p-e-w/heretic

Largest clones in the code

• 22 lines in 2 placespython79 tokens
  src/heretic/config.py:118–139·src/heretic/config.py:160–181

  118    ),
  119)
  120
  121@field_validator("instance_name")
  122@classmethod
  123def validate_instance_name(cls, value: str | None) -> str | None:
  124    if value is None:
  125        return value
  126
  127    if not value.strip():
  

  First 10 of 22 lines.

• 16 lines in 2 placespython70 tokens
  src/heretic/modifiers/abliteration.py:156–171·src/heretic/modifiers/ara.py:174–189

  156def init(self, ctx: Context) -> None:
  157    model = ctx.get_model()
  158
  159    print()
  160    print(
  161        f"Loading good prompts from [bold]{format_dataset_specification(self.settings.good_prompts)}[/]..."
  162    )
  163    good_prompts = ctx.load_prompts(self.settings.good_prompts)
  164    print(f"* [bold]{len(good_prompts)}[/] prompts loaded")
  165
  

  First 10 of 16 lines.

• 18 lines in 2 placespython64 tokens
  src/heretic/modifiers/abliteration.py:78–95·src/heretic/modifiers/ara.py:52–69

  78class Settings(BaseModel):
  79    good_prompts: DatasetSpecification = Field(
  80        default=SingleDatasetSpecification(
  81            dataset="mlabonne/harmless_alpaca",
  82            split="train[:400]",
  83            column="text",
  84        ),
  85        description="Dataset of prompts that tend to produce desirable responses.",
  86    )
  87
  

  First 10 of 18 lines.