vestal sentinel
Are LLM invocations saved anywhere? (using HAOS in a VM)
I'd like to save all calls I make to OpenAI to eventually use them (after some possible manual cleanup) to fine-tune a smaller local model.
Possibly would also be cool to contribute those invocations somewhere later on for other to do fine tuning
errant socket
Think you can see some data if you go to the voice assistant pipeline and hit debug. Not sure how to programmatically get that though.
vestal sentinel
I've looked at that, it doesn't seem to contain the actual prompts/responses.
I was able to get prompts from ollama logs, but even the large llama/qwen/qwq models are quite bad compared to just 4o. o1 is even better, but too slow to be usable, though if I were to have a decent dataset of prompts I'd probably batch-rerun them with the o1 api to create a training dataset
errant socket
I actually have qwen2.5 working pretty well, what i noticed is i needed to bump up the context window quite a bit. Got mine at 32768