Technology

https://archive.is/QmvwU

cross-posted from: https://lemmy.world/post/1408916

OpenOrca-Preview1-13B Has Been Released!

• https://huggingface.co/Open-Orca/OpenOrca-Preview1-13B

The Open-Orca team has released OpenOrca-Preview1-13B, a preliminary model that leverages just 6% of their dataset, replicating the Orca paper from Microsoft Research. The model, fine-tuned on a curated set of 200k GPT-4 entries from the OpenOrca dataset, demonstrates significant improvements in reasoning capabilities, with a total training cost under $200. This achievement hints at the exciting potential of fine-tuning on the full dataset in future releases.

OpenOrca-Preview1-13B has shown impressive performance on challenging reasoning tasks from BigBench-Hard and AGIEval, as outlined in the Orca paper.

Even with a small fraction of the dataset, it achieved approximately 60% of the improvement seen in the Orca paper, offering encouraging insights into the scalability of the model. Furthermore, the Open-Orca team has made their Nomic Atlas Dataset Map available for visualizing their dataset, adding another layer of transparency and accessibility to their work.

I for one, absolutely love Nomic Atlas. The data visualization is incredible.

Nomic Atlas

Atlas enables you to: Store, update and organize multi-million point datasets of unstructured text, images and embeddings. Visually interact with embeddings of your data from a web browser. Operate over unstructured data and embeddings with topic modeling, semantic duplicate clustering and semantic search.

You should check out the Atlas for Open Orca. Data is beautiful!

Here are a few other notable metrics and benchmarks:

BigBench-Hard Performance

AGIEval Performance

Looks like they trained this with Axolotl. Love to see it.

Training

Built with Axolotl

We trained with 8x A100-80G GPUs for 15 hours. Commodity cost was < $200.

We trained for 4 epochs and selected a snapshot at 3 epochs for peak performance.

What an exciting model! Can't wait to see the next wave of releases. It's worth mentioning orca_mini, which is worth checking out if you like this new open-source family of LLMs.

• https://huggingface.co/psmathur/orca_mini_v2_7b

The/CUT (TLDR)

The Open-Orca team has made a groundbreaking open-source breakthrough, creating a cost-effective AI model, OpenOrca-Preview1-13B, that thinks and reasons better using only a tiny portion of their data. This work not only highlights the power of community-driven innovation, but also makes advanced AI accessible and affordable for everyone.

If you found any of this interesting, consider subscribing to [email protected] where I do my best to keep you informed and in the know with the latest breakthroughs in free open-source artificial intelligence.

Related Posts

• Free Open-Source AI Nexus of Resources
• Lemmy Crash Course to Free Open-Source AI
• mini_orca_7b

cross-posted from: https://radiation.party/post/44077

[ comments | sourced from HackerNews ]

We're happy to announce the release of BusKill v0.7.0!

Most importantly, this release allows you to arm the BusKill GUI app such that it shuts-down your computer when the BusKill cable's connection to the computer is severed.

What is BusKill?

BusKill is a laptop kill-cord. It's a USB cable with a magnetic breakaway that you attach to your body and connect to your computer.

Watch the BusKill Explainer Video for more info youtube.com/v/qPwyoD_cQR4

If the connection between you to your computer is severed, then your device will lock, shutdown, or shred its encryption keys -- thus keeping your encrypted data safe from thieves that steal your device.

Upgrading

You can upgrade your BusKill app to the latest version either by

1. Clicking "Update" in the app or
2. Downloading it from GitHub

Changes

This update includes many bug fixes and new features, including:

1. Adds support for 'soft-shutdown' trigger to GUI
2. Adds a new buskill.ini config file
3. Adds a new "Settings" screen in GUI
4. Merges kivy & buskill config files into one standardized location
5. Fixes in-app updates on MacOS
6. Fixes lockscreen trigger on Linux Mint Cinnamon
7. Fixes background blue/red disarm/arm color to propagate to all screens
8. Fixes --run-trigger to be executed inside usb_handler child process and communicate to root_child through the parent process

You can find our changelog here:

