xinference 0.8.1__py3-none-any.whl → 0.8.3__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.
Potentially problematic release.
This version of xinference might be problematic. Click here for more details.
- xinference/_version.py +3 -3
- xinference/api/oauth2/auth_service.py +132 -0
- xinference/api/restful_api.py +282 -78
- xinference/client/handlers.py +3 -0
- xinference/client/restful/restful_client.py +108 -75
- xinference/constants.py +14 -4
- xinference/core/cache_tracker.py +102 -0
- xinference/core/chat_interface.py +10 -4
- xinference/core/event.py +56 -0
- xinference/core/model.py +44 -0
- xinference/core/resource.py +19 -12
- xinference/core/status_guard.py +4 -0
- xinference/core/supervisor.py +278 -87
- xinference/core/utils.py +68 -3
- xinference/core/worker.py +98 -8
- xinference/deploy/cmdline.py +6 -3
- xinference/deploy/local.py +2 -2
- xinference/deploy/supervisor.py +2 -2
- xinference/model/audio/__init__.py +27 -0
- xinference/model/audio/core.py +161 -0
- xinference/model/audio/model_spec.json +79 -0
- xinference/model/audio/utils.py +18 -0
- xinference/model/audio/whisper.py +132 -0
- xinference/model/core.py +18 -13
- xinference/model/embedding/__init__.py +27 -2
- xinference/model/embedding/core.py +43 -3
- xinference/model/embedding/model_spec.json +24 -0
- xinference/model/embedding/model_spec_modelscope.json +24 -0
- xinference/model/embedding/utils.py +18 -0
- xinference/model/image/__init__.py +12 -1
- xinference/model/image/core.py +63 -9
- xinference/model/image/utils.py +26 -0
- xinference/model/llm/__init__.py +20 -1
- xinference/model/llm/core.py +43 -2
- xinference/model/llm/ggml/chatglm.py +15 -6
- xinference/model/llm/llm_family.json +197 -6
- xinference/model/llm/llm_family.py +9 -7
- xinference/model/llm/llm_family_modelscope.json +189 -4
- xinference/model/llm/pytorch/chatglm.py +3 -3
- xinference/model/llm/pytorch/core.py +4 -2
- xinference/model/{multimodal → llm/pytorch}/qwen_vl.py +10 -8
- xinference/model/llm/pytorch/utils.py +21 -9
- xinference/model/llm/pytorch/yi_vl.py +246 -0
- xinference/model/llm/utils.py +57 -4
- xinference/model/llm/vllm/core.py +5 -4
- xinference/model/rerank/__init__.py +25 -2
- xinference/model/rerank/core.py +51 -9
- xinference/model/rerank/model_spec.json +6 -0
- xinference/model/rerank/model_spec_modelscope.json +7 -0
- xinference/{api/oauth2/common.py → model/rerank/utils.py} +6 -2
- xinference/model/utils.py +5 -3
- xinference/thirdparty/__init__.py +0 -0
- xinference/thirdparty/llava/__init__.py +1 -0
- xinference/thirdparty/llava/conversation.py +205 -0
- xinference/thirdparty/llava/mm_utils.py +122 -0
- xinference/thirdparty/llava/model/__init__.py +1 -0
- xinference/thirdparty/llava/model/clip_encoder/__init__.py +0 -0
- xinference/thirdparty/llava/model/clip_encoder/builder.py +11 -0
- xinference/thirdparty/llava/model/clip_encoder/clip_encoder.py +86 -0
- xinference/thirdparty/llava/model/constants.py +6 -0
- xinference/thirdparty/llava/model/llava_arch.py +385 -0
- xinference/thirdparty/llava/model/llava_llama.py +163 -0
- xinference/thirdparty/llava/model/multimodal_projector/__init__.py +0 -0
- xinference/thirdparty/llava/model/multimodal_projector/builder.py +64 -0
- xinference/types.py +1 -1
- xinference/web/ui/build/asset-manifest.json +3 -3
- xinference/web/ui/build/index.html +1 -1
- xinference/web/ui/build/static/js/main.15822aeb.js +3 -0
- xinference/web/ui/build/static/js/main.15822aeb.js.map +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/139e5e4adf436923107d2b02994c7ff6dba2aac1989e9b6638984f0dfe782c4a.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/52aa27272b4b9968f62666262b47661cb1992336a2aff3b13994cc36877b3ec3.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/64accc515dc6cd584a2873796cd7da6f93de57f7e465eb5423cca9a2f3fe3eff.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/65ca3ba225b8c8dac907210545b51f2fcdb2591f0feeb7195f1c037f2bc956a0.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/b80db1012318b97c329c4e3e72454f7512fb107e57c444b437dbe4ba1a3faa5a.json +1 -0
- {xinference-0.8.1.dist-info → xinference-0.8.3.dist-info}/METADATA +33 -23
- {xinference-0.8.1.dist-info → xinference-0.8.3.dist-info}/RECORD +81 -64
- xinference/api/oauth2/core.py +0 -93
- xinference/model/multimodal/__init__.py +0 -52
- xinference/model/multimodal/core.py +0 -467
- xinference/model/multimodal/model_spec.json +0 -43
- xinference/model/multimodal/model_spec_modelscope.json +0 -45
- xinference/web/ui/build/static/js/main.b83095c2.js +0 -3
- xinference/web/ui/build/static/js/main.b83095c2.js.map +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/101923c539819f26ad11fbcbd6f6e56436b285efbb090dcc7dd648c6e924c4a8.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/4942da6bc03bf7373af068e22f916341aabc5b5df855d73c1d348c696724ce37.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/52a6136cb2dbbf9c51d461724d9b283ebe74a73fb19d5df7ba8e13c42bd7174d.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/71493aadd34d568fbe605cacaba220aa69bd09273251ee4ba27930f8d01fccd8.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/8b071db2a5a9ef68dc14d5f606540bd23d9785e365a11997c510656764d2dccf.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/a4d72d3b806ba061919115f0c513738726872e3c79cf258f007519d3f91d1a16.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/f037ffef5992af0892d6d991053c1dace364cd39a3f11f1a41f92776e8a59459.json +0 -1
- /xinference/web/ui/build/static/js/{main.b83095c2.js.LICENSE.txt → main.15822aeb.js.LICENSE.txt} +0 -0
- {xinference-0.8.1.dist-info → xinference-0.8.3.dist-info}/LICENSE +0 -0
- {xinference-0.8.1.dist-info → xinference-0.8.3.dist-info}/WHEEL +0 -0
- {xinference-0.8.1.dist-info → xinference-0.8.3.dist-info}/entry_points.txt +0 -0
- {xinference-0.8.1.dist-info → xinference-0.8.3.dist-info}/top_level.txt +0 -0
xinference/_version.py
CHANGED
|
@@ -8,11 +8,11 @@ import json
|
|
|
8
8
|
|
|
9
9
|
version_json = '''
|
|
10
10
|
{
|
|
11
|
-
"date": "2024-
|
|
11
|
+
"date": "2024-02-02T12:27:24+0800",
|
|
12
12
|
"dirty": false,
|
|
13
13
|
"error": null,
|
|
14
|
-
"full-revisionid": "
|
|
15
|
-
"version": "0.8.
