LMX-Omni-52B-Halo

LMX-Omni-52B-Halo

LMX-Omni-52B-Halo is a Lemonade Mix (LMX) virtual model that offers true all-to-all omni-modality: chat, vision, image generation and editing, speech transcription, and text-to-speech through a single OpenAI-compatible model. It is a curated bundle of component models built around a large MoE LLM and targeted at Strix Halo class hardware, served by Lemonade.

Components

Total download: ~44.8 GB

Usage

Use it in the Lemonade app

Install Lemonade, then:

lemonade pull LMX-Omni-52B-Halo
lemonade run LMX-Omni-52B-Halo

Pulling the collection downloads every component above. Once downloaded, the model appears in the Lemonade app's chat, where you can talk to it, generate and edit images, and hear spoken replies.

Use it in any OpenAI-compatible app

Any app that calls /chat/completions and can render multimedia output, such as Open WebUI or AnythingLLM, can use the model by name. The server runs an internal tool-calling loop across the components and embeds generated images and audio directly in the assistant message, so to the app it behaves as a genuine OpenAI-compatible chat model.

Build your own app

LMX models are made for building apps with native omni-modal interactions. Apps that want full control can run the omni tool-calling loop themselves, addressing the planner LLM directly with the same tool definitions. See Lemonade Omni Models for the server-side and client-side orchestration contracts, and Embeddable Lemonade to ship the whole experience inside your own app.

Collection definition

This repository is the source of truth for the collection definition: the self-contained manifest LMX-Omni-52B-Halo.json carries each component's checkpoint, recipe, labels, and defaults, so the collection is fully reproducible from this repo alone.