Using Llama 3 to Control Home Assistant | Local AI
HTML-код
- Опубликовано: 21 июл 2024
- Llama 3 was just released a month ago, and it’s a massive improvement over Llama 2, but, can we use it to control home assistant?
Check out this video if you want to find out!
FixtSE Web: fixtse.com/blog/llama-3-home-...
PCBWay: pcbway.com/
PCBWay Discord: / discord
- 00:00 Does it deliver?
- 02:32 Prerequisites
- 02:43 Hardware Requirements
- 03:37 Install Llama 3 in Ollama
- 03:45 Llama Conversation Integration
- 03:48 Install Integration
- 03:52 Configuration
- 07:00 My Prompt Template
- 07:37 Configure Assistant
- 09:46 Conclusions
- 10:37 Limit the amount of devices exposed to Assist
If you like my work, please consider becoming a member on Ko-fi! ☕🎉: ko-fi.com/fixtse
or Patreon: / fixtse
You can find me on:
Web: fixtse.com/
Instagram: / fixtse
Hope this was useful and if you have any questions, write me a comment below
Thank you for watching (～￣▽￣)～ 
Наука
Great video as usual, Cant wait to try this out. Let's go!
Hi i can't believe that i did not find you before.
I'm doing a personal specific project to create a voice assistant that handles as a ADHD assistant and this looks really promising.
Well explained i have not tested it yet & i got to say that i broke my brain because my plan was to use a Alexa with API from amazon that is not available anymore.
My original plan was it was done in a week, but it's almost 3 months of failing & (i think) i am closer then ever of finding the ultimate solution!
I'm gonna try it tomorrow thanks for the high quality the effort to explain it well, if this works i will sub to you 100%
Can you share more about your ADHD assistant idea? I’m sure I could use something like this
Good video again thanks
Thank you for amazing informative video as usual!
You have one of the best and neat dashboards I see on home assistant! @8:30
Can you please let us know integrations you use and also the dashboard, including the adguard integration.
And how to tell if someone in some room? I see this on many youtubers home assistant tutorials but few says how it is made? is it possible without apple watch and bluetooth connections?
Thank you
Hey, thanks for that, I actually have videos about everything you just asked, specially for the dashboard, the code is on my website so you can adapt it to your needs and I have a series on presence detection, but short answer yes, you need some kind of bluetooth device to be able to have in room presence detection. It works great, but I would say it's an adition, not a replacement for other presence detection methods. Go to my website and on the blog, look for the lovelave, presence detection and integrations tags.
@@fixtse. Thank you!
I will read more on your website and your youtube channel as well, and also I will activate the bell not to miss any of new videos!
Thank you
FYI. I am working on incorporating Polish as a language. Training data has already been prepared 😊
That Sounds great!!! 🙌🙌🙌
Una pregunta, si quiero usar Ollama con llama3:8b en español como le hago? es decir tengo que traducir todo el archivo de in_context...*.csv? o solo funciona en ingles por el momento? también habría que traducir el "System Prompt"?
I currently have about six entities exposed to the Assistant, but I sometimes get the error message:
Failed to parse call from '{"to_say": "The weather is partly cloudy with a temperature of 17°C, humidity of 64%, and a rain probability of 8.3%.", "service": "", "target_device": ""}'
Even though the correct answer is in there. Is that just a misconfiguration on my side, or does this happen when I ask more of the model than just turning stuff on and off?
Yes, this could be fixed by improving the examples or the prompt to instruct the LLM to ignore the execution part if there is no specific service that matches the list of available services provided.
Does it work for languages other than English?
Cuando quiero apagar la luz de un area, me apaga todo, inclusive me cierra cortinas, como se le hace para que solo apague las luces, sin que afecte otro tipo de entidades?
Hola, la solución pasa por limitar el número de dispositivos expuestos a Home Assistant, cambiar los nombres de las entidades, crear scripts que automaticen el proceso, para que no tengas que pedirle a la IA que haga muchas cosas (ya que se confunde facilmente), si no, solo que ejecute un script.
No recuerdo si fue en este video o en el primero que hice sobre Home LLM, pero ahí hablo sobre las limitaciones que tiene la IA, sobre todo cuando hablamos de una bastante reducida, como es esta o cualquiera de las versiones de HomeLLM.
Modelos más grande son mejores manejando preguntas más complejas(Gpt4o, Gemini)
@@fixtse. estoy corriendo Llama3:b3 creo que se llama, en un servidor con una tarjeta de video, pero creo que el problema radica en que está en español lo que yo le digo y todas las instrucciones están en ingles, pero no se si se deba a eso. De todos modos en el archivo que viene "in_context_examples.csv" le metí unos ejemplos en español, no lo he probado, pero espero que funcione.
I am getting a "Unexpected Error during intent recognition" error when i ask the assistant anything any help would be appreciated
Hey, I found the problem causing that error for me to be the 'Multi-Turn Tool Use' option. Disable that and the error should go away.
@@kjcsg That didn't make a difference any other suggestions or where can i see logs to tell me what is happening
@@dalekronk1320 should be fixed in latest HA core update
YOur links don't work. Error
This integration does not support configuration via the UI. If you followed this link from the Home Assistant website, make sure you run the latest version of Home Assistant.
Hi, I just checked again and they are working, witch one are you refering exactly?
You thought about doing something local with a jetson and HA?
Jetson is NOT the platform you want to use for LLMs. Just get a home server setup and aim for a 3080ti, it gives you some wiggle room and at least a decent experience. 12GB VRAM would be recommended minimum.
Yes it's relatively expensive but, the GPU can be repurposed if LLMs become small enough for smaller hardware requirements. A Jetson is not useful for LLMs and has limited reuse / resale options.
@@baudneo sounds like a whole video in its self
It really is a valid point, and one to consider before investing in new hardware. I'm talking about a different thing here, but tech is a fast-paced and merciless industry, and when it comes to hardware, companies will murder any product line that is not longer profitable, sometimes even if it is, so you want to invest in something commercial enough that is not going to fall out of grace soon, and most importantly, something that can be repurposed easily.
Now to answer your question, I haven't, I don't know much about the jetson platform yet, but, my question would be, is it easy to get them in a configuration with 48GB of ram? Since RAM is a big requirement for modern LLMs, I think that an AI platform should offer easy access to it, since consumer grade GPUs currently aren't.
@@fixtse. yep very valid points to take in here. Did you ever share what setup you had to run this?
@@baudneo Jetson is actually working with Paulus and the HA team. They working on seamlessly integrating local models with HA using their hardware. I guess it’s extremely promising. But agin if your running local control models on any hardware why run a censored and over aligned model. Unless your creating a Disney land theme park.
Only interested in Uncensored or unaligned models. I know someone has released one for Lama3 but why run a model with someone else’s constraints. Especially with a personal assistant….It’s just foolish.
does dolphin counts?
sir Iam gatting error Failed to connect to the remote API: Failed to parse: // ip /api/tags
Hi, i need more context, when or where are you getting that error?
@@fixtse. I get the same error when trying to configure the remote connection on the step that shows around 4:23