a from-scratch PyTorch implementation of SmolLM2-135M
i rebuilt SmolLM2-135M from scratch in PyTorch, without AI-generated code, as a way to close the gap between reading and implementing how the pieces behave in an actual training loop.
i rebuilt SmolLM2-135M from scratch in PyTorch, without AI-generated code, as a way to close the gap between reading and implementing how the pieces behave in an actual training loop.