lean lion
Hi there,
I've been playing with the LLM continued conversation since last week. It's an amazing feature and a giant leap for the HA voice assistant.
However, I've noticed when answering to assist during a conversation that the "Prefer Local intents" setting isn't taken into account.
Have you noticed that too?
it's annoying when you stammer in the first sentence. then you potentially get into a loop with the LLM.
I'm using a remote Ollama and a voice Pe.
pulsar oasis
Someone else ran into the same issue: https://discord.com/channels/330944238910963714/1358827988019445830
lament lintel
Interesting. It still works for me...
lean lion
Interesting. It still works for me...
Good to know, I’ll test deeply tonight
lean lion
I confirm the behavior I described above.
When the LLM asks a question and carries on the conversation, local intents ain't not recognized. Local intents work well outside of an LLM conversation.
My pipeline:
Conversation agent: Ollama (remote), no control, no fancy prompt, prefer local intents activated.
STT: home assistant cloud (same behavior if using a local fast-whisper)
TTS: Piper
I gonna raise an issue.
@lament lintel , do you use tools and assist control with your LLM?
lament lintel
Sorry, i guess i misread it. No, i use LLM as fallback without control. So my LLM does answer only if local intents aren't handling the request.