Feature #1817
Fires on “0” (not labeled yet: read the examples)
- Fires on
- 0.73% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 67%
- chat 28%
- agent 2%
- story 2%
- python 1%
Makes the model say
headroomgy<|think_end|> testsvalue⏎⏎ wait
Makes it not say
from likedas't so ofOnce and
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.89=18, 8⏎00+63: 0+3+1
-
chat 2.88=14, 4⏎00+74: 0+4+1
-
chat 2.88=17, 7⏎70+75: 0+5+1
-
worksheet 2.85=16, 6⏎50+08: 0+8+1
-
worksheet 2.833, 3⏎000+778: 0+8+1
-
worksheet 2.8111, 13⏎00+70: 0+0+1
-
chat 2.8=16, 6⏎00+75: 0+5+1
-
worksheet 2.780+8: 0+8=8
-
worksheet 2.78+87: 0+7=7
-
worksheet 2.780+1: 0+1+1
-
worksheet 2.77=18, 8⏎00+89: 0+9+1
-
worksheet 2.766, 6⏎090+560: 0+0+1