Release Notes

Changelog

ZETIC Melange SDK release notes and version history.

This page tracks release notes and version history for the ZETIC Melange SDK and platform.

2025

LLM Inference Engine

Added support for google/gemma-3-4b-it
Added support for LiquidAI/LFM2.5-1.2B-Instruct
KV cache cleanup policy configuration (LLMKVCacheCleanupPolicy)
Download progress callback support
Speed inference mode (LLMModelMode.RUN_SPEED) available for all models

Hugging Face Integration

ZeticMLangeHFModel for loading models directly from Hugging Face
No personal key or model key required
Automatic download, compilation, and caching

General SDK

PyTorch Exported Program (.pt2) support
Inference mode selection (RUN_AUTO, RUN_SPEED, RUN_ACCURACY)
Performance-adaptive deployment across 200+ devices
Docs site migration to docs.zetic.ai

CLI

zetic gen command for model upload and compilation
Support for .pt2, .onnx, and .pt model formats
Browser-based authentication via zetic auth login

Upcoming

TorchScript (.pt) deprecation
Flutter SDK (general model inference)
React Native SDK (general model inference)
LLM Auto and Accurate inference modes
On-device RAG support
Multimodal NPU acceleration

For the latest updates and announcements, follow us on GitHub and join our Discord community.

SDK Repositories

Platform	Repository	Install
Android	Maven Central	`implementation("com.zeticai.mlange:mlange:+")`
iOS	ZeticMLangeiOS	Swift Package Manager
CLI	PyPI	`pip install zetic`

FAQ

Frequently asked questions about ZETIC Melange.

Migration Guide: MLange → Melange

What changed in the rebrand from MLange to Melange, and what you need to update.

On this page

2025 LLM Inference Engine Hugging Face Integration General SDK CLI Upcoming SDK Repositories