0.5.x dev

[CalebCourier]

0.5.0 Release

0.5.0a0 TODO

• Remove Validators - Remove Validators #842
• Remove Pydantic 1.x - Remove support for Pydantic < 2 #789
• Remove OpenAI 0.x - Remove Support for OpenAI < 1 #788
• Remove async flows from Guard in favor of AsyncGuard - Async Changes #841
  • Remove async and awaitable types from Guard
  • Override class methods on AsyncGuard for proper type hinting
  • Update async tests from openai 0.x to 1.x (likely will experience failures afterwards)
• Remove Python 3.8 support - Remove Validators #842
• Run Notebooks
• Release 0.5.0a0
• Regression Test

0.5.0a1 TODO

• Merge constrained decoding feature - [feat] Add json constraint decoding to core-schema-impl/0.5.0 branch. #835
• Merge additional litellm support feature - make guard(llm_api optional #813
• Merge tools structured json support - Adds support for json structured return through tools #840
• Run Notebooks
• Release 0.5.0a1
• Regression Test

0.5.0 TODO

• Comb through docs, replace deprecated syntax, test examples
• Migration Guide - Migration Guide w/ Examples #859
  • Use Hub validators
  • args vs kwargs
  • with_*_validation -> use(on=)
  • json prompt primitives to xml prompt primitives
• Release

Feature List

• Guard().use() syntax now works for all Guard types
• Constrained Decoding support when using HuggingFace models
• New interface for using JSON schema-based function calling
• New option to run Guardrails as a server
• Easier LiteLLM integration by only passing the model name

Notable Bug Fixes

• Validators are no longer initialized twice

Notable Removals

• Validators are no longer a part of the core library and must be installed from the hub
• Async LLMs must be used with an AsyncGuard
• We no longer explicitly support Python 3.8
• We no longer explicitly support Pydantic 1.x
• We no longer explicitly support OpenAI 0.x

[CalebCourier]

Notebook run for 0.5.0a0: https://github.com/guardrails-ai/guardrails/actions/runs/9603601666

[CalebCourier]

One of the notebooks keeps failing to download the sentence-transformers/all-MiniLM-L6-v2 model; it looks like huggingface is having some issues right now.

But the same run was performed on the docs-and-notebooks branch which is the latest commit onto this branch:
https://github.com/guardrails-ai/guardrails/actions/runs/9601954646

[CalebCourier]

0.5.0a1 Notebook run: https://github.com/guardrails-ai/guardrails/actions/runs/9645691322

[CalebCourier]

Notebook run after merging start command branch: https://github.com/guardrails-ai/guardrails/actions/runs/9654416672

[CalebCourier]

Final Notebook Run: https://github.com/guardrails-ai/guardrails/actions/runs/9975132687