""" title: Ollama Model Unloader description: Ollama Model Unloader Button author: pkeffect author_url: https://github.com/pkeffect/ project_url: https://github.com/pkeffect/open-webui-tools/tree/main/ollama_unload_model funding_url: https://github.com/open-webui required_open_webui_version: 0.6.0 version: 0.0.5 date: 2025-04-12 license: MIT changelog: - 0.0.5 - Added more error handling and support for remote Ollama servers - 0.0.4 - Refactor - 0.0.3 - Added error reporting and better checks - 0.0.2 - Added status and progress - 0.0.1 - Initial upload to openwebui community. features: - For local and remote Ollama hosts and custom ports (editable in valves) - Searches multiple locations for Ollama servers (automatic) - Displays progress and results important: - Unloading models might not instantly release system RAM if models have spilled over from VRAM. Monitor your system resources - High CPU usage might persist temporarily after unloading if system RAM is heavily utilized - Avoid unloading models while they are actively generating text to prevent potential issues - If issues persist after unloading, consider restarting your Ollama service manually - For remote Ollama instances, ensure the Ollama API is accessible over your network and that any necessary firewalls are configured to allow connections icon_url: data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHdpZHRoPSIyNCIgaGVpZ2h0PSIyNCIgdmlld0JveD0iMCAwIDI0IDI0Ij48cGF0aCBmaWxsPSJjdXJyZW50Q29sb3IiIGQ9Ik0xMiAyQTEwIDEwIDAgMCAwIDIgMTJhMTAgMTAgMCAwIDAgMTAgMTBhMTAgMTAgMCAwIDAgMTAtMTBBMTAgMTAgMCAwIDAgMTIgMm0wIDJjNC40MSAwIDggMy41OSA4IDhzLTMuNTkgOC04IDhzLTgtMy41OS04LThzMy41OS04IDgtOE05IDl2Nmg2VjkiLz48L3N2Zz4= requirements: """ import requests import time import asyncio from concurrent.futures import ThreadPoolExecutor from pydantic import BaseModel, Field from typing import Callable, Any, Dict, List import threading import json # Constants TIMEOUT = 15 DEFAULT_OLLAMA_HOSTS = ["localhost", "127.0.0.1", "ollama", "host.docker.internal"] DEFAULT_OLLAMA_PORT = 11434 requests.adapters.DEFAULT_TIMEOUT = TIMEOUT class OllamaAPIClient: def __init__(self, base_url="http://localhost:11434"): self.base_url = base_url def get_running_models(self) -> List[Dict]: try: response = requests.get(f"{self.base_url}/api/ps", timeout=TIMEOUT) response.raise_for_status() return ( response.json().get("models", []) if response.status_code == 200 else [] ) except requests.exceptions.RequestException as e: print(f"Request error during get_running_models: {e}") return [] except json.JSONDecodeError as e: print( f"JSON decode error during get_running_models: {e}, Response text: {response.text if 'response' in locals() else 'No response'}" ) return [] except Exception as e: print(f"Unexpected error during get_running_models: {e}") return [] def unload_model(self, model_name: str) -> Dict: result = {"success": False, "model": model_name, "message": ""} try: if not any(m.get("model") == model_name for m in self.get_running_models()): result["message"] = "Model not found in running models list." return result response = requests.post( f"{self.base_url}/api/generate", json={"model": model_name, "prompt": " ", "keep_alive": 0}, timeout=TIMEOUT, ) response.raise_for_status() if response.status_code in [200, 204]: time.sleep(2) if not any( m.get("model") == model_name for m in self.get_running_models() ): result["success"] = True else: result["success"] = False result["message"] = "Model still running after unload attempt." else: result["message"] = ( f"Unload request failed with status code: {response.status_code}" ) except requests.exceptions.RequestException as e: result["message"] = f"Request error during unload_model: {e}" print(f"Request error during unload_model for {model_name}: {e}") except json.JSONDecodeError as e: result["message"] = ( f"JSON decode error during unload_model: {e}, Response text: {response.text if 'response' in locals() else 'No response'}" ) print( f"JSON decode error during unload_model for {model_name}: {e}, Response text: {response.text if 'response' in locals() else 'No response'}" ) except Exception as e: result["message"] = f"Unexpected error during unload_model: {e}" print(f"Unexpected error during unload_model for {model_name}: {e}") return result class OllamaUnloader: @staticmethod def run_stop_command( status_callback=None, ollama_hosts=None, ollama_port=DEFAULT_OLLAMA_PORT ) -> str: ollama_hosts = ollama_hosts or DEFAULT_OLLAMA_HOSTS total_unloaded = 0 total_failed = 0 all_errors = [] for host in ollama_hosts: if status_callback: status_callback(f"Connecting to Ollama at {host}:{ollama_port}...") try: api_client = OllamaAPIClient(f"http://{host}:{ollama_port}") running_models = api_client.get_running_models() if not running_models: if status_callback: status_callback( f"No running models found on {host}:{ollama_port}" ) continue if status_callback: status_callback( f"Found {len(running_models)} running models on {host}:{ollama_port}" ) for model_info in running_models: model_name = model_info.get("model") if not model_name: continue if status_callback: status_callback(f"Attempting to unload model: {model_name}") unload_result = api_client.unload_model(model_name) if unload_result["success"]: total_unloaded += 1 if status_callback: status_callback( f"Successfully unloaded model: {model_name}" ) else: total_failed += 1 error_message = unload_result.get("message", "Unknown error") all_errors.append( f"Failed to unload model '{model_name}': {error_message}" ) if status_callback: status_callback( f"Failed to unload model: {model_name} - {error_message}" ) except requests.exceptions.ConnectionError as e: error_msg = f"Connection error to Ollama at {host}:{ollama_port}: {e}" all_errors.append(error_msg) if status_callback: status_callback(error_msg) continue except Exception as e: error_msg = ( f"Unexpected error while processing host {host}:{ollama_port}: {e}" ) all_errors.append(error_msg) if status_callback: status_callback(error_msg) continue if total_unloaded > 0 and total_failed == 0: return f"Successfully unloaded {total_unloaded} Ollama models." elif total_unloaded > 0: error_summary = "\n".join(all_errors) return f"Partially successful: Unloaded {total_unloaded} models, failed to unload {total_failed} models.