fix(openai): structured output parsing with openai >= 1.50 (#957)

langfuse/openai.py

@@ -19,9 +19,11 @@
 
 import copy
 import logging
+from inspect import isclass
 import types
 
 from collections import defaultdict
+from dataclasses import dataclass
 from typing import List, Optional
 
 import openai.resources
@@ -34,6 +36,7 @@
 from langfuse.decorators import langfuse_context
 from langfuse.utils import _get_timestamp
 from langfuse.utils.langfuse_singleton import LangfuseSingleton
+from pydantic import BaseModel
 
 try:
     import openai
@@ -53,19 +56,14 @@
 log = logging.getLogger("langfuse")
 
 
+@dataclass
 class OpenAiDefinition:
     module: str
     object: str
     method: str
     type: str
     sync: bool
-
-    def __init__(self, module: str, object: str, method: str, type: str, sync: bool):
-        self.module = module
-        self.object = object
-        self.method = method
-        self.type = type
-        self.sync = sync
+    min_version: Optional[str] = None
 
 
 OPENAI_METHODS_V0 = [
@@ -115,6 +113,22 @@ def __init__(self, module: str, object: str, method: str, type: str, sync: bool)
         type="completion",
         sync=False,
     ),
+    OpenAiDefinition(
+        module="openai.resources.beta.chat.completions",
+        object="Completions",
+        method="parse",
+        type="chat",
+        sync=True,
+        min_version="1.50.0",
+    ),
+    OpenAiDefinition(
+        module="openai.resources.beta.chat.completions",
+        object="AsyncCompletions",
+        method="parse",
+        type="chat",
+        sync=False,
+        min_version="1.50.0",
+    ),
 ]
 
 
@@ -136,7 +150,13 @@ def __init__(
         self.args["metadata"] = (
             metadata
             if "response_format" not in kwargs
-            else {**(metadata or {}), "response_format": kwargs["response_format"]}
+            else {
+                **(metadata or {}),
+                "response_format": kwargs["response_format"].model_json_schema()
+                if isclass(kwargs["response_format"])
+                and issubclass(kwargs["response_format"], BaseModel)
+                else kwargs["response_format"],
+            }
         )
         self.args["trace_id"] = trace_id
         self.args["session_id"] = session_id
@@ -651,6 +671,11 @@ def register_tracing(self):
         resources = OPENAI_METHODS_V1 if _is_openai_v1() else OPENAI_METHODS_V0
 
         for resource in resources:
+            if resource.min_version is not None and Version(
+                openai.__version__
+            ) < Version(resource.min_version):
+                continue
+
             wrap_function_wrapper(
                 resource.module,
                 f"{resource.object}.{resource.method}",

tests/test_openai.py

@@ -1482,26 +1482,69 @@ def test_structured_output_response_format_kwarg():
 
 def test_structured_output_beta_completions_parse():
     from typing import List
+    from packaging.version import Version
 
     class CalendarEvent(BaseModel):
         name: str
         date: str
         participants: List[str]
 
-    openai.beta.chat.completions.parse(
-        model="gpt-4o-2024-08-06",
-        messages=[
+    generation_name = create_uuid()
+    api = get_api()
+
+    params = {
+        "model": "gpt-4o-2024-08-06",
+        "messages": [
             {"role": "system", "content": "Extract the event information."},
             {
                 "role": "user",
                 "content": "Alice and Bob are going to a science fair on Friday.",
             },
         ],
-        response_format=CalendarEvent,
-    )
+        "response_format": CalendarEvent,
+        "name": generation_name,
+    }
+
+    # The beta API is only wrapped for this version range. prior to that, implicitly another wrapped method was called
+    if Version(openai.__version__) < Version("1.50.0"):
+        params.pop("name")
+
+    openai.beta.chat.completions.parse(**params)
 
     openai.flush_langfuse()
 
+    if Version(openai.__version__) >= Version("1.50.0"):
+        # Check the trace and observation properties
+        generation = api.observations.get_many(name=generation_name, type="GENERATION")
+
+        assert len(generation.data) == 1
+        assert generation.data[0].name == generation_name
+        assert generation.data[0].type == "GENERATION"
+        assert generation.data[0].model == "gpt-4o-2024-08-06"
+        assert generation.data[0].start_time is not None
+        assert generation.data[0].end_time is not None
+        assert generation.data[0].start_time < generation.data[0].end_time
+
+        # Check input and output
+        assert len(generation.data[0].input) == 2
+        assert generation.data[0].input[0]["role"] == "system"
+        assert generation.data[0].input[1]["role"] == "user"
+        assert isinstance(generation.data[0].output, dict)
+        assert "name" in generation.data[0].output["content"]
+        assert "date" in generation.data[0].output["content"]
+        assert "participants" in generation.data[0].output["content"]
+
+        # Check usage
+        assert generation.data[0].usage.input is not None
+        assert generation.data[0].usage.output is not None
+        assert generation.data[0].usage.total is not None
+
+        # Check trace
+        trace = api.trace.get(generation.data[0].trace_id)
+
+        assert trace.input is not None
+        assert trace.output is not None
+
 
 @pytest.mark.asyncio
 async def test_close_async_stream():