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

Feature #2078

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

Fires on
0.51% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 90%
  • chat 6%
  • story 3%
  • python 1%
  • worksheet 1%

Makes the model say

can happen wasarse could promise:

Makes it not say

ing`.⏎print0box playingwhere

Strongest examples

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

  1. chat 3.13
    answer is 7.<|assistant_end|><|user_start|>What if you add 95
  2. agent 3.07
    <|arg|>command=python3 test_math_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  3. agent 3.01
    hell<|arg|>command=python3 test_mathlib.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent
  4. agent 3.01
    <|arg|>command=python3 test_number_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  5. agent 3.01
    <|arg|>command=python3 test_math_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  6. agent 2.98
    <|arg|>command=python3 test_math_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  7. agent 2.96
    hell<|arg|>command=python3 test_mathlib.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  8. agent 2.95
    hell<|arg|>command=python3 test_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  9. agent 2.94
    hell<|arg|>command=python3 test_strings.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  10. agent 2.93
    hell<|arg|>command=python3 test_items.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  11. agent 2.93
    hell<|arg|>command=python3 test_listlib.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎
  12. agent 2.92
    hell<|arg|>command=python3 test_items.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