixtral 8x7B Locally with llama.cpp and Benchmark MoE vs. Dense

Compile llama.cpp, run quantized Mixtral 8x7B, and measure the mixture-of-experts speed-versus-memory trade-off yourself.

Tutorial · 3w ago0

How a 20B Model Hits 120 tok/s on an iPhone

DeepGrove's Maple-Preview stacks native ternary training on MoE sparsity, and the arithmetic largely holds up.

Article · 3w ago1

Your SSD Is the New VRAM for Local LLMs

Swiftlet streams sparse Qwen experts off disk, fitting an 80B model into 4.3 GB of Mac RAM.

Article · 3w ago1

Your SSD Is the New VRAM

Swiftlet streams MoE experts from SSD, fitting an 80B Qwen in 4.3GB of RAM — and a 35B on an iPhone.

Article · 3w ago1

AirLLM's 4GB 70B Trick Is Real, and Beside the Point

Layer streaming makes giant models fit; disk bandwidth decides whether you'd ever want to run them that way.

Article · 3w ago1

A 26B Model in 2 GB of RAM, Courtesy of Your SSD

TurboFieldfare streams Gemma 4's experts from flash, and the trick will outlive the tool.

Article · 1mo ago2

Kimi K3 Turns a Year of Efficiency Tricks Into 2.8T Parameters

Moonshot's new flagship validates hybrid attention, latent MoE, and 4-bit training at frontier scale — and redefines open weights.

Article · 1mo ago1

Inkling Bets Fine-Tuning Beats Frontier Chatbots

Thinking Machines' first open-weights model is less a chatbot competitor than an infrastructure bet on customization over off-the-shelf AI.

News · 1mo ago5

Inkling Puts 975B Open Weights in Reach

Murati's lab ships a frontier MoE you can fine-tune, not just rent.

Article · 1mo ago2