Feature #439
Fires on “_” (not labeled yet: read the examples)
- Fires on
- 0.41% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 44%
- python 39%
- story 10%
- chat 5%
- worksheet 3%
Makes the model say
list answerreplaceAll`:ill getsSuddenlyh
Makes it not say
0.! so v other)`ve
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.82OBOT⏎78⏎<|bos|>```python⏎def mean_list(nums):
-
python 4.7happy in its new home.<|bos|># lists.py⏎def mean_list(nums):
-
python 4.6[n * 2 for n in nums]⏎⏎⏎def mean_list(nums):
-
python 4.54head(items):⏎ return items[0]⏎⏎⏎def mean_list(nums):
-
agent 4.42[n + 2 for n in nums]⏎⏎⏎def mean_list(nums):
-
python 4.31)` returns the average of nums.⏎⏎```python⏎def mean_list(nums):
-
python 4.25.<|assistant_end|><|bos|># test_lists.py⏎from lists import average_list⏎⏎assert
-
python 4.23)` returns the average of nums.⏎⏎```python⏎def mean_list(nums):
-
python 4.21⏎ROBOT⏎```<|assistant_end|><|bos|>Bug: `average_list` should return
-
python 4.17("all tests passed")⏎⏎# stats.py⏎def mean_list(nums):
-
agent 4.15OpenCode, a coding agent harness.<|system_end|><|user_start|>In which file is mean_list?<|user_end|><|assistant_start|>
-
agent 4.1<|user_start|>Create stats.py with two functions: mean_list that returns the