ray-embedding 0.10.1__py3-none-any.whl → 0.10.10__py3-none-any.whl

ray_embedding/deploy.py

@@ -16,6 +16,7 @@ def deploy_model(args: Dict[str, Any]) -> Application:
     model: str = args.pop("model", "")
     assert model
 
+    device: Optional[str] = args.pop("device", None)
     backend: Optional[str] = args.pop("backend", "torch")
     matryoshka_dim: Optional[int] = args.pop("matryoshka_dim", None)
     trust_remote_code: Optional[bool] = args.pop("trust_remote_code", False)
@@ -32,6 +33,7 @@ def deploy_model(args: Dict[str, Any]) -> Application:
             raise ValueError(f"Invalid torch_dtype: '{torch_dtype}'")
 
     deployment = EmbeddingModel.options(name=deployment_name).bind(model=model,
+                                                                   device=device,
                                                                    backend=backend,
                                                                    matryoshka_dim=matryoshka_dim,
                                                                    trust_remote_code=trust_remote_code,

ray_embedding/embedding_model.py

@@ -28,18 +28,23 @@ web_api = FastAPI(title=f"Ray Embeddings - OpenAI-compatible API")
 )
 @serve.ingress(web_api)
 class EmbeddingModel:
-    def __init__(self, model: str, backend: Optional[str] = "torch", matryoshka_dim: Optional[int] = None,
-                 trust_remote_code: Optional[bool] = False, model_kwargs: Dict[str, Any] = None):
+    def __init__(self, model: str, device: Optional[str] = None, backend: Optional[str] = "torch",
+                 matryoshka_dim: Optional[int] = None, trust_remote_code: Optional[bool] = False,
+                 model_kwargs: Dict[str, Any] = None):
         logging.basicConfig(level=logging.INFO)
         self.logger = logging.getLogger(__name__)
         self.model = 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"
+        self.torch_device = torch.device(self.init_device)
         self.backend = backend or "torch"
         self.matryoshka_dim = matryoshka_dim
         self.trust_remote_code = trust_remote_code or False
         self.model_kwargs = model_kwargs or {}
-        self.torch_device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         self.logger.info(f"Initializing embedding model: {self.model}")
-        self.embedding_model = SentenceTransformer(self.model, backend=self.backend, trust_remote_code=self.trust_remote_code,
+        self.embedding_model = SentenceTransformer(self.model, device=self.init_device, backend=self.backend,
+                                                   trust_remote_code=self.trust_remote_code,
                                                    model_kwargs=self.model_kwargs)
 
         self.served_model_name = os.path.basename(self.model)
@@ -93,5 +98,9 @@ class EmbeddingModel:
         """Returns the list of available models in OpenAI-compatible format."""
         return {"object": "list", "data": self.available_models}
 
-
-
+    def check_health(self):
+        if self.init_device == "cuda" and not torch.cuda.is_available():
+            # Even though CUDA was available at init time,
+            # CUDA can become unavailable - this is a known problem in AWS EC2
+            # https://github.com/ray-project/ray/issues/49594
+            raise RuntimeError("CUDA device is not available")

ray_embedding-0.10.10.dist-info/METADATA

@@ -0,0 +1,36 @@
+Metadata-Version: 2.4
+Name: ray-embedding
+Version: 0.10.10
+Summary: Deploy SentenceTransformers embedding models to a ray cluster
+Author: Crispin Almodovar
+Author-email: 
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.12
+Description-Content-Type: text/markdown
+
+# ray-embedding
+
+A Python library for deploying SentenceTransformers models to a ray cluster. 
+This tool encapsulates inference logic that uses SentenceTransformers 
+to load any compatible embedding model from the Hugging Face hub and 
+compute embeddings for input text.
+
+This library is meant to be used with the [embedding-models Ray cluster](https://bitbucket.org/docorto/embedding-models/src/dev/).
+
+Refer to this [Ray Serve deployment config](https://bitbucket.org/docorto/embedding-models/src/dev/serve-config/dev/serve-config.yaml) 
+to see how this library is used.
+
+### Supports the following backends
+
+- pytorch-gpu
+- pytorch-cpu
+
+### Planned:
+- onnx-gpu
+- onnx-cpu
+- openvino-cpu
+- fastembed-onnx-cpu
+
+

