fix

Author: shreymodi1

.github/workflows/streaming_compliance.yml

@@ -76,7 +76,6 @@ jobs:
         env:
           FIREWORKS_API_KEY: ${{ secrets.FIREWORKS_API_KEY }}
           FIREWORKS_ACCOUNT_ID: ${{ vars.FIREWORKS_ACCOUNT_ID }}
-          DISABLE_EP_SQLITE_LOG: "1"
         run: |
           . .venv/bin/activate
           mkdir -p artifacts

eval_protocol/benchmarks/test_glm_streaming_compliance.py

@@ -336,9 +336,4 @@ def test_glm_streaming_tool_call(row: EvaluationRow) -> EvaluationRow:
         ),
         metrics=metrics,
     )
-    try:
-        row.model_dump(exclude_none=True, mode="json")
-    except Exception as exc:  # pragma: no cover - debug helper
-        print("DEBUG model_dump failure", exc, row.messages)
-
     return row

eval_protocol/pytest/default_single_turn_rollout_process.py

@@ -1,4 +1,5 @@
 import asyncio
+import json
 import logging
 import os
 import time
@@ -103,6 +104,11 @@ async def process_row(row: EvaluationRow) -> EvaluationRow:
             reasoning_content = getattr(assistant_message, "reasoning_content", None)
             if reasoning_content is None:
                 reasoning_content = getattr(assistant_message, "reasoning", None)
+            if reasoning_content is not None and not isinstance(reasoning_content, str):
+                try:
+                    reasoning_content = json.dumps(reasoning_content)
+                except Exception:
+                    reasoning_content = str(reasoning_content)
             tool_calls = assistant_message.tool_calls if assistant_message.tool_calls else None
 
             converted_tool_calls = None