# LocalAI v2.24.0 — Release & Upgrade Intelligence
- **Product:** LocalAI (https://localai.io)
- **Version:** v2.24.0
- **Released:** Dec 04, 2024
- **Impact Score:** 7/10
- **Breaking Changes:** Source flagged
- **Canonical Analysis:** https://changelogs.us/localai/updates/v2.24.0/

## Executive Summary (TL;DR)
LocalAI v2.24.0. Source highlight: LocalAI release v2.24.0!

## Breaking Changes & Deprecations
- Breaking Changes 🛠 * feat(models): use rwkv from llama.cpp by @mudler in https://github.com/mudler/LocalAI/pull/4264 * feat(backends): Drop bert.cpp by @mudler in https://github.com/mudler/LocalAI/pull/4272

## New Features
- **New Backends Added**: Introducing `bark.cpp` for text-to-audio and `stablediffusion.cpp` for image generation, both powered by the ggml framework.
- **Voice Activity Detection (VAD)**: Added support for `silero-vad` to detect speech in audio streams.
- **`bark.cpp` Backend**: Transform text into realistic audio using Bark, a transformer-based text-to-audio model. Install it easily with:
- **`stablediffusion.cpp` Backend**: Create high-quality images from textual descriptions using the Stable Diffusion backend, now leveraging the ggml framework.
- **Voice Activity Detection with `silero-vad`**: Introducing support for accurate speech segment detection in audio streams. Install via:
- **Real-Time Token Usage**: Monitor token consumption dynamically during streamed outputs. This feature helps optimize performance and manage costs effectively.

## Improvements
- **Backend deprecation**: We’ve removed `rwkv.cpp` and `bert.cpp`, replacing them with enhanced functionalities in `llama.cpp` for simpler installation and better performance.
- **WebUI Improvements**: Now supports API key authentication for enhanced security.
- Calme-3 Qwen2.5 Series: Enhanced language models offering improved understanding and generation capabilities.
- Llama 3.1 and 3.2 Series: Upgraded versions of the Llama models with better performance and accuracy.

## Bug Fixes
- Performance Enhancements: Resolved issues with AVX flags and optimized binaries for accelerated performance, especially on macOS systems.
- Dependency Updates: Upgraded various dependencies to ensure compatibility, security, and performance improvements across the board.
- Parsing Corrections: Fixed parsing issues related to maddr and ExtraLLamaCPPArgs in P2P configurations.
- fix(hipblas): disable avx flags when accellerated bins are used
- chore(deps): bump sycl intel image
- fix(go.mod): add urfave/cli v2