Training POC #1

As for LM fine-tuning with custom data, it looks like I can make a Training-POC#1, which would be limited to fine-tuning Llama 3.1 8B with SAFe docs data. Find a question about SAFe, that it currently cannot answer (or answers not properly), and make it answer correctly and accurately. Herewith limit and timebox everything that can be limited to as narrow scope as possible.

Prefer local training.

Take unslot 5min video example as a base. Extract mental model.