\nErrors:\n{error_summary}" elif total_failed > 0: error_summary = "\n".join(all_errors) return f"Failed to unload {total_failed} models.\nErrors:\n{error_summary}" return "No running models found to unload across specified Ollama hosts." class Action: class Valves(BaseModel): OLLAMA_HOSTS: List[str] = Field( default=DEFAULT_OLLAMA_HOSTS, description="List of Ollama host IPs or **hostnames** to connect to. Supports both local addresses (localhost, 127.0.0.1, ollama) and remote addresses (e.g., 192.168.1.100, my-ollama-server.com).", # Updated description for remote hosts ) OLLAMA_PORT: int = Field( default=DEFAULT_OLLAMA_PORT, description="Port number for Ollama API (default: 11434)", ) WAIT_BETWEEN_UNLOADS: int = Field( default=0, description="Seconds to wait between model unloads (default: 0)" ) AUTO_CLOSE_OUTPUT: bool = Field( default=True, description="Whether to automatically close/collapse output when finished", ) AUTO_CLOSE_DELAY: int = Field( default=3, description="Seconds to wait before automatically closing output (default: 3)", ) def __init__(self): self.valves = self.Valves() # For tracking output closure self.close_task = None # For tracking event emitter self.event_emitter = None # For tracking operations self.is_operation_complete = False # For thread safety self.lock = threading.Lock() def create_message( self, type_name, description="", status="in_progress", done=False, close=False ): """Create a unified message structure for the event emitter""" message_id = f"msg_{int(time.time() * 1000)}" message = { "type": type_name, "message_id": message_id, "data": { "status": status, "description": description, "done": done, "close": close, "message_id": message_id, }, } return message async def close_emitter_output(self): """Send a final message to close/collapse the output""" await asyncio.sleep(self.valves.AUTO_CLOSE_DELAY) if self.event_emitter and self.is_operation_complete: # Try several different approaches to signal completion: # 1. Send a special close message close_msg = self.create_message( "status", description="", status="complete", done=True, close=True ) await self.event_emitter(close_msg) # 2. Send an empty message to signal end of stream (works in some systems) await self.event_emitter({}) # 3. Try a close_output event type (might work in some implementations) await self.event_emitter( {"type": "close_output", "data": {"force_close": True}} ) # 4. Try to clear all messages await self.event_emitter( {"type": "clear_all", "data": {"force_clear": True}} ) async def action( self, body: dict, user: dict = {}, __event_emitter__: Callable[[dict], Any] = None, __event_call__: Callable[[dict], Any] = None, ) -> None: if not __event_emitter__: return # Store the event emitter for later use self.event_emitter = __event_emitter__ self.is_operation_complete = False try: # Initial status with warning about active generation initial_message = self.create_message( "status", "Starting Ollama model unloader...\n**Please ensure no models are actively generating text to avoid potential issues.**\n**For remote hosts, ensure network accessibility and firewall rules are configured correctly.**", ) # Added network/firewall warning for remote hosts await __event_emitter__(initial_message) status_updates = {} def status_callback(text): status_updates["current"] = text def run_unloader(): try: result = OllamaUnloader.run_stop_command( status_callback=status_callback, ollama_hosts=self.valves.OLLAMA_HOSTS, ollama_port=self.valves.OLLAMA_PORT, ) status_updates["final_result"] = result return result except Exception as e: error_msg = f"Error in unloader thread: {str(e)}" status_updates["final_result"] = error_msg return error_msg with ThreadPoolExecutor(max_workers=1) as executor: future = executor.submit(run_unloader) last_status = None while not future.done(): if "current" in status_updates: current_status = status_updates["current"] if current_status != last_status: await __event_emitter__( self.create_message("status", current_status) ) last_status = current_status await asyncio.sleep(0.5) # Final status final_result = status_updates.get("final_result", future.result()) status_type = ( "warning" if any( x in final_result for x in ["Failed", "Error", "Partially successful"] ) else "complete" ) # Send final status with done=True final_message = self.create_message( "status", description=final_result, status=status_type, done=True ) await __event_emitter__(final_message) # Mark operation as complete self.is_operation_complete = True # Schedule auto-close if enabled if self.valves.AUTO_CLOSE_OUTPUT: # Create a task to close the output after a delay self.close_task = asyncio.create_task(self.close_emitter_output()) except Exception as e: # Error message error_message = self.create_message( "status", description=f"Error: {str(e)}", status="error", done=True ) await __event_emitter__(error_message) # Mark operation as complete even on error self.is_operation_complete = True # Schedule auto-close on error too if self.valves.AUTO_CLOSE_OUTPUT: self.close_task = asyncio.create_task(self.close_emitter_output())