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

Feature #1820

Fires on “<|tool_end|>” (not labeled yet: read the examples)

Fires on
0.84% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 88%
  • story 6%
  • python 4%
  • chat 2%
  • worksheet 0%

Makes the model say

edprint help seen⏎ feeling scom

Makes it not say

])) tests next``` plus")) forever replace

Strongest examples

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

  1. agent 3.59
    : print(backwards(TEXT))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  2. agent 3.43
    6]))⏎5: print(count_items(NUMBERS))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  3. agent 3.35
    1))⏎5: print(mean_list(NUMBERS))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  4. agent 3.29
    (exclaim("cloud"))⏎6: print(count(SCORES))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  5. agent 3.29
    3]))⏎6: print(howw_many(SCORES))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  6. agent 3.27
    9, 3]⏎4: print(tail(NUMBERS))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  7. agent 3.26
    , 1]))⏎14: print(total(NUMBERS))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  8. agent 3.24
    -8⏎4: print(magntiude(NUMBER))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  9. agent 3.24
    6⏎4: print(duoble(LIMIT))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  10. agent 3.22
    2]))⏎11: print(count_items(SCORES))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  11. agent 3.21
    WORD = "cat"⏎4: print(initial(WORD))<|tool_end|><|assistant_start|><|tool_call_start|>edit
  12. agent 3.2
    12))⏎6: print(is_odd(NUMBER))<|tool_end|><|assistant_start|><|tool_call_start|>edit