liquidai

NuTonic combines temporal satellite memory with visual-language reasoning so AI can not only see Earth from space, but explain what is changing and help people act sooner.

A self-contained CUDA inference engine for LiquidAI/LFM2.5-8B-A1B (hybrid conv + GQA-attention MoE, 8.5B params, 1B active) targeting a single RTX 3060 (12 GB). No Python, no frameworks at runtime: a single .cu engine + a header-only byte-level BPE tokenizer.

A list of Edge optimized Ollama models

Fine-tuning LiquidAI/LFM2-VL-1.6B in Colab (LoRA/4-bit) + dataset template + probe test.

Introducing liquid glass design into SaaS Webpage.

Anywhere-LFM est une application de bureau légère, pensée pour simplifier l'utilisation en local des modèles LiquidAI/LFM2 (350M à 1.2B). Elle permet à n'importe qui de discuter avec une IA générative performante, sans nécessiter de ligne de commande ou de configuration complexe.

⚡🗼Liquid AI x W&B x Lambda Hackathon —🥈 2nd Place 🧠 Building creative, human-centered AI experiences. 💻 Focused on multimodal LLMs, voice interaction & fast prototyping (vibe coding). 🌸 Exploring the intersection of Japanese culture and AI innovation.

A Vision Language Model that runs on-board satellites to automatically decide what's worth downloading to Earth

A native SwiftUI iOS app where a small language model orchestrates a time-series foundation model as a tool, entirely on the iPhone via Apple's MLX

⚡🗼Liquid AI × WAY × AMD hackathon — 🐈 NyanGuard AI is a local-first moderation demo for audio streams. 📅2026-06-06. It accepts an audio or video file, transcribes speech locally when a local ASR backend is available, replaces toxic language with Nyan Nyan, and returns a playable moderated audio file.

SFT of LiquidAI LFM2.5 models