ray-embedding 0.13.15__py3-none-any.whl → 0.14.0__py3-none-any.whl

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.
ray_embedding/deploy.py CHANGED
@@ -1,4 +1,5 @@
1
1
  import os
2
+ from typing import Any, Dict
2
3
 
3
4
  import torch
4
5
  from ray.serve import Application
@@ -6,7 +7,7 @@ from ray.serve import Application
6
7
  from ray_embedding.dto import AppConfig, ModelDeploymentConfig, DeployedModel, NodeReaperConfig
7
8
  from ray_embedding.embedding_model import EmbeddingModel
8
9
  from ray_embedding.node_reaper import NodeReaper, NODE_REAPER_DEPLOYMENT_NAME
9
- from ray_embedding.utils import node_affinity_for_head, node_affinity_for_worker
10
+ from ray_embedding.utils import get_head_node_id
10
11
  from ray_embedding.model_router import ModelRouter
11
12
 
12
13
 
@@ -56,7 +57,7 @@ def build_app(args: AppConfig) -> Application:
56
57
 
57
58
  node_reaper_config = args.node_reaper or NodeReaperConfig()
58
59
 
59
- node_reaper_kwargs = {
60
+ node_reaper_kwargs: Dict[str, Any] = {
60
61
  "ssh_user": node_reaper_config.ssh_user,
61
62
  "ssh_private_key": node_reaper_config.ssh_private_key,
62
63
  }
@@ -65,15 +66,22 @@ def build_app(args: AppConfig) -> Application:
65
66
  if node_reaper_config.reap_interval_seconds is not None:
66
67
  node_reaper_kwargs["reap_interval_seconds"] = node_reaper_config.reap_interval_seconds
67
68
 
68
- node_reaper = NodeReaper.options(
69
+ node_reaper = NodeReaper.options(
69
70
  name=NODE_REAPER_DEPLOYMENT_NAME,
70
- scheduling_strategy=node_affinity_for_head()
71
+ ray_actor_options={"num_cpus": 0.25, "resources": {"node_type:head": 1}},
72
+ autoscaling_config={"initial_replicas": 1, "min_replicas": 1, "max_replicas": 1}
71
73
  ).bind(**node_reaper_kwargs)
72
74
 
73
75
  deployed_models = {model_config.served_model_name: build_model(model_config, node_reaper) for model_config in models}
76
+ model_router_kwargs = {
77
+ "deployed_models": deployed_models,
78
+ "path_prefix": model_router.path_prefix,
79
+ "max_concurrency": model_router.max_concurrency,
80
+ "node_reaper": node_reaper
81
+ }
74
82
  router = ModelRouter.options(
75
83
  name=model_router.deployment,
76
- scheduling_strategy=node_affinity_for_worker()
77
- ).bind(deployed_models, model_router.path_prefix, node_reaper)
84
+ ray_actor_options={"num_cpus": 0.25, "resources": {"node_type:worker": 1}}
85
+ ).bind(**model_router_kwargs)
78
86
 
79
87
  return router
@@ -10,11 +10,7 @@ from ray import serve
10
10
  NODE_REAPER_DEPLOYMENT_NAME = "NodeReaper"
11
11
 
12
12
 
13
- @serve.deployment(
14
- name=NODE_REAPER_DEPLOYMENT_NAME,
15
- ray_actor_options={"num_cpus": 0.25},
16
- autoscaling_config={"initial_replicas": 1, "min_replicas": 1, "max_replicas": 1},
17
- )
13
+ @serve.deployment
18
14
  class NodeReaper:
19
15
  def __init__(
20
16
  self,
ray_embedding/utils.py CHANGED
@@ -1,4 +1,4 @@
1
- from typing import Optional
1
+ from typing import Optional, Tuple
2
2
 
3
3
  import ray
4
4
  from ray import serve
@@ -10,28 +10,17 @@ from ray_embedding.node_reaper import NODE_REAPER_DEPLOYMENT_NAME
10
10
 
11
11
 
12
12
 
13
- def get_head_node_id() -> str:
13
+ def get_head_node_id() -> Tuple[str, str]:
14
14
  try:
15
15
  nodes = state.list_nodes(filters=[("is_head_node", "=", True)])
16
16
  if not nodes:
17
17
  raise RuntimeError("Unable to locate head node for NodeReaper deployment.")
18
18
  head_node = nodes[0]
19
- return head_node["node_id"]
19
+ return head_node["node_id"], head_node["node_ip"]
20
20
  except Exception as exc:
21
21
  raise RuntimeError("Unable to locate the head node ID for NodeReaper deployment.") from exc
22
22
 
