A field guide for putting tiny language models everywhere

What it does This repository is a curated catalog of applications, tutorials, and notebooks for Liquid AI’s open-weight LFM models and the LEAP SDK. It covers desktop scripts, zero-install browser demos powered by WebGPU, native Android and iOS apps, and fine-tuning pipelines for text, vision, and audio models.

The interesting bit The project treats browsers and phones as first-class deployment targets rather than demo-day novelties. The LEAP Edge SDK is pitched as making on-device inference feel as simple as calling a cloud API, which is a refreshing stance in an industry that usually assumes you want a data center.

Key highlights

• Browser apps run entirely client-side via WebGPU and ONNX Runtime Web; no install required.
• Mobile examples span Android (Kotlin) and iOS (Swift), including chat, voice assistants, and vision-language models.
• Fine-tuning notebooks cover supervised fine-tuning, GRPO for reasoning, continued pre-training for translation, and domain-specific vision/audio adaptation.
• Desktop demos include practical tools like invoice parsing, audio transcription, and a local home assistant pipeline.
• Community contributions range from satellite wildfire detection to a grumpy Italian chef fine-tuned with DPO.

Caveats

• Many entries link out to separate repositories or Hugging Face spaces, so this repo functions more as an index than a self-contained codebase.
• The README truncates abruptly at the end of the community projects section, leaving at least one description unfinished.

Verdict Bookmark this if you are evaluating small models for edge, mobile, or browser deployment. Look elsewhere if you need a single, monolithic framework; this is a tour guide, not a library.