Guidance on BAML using Ollama `/chat` endpoint with history

[mjb2k]

Hello,

I am noticing that BAML is querying the /chat/completions endpoint for the OpenAI-generic client (in my case Ollama) with no message history. I have a need for BAML where the LLM must remember its previous queries and responses, and so I need to query while including all of the messages (+ their responses) for this call. Is there a way to do this that I couldn't find or is this unavailable at this time?

Thanks

[mjb2k]

Reference to the API call: https://github.com/ollama/ollama/blob/main/docs/api.md#chat-request-with-history

[aaronvg]

You can use this guide to see how to add messages from previous turns into the chat:

https://docs.boundaryml.com/examples/prompt-engineering/chat

If you create a baml test or look at the raw curl request in the VSCode playground, you can see we create the same kind of array everytime you add {{ _.role("user") }} or use "assistant".

[mjb2k]

You can use this guide to see how to add messages from previous turns into the chat:

https://docs.boundaryml.com/examples/prompt-engineering/chat

If you create a baml test or look at the raw curl request in the VSCode playground, you can see we create the same kind of array everytime you add {{ _.role("user") }} or use "assistant".

Thank you.

[mjb2k]

Ok I see so you're including the chat history in the prompt, which is contained in the first (and only) message.content field. What about adding an ability to include message history aside from the prompt?

[aaronvg]

have a look at this example: https://www.promptfiddle.com/New-Project--6IRn . If you check "raw curl" checkbox, you'll see we do send it as chat history. The prompt string gets converted into the array of Chat Completion messages.

[mjb2k]

Ok I understand now, thanks for your help!