OpenWebUI/pipelines
1
0
Fork 0
mirror of https://github.com/open-webui/pipelines synced 2025-06-04 03:47:04 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #438 from ther3zz/patch-3

Browse Source 
langfuse pipeline: chatID + clearer observation names + usage + trace tagging
This commit is contained in:
Timothy Jaeryang Baek 2025-03-28 15:37:00 -07:00 committed by GitHub
commit 5ccdecc4e3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 150 additions and 99 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

237
examples/filters/langfuse_filter_pipeline.py
View File

@ -1,8 +1,8 @@
"""
title: Langfuse Filter Pipeline
author: open-webui
date: 2025-02-20
version: 1.5
date: 2025-03-28
version: 1.7
license: MIT
description: A filter pipeline that uses Langfuse.
requirements: langfuse
@ -20,6 +20,7 @@ from langfuse.api.resources.commons.errors.unauthorized_error import Unauthorize
def get_last_assistant_message_obj(messages: List[dict]) -> dict:
"""Retrieve the last assistant message from the message list."""
for message in reversed(messages):
if message["role"] == "assistant":
return message
@ -33,6 +34,10 @@ class Pipeline:
secret_key: str
public_key: str
host: str
# New valve that controls whether task names are added as tags:
insert_tags: bool = True
# New valve that controls whether to use model name instead of model ID for generation
use_model_name_instead_of_id_for_generation: bool = False
debug: bool = False
def __init__(self):
@ -45,18 +50,21 @@ class Pipeline:
"secret_key": os.getenv("LANGFUSE_SECRET_KEY", "your-secret-key-here"),
"public_key": os.getenv("LANGFUSE_PUBLIC_KEY", "your-public-key-here"),
"host": os.getenv("LANGFUSE_HOST", "https://cloud.langfuse.com"),
"use_model_name_instead_of_id_for_generation": os.getenv("USE_MODEL_NAME", "false").lower() == "true",
"debug": os.getenv("DEBUG_MODE", "false").lower() == "true",
}
)
self.langfuse = None
# Keep track of the trace and the last-created generation for each chat_id
self.chat_traces = {}
self.chat_generations = {}
self.suppressed_logs = set()
# Dictionary to store model names for each chat
self.model_names = {}
# Only these tasks will be treated as LLM "generations":
self.GENERATION_TASKS = {"llm_response"}
def log(self, message: str, suppress_repeats: bool = False):
"""Logs messages to the terminal if debugging is enabled."""
if self.valves.debug:
if suppress_repeats:
if message in self.suppressed_logs:
@ -96,47 +104,44 @@ class Pipeline:
f"Langfuse error: {e} Please re-enter your Langfuse credentials in the pipeline settings."
)
def _build_tags(self, task_name: str) -> list:
"""
Builds a list of tags based on valve settings, ensuring we always add
'open-webui' and skip user_response / llm_response from becoming tags themselves.
"""
tags_list = []
if self.valves.insert_tags:
# Always add 'open-webui'
tags_list.append("open-webui")
# Add the task_name if it's not one of the excluded defaults
if task_name not in ["user_response", "llm_response"]:
tags_list.append(task_name)
return tags_list
async def inlet(self, body: dict, user: Optional[dict] = None) -> dict:
"""
Inlet handles the incoming request (usually a user message).
- If no trace exists yet for this chat_id, we create a new trace.
- If a trace does exist, we simply create a new generation for the new user message.
