Merge pull request #277 from marcklingen/fix-langfuse-filter

fix: langfuse filter pipeline cost tracking

examples/filters/langfuse_filter_pipeline.py

@@ -1,23 +1,28 @@
 """
 title: Langfuse Filter Pipeline
 author: open-webui
-date: 2024-05-30
-version: 1.3
+date: 2024-09-27
+version: 1.4
 license: MIT
 description: A filter pipeline that uses Langfuse.
 requirements: langfuse
 """
 
 from typing import List, Optional
-from schemas import OpenAIChatMessage
 import os
 import uuid
 
-from utils.pipelines.main import get_last_user_message, get_last_assistant_message
+from utils.pipelines.main import get_last_assistant_message
 from pydantic import BaseModel
 from langfuse import Langfuse
 from langfuse.api.resources.commons.errors.unauthorized_error import UnauthorizedError
 
+def get_last_assistant_message_obj(messages: List[dict]) -> dict:
+    for message in reversed(messages):
+        if message["role"] == "assistant":
+            return message
+    return {}
+
 
 class Pipeline:
     class Valves(BaseModel):
@@ -109,21 +114,28 @@ async def inlet(self, body: dict, user: Optional[dict] = None) -> dict:
 
     async def outlet(self, body: dict, user: Optional[dict] = None) -> dict:
         print(f"outlet:{__name__}")
+        print(f"Received body: {body}")
         if body["chat_id"] not in self.chat_generations:
             return body
 
         generation = self.chat_generations[body["chat_id"]]
         assistant_message = get_last_assistant_message(body["messages"])
 
-        # Extract usage information
-        info = assistant_message.get("info", {})
+
+        # Extract usage information for models that support it
         usage = None
-        if "prompt_tokens" in info and "completion_tokens" in info:
-            usage = {
-                "input": info["prompt_tokens"],
-                "output": info["completion_tokens"],
-                "unit": "TOKENS",
-            }
+        assistant_message_obj = get_last_assistant_message_obj(body["messages"])
+        if assistant_message_obj:
+            info = assistant_message_obj.get("info", {})
+            if isinstance(info, dict):
+                input_tokens = info.get("prompt_eval_count") or info.get("prompt_tokens")
+                output_tokens = info.get("eval_count") or info.get("completion_tokens")
+                if input_tokens is not None and output_tokens is not None:
+                    usage = {
+                        "input": input_tokens,
+                        "output": output_tokens,
+                        "unit": "TOKENS",
+                    }
 
         # Update generation
         generation.end(