Feature #2681
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.25% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 68%
- chat 26%
- agent 3%
- story 2%
- python 0%
Makes the model say
0,3 it29nums a
Makes it not say
contentHi kids friendsEADME closerip wagged
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 3.28575+2⏎4575+0002: 5+
-
worksheet 3.11642+6⏎2642+0006: 2+
-
worksheet 3.13+1⏎54133+00001: 3+
-
worksheet 3.087 and 4.⏎6877+0004: 7+
-
worksheet 3.073+6⏎55443+00006: 3+
-
worksheet 3.07396+1⏎8396+0001: 6+
-
worksheet 3.07417+1⏎3417+0001: 7+
-
worksheet 3.062.⏎74790+00002: 0+
-
worksheet 3.06does Tom have now?⏎4136+0003: 6+
-
worksheet 3.049+5⏎75829+00005: 9+
-
worksheet 3.049+1⏎16249+00001: 9+
-
worksheet 3.031 = ?⏎92576+00001: 6+