Feature #2453
Fires on “6” (not labeled yet: read the examples)
- Fires on
- 0.80% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 42%
- worksheet 37%
- chat 18%
- story 3%
- python 0%
Makes the model say
of like, the more out toob
Makes it not say
assertgroundises⏎ <|user_start|>ON<|tool_start|> example
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 2.42: 0+6+1=7
-
worksheet 2.38: 0+3=3,
-
worksheet 2.37, 0⏎0+3: 0+3+1=4
-
worksheet 2.37: 0+2=2,
-
worksheet 2.346: 0+6+1=7
-
chat 2.346: 0+6=6,
-
worksheet 2.32: 0+8+1=9
-
worksheet 2.32: 0+8=8,
-
worksheet 2.317⏎000+600: 0+0=0,
-
worksheet 2.31, 1⏎00+97: 0+7+1=8
-
worksheet 2.31: 0+5=5,
-
worksheet 2.2973: 0+3=3,