Merge pull request #1653 from open-webui/litellm-as-subprocess

fix: litellm as subprocess

backend/apps/litellm/main.py

-import logging
-
-from litellm.proxy.proxy_server import ProxyConfig, initialize
-from litellm.proxy.proxy_server import app
+from fastapi import FastAPI, Depends, HTTPException
+from fastapi.routing import APIRoute
+from fastapi.middleware.cors import CORSMiddleware
 
+import logging
 from fastapi import FastAPI, Request, Depends, status, Response
 from fastapi.responses import JSONResponse
 
 from starlette.middleware.base import BaseHTTPMiddleware, RequestResponseEndpoint
 from starlette.responses import StreamingResponse
 import json
+import time
+import requests
 
-from utils.utils import get_http_authorization_cred, get_current_user
+from pydantic import BaseModel
+from typing import Optional, List
+
+from utils.utils import get_verified_user, get_current_user, get_admin_user
 from config import SRC_LOG_LEVELS, ENV
+from constants import MESSAGES
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["LITELLM"])
 
 
-from config import (
-    MODEL_FILTER_ENABLED,
-    MODEL_FILTER_LIST,
+from config import MODEL_FILTER_ENABLED, MODEL_FILTER_LIST, DATA_DIR
+
+
+import asyncio
+import subprocess
+import yaml
+
+app = FastAPI()
+
+origins = ["*"]
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=origins,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
 )
 
 
-proxy_config = ProxyConfig()
+LITELLM_CONFIG_DIR = f"{DATA_DIR}/litellm/config.yaml"
 
+with open(LITELLM_CONFIG_DIR, "r") as file:
+    litellm_config = yaml.safe_load(file)
 
-async def config():
-    router, model_list, general_settings = await proxy_config.load_config(
-        router=None, config_file_path="./data/litellm/config.yaml"
+app.state.CONFIG = litellm_config
+
+# Global variable to store the subprocess reference
+background_process = None
+
+
+async def run_background_process(command):
+    global background_process
+    log.info("run_background_process")
+
+    try:
+        # Log the command to be executed
+        log.info(f"Executing command: {command}")
+        # Execute the command and create a subprocess
+        process = await asyncio.create_subprocess_exec(
+            *command.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE
         )
+        background_process = process
+        log.info("Subprocess started successfully.")
+
+        # Capture STDERR for debugging purposes
+        stderr_output = await process.stderr.read()
+        stderr_text = stderr_output.decode().strip()
+        if stderr_text:
+            log.info(f"Subprocess STDERR: {stderr_text}")
+
+        # log.info output line by line
+        async for line in process.stdout:
+            log.info(line.decode().strip())
+
+        # Wait for the process to finish
+        returncode = await process.wait()
+        log.info(f"Subprocess exited with return code {returncode}")
+    except Exception as e:
+        log.error(f"Failed to start subprocess: {e}")
+        raise  # Optionally re-raise the exception if you want it to propagate
 
-    await initialize(config="./data/litellm/config.yaml", telemetry=False)
 
+async def start_litellm_background():
+    log.info("start_litellm_background")
+    # Command to run in the background
+    command = (
+        "litellm --port 14365 --telemetry False --config ./data/litellm/config.yaml"
+    )
+
+    await run_background_process(command)
 
-async def startup():
-    await config()
+
+async def shutdown_litellm_background():
+    log.info("shutdown_litellm_background")
+    global background_process
+    if background_process:
+        background_process.terminate()
+        await background_process.wait()  # Ensure the process has terminated
+        log.info("Subprocess terminated")
+        background_process = None
 
 
 @app.on_event("startup")
-async def on_startup():
-    await startup()
+async def startup_event():
+
+    log.info("startup_event")
+    # TODO: Check config.yaml file and create one
+    asyncio.create_task(start_litellm_background())
 
 
 app.state.MODEL_FILTER_ENABLED = MODEL_FILTER_ENABLED
 app.state.MODEL_FILTER_LIST = MODEL_FILTER_LIST
 
 
-@app.middleware("http")
-async def auth_middleware(request: Request, call_next):
-    auth_header = request.headers.get("Authorization", "")
-    request.state.user = None
+@app.get("/")
+async def get_status():
+    return {"status": True}
 
+
+async def restart_litellm():
+    """
+    Endpoint to restart the litellm background service.
+    """
+    log.info("Requested restart of litellm service.")
     try:
-        user = get_current_user(get_http_authorization_cred(auth_header))
-        log.debug(f"user: {user}")
-        request.state.user = user
+        # Shut down the existing process if it is running
+        await shutdown_litellm_background()
+        log.info("litellm service shutdown complete.")
+
+        # Restart the background service
+
+        asyncio.create_task(start_litellm_background())
+        log.info("litellm service restart complete.")
+
+        return {
+            "status": "success",
+            "message": "litellm service restarted successfully.",
+        }
     except Exception as e:
-        return JSONResponse(status_code=400, content={"detail": str(e)})
+        log.info(f"Error restarting litellm service: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, detail=str(e)
+        )
+
 
