eval-protocol 0.2.76__tar.gz → 0.2.77__tar.gz

{eval_protocol-0.2.76/eval_protocol.egg-info → eval_protocol-0.2.77}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: eval-protocol
-Version: 0.2.76
+Version: 0.2.77
 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-0.2.76 → eval_protocol-0.2.77}/eval_protocol/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2025-11-02T23:42:16-0800",
+ "date": "2025-11-03T11:31:21-0800",
  "dirty": false,
  "error": null,
- "full-revisionid": "bb86fef82e6ddf66f686a71a5a80d600b9dd620c",
- "version": "0.2.76"
+ "full-revisionid": "bb05e7bbe71024a49a3e23f34bcabfc6d4b835df",
+ "version": "0.2.77"
 }
 '''  # END VERSION_JSON
 

{eval_protocol-0.2.76 → eval_protocol-0.2.77}/eval_protocol/mcp/mcp_multi_client.py

@@ -70,38 +70,6 @@ class MCPMultiClient:
                 f"Please set these variables in your environment or .env file."
             )
 
-    def _process_headers(self, headers: Dict[str, str]) -> Dict[str, str]:
-        """Process headers by substituting environment variables.
-
-        Supports environment variable substitution in the format:
-        - ${ENV_VAR} or $ENV_VAR for environment variables
-        - Raw strings are passed through unchanged
-
-        Example:
-            {"Authorization": "Bearer ${API_KEY}"}
-            -> {"Authorization": "Bearer abc123"} (if API_KEY=abc123)
-        """
-        import re
-
-        processed_headers = {}
-        for key, value in headers.items():
-            # Match ${VAR} or $VAR patterns
-            def replace_env_var(match):
-                var_name = match.group(1) or match.group(2)
-                env_value = os.environ.get(var_name)
-                if env_value is None:
-                    raise ValueError(
-                        f"Environment variable '{var_name}' referenced in header '{key}' "
-                        f"is not set. Please set it in your environment or .env file."
-                    )
-                return env_value
-
-            # Replace ${VAR} or $VAR with environment variable value
-            processed_value = re.sub(r"\$\{([^}]+)\}|\$([A-Za-z_][A-Za-z0-9_]*)", replace_env_var, value)
-            processed_headers[key] = processed_value
-
-        return processed_headers
-
     async def connect_to_servers(self):
         """Connect to all configured MCP servers"""
         if not self.config.mcpServers:
@@ -143,17 +111,8 @@ class MCPMultiClient:
             if not url:
                 raise ValueError(f"Server '{server_name}' must have a 'url' specified")
 
-            # Build headers only from the authorization field (Responses-style)
-            processed_headers: Dict[str, str] = {}
-            auth_token = getattr(server_config, "authorization", None)
-            if auth_token:
-                # Support env substitution in the authorization value as well
-                processed_headers = self._process_headers({"Authorization": auth_token})
-
-            # Connect using streamable HTTP client with auth headers
-            http_transport = await self.exit_stack.enter_async_context(
-                streamablehttp_client(url, headers=processed_headers)
-            )
+            # Connect using streamable HTTP client - manage resources manually
+            http_transport = await self.exit_stack.enter_async_context(streamablehttp_client(url))
             read_stream, write_stream, get_session_id = http_transport
             session = await self.exit_stack.enter_async_context(ClientSession(read_stream, write_stream))
         else:

{eval_protocol-0.2.76 → eval_protocol-0.2.77}/eval_protocol/models.py

@@ -1119,7 +1119,6 @@ class MCPConfigurationServerUrl(BaseModel):
     """Represents a Remote MCP configuration server."""
 
     url: str  # url to the MCP server
-    authorization: Optional[str] = None
 
 
 class MCPMultiClientConfiguration(BaseModel):

{eval_protocol-0.2.76 → eval_protocol-0.2.77}/eval_protocol/pytest/default_agent_rollout_processor.py

@@ -133,11 +133,8 @@ class Agent:
 
     async def _call_model(self, messages: list[Message], tools: Optional[List[dict[str, Any]]]) -> Message:
         # Convert Message models to plain dicts for LLM call
-        # Filter out fields that are not supported by OpenAI/LiteLLM APIs (e.g., weight, control_plane_step, reasoning_content)
         messages_payload: List[Dict[str, Any]] = [
-            message.dump_mdoel_for_chat_completion_request()
-            if hasattr(message, "dump_mdoel_for_chat_completion_request")
-            else (message.model_dump() if hasattr(message, "model_dump") else message)  # type: ignore[misc]
+            message.model_dump() if hasattr(message, "model_dump") else message  # type: ignore[misc]
             for message in messages
         ]
         # Normalize tool definitions into OpenAI-compatible dicts

{eval_protocol-0.2.76 → eval_protocol-0.2.77}/eval_protocol/pytest/default_single_turn_rollout_process.py

@@ -48,9 +48,7 @@ class SingleTurnRolloutProcessor(RolloutProcessor):
                 while messages_for_request and messages_for_request[-1].role == "assistant":
                     messages_for_request.pop()
 
