pull ollama models if not exist automatically (#512)

* pull ollama models if not exist automatically

* add ollama_url to pull_ollama_model

* fix bug

* update message for Custom llm provider and timeout for ai service

* update message

* fix intro for WREN_IBIS_CONNECTION_INFO

wren-ai-service/.env.dev.example

@@ -54,7 +54,7 @@ WREN_UI_ENDPOINT=http://localhost:3000
 WREN_IBIS_ENDPOINT=http://localhost:8000
 WREN_IBIS_SOURCE=bigquery
 WREN_IBIS_MANIFEST= # this is a base64 encoded string of the MDL
-WREN_IBIS_CONNECTION_INFO={"project_id": "", "dataset_id": "", "credentials":""}
+WREN_IBIS_CONNECTION_INFO= # this is a base64 encode string of the connection info
 
 # evaluation related
 DATASET_NAME=book_2

wren-ai-service/pyproject.toml

@@ -24,6 +24,7 @@ sf-hamilton = {version = "==1.63.0", extras = ["visualization"]}
 aiohttp = "==3.9.5"
 ollama-haystack = "==0.0.6"
 langfuse = "==2.35.0"
+ollama = "==0.2.1"
 
 [tool.poetry.group.dev.dependencies]
 pytest = "==8.2.0"

wren-ai-service/src/__main__.py

@@ -90,7 +90,8 @@ def health():
         host=server_host,
         port=server_port,
         reload=should_reload,
-        reload_dirs=["src"] if should_reload else None,
+        reload_dirs=["src"],
+        reload_includes=[".env.dev"],
         workers=1,
         loop="uvloop",
         http="httptools",

wren-ai-service/src/providers/embedder/ollama.py

@@ -12,7 +12,7 @@
 from tqdm import tqdm
 
 from src.core.provider import EmbedderProvider
-from src.providers.loader import provider
+from src.providers.loader import provider, pull_ollama_model
 from src.utils import remove_trailing_slash
 
 logger = logging.getLogger("wren-ai-service")
@@ -167,6 +167,8 @@ def __init__(
         self._url = remove_trailing_slash(url)
         self._embedding_model = embedding_model
 
+        pull_ollama_model(self._url, self._embedding_model)
+
         logger.info(f"Using Ollama Embedding Model: {self._embedding_model}")
         logger.info(f"Using Ollama URL: {self._url}")
 

wren-ai-service/src/providers/engine/wren.py

@@ -1,3 +1,4 @@
+import base64
 import logging
 import os
 from typing import Any, Dict, Optional, Tuple
@@ -54,8 +55,10 @@ async def dry_run_sql(
             "source": os.getenv("WREN_IBIS_SOURCE"),
             "manifest": os.getenv("WREN_IBIS_MANIFEST"),
             "connection_info": orjson.loads(
-                os.getenv("WREN_IBIS_CONNECTION_INFO", "{}")
-            ),
+                base64.b64decode(os.getenv("WREN_IBIS_CONNECTION_INFO"))
+            )
+            if os.getenv("WREN_IBIS_CONNECTION_INFO")
+            else {},
         },
     ) -> Tuple[bool, Optional[Dict[str, Any]]]:
         async with session.post(

wren-ai-service/src/providers/llm/ollama.py

@@ -9,7 +9,7 @@
 from haystack_integrations.components.generators.ollama import OllamaGenerator
 
 from src.core.provider import LLMProvider
-from src.providers.loader import provider
+from src.providers.loader import provider, pull_ollama_model
 from src.utils import remove_trailing_slash
 
 logger = logging.getLogger("wren-ai-service")
@@ -130,6 +130,8 @@ def __init__(
         self._url = remove_trailing_slash(url)
         self._generation_model = generation_model
 
+        pull_ollama_model(self._url, self._generation_model)
+
         logger.info(f"Using Ollama LLM: {self._generation_model}")
         logger.info(f"Using Ollama URL: {self._url}")
 

wren-ai-service/src/providers/loader.py

@@ -2,6 +2,8 @@
 import logging
 import pkgutil
 
+from ollama import Client
+
 logger = logging.getLogger("wren-ai-service")
 
 
@@ -98,3 +100,19 @@ def get_default_embedding_model_dim(embedder_provider: str):
     return importlib.import_module(
         f"src.providers.embedder.{file_name}"
     ).EMBEDDING_MODEL_DIMENSION
+
+
+def pull_ollama_model(url: str, model_name: str):
+    client = Client(host=url)
+    models = client.list()["models"]
+    if model_name not in models:
+        logger.info(f"Pulling Ollama model {model_name}")
+        percentage = 0
+        for progress in client.pull(model_name, stream=True):
+            if "completed" in progress and "total" in progress:
+                new_percentage = int(progress["completed"] / progress["total"] * 100)
+                if new_percentage > percentage:
+                    percentage = new_percentage
+                    logger.info(f"Pulling Ollama model {model_name}: {percentage}%")
+    else:
+        logger.info(f"Ollama model {model_name} already exists")

wren-launcher/commands/launch.go

@@ -210,10 +210,12 @@ func Launch() {
 		panic(err)
 	}
 
-	// wait for 10 seconds
 	pterm.Info.Println("Wren AI is starting, please wait for a moment...")
+	if llmProvider == "Custom" {
+		pterm.Info.Println("If you choose Ollama as LLM provider, please make sure you have started the Ollama service first. Also, Wren AI will automatically pull your chosen models if you have not done so. You can check the progress by executing `docker logs -f wrenai-wren-ai-service-1` in the terminal.")
+	}
 	url := fmt.Sprintf("http://localhost:%d", uiPort)
-	// wait until checking if CheckWrenAIStarted return without error
+	// wait until checking if CheckUIServiceStarted return without error
 	// if timeout 2 minutes, panic
 	timeoutTime := time.Now().Add(2 * time.Minute)
 	for {
@@ -230,6 +232,9 @@ func Launch() {
 		time.Sleep(5 * time.Second)
 	}
 
+	// wait until checking if CheckWrenAIStarted return without error
+	// if timeout 30 minutes, panic
+	timeoutTime = time.Now().Add(30 * time.Minute)
 	for {
 		if time.Now().After(timeoutTime) {
 			panic("Timeout")