Merge pull request #473 from Scale3-Labs/release-3.5.0

Release 3.5.0

Author: alizenhom

src/langtrace_python_sdk/constants/instrumentation/common.py

@@ -30,7 +30,7 @@
     "QDRANT": "Qdrant",
     "WEAVIATE": "Weaviate",
     "OLLAMA": "Ollama",
-    "VERTEXAI": "VertexAI",
+    "VERTEXAI": "Vertex AI",
     "GEMINI": "Gemini",
     "MISTRAL": "Mistral",
     "EMBEDCHAIN": "Embedchain",

src/langtrace_python_sdk/constants/instrumentation/vertexai.py

@@ -39,4 +39,24 @@
         "method": "ChatSession",
         "operation": "send_message_streaming",
     },
+    "PREDICTION_SERVICE_BETA_GENERATE_CONTENT": {
+        "module": "google.cloud.aiplatform_v1beta1.services.prediction_service.client",
+        "method": "PredictionServiceClient",
+        "operation": "generate_content",
+    },
+    "PREDICTION_SERVICE_GENERATE_CONTENT": {
+        "module": "google.cloud.aiplatform_v1.services.prediction_service.client",
+        "method": "PredictionServiceClient",
+        "operation": "generate_content",
+    },
+    "PREDICTION_SERVICE_BETA_STREAM_GENERATE_CONTENT": {
+        "module": "google.cloud.aiplatform_v1beta1.services.prediction_service.client",
+        "method": "PredictionServiceClient",
+        "operation": "stream_generate_content",
+    },
+    "PREDICTION_SERVICE_STREAM_GENERATE_CONTENT": {
+        "module": "google.cloud.aiplatform_v1.services.prediction_service.client",
+        "method": "PredictionServiceClient",
+        "operation": "stream_generate_content",
+    },
 }

src/langtrace_python_sdk/instrumentation/vertexai/patch.py

@@ -27,12 +27,13 @@ def patch_vertexai(name, version, tracer: Tracer):
     def traced_method(wrapped, instance, args, kwargs):
         service_provider = SERVICE_PROVIDERS["VERTEXAI"]
         prompts = serialize_prompts(args, kwargs)
+
         span_attributes = {
             **get_langtrace_attributes(version, service_provider),
             **get_llm_request_attributes(
                 kwargs,
                 prompts=prompts,
-                model=get_llm_model(instance),
+                model=get_llm_model(instance, kwargs),
             ),
             **get_llm_url(instance),
             SpanAttributes.LLM_PATH: "",
@@ -77,6 +78,10 @@ def set_response_attributes(span: Span, result):
     if hasattr(result, "text"):
         set_event_completion(span, [{"role": "assistant", "content": result.text}])
 
+    if hasattr(result, "candidates"):
+        parts = result.candidates[0].content.parts
+        set_event_completion(span, [{"role": "assistant", "content": parts[0].text}])
+
     if hasattr(result, "usage_metadata") and result.usage_metadata is not None:
         usage = result.usage_metadata
         input_tokens = usage.prompt_token_count
@@ -96,17 +101,23 @@ def set_response_attributes(span: Span, result):
 
 
 def is_streaming_response(response):
-    return isinstance(response, types.GeneratorType) or isinstance(
-        response, types.AsyncGeneratorType
+    return (
+        isinstance(response, types.GeneratorType)
+        or isinstance(response, types.AsyncGeneratorType)
+        or str(type(response).__name__) == "_StreamingResponseIterator"
     )
 
 
-def get_llm_model(instance):
+def get_llm_model(instance, kwargs):
+    if "request" in kwargs:
+        return kwargs.get("request").model.split("/")[-1]
+
     if hasattr(instance, "_model_name"):
         return instance._model_name.replace("publishers/google/models/", "")
     return getattr(instance, "_model_id", "unknown")
 
