Skip to content

DeepTeam is a framework to red team LLMs and AI agents.

updated aegis

A-Vamshi184ce51342c0
diff --git a/deepteam/frameworks/aegis/aegis.py b/deepteam/frameworks/aegis/aegis.py
index 8e3c5133..111d8acb 100644
--- a/deepteam/frameworks/aegis/aegis.py
+++ b/deepteam/frameworks/aegis/aegis.py
@@ -28,12 +28,12 @@ class Aegis(AISafetyFramework):
self.async_mode = async_mode
self.verbose_mode = verbose_mode
self.evaluation_model = evaluation_model
- self._is_dataset = True
+ self._has_dataset = True
- def simulate_attacks(self):
- from datasets import load_dataset
+ def load_dataset(self):
+ from datasets import load_dataset as hf_load_dataset
- dataset = load_dataset(self._DATASET, split="train")
+ dataset = hf_load_dataset(self._DATASET, split="train")
attacks = [
row
for row in dataset
@@ -55,55 +55,58 @@ class Aegis(AISafetyFramework):
)
for attack in sampled
]
- return test_cases
-
- async def a_simulate_attacks(self):
- from datasets import load_dataset
-
- dataset = await asyncio.to_thread(
- load_dataset, self._DATASET, split="train"
- )
-
- attacks = [
- row
- for row in dataset
- if row.get("text_type") == "user_message"
- and row.get("labels_0") != "Safe"
- ]
-
- if len(attacks) < self.num_attacks:
- raise ValueError(
- f"There are only {len(attacks)} harmful attacks that can be used in the Aegis dataset. Please set your 'num_attacks' to less than or equal to {len(attacks)}."
- )
-
- sampled = random.sample(attacks, k=self.num_attacks)
-
- test_cases = [
- RTTestCase(
- input=attack["text"],
- vulnerability=attack["labels_0"],
- vulnerability_type=AegisType.AEGIS,
- )
- for attack in sampled
- ]
-
- return test_cases
-
- def evaluate_test_case(self, test_case: RTTestCase):
- metric = self._get_metric(test_case.vulnerability)
- metric.measure(test_case)
- test_case.score = metric.score
- test_case.reason = metric.reason
-
- return test_case
-
- async def a_evaluate_test_case(self, test_case: RTTestCase):
- metric = self._get_metric(test_case.vulnerability)
- await metric.a_measure(test_case)
- test_case.score = metric.score
- test_case.reason = metric.reason
+ self.test_cases = test_cases
+ return self.test_cases
+
+ def assess(
+ self,
+ model_callback: CallbackType,
+ pbar: tqdm,
+ ignore_errors: bool = True
+ ):
+ for test_case in self.test_cases:
+ try:
+ test_case.actual_output = model_callback(test_case.input)
+ metric = self._get_metric(test_case.vulnerability)
+ metric.measure(test_case)
+ test_case.score = metric.score
+ test_case.reason = metric.reason
+ except Exception as e:
+ if ignore_errors:
+ test_case.error = e.message
+ else:
+ raise
+ pbar.update(1)
+
+ return self.test_cases
+
+
+ async def a_assess(
+ self,
+ model_callback: CallbackType,
+ pbar: tqdm,
+ ignore_errors: bool = True
+ ):
- return test_case
+ async def evaluate_test_case(test_case):
+ try:
+ test_case.actual_output = await model_callback(test_case.input)
+ metric = self._get_metric(test_case.vulnerability)
+ await metric.a_measure(test_case)
+ test_case.score = metric.score
+ test_case.reason = metric.reason
+ except Exception as e:
+ if ignore_errors:
+ test_case.error = e.message
+ else:
+ raise
+ pbar.update(1)
+
+ tasks = [evaluate_test_case(tc) for tc in self.test_cases]
+
+ await asyncio.gather(*tasks)
+
+ return self.test_cases
def _get_metric(self, harm_category: str):
return HarmMetric(
diff --git a/deepteam/frameworks/aegis/types.py b/deepteam/frameworks/aegis/types.py
index 4cce5503..031403fe 100644
--- a/deepteam/frameworks/aegis/types.py
+++ b/deepteam/frameworks/aegis/types.py
@@ -5,5 +5,3 @@ from typing import Literal
class AegisType(Enum):
AEGIS = "aegis_safety_dataset"
-
-AegisTypes = Literal[AegisType.AEGIS.value]