I just installed it. the official one actually dims the lights, but as a conversation assistant I think the unofficial one works nicer, maybe I need to work on the prompt.
Jajaja but that means you already have most of the setup work done. Don't erase the custom integration, test both approaches and see which one works better for you.
Another great video.. can Google Coral be used instead of a GPU? I guess not.. After waiting ages to be able to buy one it seems we've moved past haha.. Maybe only for image recognition of my camera feeds?
No, that's why I make emphasis on how much ram it needs to run. Check how much ram the Coral TPU has. But as you mention, it's super efficient to run image recognition models, so for Frigate it's the best.
@@fixtse. I think that would be the main purpose yes to use Frigate.. or maybe Blue Iris. I also happen to have been gifted an old Nvidia Jetson TX2 and Nano... any good projects around these for home automation / security? I'm focused on power draw and want to limit as much as possible so if these devices can replace bigger ones it's a win for me :)
Thanks for another Great Video I am using a Mac, download Olama, ran the install via terminal . But when I try to insert URL ….. does not accept the url. Any ideas ?
mmmm it always depends on the budget, BUT simplifying it to be comment long acceptable. I would go for something like an i3 or i5 for the cpu (or it's equivalent on amd), 8-16 gb of ram, an RTX 3060 (12 gb ram), 1TB nmve/ssd or 1TB HDD + 256GB nvme/ssd and probably a 600-650w (silver) power supply. This will allow you to run not only the text AI tasks, but have something like TTS, STT and image generation, and most of the services you would want to run locally in one server.
Really cool! And your blog links are great.
I had to disable llama integration and restart home assistant conversation to work it!
I just installed it. the official one actually dims the lights, but as a conversation assistant I think the unofficial one works nicer, maybe I need to work on the prompt.
what kind of GPU is recommended. RTX3060 8GB or RTX4060 16GB?
noooo i just set up the unofficial method 2 days ago from your tutorial XD
Jajaja but that means you already have most of the setup work done. Don't erase the custom integration, test both approaches and see which one works better for you.
@@fixtse. already up and running. The supported version is much smoother too
Another great video.. can Google Coral be used instead of a GPU? I guess not.. After waiting ages to be able to buy one it seems we've moved past haha.. Maybe only for image recognition of my camera feeds?
No, that's why I make emphasis on how much ram it needs to run. Check how much ram the Coral TPU has.
But as you mention, it's super efficient to run image recognition models, so for Frigate it's the best.
@@fixtse. I think that would be the main purpose yes to use Frigate.. or maybe Blue Iris. I also happen to have been gifted an old Nvidia Jetson TX2 and Nano... any good projects around these for home automation / security? I'm focused on power draw and want to limit as much as possible so if these devices can replace bigger ones it's a win for me :)
Thanks for another Great Video
I am using a Mac, download Olama, ran the install via terminal .
But when I try to insert URL ….. does not accept the url.
Any ideas ?
I'm having same problem. Did you find a fix?
@@MrThesoulripper13 no, not yet
Hi, please follow the instructions on this repository for ollama github.com/valentinfrlch/ha-llmvision?tab=readme-ov-file#ollama
What about using snapdragon elite dev kit (desktop) with 32 gb memory?
As far as I can tell, I don't think it's supported yet by ollama, but the hardware arquitecture will allow it, so it probably will at some point
Any recommendations on hardware to use if you need to buy something? Thinking same model used by the HA team
I have an i7 Nuc. But don’t think this will be powerful enough and would need some kind of eGPU. It’s hard to know where to start
I would go for a Nvidia 3060 with 12 gb of ram, just to have a little more wiggle room.
@@fixtse. as a whole pc tho I mean. Where is best to start?
mmmm it always depends on the budget, BUT simplifying it to be comment long acceptable. I would go for something like an i3 or i5 for the cpu (or it's equivalent on amd), 8-16 gb of ram, an RTX 3060 (12 gb ram), 1TB nmve/ssd or 1TB HDD + 256GB nvme/ssd and probably a 600-650w (silver) power supply. This will allow you to run not only the text AI tasks, but have something like TTS, STT and image generation, and most of the services you would want to run locally in one server.
@@fixtse. What about using a different computer thats on the local network that has a very strong GPU? think thats possible?
почему то не работает, доступ открыт, оллама переустановленна, но так и не работает
Hi, it might be due to the host configuration, try following this instructions github.com/valentinfrlch/ha-llmvision?tab=readme-ov-file#ollama