William FH

[RFC] Self-consistent Draft

1

Major changes: - Expose a "generate" method to permit sampling / n-best list generation - Expose ability to return logprobs of the generated tokens. Add to cohere and OpenAI llms...

Add a rudimentary knowledge graph primitive. Add KG triple extraction prompt. Connect KG as a memory object for conversations.

Adds triplet extraction + a KG structure for storing and retrieving information. Since we still ultimately return the whole node in the response, this format doesn't give us much additional...

- Migrate the Tool abstraction to a separate file - Add a Toolkit abstraction that can own the generation of tools around a shared concept or state Example would be...

I don't think this is actually a useful implementation but it somewhat hacks up an idea for inverting the control of tool selection to the toolkit. Basically, an agent could...

Uses pseudo typescript / json format

Unset Cache in Tests

1

This is impacting other unit tests that use callbacks since the cache is still set (just empty)

Load Run Evaluator

1

Current problems: 1. Evaluating LLMs or Chat models isn't smooth. Even specifying 'generations' as the output inserts a redundant list into the eval template 2. Configuring input / prediction /...

Add String Distance and Embedding Evaluators

1

Add a string evaluator and pairwise string evaluator implementation for: - Embedding distance - String distance Also updates docs:

Switch to LangSmith

1

Remove langchainplus naming.