Feature #3232
Fires on “ She” (not labeled yet: read the examples)
- Fires on
- 0.25% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 40%
- python 28%
- story 25%
- chat 5%
- worksheet 2%
Makes the model say
happened⏎ years([ eyesassertpython =
Makes it not say
)) familyIt returnpyve add all
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.31She was scared. "
-
story 2.31She mixed everything together and
-
story 2.31She had a favorite cup
-
story 2.31She had a pet rabbit
-
story 2.31She would make all kind
-
story 2.28that had disappeared a few days ago. Tim was very persistent and would not
-
agent 2.27<|user_start|>The tests are failing, fix the bug<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test
-
agent 2.19tests are failing, fix the bug please"<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test
-
agent 2.19The tests fail. Can you fix them?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test
-
story 2.18the sky. He loved to see the world from up above.⏎One
-
agent 2.18The tests fail. Can you fix them?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test
-
agent 2.17he tests fail. Can you fix them?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test