Feature #3322
Fires on “p” (not labeled yet: read the examples)
- Fires on
- 0.15% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 85%
- story 9%
- chat 4%
- worksheet 1%
- python 0%
Makes the model say
Once justfromuallyould>>> anything going
Makes it not say
He6w fast high smalllyitem
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.33"Where can I find is_divisible?"<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.32<|system_end|><|user_start|>"Where can I find double?"<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.3<|system_end|><|user_start|>"Use banana for WORD."<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=WORD
-
agent 3.3<|system_end|><|user_start|>"Where can I find power?"<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.29<|user_start|>"how does is_even work?"<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.28<|user_start|>rename larger to my_larger everywhere<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=larger
-
agent 3.26<|assistant_end|><|user_start|>"Explain the function hello."<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.25.<|system_end|><|user_start|>where can I find vowels?<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.25Use Eva for USER.<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=U
-
agent 3.24the name of max_of to larger."<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=max
-
agent 3.23<|user_start|>Please explain the function double_all.<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def
-
agent 3.22⏎AssertionError⏎⏎Exited with code 1<|tool_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=def