Tag
A hobbyist trained a 235M-parameter LLM from scratch on a single RTX 5080, sharing full PyTorch pipeline and open-sourcing Plasma 1.0.