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

Feature #1853

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

Fires on
0.29% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 54%
  • story 36%
  • chat 8%
  • worksheet 1%
  • python 0%

Makes the model say

ishhello scratchally candemoer

Makes it not say

of always from," napIn])) today

Strongest examples

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

  1. story 3.36
    asked him to forgive him. He said he wanted to play with
  2. story 3.25
    their study session. Tom was sorry, but he said he needed the micr
  3. agent 3.21
    Read file main.py, lines 1-6⏎1: from tools import is_
  4. story 3.2
    . He hugs mom and mom hugs him back. She says she is sorry too and
  5. agent 3.2
    Found 6 matches⏎run.py:⏎ Line 1: from stats import evens⏎⏎
  6. agent 3.2
    Found 3 matches⏎script.py:⏎ Line 1: from items import sum_
  7. agent 3.15
    Found 6 matches⏎main.py:⏎ Line 1: from helpers import count⏎⏎
  8. agent 3.14
    Found 5 matches⏎hello.py:⏎ Line 1: from mathlib import odd⏎⏎
  9. agent 3.12
    Read file demo.py, lines 1-3⏎1: from words import echo⏎
  10. agent 3.12
    Found 6 matches⏎main.py:⏎ Line 1: from arith import product,
  11. agent 3.12
    Read file main.py, lines 1-3⏎1: from stats import biggest⏎
  12. agent 3.11
    Found 3 matches⏎hello.py:⏎ Line 1: from lib import raise_