ray-embedding 0.11.6__tar.gz → 0.11.8__tar.gz

{ray_embedding-0.11.6 → ray_embedding-0.11.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ray-embedding
-Version: 0.11.6
+Version: 0.11.8
 Summary: Deploy SentenceTransformers embedding models to a ray cluster
 Author: Crispin Almodovar
 Author-email: 

{ray_embedding-0.11.6 → ray_embedding-0.11.8}/ray_embedding/deploy.py

@@ -1,17 +1,19 @@
+import os
 from typing import Dict, Any, Optional
 from ray.serve import Application
 from ray.serve.handle import DeploymentHandle
 
-from ray_embedding.dto import AppConfig, ModelDeploymentConfig
+from ray_embedding.dto import AppConfig, ModelDeploymentConfig, DeployedModel
 from ray_embedding.embedding_model import EmbeddingModel
 import torch
 
 from ray_embedding.model_router import ModelRouter
 
 
-def build_model(model_config: ModelDeploymentConfig) -> DeploymentHandle:
+def build_model(model_config: ModelDeploymentConfig) -> DeployedModel:
     deployment_name = model_config.deployment
     model = model_config.model
+    served_model_name = model_config.served_model_name or os.path.basename(model)
     device = model_config.device
     backend = model_config.backend or "torch"
     matryoshka_dim = model_config.matryoshka_dim
@@ -29,18 +31,24 @@ def build_model(model_config: ModelDeploymentConfig) -> DeploymentHandle:
             raise ValueError(f"Invalid torch_dtype: '{torch_dtype}'")
 
     deployment = EmbeddingModel.options(name=deployment_name).bind(model=model,
+                                                                   served_model_name=served_model_name,
                                                                    device=device,
                                                                    backend=backend,
                                                                    matryoshka_dim=matryoshka_dim,
                                                                    trust_remote_code=trust_remote_code,
                                                                    model_kwargs=model_kwargs
                                                                    )
-    return deployment
+    return DeployedModel(model=served_model_name,
+                         deployment_handle=deployment,
+                         batch_size=model_config.batch_size,
+                         num_retries=model_config.num_retries
+                         )
+
 
 def build_app(args: AppConfig) -> Application:
     model_router, models = args.model_router, args.models
     assert model_router and models
 
-    served_models = {model_config.model: build_model(model_config) for model_config in models}
-    router = ModelRouter.options(name=model_router.deployment).bind(served_models)
+    deployed_models = {model_config.served_model_name: build_model(model_config) for model_config in models}
+    router = ModelRouter.options(name=model_router.deployment).bind(deployed_models)
     return router

{ray_embedding-0.11.6 → ray_embedding-0.11.8}/ray_embedding/embedding_model.py

@@ -18,6 +18,7 @@ class EmbeddingModel:
         logging.basicConfig(level=logging.INFO)
         self.logger = logging.getLogger(self.__class__.__name__)
         self.model = model
+        self.served_model_name = served_model_name or os.path.basename(self.model)
         self.init_device = device
         if self.init_device is None or self.init_device == "auto":
             self.init_device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -33,7 +34,6 @@ class EmbeddingModel:
                                                    trust_remote_code=self.trust_remote_code,
                                                    model_kwargs=self.model_kwargs)
 
-        self.served_model_name = served_model_name or os.path.basename(self.model)
         self.logger.info(f"Successfully initialized embedding model {self.model} using device {self.torch_device}")
 
     async def __call__(self, text: Union[str, List[str]], dimensions: Optional[int] = None):

{ray_embedding-0.11.6 → ray_embedding-0.11.8}/ray_embedding/model_router.py

@@ -13,33 +13,32 @@ web_api = FastAPI(title="Ray Embeddings - OpenAI-compatible API")
 @serve.deployment
 @serve.ingress(web_api)
 class ModelRouter:
-    def __init__(self, served_models: Dict[str, DeployedModel]):
-        assert served_models, "models cannot be empty"
+    def __init__(self, deployed_models: Dict[str, DeployedModel]):
+        assert deployed_models, "models cannot be empty"
         logging.basicConfig(level=logging.INFO)
         self.logger = logging.getLogger(self.__class__.__name__)
-        self.served_models = served_models
+        self.deployed_models = deployed_models
         self.available_models = [
             {"id": str(item),
              "object": "model",
              "created": int(time.time()),
              "owned_by": "openai",
-             "permission": []} for item in self.served_models.keys()
+             "permission": []} for item in self.deployed_models.keys()
         ]
         self.logger.info(f"Successfully registered models: {self.available_models}")
 
     async def _compute_embeddings_from_resized_batches(self, model: str, inputs: List[str], dimensions: Optional[int] = None):
-        assert model in self.served_models
-        self.logger.info(f"Retrieving model handle for: {model} ")
-        model_handle = self.served_models[model].deployment_handle
-        self.logger.info(f"Model handle: {model_handle} ")
-        batch_size = self.served_models[model].batch_size
-        num_retries = self.served_models[model].num_retries
+        assert model in self.deployed_models
+        deployed_model = self.deployed_models[model]
+        model_handle = deployed_model.deployment_handle
+        batch_size = deployed_model.batch_size
+        num_retries = deployed_model.num_retries
 
         # Resize the inputs into batch_size items, and dispatch in parallel
         batches = [inputs[i:i+batch_size] for i in range(0, len(inputs), batch_size)]
         if len(inputs) > batch_size:
-            self.logger.info(f"Original input is greater than {batch_size}. "
-                             f"It was resized to {len(batches)} mini-batches of size {batch_size}")
+            self.logger.info(f"Original input (length {len(inputs)} was resized "
+                             f"to {len(batches)} mini-batches of length {batch_size}")
         tasks = [model_handle.remote(batch, dimensions) for batch in batches]
         all_results = await asyncio.gather(*tasks, return_exceptions=True)
 
@@ -68,7 +67,7 @@ class ModelRouter:
     async def compute_embeddings(self, request: EmbeddingRequest):
         try:
             inputs = request.input if isinstance(request.input, list) else [request.input]
-            self.logger.info(f"Received input of size {len(inputs)} text chunks")
+            self.logger.info(f"Computing embeddings for a batch of {len(inputs)} texts using model: {request.model}")
             embeddings = await self._compute_embeddings_from_resized_batches(request.model, inputs, request.dimensions)
             response_data = [
                 {"index": idx, "embedding": emb}

{ray_embedding-0.11.6 → ray_embedding-0.11.8}/ray_embedding.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ray-embedding
-Version: 0.11.6
+Version: 0.11.8
 Summary: Deploy SentenceTransformers embedding models to a ray cluster
 Author: Crispin Almodovar
 Author-email: 

{ray_embedding-0.11.6 → ray_embedding-0.11.8}/setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = ray-embedding
-version = 0.11.6
+version = 0.11.8
 author = Crispin Almodovar
 author_email = 
 description = Deploy SentenceTransformers embedding models to a ray cluster