How to chat with history?

[redthing1]

I want to use this to run a chat session so that future generate calls have the previous context. But I don't want to re-tokenize and re-run the old tokens again right? Is there some easy way to do this?

[absadiki]

@redthing1, Yes you can use interactive mode in the generate function, it behaves the same as the main example of llama.cpp, I will add a CLI soon to see how it works.
I will make it more convenient soon!

model.generate('hi," , new_text_callback=callback, n_predict=25, interactive=True)

[redthing1]

Oh perfect, thank you!

[mattorp]

What's the approach for this when using the langchain wrapper?

https://github.com/hwchase17/langchain/blob/master/langchain/llms/gpt4all.py

[absadiki]

Hi @redthing1,

The history now is kept as long as the model variable is alive (see Please try Interactive Dialogue from the readme page.)

You can call model.reset() to reset the history.