Liquid4All/cookbook

What it solves

This repository provides a comprehensive collection of examples, tutorials, and applications designed to help developers deploy and customize Liquid AI's open-weight Liquid Foundation Models (LFMs) and the LEAP SDK across various platforms, including laptops, mobile devices, and edge hardware.

How it works

The cookbook organizes implementation patterns into several categories:

• Platform-Specific Apps: Ready-to-use code for desktop (Python/CLI), browser (WebGPU/ONNX Runtime Web), and mobile (Kotlin for Android, Swift for iOS) applications.
• Fine-Tuning Pipelines: Colab notebooks and scripts using frameworks like Unsloth, TRL, and Halo to perform Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Group Relative Policy Optimization (GRPO) on text, vision, and audio models.
• Deployment Guides: Step-by-step notebooks demonstrating specific model capabilities such as image understanding and clinical data interpretation.
• Integration Examples: Showcases of how LFMs are integrated into third-party and community-built projects, ranging from RAG apps to on-device voice assistants.

Who it’s for

It is intended for AI developers and engineers who want to implement small, efficient language models on-device or at the edge, as well as those looking to fine-tune these models for specialized domains like satellite imagery, audio transcription, or medical data.

Highlights

• Cross-Platform Support: Examples for native iOS/Android apps, WebGPU browser apps, and desktop CLI tools.
• Multimodal Capabilities: Implementations for vision-language models (VLM), audio-to-text transcription, and text-based reasoning.
• Advanced Training Techniques: Tutorials for memory-efficient fine-tuning (LoRA), reinforcement learning (GRPO), and eliminating "doom loops" via FTPO.
• Edge-First Focus: Heavy emphasis on local inference to ensure privacy and reduce latency on mobile and edge devices.