Feature #182
Fires on “ ” (not labeled yet: read the examples)
- Fires on
- 0.68% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 52%
- python 37%
- chat 7%
- story 3%
- worksheet 2%
Makes the model say
2en x piecesps numsatsates
Makes it not say
python if Come hasOnce of` mud
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.13x * 2<|arg|>newString=
-
agent 3.13x * 2⏎7:
-
python 2.7def double(x):⏎ return x * 2⏎```⏎
-
agent 2.7* 2 for n in
-
agent 2.54n * 2 for n in
-
agent 2.51= return [n * 2 for n in
-
agent 2.44= return sum(nums) / 2<|arg|>newString=
-
python 2.41squared` should return x squared, but it has `return x * 2`.⏎⏎
-
agent 2.38mean(a, b):⏎ return (a + b) / 2<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 2.38average of nums, but it has `return sum(nums) / 2`.⏎⏎
-
agent 2.35mean(a, b):⏎ return (a + b) / 2<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 2.34average of nums, but it has `return sum(nums) / 2`.⏎⏎