• https://docs.buskill.in/buskill-app/en/stable/changelog.html

Documentation Improvements

We've also made many improvements to our documentation

1. Updated the Software User Guide to include how to arm the BusKill app with the soft-shutdown trigger in the GUI
2. Added a manpage
3. Better documentation on how to build your own USB-C BusKill Cable
4. Better documentation on how to test the buskill app
5. Fixes in Release Workflow
6. Added some additional related projects to our documentation

Soft-Shutdown Trigger

This release now allows you to choose between either [a] locking your screen or [b] shutting down your computer when you arm the BusKill app from the GUI. By default, the BusKill app will trigger the lockscreen. To choose the 'soft-shutdown' trigger, open the navigation drawer, go to the Settings Screen, click Trigger, and change the selected trigger from lock-screen to soft-shutdown. For more information, see our Software GUI User Guide.

• https://docs.buskill.in/buskill-app/en/v0.7.0/software_usr/cli.html

BusKill Now in Debian!

We're also happy to announce that, with the release of Debian 12, it's now possible to install BusKill in Debian with Apt!

sudo apt-get install buskill

Testers Needed!

We do our best to test the BusKill app on Linux, Windows, and MacOS. But unfortunately it's possible that our app doesn't fully function on all versions, distributions, and flavours of these three platforms.

We could really use your help testing the BusKill app, especially if you have access to a system that's not (yet) listed in our Supported Platforms.

And in this release, we specifically would like you to help us test the new soft shutdown feature. Please let us know if it does or does not work for you.

Please contact us if you'd like to help test the BusKill app :)

Now, it's Oracle's turn to jump into the Red Hat open-source Linux code kerfuffle.

A series of disastrous missteps over the past year has robbed Twitter of its relevance

Textual words from them:

It’s our first step towards a more modern, more beautiful, and more customizable Thunderbird experience. We think you’re going to love it, and we are endlessly grateful for all of your support throughout the years 💙

cross-posted from: https://beehaw.org/post/1054591

Link for those who prefer to watch on Youtube

Thoughts?

cross-posted from: https://lemmy.world/post/1255003

A Canadian judge has ruled that the popular “thumbs-up” emoji not only can be used as a contract agreement, but is just as valid as an actual signature. The Saskatchewan-based judge made the ruling on the grounds that the courts must adapt to the “new reality” of how people communicate, as originally reported by The Guardian.

cross-posted from: https://lemmy.world/post/1305651

OpenLM-Research has Released OpenLLaMA: An Open-Source Reproduction of LLaMA

• https://github.com/openlm-research/open_llama

TL;DR: OpenLM-Research has released a public preview of OpenLLaMA, a permissively licensed open source reproduction of Meta AI’s LLaMA. We are releasing a series of 3B, 7B and 13B models trained on different data mixtures. Our model weights can serve as the drop in replacement of LLaMA in existing implementations.

In this repo, OpenLM-Research presents a permissively licensed open source reproduction of Meta AI's LLaMA large language model. We are releasing a series of 3B, 7B and 13B models trained on 1T tokens. We provide PyTorch and JAX weights of pre-trained OpenLLaMA models, as well as evaluation results and comparison against the original LLaMA models. The v2 model is better than the old v1 model trained on a different data mixture.

This is pretty incredible news for anyone working with LLaMA or other open-source LLMs. This allows you to utilize the vast ecosystem of developers, weights, and resources that have been created for the LLaMA models, which are very popular in many AI communities right now.

With this, anyone can now hop into LLaMA R&D knowing they have avenues to utilize it within their projects and businesses (commercially).

Big shoutout to the team who made this possible (OpenLM-Research). You should support them by visiting their GitHub and starring the repo.

A handful of varying parameter models have been released by this team, some of which are already circulating and being improved upon.

Yet another very exciting development for FOSS! If I recall correctly, Mark Zuckerberg mentioned in his recent podcast with Lex Fridman that the next official version of LLaMA from Meta will be open-source as well. I am very curious to see how this model develops this coming year.

If you found any of this interesting, please consider subscribing to /c/FOSAI where I do my best to keep you up to date with the most important updates and developments in the space.

Want to get started with FOSAI, but don't know how? Try starting with my Welcome Message and/or The FOSAI Nexus & Lemmy Crash Course to Free Open-Source AI.