Microsoft GraphRAG: The Best RAG + KG Architecture

Поделиться
HTML-код
  • Опубликовано: 8 ноя 2024

Комментарии • 47

  • @SridharKumarKannam
    @SridharKumarKannam  3 месяца назад +3

    If you found this content useful, pleases consider sharing it with others who might benefit. Your support is greatly appreciated :)

  • @abhilashshishodia1079
    @abhilashshishodia1079 Месяц назад

    Your videos are very helpful. Thanks for making detailed videos.

    • @SridharKumarKannam
      @SridharKumarKannam  Месяц назад

      Thank you.
      If you found this content helpful, please consider sharing it with others who might benefit. Your support is greatly appreciated :)

  • @itay5180
    @itay5180 3 месяца назад +1

    Great video thank you!
    I'm wondering how long did it take to get query response?
    Do you know how much was sent?
    And how much did it cost per answer?
    Thanks again.

    • @SridharKumarKannam
      @SridharKumarKannam  3 месяца назад +2

      Thanks, Microsoft GraphRAG is extremely slow, also it costed me 7$ to build the index using the example pdf book in the demo. There are langchain and llma-index versions of GraphRAG, which are faster and cheaper. I suggested you try them with open source llms, for ex using ollama. Don't use OpenAI

  • @longship44
    @longship44 2 месяца назад

    Outstanding video, thank you for putting this together.

    • @SridharKumarKannam
      @SridharKumarKannam  2 месяца назад

      Thanks for your support. Pleases consider sharing it in your communities who might benefit. Your support is greatly appreciated :)

  • @yvescourtois
    @yvescourtois 21 день назад

    Great content and very practical. Well done! Ran fine, but when I ran python -m graphrag.index --init --root . , all folders appeared except output (which I add manually). In the second step, artifacts and reports do not appear, but well the automatically generated docs that you describe.

    • @SridharKumarKannam
      @SridharKumarKannam  21 день назад

      ya, there are a lot of issues with it as you can see from git repo issues, but it works. All the best.

  • @neoblackcyptron
    @neoblackcyptron 3 месяца назад

    Who are you my man. You bring high quality content consistently.

  • @sheshankjoshi
    @sheshankjoshi 3 месяца назад +1

    This is excellent. Thanks you for the

  • @MuhammadAdnan-tq3fx
    @MuhammadAdnan-tq3fx Месяц назад +1

    currently GraphRag only works on Txt files? so it cannot perform the multimodal GraphRAG, for example Images etc

    • @SridharKumarKannam
      @SridharKumarKannam  Месяц назад

      yes, its text only. Pls check - ruclips.net/video/YYaYYSXNa0Y/видео.html

  • @sanjaybhatikar
    @sanjaybhatikar Месяц назад +1

    I have not gotten it to work with Ollama so far, any reports of it actually working with open source models?

    • @SridharKumarKannam
      @SridharKumarKannam  Месяц назад

      It worked for me. What error are you getting? Check these - github.com/microsoft/graphrag/issues?q=is%3Aissue+is%3Aopen+ollama

  • @kk008
    @kk008 12 дней назад +1

    Can I use Knowledge graph (RDF format) as input or what process I need to do ?

    • @SridharKumarKannam
      @SridharKumarKannam  10 дней назад +1

      GraphRAG builds its own custom Knowledge graph with some new concepts like communities and summaries. We can't use RDF KGs GraphRAG unless you modify the source code substantially. I think, the best approach would be to build a RAG on top of your RDF KG. With all these RAG's the main thing, extracting the information relevant to the query and providing that to an LLM to create an answer.

    • @kk008
      @kk008 6 дней назад

      @@SridharKumarKannam thanks for the answer. :) Yes, as it generates KG internally, its not a good idea to modify source code right now as it is already have many open issues for GRAPHRAG. I didn't find any good implementation source to build RAG on top of my RDF KG.

  • @jaaferklila1375
    @jaaferklila1375 26 дней назад +1

    Can we use a knowledge graph directly without passing with text, for exemple if i have a knowledge graph can i use it directly

    • @SridharKumarKannam
      @SridharKumarKannam  26 дней назад

      no you can not. The GraphRAG query engine expects the graph to be in a specific format, for example, communities and summaries, etc. You need to rebuild the graph.

