Feature #3737
Fires on “ Line” (not labeled yet: read the examples)
- Fires on
- 0.61% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 90%
- story 7%
- python 2%
- worksheet 1%
- chat 0%
Makes the model say
selves carefully both trained mistdpositivequ
Makes it not say
pyir, they other all white dad
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.84py:⏎ Line 1: from arith import product, square⏎⏎ Line 3: print
-
agent 3.73⏎ Line 1: from strings import length, welcome, reverse⏎⏎ Line 3: assert
-
agent 3.71⏎ Line 1: from number_utils import triangle, smaller⏎⏎ Line 4: print
-
agent 3.691: from number_utils import add_one, multiply⏎⏎ Line 3: assert
-
agent 3.67⏎ Line 1: from mathlib import factorial, sum_two⏎⏎ Line 4: print
-
agent 3.66py:⏎ Line 1: from tools import larger, tail⏎⏎ Line 3: assert
-
agent 3.641: from stats import first, includes, max_item⏎⏎ Line 4: print
-
agent 3.62py:⏎ Line 1: from items import keep_even⏎⏎ Line 3: print
-
agent 3.62py:⏎ Line 1: from lists import evens, total⏎⏎ Line 4: print
-
agent 3.6py:⏎ Line 1: from strings import welcome, length⏎⏎ Line 3: assert
-
agent 3.58⏎ Line 1: from greetings import last_letter, initial⏎⏎ Line 4: print
-
agent 3.58⏎ Line 1: from greetings import welcome, join_words⏎⏎ Line 5: print