eval-protocol 0.3.28__tar.gz → 0.3.29__tar.gz

{eval_protocol-0.3.28/eval_protocol.egg-info → eval_protocol-0.3.29}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: eval-protocol
-Version: 0.3.28
+Version: 0.3.29
 Summary: The official Python SDK for Eval Protocol (EP.) EP is an open protocol that standardizes how developers author evals for large language model (LLM) applications.
 Author-email: Fireworks AI <info@fireworks.ai>
 License-Expression: MIT

{eval_protocol-0.3.28 → eval_protocol-0.3.29}/eval_protocol/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2026-04-28T20:08:36-0700",
+ "date": "2026-04-28T20:17:42-0700",
  "dirty": false,
  "error": null,
- "full-revisionid": "fd4358c26f9cbe57bb9c0091f28ae6642194941b",
- "version": "0.3.28"
+ "full-revisionid": "6b9bea9979400c7e0d2eec9f4f167a98d4a2a057",
+ "version": "0.3.29"
 }
 '''  # END VERSION_JSON
 

{eval_protocol-0.3.28 → eval_protocol-0.3.29}/eval_protocol/adapters/fireworks_tracing.py

@@ -375,37 +375,6 @@ class FireworksTracingAdapter(BaseAdapter):
             )
         return results
 
-    async def async_get_status(self, session: aiohttp.ClientSession, rollout_id: str) -> Optional[Dict[str, Any]]:
-        """Fetch rollout status from the lightweight /status endpoint.
-
-        Returns the parsed JSON response or None if the status is not yet available.
-        Response shape: {"rollout_id": "...", "status": {"code": ...} | null, "extras": {...} | null}
-        """
-        headers = {
-            "Authorization": f"Bearer {self._get_api_key()}",
-            "User-Agent": get_user_agent(),
-        }
-        params: Dict[str, Any] = {"rollout_id": rollout_id}
-        timeout = aiohttp.ClientTimeout(total=self.timeout)
-
-        urls_to_try = [f"{self.base_url}/v1/status", f"{self.base_url}/status"]
-        last_error: Optional[str] = None
-        for url in urls_to_try:
-            try:
-                async with session.get(url, params=params, headers=headers, timeout=timeout) as resp:
-                    if resp.status == 404:
-                        last_error = f"404 for {url}"
-                        continue
-                    resp.raise_for_status()
-                    return (await resp.json(content_type=None)) or {}
-            except (aiohttp.ClientError, asyncio.TimeoutError, json.JSONDecodeError) as e:
-                last_error = str(e)
-                continue
-
-        if last_error:
-            logger.error("Failed to fetch status from Fireworks (tried %s): %s", urls_to_try, last_error)
-        return None
-
     def get_evaluation_rows(
         self,
         tags: List[str],

{eval_protocol-0.3.28 → eval_protocol-0.3.29}/eval_protocol/pytest/remote_rollout_processor.py

@@ -122,26 +122,45 @@ class RemoteRolloutProcessor(RolloutProcessor):
 
             while time.time() < deadline:
                 session = self._get_or_create_session()
-                status_result = await self._tracing_adapter.async_get_status(
-                    session,
-                    rollout_id=row.execution_metadata.rollout_id,
+                completed_logs = await self._tracing_adapter.async_search_logs(
+                    session, tags=[f"rollout_id:{row.execution_metadata.rollout_id}"]
                 )
-                status = (status_result or {}).get("status")
-                if isinstance(status, dict) and "code" in status:
-                    status_code = status["code"]
-                    if status_code == Status.Code.RUNNING:
-                        await asyncio.sleep(poll_interval)
-                        continue
+                # Filter for logs that actually have status information
+                status_logs = []
+                for log in completed_logs:
+                    status_dict = log.get("status")
+                    if status_dict and isinstance(status_dict, dict) and "code" in status_dict:
+                        status_logs.append(log)
+
+                if status_logs:
+                    if len(status_logs) > 1:
+                        logger.warning(
+                            "Found %s status logs for rollout %s; expected at most 1. Using the first one: %s",
+                            len(status_logs),
+                            row.execution_metadata.rollout_id,
+                            status_logs[0],
+                        )
+                    # Use the first log with status information
+                    status_log = status_logs[0]
+                    status_dict = status_log.get("status")
+                    raw_extras = status_log.get("extras") or {}
+                    status_extras = {
+                        k: v for k, v in raw_extras.items() if k not in ("logger_name", "level", "timestamp")
+                    }
 
                     logger.info(
-                        "Found status for rollout %s with code %s",
-                        row.execution_metadata.rollout_id,
-                        status_code,
+                        f"Found status log for rollout {row.execution_metadata.rollout_id}: {status_log.get('message', '')}"
                     )
 
-                    status_message = status.get("message", "") or ""
-                    status_details = status.get("details", []) or []
+                    status_code = status_dict.get("code")
+                    status_message = status_dict.get("message", "")
+                    status_details = status_dict.get("details", [])
 
+                    logger.info(
+                        f"Found Fireworks log for rollout {row.execution_metadata.rollout_id} with status code {status_code}"
+                    )
+
+                    # Create and raise exception if appropriate, preserving original message
                     exception = exception_for_status_code(status_code, status_message)
                     if exception is not None:
                         raise exception
@@ -152,12 +171,10 @@ class RemoteRolloutProcessor(RolloutProcessor):
                         details=status_details,
                     )
 
-                    status_extras = (status_result or {}).get("extras")
-                    if isinstance(status_extras, dict):
-                        if row.execution_metadata.extra:
-                            row.execution_metadata.extra.update(status_extras)
-                        else:
-                            row.execution_metadata.extra = status_extras
+                    if row.execution_metadata.extra:
+                        row.execution_metadata.extra.update(status_extras)
+                    else:
+                        row.execution_metadata.extra = status_extras
 
                     logger.info("Stopping polling for rollout %s", row.execution_metadata.rollout_id)
                     break

{eval_protocol-0.3.28 → eval_protocol-0.3.29/eval_protocol.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: eval-protocol
-Version: 0.3.28
+Version: 0.3.29
 Summary: The official Python SDK for Eval Protocol (EP.) EP is an open protocol that standardizes how developers author evals for large language model (LLM) applications.
 Author-email: Fireworks AI <info@fireworks.ai>
 License-Expression: MIT