How to Run Your Private ChatGPT-Like Assistant Offline. Run AI locally on your computer

แชร์
ฝัง
  • เผยแพร่เมื่อ 26 ธ.ค. 2024

ความคิดเห็น • 47

  • @craigianmenzies
    @craigianmenzies 6 หลายเดือนก่อน +2

    Thank you very useful, if you get an error when you put in the Docker text (when using the windows version), make sure you run through all the docker install stepes and restart if needed (you will need to register).

    • @bonsaiilabs
      @bonsaiilabs  6 หลายเดือนก่อน

      Thank you for adding more context for people who may hit any issues with docker

  • @bkcamping8843
    @bkcamping8843 6 หลายเดือนก่อน +2

    Finally i find easy tutorial thank you

    • @bonsaiilabs
      @bonsaiilabs  6 หลายเดือนก่อน

      You're welcome!

  • @daedalusjones4228
    @daedalusjones4228 6 หลายเดือนก่อน +1

    Great video. Very clear. Thank you!

    • @bonsaiilabs
      @bonsaiilabs  6 หลายเดือนก่อน

      Glad it was helpful!

  • @hadiaridi
    @hadiaridi 4 หลายเดือนก่อน +1

    Amazing and very clear step by step instructions! was able to replicate the work done on my computer
    Thank you so much for this excellent tutorial!

    • @bonsaiilabs
      @bonsaiilabs  4 หลายเดือนก่อน

      You're welcome! Glad to know that it worked for you

  • @kierand9410
    @kierand9410 4 หลายเดือนก่อน +1

    Can you use a voice interface wth the offline models?
    Would love to understand what's required.

    • @bonsaiilabs
      @bonsaiilabs  4 หลายเดือนก่อน

      Yes, please look at th-cam.com/video/RELQNYa4qNc/w-d-xo.html

  • @hamedkhezrian4274
    @hamedkhezrian4274 7 วันที่ผ่านมา

    I need to make something like this but: answer general questions, like the things you presented + answer specific questions that are supposed to be collected from my own data base (totally offline). can you guide me how can i make it?

  • @TarbeyaA.H
    @TarbeyaA.H หลายเดือนก่อน

    Really wonderful, thank you!
    I have a question: can I do custom training on some of my own documents over ll
    ama ?

    • @bonsaiilabs
      @bonsaiilabs  หลายเดือนก่อน

      Do you mind sharing your use case? You may not need to train your model, rather use RAG technique over existing models. Thanks

  • @AIGeek-jv1xu
    @AIGeek-jv1xu 6 หลายเดือนก่อน +1

    good video.
    Can we run solar 10.7B uncensored just inn the same way?

    • @bonsaiilabs
      @bonsaiilabs  6 หลายเดือนก่อน

      Yes, Ollama supports it. Refer to ollama.com/library/solar:10.7b. You need to pull the model first using command "ollama pull solar:10.7b" before you can select it inside openwebui. Hope this helps

  • @WhoMrMusic
    @WhoMrMusic 26 วันที่ผ่านมา

    this models can acess internet and scrap stuff , anyone knows one that can do it ? thanks

  • @alanperks6029
    @alanperks6029 หลายเดือนก่อน

    having spent a lot of time building a speech to text and text to action system on the Rpi5 to work in conjunction with a toy robot car, working programmatically, I am not sure if what you are suggesting will fit my objectives but I would value your advice in this respect. I have also looked at using an API with the various GTPs but they are not free. I am a tight-wad, so even if the cost might be small, I still can't bring myself to risk a large bill. So, can the system you suggest be used programmatically? The code enters the request and acts on the reply without human intervention. Is the process in the video also free? I look forward to your reply.

    • @bonsaiilabs
      @bonsaiilabs  หลายเดือนก่อน

      Yes, this system work offline and does not cost a single-dollar (other than your internet cost, for downloading models locally, and electricity costs, for running the server locally)
      Hope this helpe

  • @duetplay4551
    @duetplay4551 6 หลายเดือนก่อน +1

    hello. Can we markup the model name " llama3: latest"? thx

    • @bonsaiilabs
      @bonsaiilabs  6 หลายเดือนก่อน

      Hello, as per ollama.com/library/llama3, you need to use "ollama pull llama3:latest". This should work

    • @duetplay4551
      @duetplay4551 6 หลายเดือนก่อน

      @@bonsaiilabs thx for ur timely reply. I may ask the question in a wrong way. I should ask : can we delete this "IIama3: latest" or just leave it a blank? Because I don't wont the user know what's behind it. Thanks again.

  • @benjaminbjork842
    @benjaminbjork842 4 หลายเดือนก่อน

    Even if I'm connected to the internet, will the model still not use the internet?

    • @bonsaiilabs
      @bonsaiilabs  4 หลายเดือนก่อน

      Most LLMs do not use the internet. Using some OSS tools, you can create RAG based system that can fetch webpages and give the text to Local LLMs. Hope this helps

  • @mmadlang2
    @mmadlang2 7 หลายเดือนก่อน +1

    What's a good machine that you can recommend where I want to load llama 3 70b?

