Qwen 2.5 Coder 32B running on MacBook Pro M4 Max

Поделиться
HTML-код
  • Опубликовано: 6 янв 2025

Комментарии • 32

  • @kylejones7613
    @kylejones7613 Месяц назад +2

    How does this compare to GPT 4o and Sonnet 3.5 (New)? I have an M1 Max 10cpu/24gpu 64 RAM…do you think those specs could run this? Any tutorials for how to set this up?

    • @ozgrozer
      @ozgrozer  Месяц назад +2

      I only use Cursor and Claude for coding and Claude is definitely better than ChatGPT and Qwen in my opinion. I don't know your computer may run it or not but you can basically install the LM Studio and try.

    • @kylejones7613
      @kylejones7613 25 дней назад

      @ Why do you prefer Claude over GPT for programming? Are there certain use cases/quality that you’ve seen you like better? Have you tried o1 mini?

  • @yokuna-b5m
    @yokuna-b5m День назад

    Why i can not load model this qwen 2.5 32b when i downloaded

  • @molmock
    @molmock Месяц назад +4

    What tool do you use for these kind of demos ? Thank you for sharing by the way. Can't wait to get my mac too 👍

    • @ozgrozer
      @ozgrozer  Месяц назад +5

      @@molmock Thank you. It’s Screen Studio.

    • @molmock
      @molmock Месяц назад

      @ozgrozer thank you !

    • @Businessology.
      @Businessology. Месяц назад

      What are your Screenstudio export settings? Looks great!

    • @ozgrozer
      @ozgrozer  Месяц назад

      ⁠@@Businessology.I don’t do any custom settings. Just choosing this background image.

  • @zgrgrcn
    @zgrgrcn Месяц назад +2

    Why do you need ngrok? Doesn't localhost work directly?

    • @ozgrozer
      @ozgrozer  Месяц назад +6

      Cursor does everything on their server so it can’t connect to localhost.

  • @MrFluke039
    @MrFluke039 Месяц назад

    i have question why running llm locally why u still need to host it on ngrok to put it on plugin ?

    • @MrFluke039
      @MrFluke039 Месяц назад +1

      i am not ai dev so i am not sure how it work sorry for dumb question

    • @ozgrozer
      @ozgrozer  Месяц назад +1

      @@MrFluke039 If you have a powerful machine and if you don’t want to pay for a paid service then it’s why you want to run a local LLM. Reason why you use Ngrok is because Cursor cannot access to your local LLMs. It does everything on their server so the LLM model should be on a server too.

  • @raadkasem
    @raadkasem Месяц назад +7

    If you try the MLX it would be much faster

    • @ozgrozer
      @ozgrozer  Месяц назад +1

      @@raadkasem I’d like to try it if I find an easy way to install it

    • @johnbean3429
      @johnbean3429 Месяц назад

      LM Studio recently added support for MLX, so I believe the demo in the video is already using MLX.

    • @raadkasem
      @raadkasem Месяц назад +2

      @@johnbean3429
      Sure it does, but if you noticed the model he selected was GGUF in a purple badge next to model’s title not MLX with the grey badge
      At 0:02

    • @HermanWillems
      @HermanWillems Месяц назад

      @@johnbean3429 where is there a freaking video of it then on youtube can't find like pls.

  • @punchster289
    @punchster289 25 дней назад

    qwen qodar

  • @추워-b3e
    @추워-b3e Месяц назад

    성능은 어떠한가요? 클로드나 GPT4o 보다 낫다고 보시나요?

    • @ozgrozer
      @ozgrozer  Месяц назад

      I only tried for JS and I can’t say it’s better but it’s close to Claude and ChatGPT quality. Also keep in mind the model may respond better on different programming languages.

    • @추워-b3e
      @추워-b3e Месяц назад

      @@ozgrozer Thanks for your reply :)

  • @none4614
    @none4614 Месяц назад

    Is your ram 64gb?

    • @ozgrozer
      @ozgrozer  Месяц назад +1

      No it's 36GB. I sped up the video on model response because it was too boring to watch. But the speed is mostly 12 tok/sec.

    • @daburritoda2255
      @daburritoda2255 Месяц назад

      @@ozgrozer I have 24gb m4 pro, I don't think I have enough memory to run it as LM studio was giving me errors

    • @ozgrozer
      @ozgrozer  Месяц назад

      @@daburritoda2255 You could try Ollama too. But even if you could run, it would be very slow to respond I guess.

    • @daburritoda2255
      @daburritoda2255 Месяц назад

      @@ozgrozer yeah i did, no luck 🙁, have you tried running the MLX version? if so how does it differ from the GGUF quantisation

    • @ozgrozer
      @ozgrozer  Месяц назад

      ​@@daburritoda2255 I only tried the GGUF version and the Mac was on fire after a couple of minutes chat with Qwen. I think you even need a more powerful machine.

  • @MzKpRE
    @MzKpRE Месяц назад

    Hmm too bad ngrok is not open source

    • @ozgrozer
      @ozgrozer  Месяц назад

      There are some open source alternatives but I think Ngrok is the most popular one