Feature #2291
Fires on “ recent” (not labeled yet: read the examples)
- Fires on
- 0.85% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 49%
- agent 19%
- python 13%
- chat 12%
- worksheet 7%
Makes the model say
like liked wanted strong walking sailse tall
Makes it not say
<|think_end|>oldString{command code voicepypo
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.21there was a mighty big farm. On this farm, there were many animals who liked to play
-
agent 2.17Traceback (most recent call last):⏎
-
story 2.15The tree was in a big forest. The forest had many trees and animals. Tim loved to
-
story 2.1the snow all day.<|bos|>Once upon a time, there were two friends called Ant and
-
agent 2.1(most recent call last):⏎
-
agent 2.1(most recent call last):⏎
-
agent 2.1(most recent call last):⏎
-
story 2.1`.⏎<|bos|>Once upon a time, in a beautiful land, there were many colors. The
-
story 2.05a time, there was a colorful farm. On this farm, there were many animals. The
-
story 2.04and add numbers.<|assistant_end|><|bos|>Once upon a time, there were two friends called Drew and
-
story 2.04The next day, the
-
chat 2.04The answer is 8