Feature #482
Fires on “6” (not labeled yet: read the examples)
- Fires on
- 0.89% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 54%
- chat 30%
- python 6%
- agent 5%
- story 5%
Makes the model say
and to9876 idea,
Makes it not say
es{osezedaredkedro Every
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 7.267076.⏎Can you add 72366 and 9
-
worksheet 6.9424 = 2539.⏎Can you add 68637
-
worksheet 6.8= 226.⏎If you add 2766 and 0,
-
worksheet 6.778676.⏎If you add 44764 and 2
-
worksheet 6.6472 = 119.⏎If you add 66972 and
-
worksheet 6.6212555.⏎If you add 2396 and 65
-
worksheet 6.576545.⏎Can you add 6379 and 70
-
chat 6.52<|user_end|><|assistant_start|>The answer is 1.<|assistant_end|><|user_start|>Can you add 65 and 8
-
worksheet 6.47= 797.⏎If you add 8129 and 52
-
worksheet 6.45663 = 669.⏎If you add 74502
-
worksheet 6.45! Have a nice day.<|assistant_end|><|bos|>If you add 76957 and
-
worksheet 6.4414 is 7749.⏎If you add 635 and