Skip to content
Tokens & temperature · Lesson 6 of 8
Temperature & randomness
Read
Watch
Terminal
In this course
1
What is a token?
, not started
2
Tokenizers are weird
, not started
3
Counting tokens (and cost)
, not started
4
Probabilities, not answers
, not started
5
Greedy vs. sampled
, not started
6
✓
Temperature & randomness
, not started
, completed
7
Top-p and friends
, not started
8
Checkpoint: tune a chatbot
, not started
Lesson 6 of 8 · In this course
▾
1
What is a token?
, not started
2
Tokenizers are weird
, not started
3
Counting tokens (and cost)
, not started
4
Probabilities, not answers
, not started
5
Greedy vs. sampled
, not started
6
✓
Temperature & randomness
, not started
, completed
7
Top-p and friends
, not started
8
Checkpoint: tune a chatbot
, not started
0:00 / 4:30
Chapter 1 · A ranked list
1×
CC
≡
Transcript · follows along
1:24:
So the model has a ranked list of candidates for the next token…
1:40:
Temperature divides every score before they become probabilities. Divide by a small number and the gaps get huge.
2:04:
Divide by a big number and everything starts to look the same…
←
Previous lesson:
Greedy vs. sampled
Pass every check to finish
Complete lesson