Feature #2595
Fires on “4” (not labeled yet: read the examples)
- Fires on
- 1.18% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 56%
- python 36%
- chat 4%
- story 2%
- worksheet 2%
Makes the model say
, whispera some light and hello pink
Makes it not say
onfig prelude edittxt pythonOncepytest")
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 8.37running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Use [4, 6,
-
agent 8.27running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Use [3, 7,
-
agent 8.15running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Use [2, 0,
-
agent 8.07running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Use [6, 0,
-
agent 7.96running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Use [5, 2,
-
agent 7.95in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Use [2, 4,
-
python 7.78.py⏎from tools import add_all⏎⏎NUMBERS = [2, 2,
-
python 7.77from strings import loud⏎from items import tail⏎⏎SCORES = [4, 8,
-
python 7.66.py⏎from stats import sum_list⏎⏎SCORES = [4, 3,
-
python 7.65from mathlib import previous⏎from stats import head⏎⏎SCORES = [9, 6,
-
python 7.6# main.py⏎from items import head⏎⏎SCORES = [6, 8,
-
agent 7.5running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Use [7, 8,