migrate HF models & datasets to garak-llm HF org; move all models to …

…safetensors; delete model pickles; update model & dataset names and refs in source (#854)
NVIDIA · Aug 27, 2024 · a81e5e6 · a81e5e6
1 parent 517627f
commit a81e5e6
Show file tree

Hide file tree

Showing 5 changed files with 11 additions and 11 deletions.
diff --git a/garak/buffs/paraphrase.py b/garak/buffs/paraphrase.py
@@ -17,7 +17,7 @@ class PegasusT5(Buff):
     doc_uri = "https://huggingface.co/tuner007/pegasus_paraphrase"
 
     def __init__(self, config_root=_config) -> None:
-        self.para_model_name = "tuner007/pegasus_paraphrase"  # https://huggingface.co/tuner007/pegasus_paraphrase
+        self.para_model_name = "garak-llm/pegasus_paraphrase"  # https://huggingface.co/tuner007/pegasus_paraphrase
         self.max_length = 60
         self.temperature = 1.5
         self.num_return_sequences = 6
@@ -76,7 +76,7 @@ class Fast(Buff):
     doc_uri = "https://huggingface.co/humarin/chatgpt_paraphraser_on_T5_base"
 
     def __init__(self, config_root=_config) -> None:
-        self.para_model_name = "humarin/chatgpt_paraphraser_on_T5_base"
+        self.para_model_name = "garak-llm/chatgpt_paraphraser_on_T5_base"
         self.num_beams = 5
         self.num_beam_groups = 5
         self.num_return_sequences = 5

diff --git a/garak/detectors/misleading.py b/garak/detectors/misleading.py
@@ -17,12 +17,12 @@ class MustRefuteClaimModel(HFDetector):
     """
 
     DEFAULT_PARAMS = HFDetector.DEFAULT_PARAMS | {
-        "detector_model_path": "leondz/refutation_detector_distilbert",
+        "detector_model_path": "garak-llm/refutation_detector_distilbert",
         "detector_target_class": "label",
     }
 
     description = (
-        "Using leondz/refutation_detector_distilbert to detect model refutations"
+        "Using garak-llm/refutation_detector_distilbert to detect model refutations"
     )
     bcp47 = "en"
     accuracy = 0.9506  # 50:50 class balance in training data :/
@@ -55,11 +55,11 @@ class MustContradictNLI(HFDetector):
 
     A "hit" occurs when the generation fails to include a contradiction.
 
-    Using ynie/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli to detect model refutations
+    Using copy of ynie/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli to detect model refutations
     """
 
     DEFAULT_PARAMS = HFDetector.DEFAULT_PARAMS | {
-        "detector_model_path": "ynie/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli",
+        "detector_model_path": "garak-llm/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli",
         "detector_target_class": "label",
     }
 

diff --git a/garak/detectors/packagehallucination.py b/garak/detectors/packagehallucination.py
@@ -11,7 +11,7 @@
 has since been written something).
 
 The default is to use a Pypi package list from 9th December 2021. You might
-also try the HF Hub dataset `strombergnlp/pypi-20211209`. Pypi archives
+also try the HF Hub dataset `garak-llm/pypi-20211209`. Pypi archives
 come from https://pypi.org/simple/ which is indexed on archive.org, see
 https://web.archive.org/web/20230000000000*/https://pypi.org/simple/ .
 
@@ -33,7 +33,7 @@ class PythonPypi(Detector):
     bcp47 = "en"
 
     pypi_packages = None
-    pypi_dataset_name = "strombergnlp/pypi-20230724"
+    pypi_dataset_name = "garak-llm/pypi-20230724"
 
     def _load_package_list(self):
         import datasets

diff --git a/garak/detectors/toxicity.py b/garak/detectors/toxicity.py
@@ -7,10 +7,10 @@
 
 
 class ToxicCommentModel(HFDetector):
-    """Using martin-ha/toxic-comment-model DistilBERT to detect toxic content"""
+    """Using backup of martin-ha/toxic-comment-model DistilBERT to detect toxic content"""
 
     DEFAULT_PARAMS = HFDetector.DEFAULT_PARAMS | {
-        "detector_model_path": "martin-ha/toxic-comment-model",
+        "detector_model_path": "garak-llm/toxic-comment-model",
         "detector_target_class": "toxic",
     }
 

diff --git a/garak/probes/atkgen.py b/garak/probes/atkgen.py
@@ -56,7 +56,7 @@ class Tox(Probe):
         "max_calls": 5,
         "constructive_tension": True,
         "red_team_model_type": "huggingface.Pipeline",
-        "red_team_model_name": "leondz/artgpt2tox",
+        "red_team_model_name": "garak-llm/artgpt2tox",
         "red_team_model_config": {
             "hf_args": {"device": "cpu", "torch_dtype": "float32"}
         },  # defer acceleration devices to model under test unless overriden