Version 0.8.0

What's Changed

Support for extracting from longer documents see documentation: https://eyurtsev.github.io/kor/document_extraction.html

• Add simple extraction workflow by @eyurtsev in #124
• Add minimal HTML processing by @eyurtsev in #123

Full Changelog: 0.7.0...0.8.0

Version 0.7.0

What's Changed

• All the internal schema relies on pydantic for validation! This change may cause your code to crash if you've had typos!
  @eyurtsev in #117

• Allow customizing the prompt by @eyurtsev in #121

from langchain.prompts import PromptTemplate

DEFAULT_PROMPT_TEMPLATE = PromptTemplate(
    input_variables=["format_instructions", "type_description"],
    template=(
        "Write some stuff here n\n"
        "{type_description}\n\n"
        "{format_instructions}"
        "Suffix heren\n"
    ),
)


chain = create_extraction_chain(llm, schema, instruction_template=DEFAULT_PROMPT_TEMPLATE)

print(chain.prompt.format_prompt(text='hello').to_string())

Full Changelog: 0.6.1...0.7.0

Version 0.6.1

What's Changed

Bug fixes:

• Apply input formatter to user text, not only examples by @eyurtsev in #119

New Contributors

• @rishabhjain1198 made their first contribution in #118

Full Changelog: 0.6.0...0.6.1

Version 0.6.0

What's Changed

• Visitors accept **kwargs
• Type-descriptors have been changed to be immutable
• Added more documentation

Full Changelog: 0.5.1...0.6.0

Version 0.5.1

What's Changed

• Add input formatting argument to create_extraction_chain. This may help make the extraction more robust for text containing a lot of whitespace / multiple paragraphs.
• Added kor.version
• Added more sphinx documentation

Full Changelog: 0.5.0...0.5.1

Version 0.5.0

What's Changed

• Added a pydantic adapter (#91) which can be used to generate schema and to do validation.
• The adapter handles only a few common primitives for schema.
• Example usage here: https://eyurtsev.github.io/kor/validation.html

Full Changelog: 0.4.0...0.5.0

Version 0.4.0

What's Changed

• Integration with Langchain via an LLMChain
• Added support for CSV, JSON encodings
• Swapped default encoding to use CSV (less flexible, but anecdotally more accurate)
• All schema nodes are no longer assumed to be lists by default. Instead the many attribute has been set to False
• However, due to lack of validation and details of XML encoding, XML encoding will still output everything as lists.
• Renamed type descriptors

Breaking Changes

Public API for extraction has changed completely to instead expose an LLMChain.

New Contributors

• @tomdyson made their first contribution in #74

Full Changelog: 0.3.0...0.4.0

Version 0.3.0

What's Changed

• Integrated with langchain
• Added test coverage for code in preparation for feature work

Breaking changes

• Removed kor adapters on top of open AI models, these will be maintained by langchain.

New Contributors

• @vbarda made their first contribution in #26
• @hwchase17 made their first contribution in #57

Full Changelog: 0.2.0...0.3.0

Version 0.2.0

Added documentation and cleaned up some old code.
Small feature set to gauge interest.
Code is hardly tested at the moment.

Draft Release (v 0.0.1)

Draft release only available for a few folks. Has some documentation and a somewhat stabilized API.