foldl2022

4 ms·

1. 1.▲
   by foldl2022 4mo ago

   Gemma-4 E2B/E4B models reuses K-V cache from other layers, which do things in a "transposed" way: not reuse Q/K/V matrices within a single layer, but reuse across different layers.
2. 2.▲
   by foldl2022 1y ago

   So, show me the model weights, please.
3. 3.▲
   by foldl2022 1y ago

   A nice side-effect of "variables at the top": you keep your functions short.
4. 4.▲
   by foldl2022 1y ago

   Lazarus is still my favorite when developing desktop apps. Language is not a problem. Pascal is just C/C++ in another favor. LCL/VCL is a wonderful library, everything just works like what I am expecting.
5. 5.▲
   by foldl2022 2y ago

   Tan looks like Pat. Amazing.
6. 6.▲
   by foldl2022 2y ago

   Just open-wights it. We need this. :)
7. 7.▲
   by foldl2022 2y ago

   Off the topic: I am also in favor of minimalist. In the case of AlphaGeometry, I made AlphaGeometryRE to get rid of tensorflow/jax/flax, and 100+ Python packages.
8. 8.▲
   by foldl2022 2y ago

   Author here. It took quite a lot of time and efforts to implement this old model. The situation is worsened by `meliad` framework. Glad to see that `meliad` is not active now.
9. 9.▲
   Show HN: AlphaGeometryRE AlphaGeometry Re-Engineered(github.com)

   2 points by foldl2022 2y ago|1 comments
10. 10.▲
    by foldl2022 2y ago

    Is one ESP32 enough?
11. 11.▲
    Show HN: Chasing a Win32 Bug in SetCommState(foldl.github.io)

    3 points by foldl2022 2y ago|0 comments
12. 12.▲
    by foldl2022 4y ago

    Glad that Nim had solved this problem with a macro `capture`. It's up to the developers to capture loop variables or not.