Fixes traceloop#535 by validating for LegacyAPIResponse or anything w…

…hich has the .parse() method and then recursively calling the result so that span attributes are still captured

packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/__init__.py

@@ -166,5 +166,9 @@ def is_streaming_response(response):
 def model_as_dict(model):
     if version("pydantic") < "2.0.0":
         return model.dict()
-
-    return model.model_dump()
+    if hasattr(model, "model_dump"):
+        return model.model_dump()
+    elif hasattr(model, "parse"):  # Raw API response
+        return model_as_dict(model.parse())
+    else:
+        return model

packages/opentelemetry-instrumentation-openai/tests/test_embeddings.py

@@ -25,7 +25,7 @@ def test_embeddings(exporter, openai_client):
 
 @pytest.mark.vcr
 def test_embeddings_with_raw_response(exporter, openai_client):
-    openai_client.embeddings.with_raw_response.create(
+    response = openai_client.embeddings.with_raw_response.create(
         input="Tell me a joke about opentelemetry",
         model="text-embedding-ada-002",
     )
@@ -39,6 +39,13 @@ def test_embeddings_with_raw_response(exporter, openai_client):
         == "Tell me a joke about opentelemetry"
     )
 
+    assert open_ai_span.attributes["llm.request.model"] == "text-embedding-ada-002"
+    assert open_ai_span.attributes["llm.usage.prompt_tokens"] == 8
+    assert open_ai_span.attributes["openai.api_base"] == "https://api.openai.com/v1/"
+
+    parsed_response = response.parse()
+    assert parsed_response.data[0]
+
 
 @pytest.mark.vcr
 def test_azure_openai_embeddings(exporter):