"""
if self.valves.debug:
print(f"[DEBUG] Received request: {json.dumps(body, indent=2)}")
self.log(f"Inlet function called with body: {body} and user: {user}")
metadata = body.get("metadata", {})
# ---------------------------------------------------------
# Prepend the system prompt from metadata to the system message:
model_info = metadata.get("model", {})
params_info = model_info.get("params", {})
system_prompt = params_info.get("system", "")
if system_prompt:
for msg in body["messages"]:
if msg.get("role") == "system":
# Only prepend if it hasn't already been prepended:
if not msg["content"].startswith("System Prompt:"):
msg["content"] = f"System Prompt:\n{system_prompt}\n\n{msg['content']}"
break
# ---------------------------------------------------------
# Fix SYSTEM MESSAGE prefix issue: Only apply for "task_generation"
if "chat_id" not in metadata:
if "task_generation" in metadata.get("type", "").lower():
chat_id = f"SYSTEM MESSAGE {uuid.uuid4()}"
self.log(f"Task Generation detected, assigned SYSTEM MESSAGE ID: {chat_id}")
else:
chat_id = str(uuid.uuid4()) # Regular chat messages
self.log(f"Assigned normal chat_id: {chat_id}")
chat_id = metadata.get("chat_id", str(uuid.uuid4()))
metadata["chat_id"] = chat_id
body["metadata"] = metadata
# Extract and store both model name and ID if available
model_info = metadata.get("model", {})
model_id = body.get("model")
# Store model information for this chat
if chat_id not in self.model_names:
self.model_names[chat_id] = {"id": model_id}
else:
chat_id = metadata["chat_id"]
self.model_names[chat_id]["id"] = model_id
if isinstance(model_info, dict) and "name" in model_info:
self.model_names[chat_id]["name"] = model_info["name"]
self.log(f"Stored model info - name: '{model_info['name']}', id: '{model_id}' for chat_id: {chat_id}")
required_keys = ["model", "messages"]
missing_keys = [key for key in required_keys if key not in body]
@ -146,100 +151,108 @@ class Pipeline:
raise ValueError(error_message)
user_email = user.get("email") if user else None
# Defaulting to 'user_response' if no task is provided
task_name = metadata.get("task", "user_response")
# Build tags
tags_list = self._build_tags(task_name)
# Check if we already have a trace for this chat
if chat_id not in self.chat_traces:
# Create a new trace and generation
self.log(f"Creating new chat trace for chat_id: {chat_id}")
self.log(f"Creating new trace for chat_id: {chat_id}")
trace_payload = {
"name": f"filter:{__name__}",
"name": f"chat:{chat_id}",
"input": body,
"user_id": user_email,
"metadata": {"chat_id": chat_id},
"metadata": metadata,
"session_id": chat_id,
}
if tags_list:
trace_payload["tags"] = tags_list
if self.valves.debug:
print(f"[DEBUG] Langfuse trace request: {json.dumps(trace_payload, indent=2)}")
trace = self.langfuse.trace(**trace_payload)
self.chat_traces[chat_id] = trace
else:
trace = self.chat_traces[chat_id]
self.log(f"Reusing existing trace for chat_id: {chat_id}")
if tags_list:
trace.update(tags=tags_list)
# Update metadata with type
metadata["type"] = task_name
metadata["interface"] = "open-webui"
# If it's a task that is considered an LLM generation
if task_name in self.GENERATION_TASKS:
# Determine which model value to use based on the use_model_name valve
model_id = self.model_names.get(chat_id, {}).get("id", body["model"])
model_name = self.model_names.get(chat_id, {}).get("name", "unknown")
# Pick primary model identifier based on valve setting
model_value = model_name if self.valves.use_model_name_instead_of_id_for_generation else model_id
# Add both values to metadata regardless of valve setting
metadata["model_id"] = model_id
metadata["model_name"] = model_name
generation_payload = {
"name": chat_id,
"model": body["model"],
"name": f"{task_name}:{str(uuid.uuid4())}",
"model": model_value,
"input": body["messages"],
"metadata": {"interface": "open-webui"},
"metadata": metadata,
}
if tags_list:
generation_payload["tags"] = tags_list
if self.valves.debug:
print(f"[DEBUG] Langfuse generation request: {json.dumps(generation_payload, indent=2)}")
generation = trace.generation(**generation_payload)
self.chat_traces[chat_id] = trace
self.chat_generations[chat_id] = generation
self.log(f"Trace and generation objects successfully created for chat_id: {chat_id}")
trace.generation(**generation_payload)
else:
# Re-use existing trace but create a new generation for each new message
self.log(f"Re-using existing chat trace for chat_id: {chat_id}")
trace = self.chat_traces[chat_id]
new_generation_payload = {
"name": f"{chat_id}:{str(uuid.uuid4())}",
"model": body["model"],
# Otherwise, log it as an event
event_payload = {
"name": f"{task_name}:{str(uuid.uuid4())}",
"metadata": metadata,
"input": body["messages"],
"metadata": {"interface": "open-webui"},
}
if self.valves.debug:
print(f"[DEBUG] Langfuse new_generation request: {json.dumps(new_generation_payload, indent=2)}")
if tags_list:
event_payload["tags"] = tags_list
new_generation = trace.generation(**new_generation_payload)
self.chat_generations[chat_id] = new_generation
if self.valves.debug:
print(f"[DEBUG] Langfuse event request: {json.dumps(event_payload, indent=2)}")
trace.event(**event_payload)
return body
async def outlet(self, body: dict, user: Optional[dict] = None) -> dict:
"""
Outlet handles the response body (usually the assistant message).
