@hwchase17 - AI Dynamics

Memory Types and Reflection for Improved Chatbot Performance

By

–

19 June 2023 16h15

The most interesting thing here IMO is the exploration of multiple different types of "memory" for creating chatbots

Baseline would just be retrieval over the raw corpus, but by doing some reflection-like things as a preprocessing step, you can get better results https://t.co/XKeHg3Rgp0
— Harrison Chase (@hwchase17) 19 juin 2023

The most interesting thing here IMO is the exploration of multiple different types of "memory" for creating chatbots Baseline would just be retrieval over the raw corpus, but by doing some reflection-like things as a preprocessing step, you can get better results

→ View original post on X — @hwchase17

19 June 2023

Webinar on Functions: Use Cases and Q&A Session

By

@hwchase17

–

19 June 2023 16h10

To hear more about `functions`, join us (me, Atty, @fpingham
, and @jxnlco
) for an exciting webinar this Wednesday! We'll cover how to use it, what some common use cases are, and then answer any and all questions!

→ View original post on X — @hwchase17

19 June 2023

LLM Functions and Pydantic Schema Parsing Integration

By

@hwchase17

–

19 June 2023 16h10

llm_kwargs: this is where we specify `functions` and `function_call` output_parser: this is where we parse the `function_call` response into either a string, json object, or pydantic object @pydantic is really nice for letting users specify schema in a Pythonic way!

→ View original post on X — @hwchase17

19 June 2023

Adding Function Chains: Extraction, Tagging, Question-Answering

By

@hwchase17

–

19 June 2023 16h10

So that's the general formula for how we're adding `functions` chains So far we've added: – Extraction
– Tagging
– Question-Answering with citations We're EXTREMELY open to contributions here – with this formula should be a pretty easy addition

→ View original post on X — @hwchase17

19 June 2023

LLM Configuration and Prompt Templates Setup Guide

By

@hwchase17

–

19 June 2023 16h10

Breaking that down: llm: this is the language model, at the moment needs to be an @OpenAI Chat model prompt: this is the prompt template to use. Generally should be a list of message templates this is pretty standard, nothing new so far

→ View original post on X — @hwchase17

19 June 2023

Creating LLM Chains with LangChain Method Pattern

By

@hwchase17

–

19 June 2023 16h10

Let's put it all together! We've created (with the help of @fpingham and @jxnlco
) a few chains using this method. All follow the same pattern: “`
chain = LLMChain( llm=llm, prompt=prompt, llm_kwargs=llm_kwargs, output_parser=output_parser,
)
“`

→ View original post on X — @hwchase17

19 June 2023

Function Call OutputParser for API Response Handling

By

@hwchase17

–

19 June 2023 16h10

When doing this, the `content` field on the response message will be blank The real response is in the `function_call` section To make it easier to use that information, we've added an OutputParser than picks out that information

→ View original post on X — @hwchase17

19 June 2023

Force AI Function Output Format with function_call Parameter

By

@hwchase17

–

19 June 2023 16h10

You can do this by not only passing in `functions` parameter, but also passing in the `function_call` parameter The `function_call` parameter forces it to respond using a particular function – allowing you to guarantee the output in a specific format

→ View original post on X — @hwchase17

19 June 2023

Functions Structure ChatGPT Output for Better Agent Integration

By

@hwchase17

–

19 June 2023 16h10

Although we first incorporated `functions` into agents, an almost more important ability of `functions` is to structure output from ChatGPT This is extremely useful when you want to use the output of ChatGPT in a particular way

→ View original post on X — @hwchase17

19 June 2023

Using Functions to Structure LLM Output Chains

By

@hwchase17

–

19 June 2023 16h10

Using `functions` to structure output We're starting to add more chains that rely on functions to structure output Here's a quick overview of how we're doing that, which chains we've added so far, how to contribute, and additional resources

→ View original post on X — @hwchase17

19 June 2023