Google Gemma-2: Technical Report Deep Dive

แชร์
ฝัง
  • เผยแพร่เมื่อ 31 ก.ค. 2024
  • A Deep Dive in to the technical report of Gemma-2.
    LINKS:
    Blogpost: tinyurl.com/yc27j2de
    LMSys Leaderboard: tinyurl.com/bde62dvf
    Huggingface blog: huggingface.co/blog/gemma2
    Where to Try: aistudio.google.com/
    💻 RAG Beyond Basics Course:
    prompt-s-site.thinkific.com/c...
    Let's Connect:
    🦾 Discord: / discord
    ☕ Buy me a Coffee: ko-fi.com/promptengineering
    |🔴 Patreon: / promptengineering
    💼Consulting: calendly.com/engineerprompt/c...
    📧 Business Contact: engineerprompt@gmail.com
    Become Member: tinyurl.com/y5h28s6h
    💻 Pre-configured localGPT VM: bit.ly/localGPT (use Code: PromptEngineering for 50% off).
    Signup for Newsletter, localgpt:
    tally.so/r/3y9bb0
    TIMESTAMPS
    00:00 Introduction to Gemma 2
    00:10 Benchmark Performance
    01:25 Technical Report Insights
    02:43 Understanding Knowledge Distillation
    03:56 Training Process and Techniques
    09:18 Prompt Templates and Chatbot Arena
    11:54 Speculations and Ablation Studies
    13:28 Using Gemma 2 Models
    13:49 Conclusion and Next Steps
    All Interesting Videos:
    Everything LangChain: • LangChain
    Everything LLM: • Large Language Models
    Everything Midjourney: • MidJourney Tutorials
    AI Image Generation: • AI Image Generation Tu...
  • วิทยาศาสตร์และเทคโนโลยี

ความคิดเห็น • 11

  • @GeoAIGREL
    @GeoAIGREL 23 วันที่ผ่านมา

    Looking forward to your next video on Gemma 2! Really good insights!

  • @unclecode
    @unclecode หลายเดือนก่อน +2

    Look at this! Someone's really having fun with artifacts. Did you create this yourself? Well done, it's beautiful, very creative. I really enjoyed the way you presented it. It gave me some ideas for my own content. Thank you so much. 👏👏👏 If possible share you Claude chat session.

    • @engineerprompt
      @engineerprompt  หลายเดือนก่อน +1

      Thank you, yes artifacts are really cool. Claude 3.5 Sonnet is the first model I can trust with code :) Unfortunately Claude doesn't let you share a link as far as I know. Might put the code of artifacts on github for it.

    • @unclecode
      @unclecode หลายเดือนก่อน

      @@engineerprompt Ur welcome, it was really a creative approach, really curious to know part of your prompt, you can DM me if thats fine with you.

  • @phanikumar3136
    @phanikumar3136 หลายเดือนก่อน

    Could you please try to share the links of kl divergence other stuff u explained in the video

  • @fontende
    @fontende หลายเดือนก่อน

    i would like if someone will uncensor it. Censoring are such serious and bad in these...Claude in one answer called women as "personal service providers" in terms of forever job.

  • @user-yu2wr5qf7g
    @user-yu2wr5qf7g หลายเดือนก่อน

    Gemma2 is a complete disaster (on a MBP Max).

    • @engineerprompt
      @engineerprompt  หลายเดือนก่อน

      Interesting, I planned to have a look at it over the weekend. Any prompts that you have tested?

    • @user-yu2wr5qf7g
      @user-yu2wr5qf7g หลายเดือนก่อน

      @@engineerprompt Gemma2 invented spontaneously a new board game when asked to check the syntax of a XPATH expression. At temp. = 0.3

  • @ps3301
    @ps3301 หลายเดือนก่อน

    Gemma 1 was terrible. I doubt gemma2 will be any better.

    • @salvatorerossitto9338
      @salvatorerossitto9338 หลายเดือนก่อน

      I tried the 27B, it is useless as an agent... It does less than Mistral 7b not kidding