23
23
 
24
- HEAD_NODE_ID = get_head_node_id()
25
-
26
-
27
- def node_affinity_for_head() -> NodeAffinitySchedulingStrategy:
28
- return NodeAffinitySchedulingStrategy(node_id=HEAD_NODE_ID, soft=False)
29
-
30
-
31
- def node_affinity_for_worker() -> NodeAffinitySchedulingStrategy:
32
- return NodeAffinitySchedulingStrategy(node_id=NotIn(HEAD_NODE_ID), soft=False)
33
-
34
-
35
24
  def get_node_reaper_handle() -> DeploymentHandle:
36
25
  try:
37
26
  return serve.context.get_deployment_handle(NODE_REAPER_DEPLOYMENT_NAME)
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: ray-embedding
3
- Version: 0.13.15
3
+ Version: 0.14.0
4
4
  Summary: Deploy SentenceTransformers embedding models to a ray cluster
5
5
  Author: Crispin Almodovar
6
6
  Author-email:
@@ -0,0 +1,11 @@
1
+ ray_embedding/__init__.py,sha256=YS5LAZfRIwwVvE3C9g7hsauvjgIkqKtHyxkwMFFfAGY,46
2
+ ray_embedding/deploy.py,sha256=NYpGDGF8y1rh3Thts-NC4nb8anXQJDC0dFZC18_R2f8,4170
3
+ ray_embedding/dto.py,sha256=6JuAcD6pLfzUL48HfyPnZI7Hb-o66KFM5UtYZOOgwc8,1739
4
+ ray_embedding/embedding_model.py,sha256=Zr5lxVuy60y8-JgsOmKDD44FZlbTL1tiiY-3_72sTR4,4905
5
+ ray_embedding/model_router.py,sha256=W2c0hvqwDe1iCfNx4ee2UT7wKduywMP8dY0Ggb8xBvU,6658
6
+ ray_embedding/node_reaper.py,sha256=ISwSHnQs22B_f3PihND3KYTLkJSDbg1JWIAaKS-qCm0,4800
7
+ ray_embedding/utils.py,sha256=cbdI7q6xSvbl31ZthdM8mz55VrN8pubkoD6RqKGYLUc,1898
8
+ ray_embedding-0.14.0.dist-info/METADATA,sha256=uJlMttPN4bYVZbvLi5g37dCQRyzEzHrj7eqK1fYiv_w,1094
9
+ ray_embedding-0.14.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
10
+ ray_embedding-0.14.0.dist-info/top_level.txt,sha256=ziCblpJq1YsrryshFqxTRuRMgNuO1_tgvAAkGShATNA,14
11
+ ray_embedding-0.14.0.dist-info/RECORD,,
@@ -1,11 +0,0 @@
1
- ray_embedding/__init__.py,sha256=YS5LAZfRIwwVvE3C9g7hsauvjgIkqKtHyxkwMFFfAGY,46
2
- ray_embedding/deploy.py,sha256=QWbDjApgVWaMy_uWOC67zSuwrpj44vZx1ws1ersq-lw,3824
3
- ray_embedding/dto.py,sha256=6JuAcD6pLfzUL48HfyPnZI7Hb-o66KFM5UtYZOOgwc8,1739
4
- ray_embedding/embedding_model.py,sha256=Zr5lxVuy60y8-JgsOmKDD44FZlbTL1tiiY-3_72sTR4,4905
5
- ray_embedding/model_router.py,sha256=W2c0hvqwDe1iCfNx4ee2UT7wKduywMP8dY0Ggb8xBvU,6658
6
- ray_embedding/node_reaper.py,sha256=0lQvq94pdqZE21JeahIyc3LXA4xKtSEv8oELJ1y25QI,4969
7
- ray_embedding/utils.py,sha256=IE0uyBRIW9HVQjO1I46qvtxGQfdorPwU4FsAlYHjI9g,2186
8
- ray_embedding-0.13.15.dist-info/METADATA,sha256=je1cLWUJfUQ8skQDOTzktBgFpvot7nz8Dsy01gDo2mM,1095
9
- ray_embedding-0.13.15.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
10
- ray_embedding-0.13.15.dist-info/top_level.txt,sha256=ziCblpJq1YsrryshFqxTRuRMgNuO1_tgvAAkGShATNA,14
11
- ray_embedding-0.13.15.dist-info/RECORD,,