Feature #3906
Fires on “ best” (not labeled yet: read the examples)
- Fires on
- 0.54% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 38%
- chat 26%
- agent 16%
- worksheet 12%
- python 8%
Makes the model say
<|bos|> of one they-i friends
Makes it not say
NoBut ButWriteexpressionsed trained please
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.5as fast as she had
-
chat 3.5as they finished, something
-
worksheet 3.4603 more stickers. How many stickers does Max have
-
story 3.42best treasure hunters
-
story 3.42best friends, called Jack
-
story 3.42best of friends forever.
-
chat 3.31how to tell stories and
-
chat 3.31how to tell little stories
-
chat 3.31how to tell stories and
-
story 3.27most flowers wins."
-
chat 3.25more flowers. How many flowers does Anna have
-
story 3.22so proud that she was