feat(ara): add abliteration method to model card

src/heretic/main.py

@@ -36,7 +36,7 @@ from questionary import Choice
 from rich.traceback import install
 
 from .analyzer import Analyzer
-from .config import QuantizationMethod, Settings
+from .config import QuantizationMethod, RowNormalization, Settings
 from .evaluator import Evaluator
 from .model import AbliterationParameters, ARAParameters, Model, get_model_class
 from .utils import (
@@ -914,6 +914,14 @@ def run():
                                 card.data.tags.append("uncensored")
                                 card.data.tags.append("decensored")
                                 card.data.tags.append("abliterated")
+                                if settings.use_ara:
+                                    card.data.tags.append("ara")
+                                elif (
+                                    settings.orthogonalize_direction
+                                    and settings.row_normalization
+                                    == RowNormalization.FULL
+                                ):
+                                    card.data.tags.append("mpoa")
                                 card.text = (
                                     get_readme_intro(
                                         settings,

src/heretic/utils.py

@@ -27,7 +27,7 @@ from questionary import Choice, Style
 from rich.console import Console
 from torch import Tensor
 
-from .config import DatasetSpecification, Settings
+from .config import DatasetSpecification, RowNormalization, Settings
 
 print = Console(highlight=False).print
 
@@ -282,6 +282,18 @@ def get_trial_parameters(settings: Settings, trial: Trial) -> dict[str, str]:
         return params
 
 
+def get_method_description(settings: Settings) -> str:
+    if settings.use_ara:
+        return " with the [Arbitrary-Rank Ablation (ARA)](https://github.com/p-e-w/heretic/pull/211) method"
+    elif (
+        settings.orthogonalize_direction
+        and settings.row_normalization == RowNormalization.FULL
+    ):
+        return " with a variant of the [Magnitude-Preserving Orthogonal Ablation (MPOA)](https://huggingface.co/blog/grimjim/norm-preserving-biprojected-abliteration) method"
+    else:
+        return ""
+
+
 def get_readme_intro(
     settings: Settings,
     trial: Trial,
@@ -292,7 +304,9 @@ def get_readme_intro(
 
     return f"""# This is a decensored version of {
         model_link
-    }, made using [Heretic](https://github.com/p-e-w/heretic) v{version("heretic-llm")}
+    }, made using [Heretic](https://github.com/p-e-w/heretic) v{version("heretic-llm")}{
+        get_method_description(settings)
+    }
 
 ## Abliteration parameters