ray-embedding 0.12.5__tar.gz → 0.12.7__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of ray-embedding might be problematic. Click here for more details.
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/PKG-INFO +1 -1
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding/deploy.py +2 -1
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding/embedding_model.py +2 -1
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding.egg-info/PKG-INFO +1 -1
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/setup.cfg +1 -1
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/README.md +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/pyproject.toml +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding/__init__.py +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding/dto.py +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding/model_router.py +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding.egg-info/SOURCES.txt +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding.egg-info/dependency_links.txt +0 -0
- {ray_embedding-0.12.5 → ray_embedding-0.12.7}/ray_embedding.egg-info/top_level.txt +0 -0
|
@@ -36,7 +36,8 @@ def build_model(model_config: ModelDeploymentConfig) -> DeployedModel:
|
|
|
36
36
|
backend=backend,
|
|
37
37
|
matryoshka_dim=matryoshka_dim,
|
|
38
38
|
trust_remote_code=trust_remote_code,
|
|
39
|
-
model_kwargs=model_kwargs
|
|
39
|
+
model_kwargs=model_kwargs,
|
|
40
|
+
cuda_memory_flush_threshold=cuda_memory_flush_threshold
|
|
40
41
|
)
|
|
41
42
|
return DeployedModel(model=served_model_name,
|
|
42
43
|
deployment_handle=deployment,
|
|
@@ -20,6 +20,7 @@ class EmbeddingModel:
|
|
|
20
20
|
self.model = model
|
|
21
21
|
self.served_model_name = served_model_name or os.path.basename(self.model)
|
|
22
22
|
self.init_device = device
|
|
23
|
+
self.cuda_memory_flush_threshold = cuda_memory_flush_threshold
|
|
23
24
|
if self.init_device is None or self.init_device == "auto":
|
|
24
25
|
self.init_device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
25
26
|
if self.init_device == "cuda":
|
|
@@ -29,7 +30,7 @@ class EmbeddingModel:
|
|
|
29
30
|
self.matryoshka_dim = matryoshka_dim
|
|
30
31
|
self.trust_remote_code = trust_remote_code or False
|
|
31
32
|
self.model_kwargs = model_kwargs or {}
|
|
32
|
-
|
|
33
|
+
|
|
33
34
|
self.logger.info(f"Initializing embedding model: {self.model}")
|
|
34
35
|
self.embedding_model = SentenceTransformer(self.model, device=self.init_device, backend=self.backend,
|
|
35
36
|
trust_remote_code=self.trust_remote_code,
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|