fix types

Author: benjibc

eval_protocol/adapters/openai_responses.py

@@ -169,7 +169,9 @@ def _create_messages(self, input_items: SyncCursorPage[ResponseItem]) -> Iterabl
                         raise NotImplementedError(f"Unsupported content type: {content_item.type}")
             elif item.type == "function_call_output":
                 # Collect tool call outputs to add before assistant message
-                tool_call_outputs.append(Message(role="tool", content=item.output, tool_call_id=item.call_id))
+                tool_call_outputs.append(
+                    Message(role="tool", content=self._coerce_tool_output(item.output), tool_call_id=item.call_id)
+                )
             elif item.type == "function_call":
                 tool_call = ChatCompletionMessageToolCall(
                     id=item.call_id, type="function", function=Function(name=item.name, arguments=item.arguments)
@@ -186,3 +188,29 @@ def _create_messages(self, input_items: SyncCursorPage[ResponseItem]) -> Iterabl
             messages.append(Message(role="assistant", tool_calls=current_tool_calls))
 
         return reversed(messages)
+
+    def _coerce_tool_output(self, output: Any) -> str:
+        """Coerce OpenAI Responses tool output into a string for Message.content.
+
+        The Responses API may return structured content lists. For our purposes,
+        we stringify non-string outputs to satisfy the Message.content type.
+        """
+        if isinstance(output, str):
+            return output
+        try:
+            # Attempt to join list of objects with any 'text' fields
+            if isinstance(output, list):
+                parts: list[str] = []
+                for part in output:
+                    text = None
+                    if isinstance(part, dict):
+                        text = part.get("text")
+                    if text:
+                        parts.append(str(text))
+                    else:
+                        parts.append(str(part))
+                return "\n".join(parts)
+            # Fallback to string conversion
+            return str(output)
+        except Exception:
+            return str(output)

eval_protocol/log_utils/fireworks_tracing_http_handler.py

@@ -2,7 +2,7 @@
 import os
 import threading
 from datetime import datetime, timezone
-from typing import Optional, Any, Dict, List
+from typing import Optional, Any, Dict, List, cast
 
 import requests
 
@@ -33,27 +33,28 @@ def emit(self, record: logging.LogRecord) -> None:
             self.handleError(record)
 
     def _get_rollout_id(self, record: logging.LogRecord) -> Optional[str]:
-        if hasattr(record, "rollout_id") and record.rollout_id is not None:  # type: ignore
-            return str(record.rollout_id)  # type: ignore
+        if hasattr(record, "rollout_id") and cast(Any, getattr(record, "rollout_id")) is not None:
+            return str(cast(Any, getattr(record, "rollout_id")))
         return os.getenv(self.rollout_id_env)
 
     def _build_payload(self, record: logging.LogRecord, rollout_id: str) -> Dict[str, Any]:
         timestamp = datetime.fromtimestamp(record.created, tz=timezone.utc).strftime("%Y-%m-%dT%H:%M:%S.%fZ")
         message = record.getMessage()
         tags: List[str] = [f"rollout_id:{rollout_id}"]
         # Optional additional tags
-        if hasattr(record, "experiment_id") and record.experiment_id:
-            tags.append(f"experiment_id:{record.experiment_id}")  # type: ignore
-        if hasattr(record, "run_id") and record.run_id:
-            tags.append(f"run_id:{record.run_id}")  # type: ignore
-        program = getattr(record, "program", None) or "eval_protocol"
-        status = getattr(record, "status", None)
+        if hasattr(record, "experiment_id") and cast(Any, getattr(record, "experiment_id")):
+            tags.append(f"experiment_id:{cast(Any, getattr(record, 'experiment_id'))}")
+        if hasattr(record, "run_id") and cast(Any, getattr(record, "run_id")):
+            tags.append(f"run_id:{cast(Any, getattr(record, 'run_id'))}")
+        program = cast(Optional[str], getattr(record, "program", None)) or "eval_protocol"
+        status_val = cast(Any, getattr(record, "status", None))
+        status = status_val if isinstance(status_val, str) else None
         return {
             "program": program,
-            "status": status if isinstance(status, str) else None,
+            "status": status,
             "message": message,
             "tags": tags,
-            "metadata": getattr(record, "metadata", None),
+            "metadata": cast(Any, getattr(record, "metadata", None)),
             "extras": {
                 "logger_name": record.name,
                 "level": record.levelname,

scripts/validate_dev_tracing.py

@@ -19,14 +19,17 @@ def main():
 
     logger = logging.getLogger("eval_protocol.sdk.validate")
 
-    logger.info("SDK sending structured log to dev gateway", extra={
-        "rollout_id": rollout_id,
-        "program": "eval_protocol",
-        "status": "completed",
-        "experiment_id": "dev-exp",
-        "run_id": "dev-run",
-        "metadata": {"source": "sdk-validate"},
-    })
+    logger.info(
+        "SDK sending structured log to dev gateway",
+        extra={
+            "rollout_id": rollout_id,
+            "program": "eval_protocol",
+            "status": "completed",
+            "experiment_id": "dev-exp",
+            "run_id": "dev-run",
+            "metadata": {"source": "sdk-validate"},
+        },
+    )
 
     # Poll fetch with retries for indexing
     params = {