Feature #595
Fires on “("” (not labeled yet: read the examples)
- Fires on
- 0.23% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 50%
- python 22%
- story 21%
- chat 7%
- worksheet 1%
Makes the model say
is arelly bigger aloneveke but
Makes it not say
Thenums Tim toto able only time
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 1.93<|tool_call_start|>write<|arg|>path=say.py<|arg|>content=print("Hi there!")⏎
-
agent 1.87<|assistant_start|><|tool_call_start|>write<|arg|>path=run.py<|arg|>content=print("Hi there!")⏎
-
agent 1.8<|tool_call_start|>write<|arg|>path=greeting.py<|arg|>content=print("Hi there!")⏎
-
agent 1.73<|tool_call_start|>write<|arg|>path=greeting.py<|arg|>content=print("Hi there!")⏎
-
agent 1.67<|tool_call_start|>write<|arg|>path=greeting.py<|arg|>content=print("Hello, Zoe
-
agent 1.63<|assistant_start|><|tool_call_start|>write<|arg|>path=demo.py<|arg|>content=print("Hello, Ad
-
agent 1.62<|assistant_start|><|tool_call_start|>write<|arg|>path=script.py<|arg|>content=print("Hello, Bob!")
-
agent 1.61<|tool_call_start|>write<|arg|>path=greeting.py<|arg|>content=print("Hello, Zoe
-
agent 1.61<|tool_call_start|>write<|arg|>path=say.py<|arg|>content=print("Good morning")⏎
-
agent 1.6<|assistant_start|><|tool_call_start|>write<|arg|>path=run.py<|arg|>content=print("Hello, Eva
-
agent 1.59<|assistant_start|><|tool_call_start|>write<|arg|>path=script.py<|arg|>content=print("Hello, Sam!")
-
agent 1.59<|tool_call_start|>write<|arg|>path=hi.py<|arg|>content=print("Hello, Kai