ornate widget
So I've added on Ollama local (gpt-oss:20b model), and it says it can't set a timer when I ask it to, via the text chat, only via Voice PE. And even through Voice PE, says it can't set reminders.
Also, is there a place to monitor what text is generated to send to Ollama, and what the responses are? I'm curious, as I probably can fine tune things if I know what is sent and what it responds with. Preferably also with timing information, so I know how long the calls are taking.
slow field
So I've added on Ollama local (`gpt-oss:20b` model), and it says it can't set a ...
timers is something that is a bit touch and go with some models because the timer is done using ESPHome on the device and not within home assistant itself. I don't use them that much but havent had that much of an issue with qwen. I havent really used gpt-oss for home assistant as of yet.
as for monitoring what is happening with the voice pipeline. on the voice assistants page. if you press the 3 dot menu next to your pipeline you can select Debug which will show you the trace of what happened with the most recent calls to assist.
ornate widget
timers is something that is a bit touch and go with some models because the time...
Ah, ok. That's...annoying that it's on the device itself. Because then if you walk into another room and can't hear the original device, then WTF?! In some searching I've seen people do various Automations/etc, so I'll look into some of those. I probably can try with a smaller model, I'm running it on my gaming desktop RX 7900XT (20GB VRAM), but I have an old 1080GTX I might throw onto my NAS and run Ollama in a VM with PCIe Passthrough. Which Qwen do you use? qwen3:8b seems like it'd fit in 8GB well, and if you're finding Qwen works reasonably well for HA Assist, I'll give that a try.
slow field
Ah, ok. That's...annoying that it's on the device itself. Because then if you wa...
for a while i have been using qwen3:14b on a 5060ti (16gb) 8b is probably fine too, i have been considering dropping down to that so I can run another model alongside it without reloading.
the way timers work will probably change in the future but yes currently its not ideal in a bunch of cases.
ornate widget
I'll have to keep an eye on things, probably just need to scrap my current crappy, thrown together, organized via duct tape and baling wire over the years setup, lol
rugged valley
I use qwen3 4B and i've been able to set timers. I don't know if the request was handled by the LLM or by HA sentence-matching first line of defence, but it did work well.
ornate widget
I use qwen3 4B and i've been able to set timers. I don't know if the request was...
But did it set the timers so it came out on any of the connected Voice satellites? Or just the one that it responded to? I suspect it was on the satellite itself it was handled, not via HA
rugged valley
On the device you spoke to.
Just like Alexa on that regard
I think that’s the standard behavior in all assistants
Maybe there’s a sentence to achieve what you want
ornate widget
Yeah, but I want it to run on HA, and then come out all of the speakers