It will finalize/end the generation created for the user request.
"""
self.log(f"Outlet function called with body: {body}")
chat_id = body.get("chat_id")
metadata = body.get("metadata", {})
# Defaulting to 'llm_response' if no task is provided
task_name = metadata.get("task", "llm_response")
# If no trace or generation exist, attempt to register again
if chat_id not in self.chat_traces or chat_id not in self.chat_generations:
self.log(f"[WARNING] No matching chat trace found for chat_id: {chat_id}, attempting to re-register.")
# Build tags
tags_list = self._build_tags(task_name)
if chat_id not in self.chat_traces:
self.log(f"[WARNING] No matching trace found for chat_id: {chat_id}, attempting to re-register.")
# Re-run inlet to register if somehow missing
return await self.inlet(body, user)
trace = self.chat_traces[chat_id]
generation = self.chat_generations[chat_id]
# Get the last assistant message from the conversation
assistant_message = get_last_assistant_message(body["messages"])
assistant_message_obj = get_last_assistant_message_obj(body["messages"])
# ---------------------------------------------------------
# If the outlet contains a sources array, append it after the "System Prompt:"
# section in the system message:
if assistant_message_obj and "sources" in assistant_message_obj and assistant_message_obj["sources"]:
for msg in body["messages"]:
if msg.get("role") == "system":
if msg["content"].startswith("System Prompt:"):
# Format the sources nicely
sources_str = "\n\n".join(
json.dumps(src, indent=2) for src in assistant_message_obj["sources"]
)
msg["content"] += f"\n\nSources:\n{sources_str}"
break
# ---------------------------------------------------------
# Extract usage if available
usage = None
if assistant_message_obj:
info = assistant_message_obj.get("info", {})
info = assistant_message_obj.get("usage", {})
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")
@ -251,20 +264,58 @@ class Pipeline:
}
self.log(f"Usage data extracted: {usage}")
# Optionally update the trace with the final assistant output
# Update the trace output with the last assistant message
trace.update(output=assistant_message)
# End the generation with the final assistant message and updated conversation
metadata["type"] = task_name
metadata["interface"] = "open-webui"
if task_name in self.GENERATION_TASKS:
# Determine which model value to use based on the use_model_name valve
model_id = self.model_names.get(chat_id, {}).get("id", body.get("model"))
model_name = self.model_names.get(chat_id, {}).get("name", "unknown")
# Pick primary model identifier based on valve setting
model_value = model_name if self.valves.use_model_name_instead_of_id_for_generation else model_id
# Add both values to metadata regardless of valve setting
metadata["model_id"] = model_id
metadata["model_name"] = model_name
# If it's an LLM generation
generation_payload = {
"input": body["messages"], # include the entire conversation
"metadata": {"interface": "open-webui"},
"name": f"{task_name}:{str(uuid.uuid4())}",
"model": model_value, # <-- Use model name or ID based on valve setting
"input": body["messages"],
"metadata": metadata,
"usage": usage,
}
if tags_list:
generation_payload["tags"] = tags_list
if self.valves.debug:
print(f"[DEBUG] Langfuse generation end request: {json.dumps(generation_payload, indent=2)}")
generation.end(**generation_payload)
trace.generation().end(**generation_payload)
self.log(f"Generation ended for chat_id: {chat_id}")
else:
# Otherwise log as an event
event_payload = {
"name": f"{task_name}:{str(uuid.uuid4())}",
"metadata": metadata,
"input": body["messages"],
}
if usage:
# If you want usage on event as well
event_payload["metadata"]["usage"] = usage
if tags_list:
event_payload["tags"] = tags_list
if self.valves.debug:
print(f"[DEBUG] Langfuse event end request: {json.dumps(event_payload, indent=2)}")
trace.event(**event_payload)
self.log(f"Event logged for chat_id: {chat_id}")
return body