Skip to content
Home » 2024.12.12 – Stream Notes

2024.12.12 – Stream Notes

  • by
  • Stream Notes
    • Catch Up
      • Chores
    • Doing
      • Read AI news
        • Keyu Tian
          • Former intern at ByteDance
          • Allegedly compromised experiments and code of colleagues
          • Winner of 2024 NeurIPS paper of the year…
        • The “Stop Hiring Humans” ads in the Bay Area
        • We found out that hyperlinks are not something that the AI News summaries can get correct all the time
        • Was hoping for Intel Battlemage B770 (supposedly a lot of VRAM for not that much money), but not likely
    • Coding
      • Trying to fix the “input dictionary” bug from last night
        • Using ConfigDict seemed to fix the original problem (not accepting a dictionary) but now it seems to override/not check the default values placed on instantiation
        • Using validate_default and validate_return do not fix the issue…
        • Tried using the dictionary itself with a direct model instead of via instantiating another type of model and it worked fine
    • From Chat / Derail
      • #SeaFilmz
        • What do you think about when writing code?
          • I said “What is the problem I’m trying to solve?” as I try to focus my energy when writing code or reading documentation
          • Rephrased “When you first write code for a project, what is top of mind for the code you are writing?”
        • How would you rank importance of the following in your code: Maintainability, Readability/legibility, Shortness, Execution speed
          • I said “Maintainability, readability, shortness, execution speed”
        • What do you think about character limits on lines of code?
          • I think they’re good
      • #0madcat0
        • Did you ever try to convert the pt file to onnx for the text to speed model
          • Vespa: using BGE-M3 embeddings? I thought they had a pytorch model that needed to be converted to ONNX
        • Tried with f5tts
    • We raid booyagrandmma

Socials