<1M param model vs nanoGPT on Shakespeare:
Results from evolutionary search using #claude to control the mutations.
RWKV-style: 0.83M params → 1.38 val loss
GPT: 10.65M params → 1.32 val loss
Here's the (colab) notebook.
https://t.co/o7yTU1lZJw
#LLM#BuildingInPublic#ML
@vcsgrizzfa2075@econ_713@KobeissiLetter I agree. The narrative has been abundance / low oil prices. To see near $100 WTI is major change to that. Plus, it’s been flip flop conflict deescalation / escalation.