A lightweight P2P-based cache system for model distributions on Kubernetes.

Name Story: the inspiration of the name Manta is coming from Dota2, called Manta Style, which will create 2 images of your hero just like peers in the P2P network.

Architecture

Note: llmaz is just one kind of integrations, Manta can be deployed and used independently.

Features Overview

• Model Preheat: Models could be preloaded to clusters, to specified nodes to accelerate the model serving.
• Model Cache: Models will be cached after downloading for faster model loading.
• Model Lifecycle Management: Manage the model lifecycle automatically with different policies, like Retain or Delete.
• Plugin Framework: Filter and Score plugins could be extended to pick up the best candidates.
• Memory Management(WIP): Manage the reserved memories for caching, together with LRU algorithm for GC.

Quick Start

Installation

Read the Installation for guidance.

Preheat Models

A sample to preload the Qwen/Qwen2.5-0.5B-Instruct model:

apiVersion: manta.io/v1alpha1
kind: Torrent
metadata:
  name: torrent-sample
spec:
  hub:
    repoID: Qwen/Qwen2.5-0.5B-Instruct

If you want to preload the model to specified nodes, use the NodeSelector:

apiVersion: manta.io/v1alpha1
kind: Torrent
metadata:
  name: torrent-sample
spec:
  hub:
    repoID: Qwen/Qwen2.5-0.5B-Instruct
  nodeSelector:
    zone: zone-a

Delete Models

If you want to remove the model weights once Torrent is deleted, set the ReclaimPolicy=Delete, default to Retain:

apiVersion: manta.io/v1alpha1
kind: Torrent
metadata:
  name: torrent-sample
spec:
  hub:
    repoID: Qwen/Qwen2.5-0.5B-Instruct
  reclaimPolicy: Delete

More details refer to the APIs.

Community

Join us for more discussions:

• Slack Channel: #manta

Contributions

All kinds of contributions are welcomed ! Please following CONTRIBUTING.md.