---
title: Kimi K3 AirLLM Demo
emoji: 🤖
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: "5.38.0"
python_version: "3.10"
app_file: app.py
pinned: false
---

# Kimi K3 + AirLLM

A Gradio demonstration of `devoppro/Kimi-K3` using AirLLM for memory-efficient inference.

The model cache is stored on the attached persistent `/data` volume.

## Model

`devoppro/Kimi-K3`

## Backend

AirLLM + Transformers

## Storage

The Space expects a persistent volume mounted at:

`/data`

The model can require a very large amount of persistent storage.

> First startup can take a long time because the Kimi K3 checkpoint consists of many large safetensor shards.