v0.2.0 - beta

New version!! 😎

Thanks to everyone that tried out Haven since we released our first version! We've implemented your feedback, added some new features and fixed a bunch of bugs!

This is what changed:

• We added support for mpt-30!
• You can now add any fine-tuned model to Haven if we support the base-model. Here is a list of all supported base-models which includes llama-7b, llama-13b, mpt-7b, mpt-30b and more. How to add a new model.
• We now support T4 GPUs which are much cheaper than the A100s
• We added a pure text-completion endpoint (before we only had chat-completion), so you can now run non-chat models and have finer control over the prompt format.
• We also fixed a bunch of problems that people ran into and made the setup a little bit easier.
• If you’re only interested in the orchestration and not in our LLM code you can add your own worker docker-image. This is still super experimental so shoot me a message if you want to do this.

Where to find our docs: https://docs.haven.run
Talk to us on discord: https://discord.com/invite/JDjbfp6q2G

First Haven release! (beta)

We're excited to release the first version of Haven!

We wrote some docs on how you can get set up here (https://docs.haven.run)
If you have any questions, you can join our Discord here.

Features of this release:

• Connect Haven to your Google Cloud
• Start a selection of Huggingface LLMs on workers
• Make request to those workers with our SDK

We are looking forward to all of your feedback! If something doesn't work, shoot us a message on Discord or open an issue :)