foldl2022·先月·議論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.
foldl2022·2 年前·議論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.