tinyllm

Tiny devices, big ideas , from bits to brilliance

An LLM model with 10 million parameters was designed for can reasoning

Fleet of tiny specialist LLMs: one shared 360M base, one LoRA adapter per specialty, embedding-centroid router with abstain path. Numbers + repro included.

A ~8M parameter LLM that talks like a tiny crab in an aquatic world. Runs in your browser, classifies intent in one forward pass.

Dynamic TinyLLM provider package for Pi

A 1,006,976-parameter transformer that runs entirely in your browser via a 13 KB WebAssembly kernel

TinyLLM for ESP32 Dev Kit V1