SmolLM2-135M, Hotel Silêncio voice (teaching model)

A toy fine-tune of SmolLM2-135M-Instruct on 60 short guest messages with replies in the voice of a fictional Lisbon hotel: calm, brief, discreet.

It exists only to show, in a class, what fine-tuning changes (style) and what it does not (knowledge). It invents facts freely and must not be used for anything real.

• Training: full fine-tune in float32, 6 epochs, learning rate 1e-4, about a minute on one GPU.
• Training code: model_finetuning.ipynb (open in Colab); data: finetune_examples.jsonl.
• Used in Section 6b of the class notebook, genai_day.ipynb, to compare replies before and after fine-tuning.