pyright round 4

Author: Benny Chen

eval_protocol/rewards/apps_coding_reward.py

@@ -84,7 +84,7 @@ def evaluate_apps_solution(messages: List[Message], ground_truth: Optional[str],
             reason="No messages provided.",
         )
 
-    raw_solution_content = messages[-1].content
+    raw_solution_content = messages[-1].content if isinstance(messages[-1].content, str) else ""
     code_solution = _extract_python_code(raw_solution_content)
 
     if not code_solution or not code_solution.strip():

eval_protocol/rewards/bfcl_reward.py

@@ -262,7 +262,8 @@ def bfcl_reward(
             assistant_message_found = True
             total_assistant_messages += 1
             # Check for any content or any tool_call
-            if (msg.content and msg.content.strip()) or msg.tool_calls:
+            content_str = msg.content if isinstance(msg.content, str) else ""
+            if (content_str and content_str.strip()) or msg.tool_calls:
                 valid_assistant_messages += 1
 
     if not assistant_message_found:

eval_protocol/rewards/length.py

@@ -93,7 +93,7 @@ def length_reward(
                     )
                 },
             )
-        text = response.content
+        text = response.content if isinstance(response.content, str) else ""
     elif isinstance(response, dict):
         if response.get("role") != "assistant" or not response.get("content"):
             return EvaluateResult(
@@ -107,7 +107,8 @@ def length_reward(
                     )
                 },
             )
-        text = response.get("content", "")
+        text_val = response.get("content", "")
+        text = text_val if isinstance(text_val, str) else ""
     else:
         return EvaluateResult(
             score=0.0,
@@ -294,7 +295,7 @@ def cosine_length_reward(
                     )
                 },
             )
-        text = response.content
+        text = response.content if isinstance(response.content, str) else ""
     elif isinstance(response, dict):
         if response.get("role") != "assistant" or not response.get("content"):
             return EvaluateResult(
@@ -308,7 +309,8 @@ def cosine_length_reward(
                     )
                 },
             )
-        text = response.get("content", "")
+        text_val = response.get("content", "")
+        text = text_val if isinstance(text_val, str) else ""
     else:
         return EvaluateResult(
             score=0.0,

eval_protocol/rewards/reasoning_steps.py

@@ -60,7 +60,7 @@ def reasoning_steps_reward(
                 )
             },
         )
-    text: str = response.content
+    text: str = response.content if isinstance(response.content, str) else ""
 
     # Default patterns for detecting reasoning steps
     default_patterns = [
@@ -199,7 +199,7 @@ def sequence_reward(
                 )
             },
         )
-    text: str = response.content
+    text: str = response.content if isinstance(response.content, str) else ""
 
     if not sequence_terms:
         sequence_terms = [

eval_protocol/rewards/repetition.py

@@ -248,7 +248,7 @@ def diversity_reward(
                     )
                 },
             )
-        text = response.content or ""
+        text = _to_text(response.content)
     elif isinstance(response, dict):
         if response.get("role") != "assistant":
             return EvaluateResult(
@@ -262,7 +262,8 @@ def diversity_reward(
                     )
                 },
             )
-        text = response.get("content", "")
+        text_val = response.get("content", "")
+        text = text_val if isinstance(text_val, str) else ""
     else:
         return EvaluateResult(
             score=0.0,