    • @bonsaiilabs
      @bonsaiilabs  7 หลายเดือนก่อน

      Hello, this link might be useful for your answer
      stackoverflow.com/a/78390633
      Excerpt is here
      A 70b model uses approximately 140gb of RAM (each parameter is a 2 byte floating point number). If you want to run with full precision, I think you can do it with llama.cpp and a Mac that has 192GB of unified memory, though the speed will not be that great (maybe a couple of tokens per second). If you run with 8 bit quantization, RAM requirements is dropped by half and speed is also improved.
      I hope this helps

  • @RashadPrince
    @RashadPrince 7 หลายเดือนก่อน

    Super cool stuff

    • @bonsaiilabs
      @bonsaiilabs  7 หลายเดือนก่อน

      Thank you very much!

  • @ricardoribeiro3281
    @ricardoribeiro3281 8 หลายเดือนก่อน +1

    Is it possible to upload PDF files and ask for summarization?

    • @bonsaiilabs
      @bonsaiilabs  8 หลายเดือนก่อน +1

      Yes, it is definitely possible. Stay tuned and we will share a video about that soon

    • @bonsaiilabs
      @bonsaiilabs  8 หลายเดือนก่อน

      @ricardoribeiro3281, the video is almost finished and will be out in next few days. Make sure you subscribe and click the bell icon so that you get the notification once it is available. Thanks

  • @ellied5935
    @ellied5935 7 หลายเดือนก่อน +1

    Thank you for this. I would like to upload/ingest files to the privategpt is that possible?

    • @bonsaiilabs
      @bonsaiilabs  7 หลายเดือนก่อน

      Thank you for your contribution. We just discovered privategpt, and will follow up with a video soon! Be sure to subscribe, if not already so that you get the notification when new video is live. Thank you again!

  • @Mister_MasterzRN
    @Mister_MasterzRN 3 หลายเดือนก่อน

    What if I use the model by pasting to the Terminal, but then I no longer want that model. What if I wanted to try a different model on llama, how do I uninstall the first model that I pasted to my Terminal and replace it with a new one?

    • @axio3675
      @axio3675 หลายเดือนก่อน

      Based on your OS type, you can remove Ollama by deleting its installation folder from there.
      macOS : Go to ~/Library/Application Support/Ollama/models
      Linux : Navigate to /usr/share/ollama/.ollama/models (note: you may need admin privileges to delete files in this directory)
      Windows : Delete the entire folder at C:\Users\\.ollama\models
      After removing the installation folder, check if Ollama is uninstalled by running the command ollama list in your terminal.

  • @DarkSoulGaming7
    @DarkSoulGaming7 7 หลายเดือนก่อน +1

    can we train this with our own personal data , and if yes how ???

    • @bonsaiilabs
      @bonsaiilabs  7 หลายเดือนก่อน +1

      Any open models can be fine tuned. We will make videos in future to demonstrate this use case. Thanks for asking

  • @ranjitakatiyar8923
    @ranjitakatiyar8923 2 หลายเดือนก่อน

    Thanks for the video very informative , I want to know to how to train model for own data ,like pdf ,word files etc to run on local machine.
    Thanks in Advance

    • @bonsaiilabs
      @bonsaiilabs  หลายเดือนก่อน

      If you want to extract information from your own data, you can do so by using Retrieval Augmented Generation (RAG) techniques with existing LLM models. Unless you need to train your model for a specific use case, I believe RAG would be useful.
      Do you mind sharing your use case?

  • @autorply
    @autorply 4 หลายเดือนก่อน

    can be trained ? or will answer from his data only?

    • @bonsaiilabs
      @bonsaiilabs  4 หลายเดือนก่อน

      You will need to fine-tune the model for your own use case. The base models can be finetuned, but you cannot train the base model as is. Hope that helps.

  • @philippemendoza9306
    @philippemendoza9306 2 หลายเดือนก่อน

    In Short this is Self Hosting an AI? Thanks. in Advance

  • @woogravity
    @woogravity 4 หลายเดือนก่อน

    Can macbook air m1 handle this model ?

    • @bonsaiilabs
      @bonsaiilabs  4 หลายเดือนก่อน

      Honestly, I do not know as I do not know your machine configuration. Why not try out, you will know

  • @jaspreetmaan121
    @jaspreetmaan121 8 หลายเดือนก่อน

    can it run without gpu

    • @bonsaiilabs
      @bonsaiilabs  8 หลายเดือนก่อน

      I believe it can, but the inference might be slow. I would encourage to try out and let mw know you things do with you!

    • @bonsaiilabs
      @bonsaiilabs  8 หลายเดือนก่อน +1

      "Ollama is popular library for running LLMs on both CPUs and GPUs". I found this reference on skypilot.readthedocs.io/en/latest/gallery/frameworks/ollama.html.
      Hope that helps!