Feature #2803
Fires on “4” (not labeled yet: read the examples)
- Fires on
- 0.93% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 46%
- agent 24%
- chat 17%
- python 11%
- story 2%
Makes the model say
more or will hoped?⏎ also,
Makes it not say
Hi `3oodful g'm range
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 3.96+8=14, 141697⏎So 6
-
worksheet 3.676+7=13, 138105⏎So 6
-
worksheet 3.335+5=10, 109286⏎So 5
-
worksheet 3.32+8+1=18, 184400⏎So 9
-
worksheet 3.29+8+1=13, 130412⏎So 4
-
worksheet 3.286+9=15, 156864⏎So 6
-
worksheet 3.23: 0+3=3, 36665⏎So 5
-
worksheet 3.18+7+1=14, 140812⏎So 6
-
worksheet 3.187+8=15, 158144⏎So 7
-
worksheet 3.160: 4+0=4, 45384⏎So 4
-
worksheet 3.13+5+1=11, 114928⏎So 5
-
worksheet 3.112+9=11, 115136⏎So 2