Skip to content
mini-lab
Features of prelude-2

Feature #2906

Fires on “ [” (not labeled yet: read the examples)

Fires on
0.58% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 48%
  • agent 45%
  • chat 4%
  • story 2%
  • worksheet 1%

Makes the model say

(nSpotopencodedemo NowSamastic

Makes it not say

butOnce from walking mostieing#

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

  1. python 6.17
    def doubled(nums):⏎ return [n * 2
  2. python 6.14
    <|assistant_end|><|bos|>```python⏎def double_all(nums):⏎ return [n * 2
  3. python 5.59
    .<|bos|># main.py⏎def evens(nums):⏎ return [n for n in
  4. python 5.57
    # items.py⏎def double_all(nums):⏎ return [n * 2
  5. python 5.54
    # main.py⏎def keep_even(nums):⏎ return [n for n in
  6. python 5.54
    # items.py⏎def double_all(nums):⏎ return [n * 2
  7. python 5.53
    # main.py⏎def only_even(nums):⏎ return [n for n in
  8. python 5.48
    # items.py⏎def only_even(nums):⏎ return [n for n in
  9. python 5.48
    ⏎⏎```python⏎def only_even(nums):⏎ return [n for n in
  10. python 5.45
    # items.py⏎def double_all(nums):⏎ return [n * 2
  11. python 5.44
    python⏎def only_even(nums):⏎ return [n for n in
  12. python 5.41
    ⏎⏎```python⏎def keep_even(nums):⏎ return [n for n in