Feature #228
Fires on “ and” (not labeled yet: read the examples)
- Fires on
- 0.95% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 80%
- chat 12%
- worksheet 4%
- python 2%
- agent 2%
Makes the model say
tiredce today room healthyitems end
Makes it not say
i])) sure "$ False multipl⏎ json
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.81fence! Tom was sad because he could not find
-
story 3.61toys and make funny shadows on the wall. They were happy and safe.<|bos|>#
-
story 3.6the room. Sue was happy and felt safe in the
-
story 3.52a little bird came to the spirit. The bird was sad because it could not choose
-
chat 3.45and peed on it. The fire went away. Tom was happy and laughed at the joke
-
story 3.44the dog. The dog was very happy and wagged his tail.
-
story 3.43Millie was very happy and very full. All
-
chat 3.43kids joined them and they all danced together. They were all happy and had a great time
-
story 3.43ense was, so they went inside to ask Lily. Lily was sad because the man was rude
-
chat 3.42was a room full of toys. The cat and the mouse were happy and played with the toys
-
chat 3.41, he saw a tiny fish in the water. The fish was sad because it could not swim
-
story 3.39also liked food. But the food was expensive. The car was sad because it could not buy