-    response = await call_next(request)
-    return response
+@app.get("/restart")
+async def restart_litellm_handler(user=Depends(get_admin_user)):
+    return await restart_litellm()
 
 
-class ModifyModelsResponseMiddleware(BaseHTTPMiddleware):
-    async def dispatch(
-        self, request: Request, call_next: RequestResponseEndpoint
-    ) -> Response:
+@app.get("/config")
+async def get_config(user=Depends(get_admin_user)):
+    return app.state.CONFIG
 
-        response = await call_next(request)
-        user = request.state.user
 
-        if "/models" in request.url.path:
-            if isinstance(response, StreamingResponse):
-                # Read the content of the streaming response
-                body = b""
-                async for chunk in response.body_iterator:
-                    body += chunk
+class LiteLLMConfigForm(BaseModel):
+    general_settings: Optional[dict] = None
+    litellm_settings: Optional[dict] = None
+    model_list: Optional[List[dict]] = None
+    router_settings: Optional[dict] = None
+
+
+@app.post("/config/update")
+async def update_config(form_data: LiteLLMConfigForm, user=Depends(get_admin_user)):
+    app.state.CONFIG = form_data.model_dump(exclude_none=True)
+
+    with open(LITELLM_CONFIG_DIR, "w") as file:
+        yaml.dump(app.state.CONFIG, file)
+
+    await restart_litellm()
+    return app.state.CONFIG
+
+
+@app.get("/models")
+@app.get("/v1/models")
+async def get_models(user=Depends(get_current_user)):
+    while not background_process:
+        await asyncio.sleep(0.1)
+
+    url = "http://localhost:14365/v1"
+    r = None
+    try:
+        r = requests.request(method="GET", url=f"{url}/models")
+        r.raise_for_status()
 
-                data = json.loads(body.decode("utf-8"))
+        data = r.json()
 
         if app.state.MODEL_FILTER_ENABLED:
             if user and user.role == "user":
                 data["data"] = list(
                     filter(
-                                lambda model: model["id"]
-                                in app.state.MODEL_FILTER_LIST,
+                        lambda model: model["id"] in app.state.MODEL_FILTER_LIST,
                         data["data"],
                     )
                 )
 
-                # Modified Flag
-                data["modified"] = True
-                return JSONResponse(content=data)
+        return data
+    except Exception as e:
+
+        log.exception(e)
+        error_detail = "Open WebUI: Server Connection Error"
+        if r is not None:
+            try:
+                res = r.json()
+                if "error" in res:
+                    error_detail = f"External: {res['error']}"
+            except:
+                error_detail = f"External: {e}"
+
+        return {
+            "data": [
+                {
+                    "id": model["model_name"],
+                    "object": "model",
+                    "created": int(time.time()),
+                    "owned_by": "openai",
+                }
+                for model in app.state.CONFIG["model_list"]
+            ],
+            "object": "list",
+        }
+
+
+@app.get("/model/info")
+async def get_model_list(user=Depends(get_admin_user)):
+    return {"data": app.state.CONFIG["model_list"]}
+
+
+class AddLiteLLMModelForm(BaseModel):
+    model_name: str
+    litellm_params: dict
+
+
+@app.post("/model/new")
+async def add_model_to_config(
+    form_data: AddLiteLLMModelForm, user=Depends(get_admin_user)
+):
+    # TODO: Validate model form
+
+    app.state.CONFIG["model_list"].append(form_data.model_dump())
+
+    with open(LITELLM_CONFIG_DIR, "w") as file:
+        yaml.dump(app.state.CONFIG, file)
+
+    await restart_litellm()
+
+    return {"message": MESSAGES.MODEL_ADDED(form_data.model_name)}
+
+
+class DeleteLiteLLMModelForm(BaseModel):
+    id: str
+
 
-        return response
+@app.post("/model/delete")
+async def delete_model_from_config(
+    form_data: DeleteLiteLLMModelForm, user=Depends(get_admin_user)
+):
+    app.state.CONFIG["model_list"] = [
+        model
+        for model in app.state.CONFIG["model_list"]
+        if model["model_name"] != form_data.id
+    ]
 
+    with open(LITELLM_CONFIG_DIR, "w") as file:
+        yaml.dump(app.state.CONFIG, file)
 