-            # Filter out fields that are not supported by OpenAI/LiteLLM APIs (e.g., weight, control_plane_step, reasoning_content)
-            # Use the Message class method that excludes unsupported fields
-            messages_payload = [message.dump_mdoel_for_chat_completion_request() for message in messages_for_request]
+            messages_payload = [message.model_dump() for message in messages_for_request]
 
             request_params = {"messages": messages_payload, **config.completion_params}
             # Ensure caching is disabled only for this request (review feedback)

{eval_protocol-0.2.76 → eval_protocol-0.2.77}/eval_protocol/pytest/remote_rollout_processor.py

@@ -123,7 +123,7 @@ class RemoteRolloutProcessor(RolloutProcessor):
                 except requests.exceptions.HTTPError as e:
                     if e.response is not None and e.response.status_code == 404:
                         # 404 means server doesn't implement /status endpoint, stop polling
-                        logger.info(
+                        logger.debug(
                             f"Server doesn't implement /status endpoint (404), stopping status polling for rollout {row.execution_metadata.rollout_id}"
                         )
                         continue_polling_status = False

{eval_protocol-0.2.76 → eval_protocol-0.2.77}/eval_protocol/pytest/tracing_utils.py

@@ -101,14 +101,11 @@ def build_init_request(
     completion_params_base_url: Optional[str] = completion_params_dict.get("base_url")
 
     # Strip non-OpenAI fields from messages
-    # Use dump_mdoel_for_chat_completion_request() to automatically exclude unsupported fields (weight, control_plane_step, reasoning_content)
+    allowed_message_fields = {"role", "content", "tool_calls", "tool_call_id", "name"}
     clean_messages = []
     for m in row.messages:
         md: Dict[str, Any]
-        if hasattr(m, "dump_mdoel_for_chat_completion_request"):
-            # Use the Message method that automatically filters unsupported fields
-            md = m.dump_mdoel_for_chat_completion_request()
-        elif hasattr(m, "model_dump"):
+        if hasattr(m, "model_dump"):
             md = m.model_dump()
         elif isinstance(m, dict):
             md = m
@@ -121,8 +118,6 @@ def build_init_request(
                 "tool_call_id": getattr(m, "tool_call_id", None),
                 "name": getattr(m, "name", None),
             }
-        # Additional filtering to ensure only allowed fields are kept (already handled by dump_mdoel_for_chat_completion_request for Message objects)
-        allowed_message_fields = {"role", "content", "tool_calls", "tool_call_id", "name"}
         clean_messages.append({k: v for k, v in md.items() if k in allowed_message_fields and v is not None})
 
     # Build final model base URL with tracing metadata

{eval_protocol-0.2.76 → eval_protocol-0.2.77/eval_protocol.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: eval-protocol
-Version: 0.2.76
+Version: 0.2.77
 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-0.2.76 → eval_protocol-0.2.77}/eval_protocol.egg-info/SOURCES.txt

@@ -316,7 +316,6 @@ tests/test_list_comparison_math_reward.py
 tests/test_logs_server.py
 tests/test_logs_server_simple.py
 tests/test_math.py
-tests/test_message_field_filtering.py
 tests/test_minimal.py
 tests/test_models.py
 tests/test_models_rl.py

eval_protocol-0.2.76/tests/test_message_field_filtering.py

@@ -1,64 +0,0 @@
-"""
-Test to verify that message fields are properly filtered before sending to API.
-
-This test verifies that unsupported fields like 'weight', 'control_plane_step',
-and 'reasoning_content' are excluded from messages when preparing API requests.
-"""
-
-from eval_protocol.models import Message
-
-
-def test_dump_model_excludes_unsupported_fields():
-    """Test that dump_mdoel_for_chat_completion_request excludes unsupported fields."""
-    # Create a message with all possible fields including unsupported ones
-    message = Message(
-        role="user",
-        content="Hello",
-        weight=0,
-        control_plane_step={"step": 1},
-        reasoning_content="Some reasoning",
-        name="test_user",
-    )
-
-    # Get the filtered dictionary
-    filtered = message.dump_mdoel_for_chat_completion_request()
-
-    # Verify unsupported fields are excluded
-    assert "weight" not in filtered, "weight field should be excluded"
-    assert "control_plane_step" not in filtered, "control_plane_step field should be excluded"
-    assert "reasoning_content" not in filtered, "reasoning_content field should be excluded"
-
-    # Verify supported fields are included
-    assert "role" in filtered, "role field should be included"
-    assert "content" in filtered, "content field should be included"
-    assert filtered["role"] == "user"
-    assert filtered["content"] == "Hello"
-
-    # Verify name is included (it's a supported field for tool calls)
-    assert "name" in filtered
-    assert filtered["name"] == "test_user"
-
-
-def test_dump_model_with_only_supported_fields():
-    """Test that supported fields are preserved."""
-    message = Message(
-        role="assistant",
-        content="I can help you",
-        tool_calls=None,
-        tool_call_id=None,
-    )
-
-    filtered = message.dump_mdoel_for_chat_completion_request()
-
-    # Should only contain supported fields
-    assert filtered["role"] == "assistant"
-    assert filtered["content"] == "I can help you"
-
-    # Should not contain unsupported fields even if None
-    assert "weight" not in filtered
-
-
-if __name__ == "__main__":
-    import pytest
-
-    pytest.main([__file__, "-v"])