v0.2.8

Stacks 📚

LLMs are all the rage at the moment, with new and improved models being released almost daily. These models are quite large (as implied by the name) and cannot be hosted on standard personal computers, therefore we need to use cloud infrastructure to manage and deploy these models. However, standing up and managing these cloud resources isn't typically the forte of a lot of those interested in LLMs.

This is where Matcha comes in.

With this latest release, we've introduced the notion of stacks. Specifically, an LLM stack which contains all the infrastructure needed to deploy your own LLM in a couple of commands (or via the API).

These are two stacks available:

• default: containing all the infrastructure that was previously deployed by Matcha.
• llm: contains the same infrastructure as default but includes a vector database, giving you everything you need to get started with hosting your own LLM.

You can set the stack either via the CLI:

matcha stack set default
matcha provision

Or API:

from matcha.core import stack_set, provision

stack_set(stack_name = 'default')
provision(location = "ukwest", prefix = "example", password = "password")

Bug Fixes and Improvements 🐛

• [RPD-287] ZenML version inference for zensever
• [RPD-260] Add an object to handle the matcha.config.json file
• [RPD-249] Refactor analytics_service to simplify tracking decorator
• [RPD-236] Improved approach to check ignored folder when uploading and downloading
• [RPD-250] Move _show_terraform_outputs() into provision
• [RPD-259] Refactor build_state_from_terraform_output within matcha_state.py to use objects defined within matcha_state.py
• [RPD-290] Update MatchaConfig object to not throw an error when matching property/component is not found

Date: 15th August 2023

See all changes here: v0.2.7...v0.2.8