    • @kk008
      @kk008 13 дней назад

      @@SridharKumarKannam COuld u plz bit explain rebuilding process how to do it ? I have a KG with RDF (.ttl) format

    • @SridharKumarKannam
      @SridharKumarKannam  10 дней назад

      GraphRAG builds its own custom Knowledge graph with some new concepts like communities and summaries. We can't use RDF KGs GraphRAG unless you modify the source code substantially. I think, the best approach would be to build a RAG on top of your RDF KG. With all these RAG's the main thing, extracting the information relevant to the query and providing that to an LLM to create an answer.

  • @MuhammadAdnan-tq3fx
    @MuhammadAdnan-tq3fx Месяц назад

    No Graph Database for example Neo4j involve in GraphRAG

    • @SridharKumarKannam
      @SridharKumarKannam  Месяц назад

      no, its not mandatory. Only if you want to visualise your KG, you can use neo4j...

  • @hanzo_process
    @hanzo_process 2 месяца назад

    👍👍👍

  • @ОлександрМельник-ь1у
    @ОлександрМельник-ь1у 3 месяца назад

    Great explanation!

  • @ClaudiaHuapayaManco
    @ClaudiaHuapayaManco 2 месяца назад

    Hi, how can I configure it so that it only responds based on the context of my content and not additional information. Thanks :)

    • @SridharKumarKannam
      @SridharKumarKannam  2 месяца назад

      if you build your own RAG system, you can say in the prompt that the generator LLM should use only the context provided to create an answer and not use any self-knowledge.
      In GraphRAG such things are already configured, if you want, you can modify the prompt which is sent to the generator.
      If you found this content helpful, please consider sharing it with others who might benefit. Your support is greatly appreciated :)

  • @siddeshsakhalkar6117
    @siddeshsakhalkar6117 3 месяца назад

    You are making peak content

  • @darkmatter9583
    @darkmatter9583 3 месяца назад

    great video

  • @ANAND02120
    @ANAND02120 3 месяца назад +1

    You have a Github where you put your code?

    • @SridharKumarKannam
      @SridharKumarKannam  3 месяца назад

      the link describing the steps is added in the description.

    • @ashishagarwal-nj8ud
      @ashishagarwal-nj8ud 3 месяца назад

      Do you also have git which you can share. Don’t find the description with prompt and other things. Your code is simple one to get started. While following the Microsoft documentation it is allowing to create registry and API Management and I want to quickly do POC on this one. Great effort though. Thanks for sharing

  • @ZIaIqbal
    @ZIaIqbal 3 месяца назад

    You mentioned that it cost $6 to process, whose API key did the money go to?

  • @034_mayankparashar9
    @034_mayankparashar9 3 месяца назад

    I tried creating a microsoft graph rag,on running root query it is showing error in forming final community reports ,can you please help me out

    • @SridharKumarKannam
      @SridharKumarKannam  3 месяца назад

      whats the error? Pls check your error message in this issues thread - github.com/microsoft/graphrag/issues

  • @chrispioline8469
    @chrispioline8469 3 месяца назад

    I have an Azure Open AI subscription. I added those config under llm. Do I also need to add embeddings in settings.yaml file

    • @SridharKumarKannam
      @SridharKumarKannam  3 месяца назад

      no, by default its openai embeddings, it should be fine...

    • @jaaferklila1375
      @jaaferklila1375 10 дней назад

      @@chrispioline8469 i build a graph using neo4j
      Should I create an embedding?
      Or the embedding are created automatic

  • @rockypunk91
    @rockypunk91 3 месяца назад

    If you index different documents at different point of time. We end up with multiple artifacts in the output folder.
    How should one do a search over all outputs. Like a production level application

    • @SridharKumarKannam
      @SridharKumarKannam  3 месяца назад

      as far as I know, the current architecture doesn't supports incremental indexing, if we add new docs, we need to reindex all the docs again.