🚀 LocalAI v2.20.0 is LIVE! 🚀
🎉 Buckle up! We've got some updates to share and fresh features that are about to take your AI game to a whole new level.
Here's what's hot in this release, a 🧵👇
LocalAI now supports Hermes2-pro-mistral with function calling! .
It is part of the All-in-One images (), ready to level up your function calling skills?
Thanks
@Teknium1
for this amazing model! works so good for function calls!
Get started with AI without any configuration with all-in-one (AIO) magic images! 🌟 The default model for CPU is even more powerful and now using , and for Intel GPU peeps, we haven't forgotten you. Thanks to
@Teknium1
for building such a great model!
👋 Hey AI fellows! we have some exciting news for you! LocalAI was just released v1.20.0 with some amazing new features and updates. Here are some highlights:
(1/7)
👇New models available in the LocalAI gallery!
- NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF () Another great release from
@Teknium1
🙌 !
- MaziyarPanahi/WizardLM2-7b () cheapeau to
@MaziyarPanahi
for the fantastic work ! 🫶
Enjoy!
🚀 Super excited to share that LocalAI v2.10.0 is here!
🎉 We've packed this update with some really cool features and necessary tweaks. It's been a journey, and I can't wait to dive into the details with you all. Let’s get into it!
a thread 1/9 🧵👇
🖼️ Highlight: This is the first release featuring a model gallery in the webUI. Discover a selection of models including SD, llama3, tts, embeddings, and more directly from the webui. You can also find lang-specialized models, uncensored, and many more!
2/6 🧵
🕶️We’ve leveled up our WebUI!
You can chat with any model in your instance directly from the same interface, an easy-to-go, no-frills, and simple, hackable interface built with
@htmx_org
and
@Alpine_JS
!
LocalAI got featured in the state of Open Source AI book!
If you want to be on the loop with AI and Open source, make sure to bookmark this gem from our friends
@premai_io
!
👉
🚀🔥 Exciting news! LocalAI v1.18.0 is here with a stellar release packed full of new features, bug fixes, and updates! 🎉🔥
A huge shoutout to our amazing community for their invaluable help in making this a fantastic community-driven release! 🙌
Let's see what's new 👇👇
Big news,
#LocalAI
fam! 🎉 Our latest release (v2.9.0) is out, and it's packed with enhancements, fixes, and a whole lotta love from our amazing community! Let's dive into what's new! 🔥🔥
🧵 Thread 1/8👇
📢 v1.14.0 is here! This release brings compatibility with the latest quantization method updates from llama.cpp while maintaining support for older models!
LocalAI supports
@huggingface
diffusers! You can generate stunning images with LocalAI by using the same OpenAI API you are used to.
You can try that out by just running "local-ai animagine-xl"!
Guess what? It's release time again! 🔥 👉 Backend updates, consolidated CUDA support (thanks to the community!), and now.. 🥁🥁 models gallery! With models gallery, sharing and installing models in LocalAI got much easier! Check out what's new in v1.13.0!
We're rolling out new features and improvements that developers have been asking for:
1. Our new model GPT-4 Turbo supports 128K context and has fresher knowledge than GPT-4. Its input and output tokens are respectively 3× and 2× less expensive than GPT-4. It’s available now to
🌐 Say Hello to LocalAGI! 🌐
Experience LocalAI functions like never before with LocalAGI. No API keys needed - just pure locally runnable assistant magic. Check it out in action, planning a San Francisco trip! 🗺️🗣️
@mudler_it
👉 Try it out: We hope you enjoy this new release and let us know what you think. You can download it from Github or use our container images. Don’t forget to check out our docs and join our Discord server for support and to reach out to the community. Happy hacking! 😊
(7/7)
Introducing Parler-TTS: an inference and training library for high-quality, controllable text-to-speech (TTS) models 🗣️
To fuel the development of open-source TTS research, we are open-sourcing all datasets, training code and our first iteration checkpoint: Parler-TTS Mini v0.1
🔥👉v1.8.0 just released! 🚀 LocalAI gets faster and more accurate embeddings support with bert.cpp. LocalAI also supports now audio transcriptions with whisper.cpp!
🙏 Thanks: We want to thank all our contributors, who helped us make this release possible. A special shout-out to
@mudler_it
, our awesome author/maintainer and the driving force behind each release. You rock! 🙌
(6/7)
1.6 is out! 🔥 rwkv backend support, token stream support added in llama and rwkv. And now thanks to the community we have examples on how to integrate with langchain! 👉
There’s a lot of awesome open-source LLM projects with OpenAI-compatible API’s over non-openai models (vllm,
@LocalAI_API
,
@anyscalecompute
LLM endpoints, etc).
