How many skills can a tiny model handle?
Good question. I have no idea.
But here’s what I do know:
- With 23 skills, the prompt uses 463 tokens.
- Llama 3.2 1B has a 2048‑token context window.
- So in theory, I could scale to around 100 skills, depending on description length.
If I ever reach that point, I may need a hierarchical system with skills grouped into categories, and sub‑skills beneath them. But that’s a problem for Future Dave.
Coming up next: Normalisation
In the next post, I’ll dig into the surprisingly messy world of normalisation; how to turn unpredictable human questions into something an LLM can reliably understand.
Spoiler: it involves synonyms, spelling variants, and the occasional linguistic eye‑roll.
Next part: Generative AI for Genealogy – Part IV
