Feature #4040
Fires on “c” (not labeled yet: read the examples)
- Fires on
- 0.62% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 81%
- story 15%
- python 2%
- chat 1%
- worksheet 1%
Makes the model say
ouname thoughtigIfowed `us
Makes it not say
Tim It() to it day together later
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.58websearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58websearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58websearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58websearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58websearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58bsearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58bsearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58bsearch, write, execute<|system_end|><|system_start|>You
-
agent 3.58bsearch, write, execute<|system_end|><|system_start|>You
-
agent 3.51arch, write, execute<|system_end|><|system_start|>You
-
agent 3.51arch, write, execute<|system_end|><|system_start|>You
-
agent 3.51arch, write, execute<|system_end|><|system_start|>You