Thanks to
@the_benbot
, you can now use *any* of them with our `OpenAILike` class 👇
(Example for vllm + mistral 🖼️)
My hardcopies of the official
@Raspberry_Pi
magazine
@TheMagPi
have arrived! And they have my Automatic Speech Recognition project based around
@NordVPN
’s Meshnet and a self-hosted
@LocalAI_API
language model inside. Huge thanks to Rob from MagPi for showing interest in it.
LocalAI plays well with many tools and integrations!
Thanks to
@sozercan
, now you can already use images with models prepared for use, don't miss out aikit!
Check out also other examples and how you can integrate LocalAI:
So, I'm playing with something.. and.. almost alive!
🧪Experiment: can you make 100% local (that works on CPU and on GPU) something like Bing or AutoGPT and works as well with OpenAI APIs? Seems you can!
a small 🧵👇
There are serious challenges in trusting OpenAI for production products. I'm not ragging on them. They're definitely SOTA and I don't blame them for this, but if your business relies on their models, and they don't support the models indefinitely, you can get the plug pulled
Mamba is a novel class of SSM (structured state space model) architecture.
You can try it out already as an API in LocaLAI just with "local-ai mamba-chat"!
Transformers power most advances in LLMs, but its core attention layer can’t scale to long context.
With
@_albertgu
, we’re releasing Mamba, an SSM architecture that matches/beats Transformers in language modeling, yet with linear scaling and 5x higher inference throughput.
1/
🎙️ Text-to-Audio generation: Now you can turn any text into speech with our new integration with go-piper. You can use different voices and languages from our gallery or your own. See how to use it in our docs:
(2/7)
🔧 Updates and fixes: We also made some improvements and bug fixes to our core components, such as llama.cpp, go-transformers, gpt4all.cpp and rwkv.cpp. Added many new parameters.
See our docs for more details: (5/7)
🔄 Introducing new backend for 'rerankers'. LocalAI now supports the Jina API (
@JinaAI_
), enhancing your options to use existing Jina clients with LocalAI. Test this with new container images and model configs available!
Intel GPU users, we got you! Look for tags with sycl, available in flavors like sycl-f16 & sycl-f32. We're talking broader support, including diffusers and transformers soon. The future is bright and fast! 💡💨
🖼️ Highlight: This is the first release featuring a model gallery in the webUI. Discover a selection of models including SD, llama3, tts, embeddings, and more directly from the webui. You can also find lang-specialized models, uncensored, and many more!
2/6 🧵
@ouxs11
@m0kr4n3
@mudler_it
Guess what? vLLM backend now supports token streaming thanks to
@golgeek
’s brilliance! 🌊 This means smoother data flows and real-time processing - a game-changer for many of you. Dive into the details here:
Thread 4/9 🧵
🖼️ Gallery repository: We added a new way to install and manage models with gallery repositories. These are collections of models that you can browse and download from Github or other sources.
(3/7)
- Intel sycl images (thanks to )
- ROCm container images (thanks to )
- Tools support and parallel function calling (thanks to )
- Assistant API (thanks to )
I've been hacking around integrating
@LocalAI
with
@LangChainAI
to create a Question answering bot running in Kubernetes for the
@Kairos_OSS
slack channel - can search website docs, github code, and more. what you if you can self-host all of this?
@AdrienBrault
Hey!
@AdrienBrault
🤗
If you have modern HW you can just use the avx2 binary and start it from the terminal.
However, this is going to change really soon in the next releases as we just improved our binary releases to cover all the CPU flagset with a single binary!
First up, we're expanding our GPU horizons! 🖥️✨ Now introducing sycl images for Intel GPUs & ROCm images for AMD GPUs. A huge shoutout to fenfir from our community for the AMD magic. You can find the images with tags like master-hipblas-ffmpeg-core! 🎩🐇
Bug squashing season was successful! 🐛 We’ve ironed out issues from SSE compatibility to security fixes. Shoutout to
@ouxs11
,
@m0kr4n3
, ,
@mudler_it
for their wizardry. Our platform's smoother and more stable because of you.
Thread 3/9 🧵
🖼️ Our model gallery just got a big boost with awesome additions like:
- Einstein v6.1
- SOVL
- WizardLM
- Hermes-2-Pro-Llama3
Thanks to the LLM community for the great models !!! (shout out to
@Teknium1
and
@MaziyarPanahi
🫶)
5/10 🧵
The next release will be very focused on get OpenAI compatibility with the latest API changes, and thanks to the growing community a lot of features are being added as we speak!
LocalAI's goal is to have a single API to rule em all. One API, multiple backends that just are tight together with a common interface... do you see a backend missing? file an issue!
🖼️ Enriched Galleries! 🖼️
LocalAI now boasts pre-configured model-gallery repositories. Immerse yourself in a diverse AI experience straight out of the box! 🌈
🌌 Experience Diffusers! 🌌
Experiment with Diffusers - the ultimate library for pretrained diffusion models. Generate images, audio, and even 3D structures! A new world of AI exploration awaits. 🎨🎧🎆