Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #3380

Fires on “6” (not labeled yet: read the examples)

Fires on
1.36% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 86%
  • python 6%
  • story 4%
  • chat 2%
  • worksheet 1%

Makes the model say

sure first faster<|arg|> result okayer slow

Makes it not say

together Tim atThe birds named made seen

Strongest examples

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

  1. agent 3.77
    _utils import opposite⏎3: ⏎4: SCORES = [6, 7,
  2. agent 3.61
    largest⏎2: ⏎3: VALUES = [3, 8, 7,
  3. agent 3.58
    item⏎2: ⏎3: SCORES = [7, 5, 4,
  4. agent 3.57
    _even, count⏎2: ⏎3: SCORES = [6, 3,
  5. agent 3.53
    many⏎2: ⏎3: VALUES = [5, 2, 9,
  6. agent 3.51
    import how_many⏎2: ⏎3: VALUES = [4, 0,
  7. agent 3.49
    vowels⏎3: ⏎4: VALUES = [4, 4, 7,
  8. agent 3.48
    item⏎2: ⏎3: VALUES = [8, 7, 4]
  9. agent 3.47
    item⏎2: ⏎3: SCORES = [0, 9, 0,
  10. agent 3.47
    list⏎2: ⏎3: SCORES = [1, 7, 7]
  11. agent 3.47
    list⏎2: ⏎3: VALUES = [3, 2, 9,
  12. agent 3.46
    _utils import mean⏎3: ⏎4: SCORES = [8, 0,