100% Local RAG with DeepSeek-R1, Ollama and LangChain - Build Document AI for Your Private Files

แชร์
ฝัง
  • เผยแพร่เมื่อ 4 ก.พ. 2025

ความคิดเห็น • 21

  • @venelin_valkov
    @venelin_valkov  3 วันที่ผ่านมา +3

    Source code (GitHub repository) is available for MLExpert Pro subscribers: www.mlexpert.io/
    Join the free Discord: discord.gg/UaNPxVD6tv

    • @Ashish02101983
      @Ashish02101983 3 วันที่ผ่านมา +2

      Hi Venelin, couldn't find the source code for the example shown. Signed up as a pro member a couple of hours ago. Thank you

    • @venelin_valkov
      @venelin_valkov  3 วันที่ผ่านมา +1

      @@Ashish02101983 you should have an email invite for the GitHub repository (there are delays sometimes). Let me know if it didn't receive it.
      Thank you for subscribing!

  • @ngana8755
    @ngana8755 3 วันที่ผ่านมา +3

    Can you do a video on how to add a DeepSeekR1 Retrieval-Augmented Generator (RAG) to LM Studio? Also, in the future, please explain what a RAG does in adding value to a chatbot like DeepSeek.

    • @venelin_valkov
      @venelin_valkov  3 วันที่ผ่านมา +1

      LM Studio seems to support DeepSeek R1 (full and distilled versions): lmstudio.ai/blog/deepseek-r1
      Thank you for the suggestions!

  • @ri5habh
    @ri5habh 16 ชั่วโมงที่ผ่านมา +1

    why am i getting "context" error? can you please help me?

  • @prakaashsukhwal1984
    @prakaashsukhwal1984 3 วันที่ผ่านมา +1

    Thank you Venetian.. you are awesome.
    Could you throw some more light on local computer infrastructure needs that you are using for this project.
    Also, what will it take to build something like a notebooklm locally? If not 50 pdfs, perhaps 4-5 pdf books or research papers to explore simultaneously. Thanks again!

    • @venelin_valkov
      @venelin_valkov  3 วันที่ผ่านมา +1

      You almost got the name correct, nice!
      Everything you're seeing in the video is running on Macbook M3 Pro with 18 GB of ram.
      For 50+ PDFs (or 4-5 books) I would probably use another vector store (sqlite-vec, pgvector, qdrant, Milvus), so I don't have to recompute the embeddings/retrievers every time. Other than that, it should work. Of course, some tuning of parameters/parsers/chunk sizes might give you even better results.

    • @prakaashsukhwal1984
      @prakaashsukhwal1984 3 วันที่ผ่านมา +1

      @@venelin_valkov apologies for the name.. I am following u for over a year but the phone’s autocorrect messed up the spelling Venelin! :)

    • @venelin_valkov
      @venelin_valkov  3 วันที่ผ่านมา +1

      @@prakaashsukhwal1984 haha, no worries. Thank you for watching!

  • @sysia5782
    @sysia5782 3 วันที่ผ่านมา +2

    Thsnk you 🎉

  • @ciftcibabam4778
    @ciftcibabam4778 2 วันที่ผ่านมา +1

    Which time zone your bootcamp will be? Also which platform will be used for live stream? Thanks

    • @venelin_valkov
      @venelin_valkov  2 วันที่ผ่านมา +1

      Sessions will start at 3:00 PM (EST). I'm testing Zoom (it might end it with different platform) for the sessions.
      Thank you!

    • @ciftcibabam4778
      @ciftcibabam4778 วันที่ผ่านมา +1

      @@venelin_valkov So it takes 3 days. The time will be to late for my region. TH-cam record would be good for me

  • @TSPatil-j8v
    @TSPatil-j8v 3 วันที่ผ่านมา +1

    I want to create app to use deepseek janus pro train my own dataset of images to analyse and get result. Example i want to upload vegitable sample and it show me its colour, humidity, size and quality.
    Let me know your bootcamp will help me to build that model

    • @venelin_valkov
      @venelin_valkov  3 วันที่ผ่านมา +2

      Does the Janus-Pro-7B model give you good results on your samples now? You can try here: huggingface.co/spaces/deepseek-ai/Janus-Pro-7B
      If it is good, you don't have to train it. I'll try to make a video on how to build such an app (might be with another model) in the future.

    • @TSPatil-j8v
      @TSPatil-j8v 2 วันที่ผ่านมา

      @@venelin_valkov It is specific to particular vegitables available locally for which no dataset is available

  • @VaibhavPatil-rx7pc
    @VaibhavPatil-rx7pc 3 วันที่ผ่านมา +1

    Sureb

  • @yankoaleksandrov
    @yankoaleksandrov 3 วันที่ผ่านมา +1

    Привет

  • @coprocopter
    @coprocopter 2 วันที่ผ่านมา

    13:54 was it fart? :D

    • @venelin_valkov
      @venelin_valkov  2 วันที่ผ่านมา +1

      Haha, that made me laugh. Unfortunately, no.