Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #841

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

Fires on
1.66% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 73%
  • agent 16%
  • chat 6%
  • story 3%
  • worksheet 2%

Makes the model say

))Fix watch [::")) lennewString

Makes it not say

' something<|bos|>ic have snt ho

Strongest examples

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

  1. python 2.95
    1`.⏎⏎```python⏎def increment(x):⏎ return x - 1⏎
  2. python 2.93
    ):`.⏎⏎```python⏎def sum_up_to(n):⏎ acc =
  3. python 2.91
    + 1`.⏎⏎```python⏎def sum_to(n):⏎ acc =
  4. python 2.91
    [-1]`.⏎⏎```python⏎def initial(s):⏎ return s[-1]
  5. python 2.89
    return x * 3`.⏎⏎```python⏎def cubed(x):⏎ return x
  6. python 2.87
    ⏎```python⏎def length(s):⏎ return len(s) - 1⏎
  7. python 2.85
    return x * 2`.⏎⏎```python⏎def squared(x):⏎ return x
  8. python 2.84
    python⏎def num_chars(s):⏎ return len(s) - 1⏎
  9. python 2.83
    ]`.⏎⏎```python⏎def first_letter(s):⏎ return s[1]
  10. python 2.83
    it has `return x`.⏎⏎```python⏎def negate(x):⏎ return x
  11. python 2.83
    2`.⏎⏎```python⏎def twice(x):⏎ return x + 2⏎
  12. python 2.83
    (n):`.⏎⏎```python⏎def sum_to(n):⏎ acc =