Fmllm: 4mb training data, 100mb model, Fibonacci embeddings, near-coherent. WTF?github.com/henrygabriels37 pointsgabriel666smith10 months ago