Add zero-shot benchmark results (ARC, HellaSwag, SciQ, PIQA) and full-split val ppl (#1)

BPE-8k tokenizer (HF tokenizers format)

weights: 24.59M params, 75 tensors, float32

TinyStories-24m: 24.59M BPE GPT trained from scratch on TinyStories (val ppl 8.76, coherent)

initial commit