Feature #1143
Fires on “Once” (not labeled yet: read the examples)
- Fires on
- 0.62% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 64%
- chat 29%
- agent 4%
- python 2%
- worksheet 0%
Makes the model say
Once upon Spot to Tomteneties
Makes it not say
Thank who happy angry faceS<|arg|> friend
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 4.02<|bos|>Once upon a time,
-
story 4.02<|bos|>Once upon a time,
-
story 4.02<|bos|>Once upon a time,
-
story 4.02<|bos|>Once upon a time,
-
story 4.02<|bos|>Once upon a time,
-
story 4.02<|bos|>Once upon a time,
-
story 3.99<|assistant_end|><|bos|>Once upon a time,
-
story 3.98the snow all day.<|bos|>Once upon a time,
-
story 3.94py`.<|assistant_end|><|bos|>Once upon a time,
-
story 3.940.<|assistant_end|><|bos|>Once upon a time,
-
story 3.93old lady was happy.<|bos|>Once upon a time,
-
story 3.9Sam was a little boy.<|assistant_end|><|bos|>Once upon a time,