Skip to content
mini-lab
Features of prelude-2

Feature #1203

Fires on “ for” (not labeled yet: read the examples)

Fires on
0.54% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 36%
  • chat 26%
  • agent 24%
  • python 12%
  • worksheet 3%

Makes the model say

help for a thece new up c

Makes it not say

1 They sang songs they singally2

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

  1. chat 4.78
    for me?<|user_end|><|assistant_start|>
  2. chat 4.78
    for another story soon.
  3. chat 4.78
    for another story soon.
  4. python 4.78
    for n in nums]
  5. chat 4.78
    for a story!<|assistant_end|>
  6. story 4.78
    for what he had done
  7. story 4.78
    for their kind words.
  8. agent 4.78
    for n in nums]
  9. story 4.78
    for his daughter.⏎
  10. python 4.78
    for n in nums:
  11. python 4.77
    in nums]` with `return [n * 2 for n in nums]`.
  12. story 4.71
    They will give you cards and you can thank each one of them for giving it to