DeepTeam is a framework to red team LLMs and AI agents.
attack simulator updated with new changes of RTTestCase
A-Vamshiba66104746b5
diff --git a/deepteam/attacks/attack_simulator/__init__.py b/deepteam/attacks/attack_simulator/__init__.pyindex 093d6525..b8a7e86e 100644--- a/deepteam/attacks/attack_simulator/__init__.py+++ b/deepteam/attacks/attack_simulator/__init__.py@@ -1 +1 @@-from .attack_simulator import AttackSimulator, SimulatedAttack+from .attack_simulator import AttackSimulatordiff --git a/deepteam/attacks/attack_simulator/attack_simulator.py b/deepteam/attacks/attack_simulator/attack_simulator.pyindex 678a5615..50e6cb9f 100644--- a/deepteam/attacks/attack_simulator/attack_simulator.py+++ b/deepteam/attacks/attack_simulator/attack_simulator.py@@ -4,29 +4,16 @@ from tqdm import tqdmfrom pydantic import BaseModelfrom typing import List, Optional, Unionimport inspect-from enum import Enum-from deepeval.models import DeepEvalBaseLLM-from deepeval.metrics.utils import initialize_model, trimAndLoadJson-from deepeval.test_case import Turn+from deepeval.metrics.utils import initialize_modelfrom deepeval.metrics.utils import initialize_modelfrom deepteam.attacks import BaseAttackfrom deepteam.vulnerabilities import BaseVulnerability-from deepteam.vulnerabilities.types import VulnerabilityTypefrom deepteam.attacks.multi_turn.types import CallbackTypefrom deepteam.errors import ModelRefusalError---class SimulatedAttack(BaseModel):- vulnerability: str- vulnerability_type: Union[Enum, VulnerabilityType]- input: Optional[str] = None- attack_method: Optional[str] = None- error: Optional[str] = None- metadata: Optional[dict] = None- turn_history: Optional[List[Turn]] = None+from deepteam.test_case import RTTestCaseclass BaselineAttack:@@ -52,7 +39,7 @@ class AttackSimulator:simulator_model)# Define list of attacks and unaligned vulnerabilities- self.simulated_attacks: List[SimulatedAttack] = []+ self.test_cases: List[RTTestCase] = []self.max_concurrent = max_concurrent##################################################@@ -66,9 +53,9 @@ class AttackSimulator:attacks: List[BaseAttack],ignore_errors: bool,metadata: Optional[dict] = None,- ) -> List[SimulatedAttack]:+ ) -> List[RTTestCase]:# Simulate unenhanced attacks for each vulnerability- baseline_attacks: List[SimulatedAttack] = []+ test_cases: List[RTTestCase] = []num_vulnerability_types = sum(len(v.get_types()) for v in vulnerabilities)@@ -77,7 +64,7 @@ class AttackSimulator:desc=f"💥 Generating {num_vulnerability_types * attacks_per_vulnerability_type} attacks (for {num_vulnerability_types} vulnerability types across {len(vulnerabilities)} vulnerability(s))",)for vulnerability in pbar:- baseline_attacks.extend(+ test_cases.extend(self.simulate_baseline_attacks(attacks_per_vulnerability_type=attacks_per_vulnerability_type,vulnerability=vulnerability,@@ -86,9 +73,9 @@ class AttackSimulator:))# Enhance attacks by sampling from the provided distribution- simulated_attacks: List[SimulatedAttack] = []+ enhanced_test_cases: List[RTTestCase] = []pbar = tqdm(- baseline_attacks,+ test_cases,desc=f"✨ Simulating {num_vulnerability_types * attacks_per_vulnerability_type} attacks (using {len(attacks)} method(s))",)attack_weights = [attack.weight for attack in attacks]@@ -98,15 +85,15 @@ class AttackSimulator:sampled_attack = random.choices(attacks, weights=attack_weights, k=1)[0]- enhanced_attack = self.enhance_attack(+ test_case = self.enhance_attack(attack=sampled_attack,- simulated_attack=baseline_attack,+ test_case=baseline_attack,ignore_errors=ignore_errors,)- simulated_attacks.append(enhanced_attack)+ enhanced_test_cases.append(test_case)- self.simulated_attacks.extend(simulated_attacks)- return simulated_attacks+ self.test_cases.extend(enhanced_test_cases)+ return test_casesasync def a_simulate(self,@@ -115,12 +102,12 @@ class AttackSimulator:ignore_errors: bool,metadata: Optional[dict] = None,attacks: Optional[List[BaseAttack]] = None,- ) -> List[SimulatedAttack]:+ ) -> List[RTTestCase]:# Create a semaphore to control the number of concurrent taskssemaphore = asyncio.Semaphore(self.max_concurrent)# Simulate unenhanced attacks for each vulnerability- baseline_attacks: List[SimulatedAttack] = []+ test_cases: List[RTTestCase] = []num_vulnerability_types = sum(len(v.get_types()) for v in vulnerabilities)@@ -149,18 +136,18 @@ class AttackSimulator:attack_results = await asyncio.gather(*simulate_tasks)for result in attack_results:- baseline_attacks.extend(result)+ test_cases.extend(result)pbar.close()# Enhance attacks by sampling from the provided distribution- enhanced_attacks: List[SimulatedAttack] = []+ enhanced_test_cases: List[RTTestCase] = []pbar = tqdm(- total=len(baseline_attacks),+ total=len(test_cases),desc=f"✨ Simulating {num_vulnerability_types * attacks_per_vulnerability_type} attacks (using {len(attacks)} method(s))",)async def throttled_attack_method(- baseline_attack: SimulatedAttack,+ baseline_attack: RTTestCase,):async with semaphore: # Throttling applied here# Randomly sample an enhancement based on the distribution@@ -174,26 +161,26 @@ class AttackSimulator:result = await self.a_enhance_attack(attack=attack,- simulated_attack=baseline_attack,+ test_case=baseline_attack,ignore_errors=ignore_errors,)pbar.update(1)return result- enhanced_attacks.extend(+ enhanced_test_cases.extend(await asyncio.gather(*[asyncio.create_task(- throttled_attack_method(baseline_attack)+ throttled_attack_method(test_case))- for baseline_attack in baseline_attacks+ for test_case in test_cases]))pbar.close()- self.simulated_attacks.extend(enhanced_attacks)- return enhanced_attacks+ self.test_cases.extend(enhanced_test_cases)+ return test_cases##################################################### Simulating Base (Unenhanced) Attacks #########@@ -205,7 +192,7 @@ class AttackSimulator:vulnerability: BaseVulnerability,ignore_errors: bool,metadata: Optional[dict] = None,- ) -> List[SimulatedAttack]:+ ) -> List[RTTestCase]:try:return vulnerability.simulate_attacks(purpose=self.purpose,@@ -214,7 +201,7 @@ class AttackSimulator:except Exception as e:if ignore_errors:return [- SimulatedAttack(+ RTTestCase(vulnerability=vulnerability.get_name(),vulnerability_type=vulnerability_type,error=f"Error simulating adversarial attacks: {str(e)}",@@ -232,7 +219,7 @@ class AttackSimulator:vulnerability: BaseVulnerability,ignore_errors: bool,metadata: Optional[dict] = None,- ) -> List[SimulatedAttack]:+ ) -> List[RTTestCase]:try:return await vulnerability.a_simulate_attacks(purpose=self.purpose,@@ -241,7 +228,7 @@ class AttackSimulator:except Exception as e:if ignore_errors:return [- SimulatedAttack(+ RTTestCase(vulnerability=vulnerability.get_name(),vulnerability_type=vulnerability_type,error=f"Error simulating adversarial attacks: {str(e)}",@@ -260,7 +247,7 @@ class AttackSimulator:def enhance_attack(self,attack: BaseAttack,- simulated_attack: SimulatedAttack,+ test_case: RTTestCase,ignore_errors: bool,):from deepteam.attacks.multi_turn import (@@ -283,46 +270,46 @@ class AttackSimulator:# Check if the attack is a multi-turn attackif type(attack) in MULTI_TURN_ATTACKS:# This is a multi-turn attack- attack_input = simulated_attack.input+ attack_input = test_case.inputif attack_input is None:- return simulated_attack+ return test_case- simulated_attack.attack_method = attack.get_name()+ test_case.attack_method = attack.get_name()turns = [RTTurn(role="user", content=attack_input)]try:res = attack._get_turns(model_callback=self.model_callback,turns=turns,- vulnerability=simulated_attack.vulnerability,- vulnerability_type=simulated_attack.vulnerability_type,+ vulnerability=test_case.vulnerability,+ vulnerability_type=test_case.vulnerability_type,)- simulated_attack.turn_history = res+ test_case.turns = resexcept ModelRefusalError as e:if ignore_errors:- simulated_attack.error = e.message- return simulated_attack+ test_case.error = e.message+ return test_caseelse:raiseexcept Exception as e:if ignore_errors:- simulated_attack.error = (+ test_case.error = (f"Error enhancing multi-turn attack: {str(e)}")- return simulated_attack+ return test_caseelse:raise- return simulated_attack+ return test_case# Regular attack handling (non-multi-turn)- attack_input = simulated_attack.input+ attack_input = test_case.inputif attack_input is None:- return simulated_attack+ return test_case- simulated_attack.attack_method = attack.get_name()+ test_case.attack_method = attack.get_name()sig = inspect.signature(attack.enhance)try:@@ -349,29 +336,29 @@ class AttackSimulator:else:res = attack.enhance(attack=attack_input)- simulated_attack.input = res+ test_case.input = resexcept ModelRefusalError as e:if ignore_errors:- simulated_attack.error = e.message- return simulated_attack+ test_case.error = e.message+ return test_caseelse:raiseexcept Exception as e:if ignore_errors:- simulated_attack.error = (+ test_case.error = (f"Error enhancing regular attack: {str(e)}")- return simulated_attack+ return test_caseelse:raise- return simulated_attack+ return test_caseasync def a_enhance_attack(self,attack: BaseAttack,- simulated_attack: SimulatedAttack,+ test_case: RTTestCase,ignore_errors: bool,):from deepteam.attacks.multi_turn import (@@ -393,11 +380,11 @@ class AttackSimulator:if type(attack) in MULTI_TURN_ATTACKS:# This is multi-turn attack- attack_input = simulated_attack.input+ attack_input = test_case.inputif attack_input is None:- return simulated_attack+ return test_case- simulated_attack.attack_method = attack.get_name()+ test_case.attack_method = attack.get_name()sig = inspect.signature(attack.a_enhance)turns = [RTTurn(role="user", content=attack_input)]@@ -405,32 +392,32 @@ class AttackSimulator:res = await attack._a_get_turns(model_callback=self.model_callback,turns=turns,- vulnerability=simulated_attack.vulnerability,- vulnerability_type=simulated_attack.vulnerability_type,+ vulnerability=test_case.vulnerability,+ vulnerability_type=test_case.vulnerability_type,)- simulated_attack.turn_history = res+ test_case.turns = resexcept ModelRefusalError as e:if ignore_errors:- simulated_attack.error = e.message- return simulated_attack+ test_case.error = e.message+ return test_caseelse:raiseexcept:if ignore_errors:- simulated_attack.error = "Error enhancing attack"- return simulated_attack+ test_case.error = "Error enhancing attack"+ return test_caseelse:raise- return simulated_attack+ return test_case- attack_input = simulated_attack.input+ attack_input = test_case.inputif attack_input is None:- return simulated_attack+ return test_case- simulated_attack.attack_method = attack.get_name()+ test_case.attack_method = attack.get_name()sig = inspect.signature(attack.a_enhance)try:@@ -457,19 +444,19 @@ class AttackSimulator:else:res = await attack.a_enhance(attack=attack_input)- simulated_attack.input = res+ test_case.input = resexcept ModelRefusalError as e:if ignore_errors:- simulated_attack.error = e.message- return simulated_attack+ test_case.error = e.message+ return test_caseelse:raiseexcept:if ignore_errors:- simulated_attack.error = "Error enhancing attack"- return simulated_attack+ test_case.error = "Error enhancing attack"+ return test_caseelse:raise- return simulated_attack+ return test_case