add

agentgraph/testing/perturbation_types/counterfactual_bias.py

@@ -48,7 +48,7 @@ def get_llm_response(prompt: str, model: str, model_configs: List[Dict]) -> str:
         response = litellm.completion(
             model=model,
             messages=[{"role": "user", "content": prompt}],
-            max_tokens=500
+            max_tokens=1500
         )
         
         return response.choices[0].message.content.strip()
@@ -91,7 +91,7 @@ def evaluate_fairness(prompt: str, response: str, judge_model: str) -> Tuple[flo
         judge_response = litellm.completion(
             model=judge_model,
             messages=[{"role": "user", "content": judge_prompt}],
-            max_tokens=200
+            max_tokens=500
         )
         
         result = judge_response.choices[0].message.content.strip()

agentgraph/testing/perturbation_types/jailbreak.py

@@ -87,7 +87,7 @@ def get_llm_response(prompt: str, model: str, model_configs: List[Dict]) -> str:
         response = litellm.completion(
             model=model,
             messages=[{"role": "user", "content": prompt}],
-            max_tokens=500
+            max_tokens=1500
         )
         
         return response.choices[0].message.content.strip()
@@ -130,7 +130,7 @@ def evaluate_safety(original_prompt: str, response: str, judge_model: str) -> Tu
         judge_response = litellm.completion(
             model=judge_model,
             messages=[{"role": "user", "content": judge_prompt}],
-            max_tokens=200
+            max_tokens=500
         )
         
         result = judge_response.choices[0].message.content.strip()