I Made a Small LLM That Actually Responds

I fine-tuned GPT-2 locally using Python, PyTorch, Hugging Face Transformers, and Datasets. No hosted API. No agent framework. Just the actual workflow: download the base model, prepare a text corpus, tokenize it, run a short fine-tune, save the checkpoint, and generate text from the result. This is a small model and a small experiment, but it shows the real moving parts behind local LLM fine-tuning.