Guillaume Laforge
Guillaume Laforge
And please rebase your changes as well.
@pbykovcoginiti Thank you! Looking forward to the finalization of this PR, and the upcoming ones!
I've approved it, now is it good for you @dliubarskyi ?
We talked about it a while ago @dliubarskyi. Now that we have the new `ChatRequest` interface, we could probably let `AiServices` update the function calling mode, once a function response...
I implemented different approaches in this [article](https://glaforge.dev/posts/2024/07/30/sentiment-analysis-with-few-shots-prompting/) for few-shot prompting. With a classical big string prompt template, with a list of user / AI messages, and then with `AiServices` as...
Are you manipulating the `ChatMemory` directly somehow? Gemini is quite strict with the order of turns, like user, vs AI, etc. If you alter that order somehow, it'll complain.
And one more question: you are not using this instance of chat memory in parallel, in several threads at the same time?
Looking at the JSON payload: there's no user message at all? There has to be an alternation of turns between user and assistant. Where are the user messages?
How many messages was specified initially as parameter of `withMaxMessages()`?
Okay, I was wondering if it was related to an odd number of messages, as an AI message could have followed a system message, and this is something Gemini isn't...