Idea
#1
by HuggingfacebrosLLM - opened
🦅 Supra-Edge-100M
Tagline: "The smartest model that fits everywhere."
🎯 Goal
A 100M parameter base and instruct model (two variants) optimized for:
📱 Phones (Android)
🍓 Raspberry Pi
💻 Old laptops
🖥️ CPUs without GPUs
🌐 Offline use
🚀 Why this is interesting
The jump from 50M → 100M is huge for reasoning quality while still being extremely lightweight.
Most people either build:
Tiny meme models (1M–10M)
Huge models (7B+)
Very few organizations focus on the 100M sweet spot. Combo for training data: 50B FineWeb-Edu tokens
6B Wikipedia tokens
3B The Stack tokens
59B tokens
We are already producing a 124M model.
AxionLab-official changed discussion status to closed
BTW: @HuggingfacebrosLLM stop using AI to impress us. ☹️