-app.add_middleware(ModifyModelsResponseMiddleware)
+    await restart_litellm()
+
+    return {"message": MESSAGES.MODEL_DELETED(form_data.id)}
+
+
+@app.api_route("/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])
+async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
+    body = await request.body()
+
+    url = "http://localhost:14365"
+
+    target_url = f"{url}/{path}"
+
+    headers = {}
+    # headers["Authorization"] = f"Bearer {key}"
+    headers["Content-Type"] = "application/json"
+
+    r = None
+
+    try:
+        r = requests.request(
+            method=request.method,
+            url=target_url,
+            data=body,
+            headers=headers,
+            stream=True,
+        )
+
+        r.raise_for_status()
+
+        # Check if response is SSE
+        if "text/event-stream" in r.headers.get("Content-Type", ""):
+            return StreamingResponse(
+                r.iter_content(chunk_size=8192),
+                status_code=r.status_code,
+                headers=dict(r.headers),
+            )
+        else:
+            response_data = r.json()
+            return response_data
+    except Exception as e:
+        log.exception(e)
+        error_detail = "Open WebUI: Server Connection Error"
+        if r is not None:
+            try:
+                res = r.json()
+                if "error" in res:
+                    error_detail = f"External: {res['error']['message'] if 'message' in res['error'] else res['error']}"
+            except:
+                error_detail = f"External: {e}"
+
+        raise HTTPException(
+            status_code=r.status_code if r else 500, detail=error_detail
+        )

backend/constants.py

@@ -3,6 +3,10 @@ from enum import Enum
 
 class MESSAGES(str, Enum):
     DEFAULT = lambda msg="": f"{msg if msg else ''}"
+    MODEL_ADDED = lambda model="": f"The model '{model}' has been added successfully."
+    MODEL_DELETED = (
+        lambda model="": f"The model '{model}' has been deleted successfully."
+    )
 
 
 class WEBHOOK_MESSAGES(str, Enum):

backend/main.py

@@ -20,12 +20,17 @@ from starlette.middleware.base import BaseHTTPMiddleware
 from apps.ollama.main import app as ollama_app
 from apps.openai.main import app as openai_app
 
-from apps.litellm.main import app as litellm_app, startup as litellm_app_startup
+from apps.litellm.main import (
+    app as litellm_app,
+    start_litellm_background,
+    shutdown_litellm_background,
+)
 from apps.audio.main import app as audio_app
 from apps.images.main import app as images_app
 from apps.rag.main import app as rag_app
 from apps.web.main import app as webui_app
 
+import asyncio
 from pydantic import BaseModel
 from typing import List
 
@@ -170,7 +175,7 @@ async def check_url(request: Request, call_next):
 
 @app.on_event("startup")
 async def on_startup():
-    await litellm_app_startup()
+    asyncio.create_task(start_litellm_background())
 
 
 app.mount("/api/v1", webui_app)
@@ -315,3 +320,8 @@ app.mount(
     SPAStaticFiles(directory=FRONTEND_BUILD_DIR, html=True),
     name="spa-static-files",
 )
+
+
+@app.on_event("shutdown")
+async def shutdown_event():
+    await shutdown_litellm_background()

backend/requirements.txt

@@ -17,7 +17,9 @@ peewee
 peewee-migrate
 bcrypt
 
-litellm==1.30.7
+litellm==1.35.17
+litellm[proxy]==1.35.17
+
 boto3
 
 argon2-cffi

src/lib/components/chat/Settings/Models.svelte

@@ -35,7 +35,7 @@
 	let liteLLMRPM = '';
 	let liteLLMMaxTokens = '';
 
-	let deleteLiteLLMModelId = '';
+	let deleteLiteLLMModelName = '';
 
 	$: liteLLMModelName = liteLLMModel;
 
@@ -472,7 +472,7 @@
 	};
 
 	const deleteLiteLLMModelHandler = async () => {
-		const res = await deleteLiteLLMModel(localStorage.token, deleteLiteLLMModelId).catch(
+		const res = await deleteLiteLLMModel(localStorage.token, deleteLiteLLMModelName).catch(
 			(error) => {
 				toast.error(error);
 				return null;
@@ -485,7 +485,7 @@
 			}
 		}
 
-		deleteLiteLLMModelId = '';
+		deleteLiteLLMModelName = '';
 		liteLLMModelInfo = await getLiteLLMModelInfo(localStorage.token);
 		models.set(await getModels());
 	};
@@ -1099,14 +1099,14 @@
 								<div class="flex-1 mr-2">
 									<select
 										class="w-full rounded-lg py-2 px-4 text-sm dark:text-gray-300 dark:bg-gray-850 outline-none"
-										bind:value={deleteLiteLLMModelId}
+										bind:value={deleteLiteLLMModelName}
 										placeholder={$i18n.t('Select a model')}
 									>
-										{#if !deleteLiteLLMModelId}
+										{#if !deleteLiteLLMModelName}
 											<option value="" disabled selected>{$i18n.t('Select a model')}</option>
 										{/if}
 										{#each liteLLMModelInfo as model}
-											<option value={model.model_info.id} class="bg-gray-100 dark:bg-gray-700"
+											<option value={model.model_name} class="bg-gray-100 dark:bg-gray-700"
 												>{model.model_name}</option
 											>
 										{/each}