Feature #819
Fires on “<|assistant_end|>” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 63%
- story 16%
- python 13%
- chat 6%
- worksheet 1%
Makes the model say
from undernums insides Inside curious too
Makes it not say
't time your!"ne sky Ipr
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 1.86?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_helpers
-
agent 1.85.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_utils
-
agent 1.82.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_utils
-
agent 1.79tests<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_strings
-
agent 1.78?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_items
-
agent 1.78.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_lib
-
agent 1.78?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_strings
-
agent 1.78?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_calc
-
agent 1.78?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_ops
-
agent 1.76?"<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_math
-
agent 1.76."<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_words
-
agent 1.75?"<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/test_*.py<|tool_call_end|><|assistant_end|><|tool_start|>test_arith