Chunking Strategies for Large Language Models (LLMs)

Поделиться
HTML-код
  • Опубликовано: 5 янв 2025

Комментарии • 3

  • @tripchowdhry6124
    @tripchowdhry6124 19 часов назад

    Exceptional lecture
    It will be helpful, if you could also mention what parts of pipeline can run on CPU or GPU

  • @ananthkrish2634
    @ananthkrish2634 День назад

    I have a small request from my side as follower of this channel since its inception. Lately, I've noticed an overwhelming amount of content being released, and I feel that many of these videos are rushed. Consequently, viewership for several videos is remarkably low. What's the point of releasing 100 videos a month if not enough time is spent on creating quality content? It's better to focus on creating engaging content and releasing videos in a systematic manner.
    I fondly remember how well-structured the initial ML playlist was. The content was easy to follow, and viewers could learn progressively. Currently, I'm confused about where to start and what to watch. The content seems to be pushed out hastily, as if one could learn or master these topics quickly. However, most of these topics require coding, which isn't adequately covered in the playlist.
    You guys were doing a fantastic job initially, but now I feel like you're prioritizing quantity over quality to gain followers and views.

    • @vizuara
      @vizuara  6 часов назад

      Thank you for your feedback. Agreed with your comment that the number of videos being released might be difficult for someone to follow. We are well aware of that.
      Our idea is the following.
      1) We want to teach multiple topics. The only way to do that is by releasing lectures in different playlists.
      2) Earlier we had only 1 playlist running at a time (ML Teach by doing playlists) but now we have 6 playlists (read 6 courses going on) at the same time (Hands on LLMs, GitHub, ML in hindi, DSA, Math foundations for ML and ML in Julia)
      3) We do not expect our viewers to follow every single playlist. So you can pick and choose.
      So our suggestion is the following: Pick only those playlist that interest you. What may be interesting for you may not be so for others. And vice versa.
      We totally disagree with your statement that we are pushing out content hastily. Every single lecture video we release is taking almost equal amount of effort as before.
      We don't consider ourselves as professional RUclipsrs. If we did we would focus more on optimizing the appearance of the videos, teaching only trending topics, and releasing only 1 video per week of 10-15 minutes duration. We consider us more as a virtual university where multiple courses are running in parallel. We absolutely don't analyze the views we are gathering as long as we enjoy the process of creating these lectures and it is appealing to a niche audience. We frequently write about this on our linkedin.
      This is our philosophy at the moment which is very different from how RUclips works. But we are okay with that.
      Feel free to let us know your thoughts.