fix: fix empty outputs in langchain

src/greptimeai/collector.py

@@ -201,7 +201,7 @@ def get_span_context(
 
         context_list = self._spans.get(key, [])
         if len(context_list) == 0:
-            logger.warning(f"get_span_context: { key } not found for { span_name= }")
+            logger.debug(f"get_span_context: { key } not found for { span_name= }")
             return None
 
         if not span_name:
@@ -232,7 +232,7 @@ def pop_span_context(
 
         context_list = self._spans.get(key, [])
         if len(context_list) == 0:
-            logger.warning(f"pop_span_context: { key } not found for { span_name= }")
+            logger.debug(f"pop_span_context: { key } not found for { span_name= }")
             return None
 
         target_context = None

src/greptimeai/langchain/__init__.py

@@ -4,6 +4,8 @@
 from langchain.schema.document import Document
 from langchain.schema.messages import BaseMessage
 
+from greptimeai import logger
+
 _SPAN_NAME_CHAIN = "langchain_chain"
 _SPAN_NAME_AGENT = "langchain_agent"
 _SPAN_NAME_LLM = "langchain_llm"
@@ -74,27 +76,26 @@ def _parse_output(raw_output: dict) -> Any:
     )
 
 
-def _parse_generation(gen: Generation) -> Optional[Dict[str, Any]]:
-    """
-    Generation, or ChatGeneration (which contains message field)
-    """
-    if not gen:
-        return None
-
-    info = gen.generation_info or {}
-    attrs = {
-        "text": gen.text,
-        # the following is OpenAI only?
-        "finish_reason": info.get("finish_reason"),
-        "log_probability": info.get("logprobs"),
-    }
+def _str_generations(gens: Sequence[Generation]) -> str:
+    def _str_generation(gen: Generation) -> Optional[str]:
+        """
+        Generation, or ChatGeneration (which contains message field)
+        """
+        if not gen:
+            return None
 
-    if isinstance(gen, ChatGeneration):
-        message: BaseMessage = gen.message
-        attrs["additional_kwargs"] = message.additional_kwargs
-        attrs["type"] = message.type
+        info = gen.generation_info or {}
+        reason = info.get("finish_reason")
+        if reason in ["function_call", "tool_calls"] and isinstance(
+            gen, ChatGeneration
+        ):
+            kwargs = gen.message.additional_kwargs
+            return f"{reason}: kwargs={kwargs}"
+        else:
+            return gen.text
 
-    return attrs
+    texts = list(filter(None, [_str_generation(gen) for gen in gens]))
+    return "\n".join(texts)
 
 
 def _parse_generations(
@@ -103,8 +104,33 @@ def _parse_generations(
     """
     parse LLMResult.generations[0] to structured fields
     """
+
+    def _parse_generation(gen: Generation) -> Optional[Dict[str, Any]]:
+        """
+        Generation, or ChatGeneration (which contains message field)
+        """
+        if not gen:
+            return None
+
+        gen.to_json()
+
+        info = gen.generation_info or {}
+        attrs = {
+            "text": gen.text,
+            # the following is OpenAI only?
+            "finish_reason": info.get("finish_reason"),
+            "log_probability": info.get("logprobs"),
+        }
+
+        if isinstance(gen, ChatGeneration):
+            message: BaseMessage = gen.message
+            attrs["additional_kwargs"] = message.additional_kwargs
+            attrs["type"] = message.type
+
+        return attrs
+
     if gens and len(gens) > 0:
-        return list(filter(None, [_parse_generation(gen) for gen in gens if gen]))
+        return list(filter(None, [_parse_generation(gen) for gen in gens]))
 
     return None
 
@@ -124,3 +150,4 @@ def _parse_doc(doc: Document) -> Dict[str, Any]:
         return [_parse_doc(doc) for doc in docs]
 
     return None
+    return None

src/greptimeai/langchain/callback.py

@@ -46,6 +46,7 @@
     _parse_generations,
     _parse_input,
     _parse_output,
+    _str_generations,
 )
 
 
@@ -271,8 +272,7 @@ def on_llm_end(
             if response and len(response.generations) > 0
             else []
         )
-        texts = [generation.text for generation in generations]
-        outputs = " ".join(texts)
+        outputs = _str_generations(generations)
 
         output = response.llm_output or {}
         model_name: Optional[str] = output.get("model_name")