crewplus 0.2.4__tar.gz → 0.2.6__tar.gz

{crewplus-0.2.4 → crewplus-0.2.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: crewplus
-Version: 0.2.4
+Version: 0.2.6
 Summary: Base services for CrewPlus AI applications
 Author-Email: Tim Liu <tim@opsmateai.com>
 License: MIT

{crewplus-0.2.4 → crewplus-0.2.6}/crewplus/services/init_services.py

@@ -3,11 +3,15 @@ from crewplus.services.model_load_balancer import ModelLoadBalancer
 
 model_balancer = None
 
-def init_load_balancer():
+def init_load_balancer(config_path: str = None):
     global model_balancer
     if model_balancer is None:
-        config_path = os.getenv("MODEL_CONFIG_PATH", "config/models_config.json")
-        model_balancer = ModelLoadBalancer(config_path)
+        # Use parameter if provided, otherwise check env var, then default
+        final_config_path = config_path or os.getenv(
+            "MODEL_CONFIG_PATH", 
+            "config/models_config.json"  # Fixed default path
+        )
+        model_balancer = ModelLoadBalancer(final_config_path)
         model_balancer.load_config()  # Load initial configuration synchronously
 
 def get_model_balancer() -> ModelLoadBalancer:

{crewplus-0.2.4 → crewplus-0.2.6}/crewplus/vectorstores/milvus/vdb_service.py

@@ -197,7 +197,7 @@ class VDBService(object):
         """
         return self._client
 
-    def get_embeddings(self, from_model_balancer: bool = False, model_type: Optional[str] = "embedding-large") -> Embeddings:
+    def get_embeddings(self, from_model_balancer: bool = False, provider: Optional[str] = "azure-openai", model_type: Optional[str] = "embedding-large") -> Embeddings:
         """
         Gets an embedding function, either from the model balancer or directly from settings.
 
@@ -211,7 +211,7 @@ class VDBService(object):
         """
         if from_model_balancer:
             model_balancer = get_model_balancer()
-            return model_balancer.get_model(model_type=model_type)
+            return model_balancer.get_model(provider=provider, model_type=model_type)
 
         embedder_config = self.settings.get("embedder")
         if not embedder_config:
@@ -248,8 +248,8 @@ class VDBService(object):
     def get_vector_store(self, collection_name: str, embeddings: Embeddings = None, metric_type: str = "L2") -> Zilliz:
         """
         Gets a vector store instance, creating it if it doesn't exist for the collection.
-
-        This method caches instances by collection name to avoid re-instantiation.
+        This method validates both the embedding function and the vector store connection
+        before caching the instance to prevent faulty instances from being reused.
 
         Args:
             collection_name (str): The name of the collection in the vector database.
@@ -272,6 +272,19 @@ class VDBService(object):
         if embeddings is None:
             embeddings = self.get_embeddings()
 
+        # 1. Validate the embedding function before proceeding.
+        try:
+            self.logger.debug(f"Testing embedding function for collection '{collection_name}'...")
+            embeddings.embed_query("validation_test_string")
+            self.logger.debug("Embedding function is valid.")
+        except Exception as e:
+            self.logger.error(
+                f"The provided embedding function is invalid and failed with error: {e}. "
+                f"Cannot create a vector store for collection '{collection_name}'."
+            )
+            raise RuntimeError(f"Invalid embedding function provided.") from e
+
+        # If embeddings are valid, proceed to create the Zilliz instance.
         index_params = self.index_params or {
             "metric_type": metric_type,
             "index_type": "AUTOINDEX",

{crewplus-0.2.4 → crewplus-0.2.6}/docs/ModelLoadBalancer.md

@@ -88,3 +88,47 @@ The model `o3mini` is identified by the deployment name `gpt-o3mini-eastus2-RPM2
 ```python
 o3mini_model = balancer.get_model(deployment_name="gpt-o3mini-eastus2-RPM25")
 ```
+
+## 5. Global Access with `init_load_balancer`
+
+The `init_load_balancer` function provides a convenient singleton pattern for accessing the `ModelLoadBalancer` throughout your application without passing the instance around.
+
+First, you initialize the balancer once at the start of your application.
+
+### Initialization
+
+You can initialize it in several ways:
+
+**1. Default Initialization**
+
+This will look for the `MODEL_CONFIG_PATH` environment variable, or use the default path `_config/models_config.json`.
+
+```python
+from crewplus.services.init_services import init_load_balancer
+
+init_load_balancer()
+```
+
+**2. Initialization with a Custom Path**
+
+You can also provide a direct path to your configuration file.
+
+```python
+from crewplus.services.init_services import init_load_balancer
+
+init_load_balancer(config_path="path/to/your/models_config.json")
+```
+
+### Getting the Balancer and Models
+
+Once initialized, you can retrieve the `ModelLoadBalancer` instance from anywhere in your code using `get_model_balancer`.
+
+```python
+from crewplus.services.init_services import get_model_balancer
+
+# Get the balancer instance
+balancer = get_model_balancer()
+
+# Get a model by deployment name
+gemini_flash_model = balancer.get_model(deployment_name="gemini-2.5-flash")
+```

{crewplus-0.2.4 → crewplus-0.2.6}/docs/VDBService.md

@@ -60,6 +60,20 @@ vdb_service = VDBService(settings=settings)
 print("VDBService initialized successfully!")
 ```
 
+**Alternative Initialization for Zilliz**
+
+For a simpler Zilliz Cloud connection, you can initialize the service directly with your endpoint and token.
+
+```python
+# Initialize directly with Zilliz credentials
+vdb_service_zilliz = VDBService(
+    endpoint="YOUR_ZILLIZ_ENDPOINT", 
+    token="YOUR_ZILLIZ_TOKEN"
+)
+
+print("VDBService for Zilliz initialized successfully!")
+```
+
 ## 3. Usage Examples
 
 ### Basic Usage: Get Vector Store with Default Embeddings
@@ -86,7 +100,8 @@ This requires the `ModelLoadBalancer` to have been initialized, as shown in the
 # 1. Get a specific embedding model from the ModelLoadBalancer
 # The service will call get_model_balancer() internally to get the initialized instance.
 embedding_model = vdb_service.get_embeddings(
-    from_model_balancer=True, 
+    from_model_balancer=True,
+    provider="azure-openai-embeddings", 
     model_type="embedding-large" # Specify the model type configured in the balancer
 )
 

{crewplus-0.2.4 → crewplus-0.2.6}/pyproject.toml

@@ -6,7 +6,7 @@ build-backend = "pdm.backend"
 
 [project]
 name = "crewplus"
-version = "0.2.4"
+version = "0.2.6"
 description = "Base services for CrewPlus AI applications"
 authors = [
     { name = "Tim Liu", email = "tim@opsmateai.com" },