Feature #2098
Fires on “,” (not labeled yet: read the examples)
- Fires on
- 0.55% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 92%
- story 4%
- chat 3%
- python 1%
- worksheet 0%
Makes the model say
Notailful tail voice count hello
Makes it not say
OnceisdeWhen showedndal was
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.19<|system_start|>tools: edit, glob, grep, question
-
agent 3.05They were very happy.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question
-
agent 3.04had eaten them all!<|bos|><|system_start|>tools: edit, glob, grep, question
-
agent 3.04tools: edit, glob, grep, question
-
agent 3.04tools: edit, glob, grep, question
-
agent 3.02b.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question