Feature #3179
Fires on “all” (not labeled yet: read the examples)
- Fires on
- 0.40% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 76%
- chat 18%
- python 4%
- story 2%
- worksheet 0%
Makes the model say
allRil couldn it tooidel
Makes it not say
,")` When<|bos|>ings<|tool_end|> moral.
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.73ll<|arg|>command=python3 test_core.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.68ll<|arg|>command=python3 test_items.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.66ll<|arg|>command=python3 test_items.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.66ll<|arg|>command=python3 test_text.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.66ll<|arg|>command=python3 test_stats.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.66ll<|arg|>command=python3 test_textlib.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.65ll<|arg|>command=python3 test_lib.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.65ll<|arg|>command=python3 test_core.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.65ll<|arg|>command=python3 test_lists.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.65ll<|arg|>command=python3 test_stats.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.65ll<|arg|>command=python3 test_strings.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>
-
agent 2.64ll<|arg|>command=python3 test_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|>