ray_embedding-0.10.10.dist-info/RECORD

@@ -0,0 +1,8 @@
+ray_embedding/__init__.py,sha256=OYJT0rVaaGzY613JqgfktsCgroDnBkGOHxR2FE9UtRU,49
+ray_embedding/deploy.py,sha256=ZGxcG4589WcRtaM6H84YJarw0m1XqHNgfOf3PLAhM5M,1995
+ray_embedding/dto.py,sha256=e91ejZbM_NB9WTjF1YnfuV71cajYIh0vOX8oV_g2OwM,595
+ray_embedding/embedding_model.py,sha256=hIszW30di-Us0TL5Wevo8gNpD-kL-bhitU4MrNHrebc,4574
+ray_embedding-0.10.10.dist-info/METADATA,sha256=5vgf2aQCm91W6V_hXKGfJJYo8a8L2b8gqzSrYscxY88,1095
+ray_embedding-0.10.10.dist-info/WHEEL,sha256=0CuiUZ_p9E4cD6NyLD6UG80LBXYyiSYZOKDm5lp32xk,91
+ray_embedding-0.10.10.dist-info/top_level.txt,sha256=ziCblpJq1YsrryshFqxTRuRMgNuO1_tgvAAkGShATNA,14
+ray_embedding-0.10.10.dist-info/RECORD,,

{ray_embedding-0.10.1.dist-info → ray_embedding-0.10.10.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.3.0)
+Generator: setuptools (80.3.1)
 Root-Is-Purelib: true
 Tag: py3-none-any
 

ray_embedding-0.10.1.dist-info/METADATA

@@ -1,28 +0,0 @@
-Metadata-Version: 2.4
-Name: ray-embedding
-Version: 0.10.1
-Summary: Deploy SentenceTransformers embedding models to a ray cluster
-Author: Crispin Almodovar
-Author-email: 
-Classifier: Programming Language :: Python :: 3
-Classifier: License :: OSI Approved :: MIT License
-Classifier: Operating System :: OS Independent
-Requires-Python: >=3.12
-Description-Content-Type: text/markdown
-
-# ray-embedding
-
-A tool for deploying SentenceTransformers models to a ray cluster.
-
-### Supports the following backends
-
-- pytorch-gpu
-- pytorch-cpu
-
-### Planned:
-- onnx-gpu
-- onnx-cpu
-- openvino-cpu
-- fastembed-onnx-cpu
-
-

ray_embedding-0.10.1.dist-info/RECORD

@@ -1,8 +0,0 @@
-ray_embedding/__init__.py,sha256=OYJT0rVaaGzY613JqgfktsCgroDnBkGOHxR2FE9UtRU,49
-ray_embedding/deploy.py,sha256=oqOhMxS5UyZ4oGhfpL7kqvrxLO8QW41sF_FHbNSJe-w,1858
-ray_embedding/dto.py,sha256=e91ejZbM_NB9WTjF1YnfuV71cajYIh0vOX8oV_g2OwM,595
-ray_embedding/embedding_model.py,sha256=54QQ5iIAnOHtg_8nqGPIfqrZVwih6qKKeEOgmZweQeQ,3931
-ray_embedding-0.10.1.dist-info/METADATA,sha256=E9sZzIgqproOjSMt1ujPha_BhuBHLtd3ipTXD8vJSAU,605
-ray_embedding-0.10.1.dist-info/WHEEL,sha256=GHB6lJx2juba1wDgXDNlMTyM13ckjBMKf-OnwgKOCtA,91
-ray_embedding-0.10.1.dist-info/top_level.txt,sha256=ziCblpJq1YsrryshFqxTRuRMgNuO1_tgvAAkGShATNA,14
-ray_embedding-0.10.1.dist-info/RECORD,,