 
+@silently_fail
 def serialize_prompts(args, kwargs):
     if args and len(args) > 0:
         prompt_parts = []
@@ -122,5 +133,24 @@ def serialize_prompts(args, kwargs):
 
         return [{"role": "user", "content": "\n".join(prompt_parts)}]
     else:
-        content = kwargs.get("prompt") or kwargs.get("message")
-        return [{"role": "user", "content": content}] if content else []
+        # Handle PredictionServiceClient for google-cloud-aiplatform.
+        if "request" in kwargs:
+            prompt = []
+            prompt_body = kwargs.get("request")
+            if prompt_body.system_instruction:
+                for part in prompt_body.system_instruction.parts:
+                    prompt.append({"role": "system", "content": part.text})
+
+            contents = prompt_body.contents
+
+            if not contents:
+                return []
+
+            for c in contents:
+                role = c.role if c.role else "user"
+                content = c.parts[0].text if c.parts else ""
+                prompt.append({"role": role, "content": content})
+            return prompt
+        else:
+            content = kwargs.get("prompt") or kwargs.get("message")
+            return [{"role": "user", "content": content}] if content else []

src/langtrace_python_sdk/utils/llm.py

@@ -395,18 +395,30 @@ def build_streaming_response(self, chunk):
             content = [chunk.text]
 
         # CohereV2
-        if (hasattr(chunk, "delta") and
-            chunk.delta is not None and
-            hasattr(chunk.delta, "message") and
-            chunk.delta.message is not None and
-            hasattr(chunk.delta.message, "content") and
-            chunk.delta.message.content is not None and
-            hasattr(chunk.delta.message.content, "text") and
-            chunk.delta.message.content.text is not None):
+        if (
+            hasattr(chunk, "delta")
+            and chunk.delta is not None
+            and hasattr(chunk.delta, "message")
+            and chunk.delta.message is not None
+            and hasattr(chunk.delta.message, "content")
+            and chunk.delta.message.content is not None
+            and hasattr(chunk.delta.message.content, "text")
+            and chunk.delta.message.content.text is not None
+        ):
             content = [chunk.delta.message.content.text]
-
+        # google-cloud-aiplatform
+        if hasattr(chunk, "candidates") and chunk.candidates is not None:
+            for candidate in chunk.candidates:
+                if hasattr(candidate, "content") and candidate.content is not None:
+                    for part in candidate.content.parts:
+                        if hasattr(part, "text") and part.text is not None:
+                            content.append(part.text)
         # Anthropic
-        if hasattr(chunk, "delta") and chunk.delta is not None and not hasattr(chunk.delta, "message"):
+        if (
+            hasattr(chunk, "delta")
+            and chunk.delta is not None
+            and not hasattr(chunk.delta, "message")
+        ):
             content = [chunk.delta.text] if hasattr(chunk.delta, "text") else []
 
         if isinstance(chunk, dict):
@@ -425,9 +437,14 @@ def set_usage_attributes(self, chunk):
 
         # CohereV2
         if hasattr(chunk, "type") and chunk.type == "message-end":
-            if (hasattr(chunk, "delta") and chunk.delta is not None and
-                hasattr(chunk.delta, "usage") and chunk.delta.usage is not None and
-                hasattr(chunk.delta.usage, "billed_units") and chunk.delta.usage.billed_units is not None):
+            if (
+                hasattr(chunk, "delta")
+                and chunk.delta is not None
+                and hasattr(chunk.delta, "usage")
+                and chunk.delta.usage is not None
+                and hasattr(chunk.delta.usage, "billed_units")
+                and chunk.delta.usage.billed_units is not None
+            ):
                 usage = chunk.delta.usage.billed_units
                 self.completion_tokens = int(usage.output_tokens)
                 self.prompt_tokens = int(usage.input_tokens)

src/langtrace_python_sdk/version.py

@@ -1 +1 @@
-__version__ = "3.4.0"
+__version__ = "3.5.0"