WebsiteHunt is broughtBrought to you by Redact Everything
FM

Flash-MoE: 397B on a Laptop

Running a big model on a small laptop.

Flash-MoE runs a 397B Mixture-of-Experts model on a laptop with a pure C/Metal inference engine. Streams a 209GB model from SSD with 4-bit quantization and no Python for production-quality output.

Discussion highlights

AI-generated summary based on Hacker News comments

🔗 Full discussion: https://news.ycombinator.com/item?id=47476422

Read full discussion on Hacker News

Comments

Loading comments...

You might also like

Labrynth AI

Labrynth AI

AI regulatory intelligence platform 🌱 Sponsored
ProjectManagementTools

ProjectManagementTools

Optimize your Projects with ProjectManagementTools 🌱 Sponsored
Introducing Claude Opus 5

Introducing Claude Opus 5

Anthropic's model for long-running agents and coding.
CA

Carton

Run any ML model from any programming language.
Smarter Quiz

Smarter Quiz

Are you smarter than a language model?
Project Vend

Project Vend

Exploring AI's role in managing a small shop
MF

Mavericks Forever

A guide to running Mac OS X Mavericks
Tokenware

Tokenware

One API for All AI Models
Needle: Tiny on-device AI

Needle: Tiny on-device AI

Tiny, open-source AI that runs on small devices.
OS

Operating System in 1,000 Lines

Build a small operating system from scratch
GC

Grok Code Fast 1

A speedy and economical reasoning model for coding.
Voicebox

Voicebox

An all-in-one generative Al model for speech
DeepSeek V4 Pro 0813

DeepSeek V4 Pro 0813

GA release of a large-scale mixture-of-experts model
Replit Code V1.5 3B

Replit Code V1.5 3B

A new code generation language model
Nativ

Nativ

Run AI locally on Apple Silicon.
JigsawStack

JigsawStack

Small models that power your tech stack
YE

You Exist In The Long Context

Thoughts on long-context AI models by Steven Johnson
Gradio

Gradio

demo your machine learning model
LLaMA

LLaMA

A foundational, 65-billion-parameter large language model
Talkie: Vintage 13B Language Model

Talkie: Vintage 13B Language Model

Exploring AI history by training models on pre-1931 text.
PrismML Bonsai 27B: On-device

PrismML Bonsai 27B: On-device

First 27B-class model that runs on a phone with multimodal reason
Molmo by Ai2

Molmo by Ai2

Multimodal Open Language Model built by Ai2
by @Micadep