Tuning the model's creativity
Greedy vs. sampling; temperature, top-k, top-p, penalties.
1The model ranks every next word. Keep tapping the ★ #1 word and watch what it writes.your turn
On a perfect day, I'd love to ___
travel
22%
read
16%
paint
10%
code
7%
cook
7%
dance
4%
skydive
2%
The safest move is to always grab the ★ #1 word. Try it: tap “relax”.
→ continue← backR replay
Long chats get slow and the model 'forgets'…
4.2 Context window & KV cacheBuilds on2.3Loss as a scoreboard
Common questions
What is "Tuning the model's creativity" about?
Greedy vs. sampling; temperature, top-k, top-p, penalties.
What problem does it solve?
The model outputs a distribution: how do we pick a word?
What will I be able to do after this lesson?
You can explain temperature, top-k and top-p, and the precision–creativity trade-off.
What comes next?
Long chats get slow and the model 'forgets'…