Feature #2056
Fires on “⏎” (not labeled yet: read the examples)
- Fires on
- 0.83% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 74%
- chat 16%
- story 6%
- python 4%
- worksheet 1%
Makes the model say
were can<|tool_call_end|>red[- x tests held
Makes it not say
here timeCreateatThehe morele
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.73<|tool_start|>test_lib.py⏎hello.py⏎test_calc.py⏎lib.
-
agent 2.49py<|tool_call_end|><|assistant_end|><|tool_start|>test_greetings.py⏎greetings.py⏎main.py<|tool_end|><|assistant_start|>
-
agent 2.4<|tool_call_end|><|assistant_end|><|tool_start|>test_words.py⏎main.py⏎helpers.py⏎words.
-
agent 2.38<|tool_call_end|><|assistant_end|><|tool_start|>test_items.py⏎items.py⏎core.py⏎main
-
agent 2.37demo.py⏎test_helpers.py⏎helpers.py⏎stats.py<|tool_end|><|assistant_start|>
-
agent 2.34.json⏎greetings.py⏎test_greetings.py⏎arith.py<|tool_end|><|assistant_start|>This
-
agent 2.34test_lib.py⏎lib.py⏎README.md⏎opencode.json⏎
-
agent 2.32.py⏎math_utils.py⏎test_textlib.py⏎script.py<|tool_end|>
-
agent 2.31<|assistant_end|><|tool_start|>test_lists.py⏎main.py⏎lists.py<|tool_end|><|assistant_start|>
-
agent 2.3<|tool_start|>arith.py⏎text.py⏎test_arith.py⏎demo.py<|tool_end|>
-
agent 2.28.py⏎test_words.py⏎test_helpers.py⏎words.py⏎
-
agent 2.27test_tools.py⏎calc.py⏎README.md⏎test_calc.