|
|
14
|
+
"full-revisionid": "749ef3ff298a94b88c1e67415819fae4fb1de75c",
|
|
15
|
+
"version": "0.8.3"
|
|
16
16
|
}
|
|
17
17
|
''' # END VERSION_JSON
|
|
18
18
|
|
|
@@ -0,0 +1,132 @@
|
|
|
1
|
+
# Copyright 2022-2024 XProbe Inc.
|
|
2
|
+
#
|
|
3
|
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
4
|
+
# you may not use this file except in compliance with the License.
|
|
5
|
+
# You may obtain a copy of the License at
|
|
6
|
+
#
|
|
7
|
+
# http://www.apache.org/licenses/LICENSE-2.0
|
|
8
|
+
#
|
|
9
|
+
# Unless required by applicable law or agreed to in writing, software
|
|
10
|
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
11
|
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
12
|
+
# See the License for the specific language governing permissions and
|
|
13
|
+
# limitations under the License.
|
|
14
|
+
from datetime import timedelta
|
|
15
|
+
from typing import List, Optional
|
|
16
|
+
|
|
17
|
+
import pydantic
|
|
18
|
+
from fastapi import Depends, HTTPException, status
|
|
19
|
+
from fastapi.security import OAuth2PasswordBearer, SecurityScopes
|
|
20
|
+
from jose import JWTError, jwt
|
|
21
|
+
from pydantic import BaseModel, ValidationError
|
|
22
|
+
from typing_extensions import Annotated
|
|
23
|
+
|
|
24
|
+
from .types import AuthStartupConfig, User
|
|
25
|
+
from .utils import create_access_token, get_password_hash, verify_password
|
|
26
|
+
|
|
27
|
+
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
|
|
28
|
+
|
|
29
|
+
|
|
30
|
+
class TokenData(BaseModel):
|
|
31
|
+
username: str
|
|
32
|
+
scopes: List[str] = []
|
|
33
|
+
|
|
34
|
+
|
|
35
|
+
class AuthService:
|
|
36
|
+
def __init__(self, auth_config_file: Optional[str]):
|
|
37
|
+
self._auth_config_file = auth_config_file
|
|
38
|
+
self._config = self.init_auth_config()
|
|
39
|
+
|
|
40
|
+
@property
|
|
41
|
+
def config(self):
|
|
42
|
+
return self._config
|
|
43
|
+
|
|
44
|
+
def init_auth_config(self):
|
|
45
|
+
if self._auth_config_file:
|
|
46
|
+
config: AuthStartupConfig = pydantic.parse_file_as(
|
|
47
|
+
path=self._auth_config_file, type_=AuthStartupConfig
|
|
48
|
+
)
|
|
49
|
+
for user in config.user_config:
|
|
50
|
+
user.password = get_password_hash(user.password)
|
|
51
|
+
return config
|
|
52
|
+
|
|
53
|
+
def __call__(
|
|
54
|
+
self,
|
|
55
|
+
security_scopes: SecurityScopes,
|
|
56
|
+
token: Annotated[str, Depends(oauth2_scheme)],
|
|
57
|
+
):
|
|
58
|
+
"""
|
|
59
|
+
Advanced dependencies. See: https://fastapi.tiangolo.com/advanced/advanced-dependencies/
|
|
60
|
+
"""
|
|
61
|
+
if security_scopes.scopes:
|
|
62
|
+
authenticate_value = f'Bearer scope="{security_scopes.scope_str}"'
|
|
63
|
+
else:
|
|
64
|
+
authenticate_value = "Bearer"
|
|
65
|
+
credentials_exception = HTTPException(
|
|
66
|
+
status_code=status.HTTP_401_UNAUTHORIZED,
|
|
67
|
+
detail="Could not validate credentials",
|
|
68
|
+
headers={"WWW-Authenticate": authenticate_value},
|
|
69
|
+
)
|
|
70
|
+
|
|
71
|
+
try:
|
|
72
|
+
assert self._config is not None
|
|
73
|
+
payload = jwt.decode(
|
|
74
|
+
token,
|
|
75
|
+
self._config.auth_config.secret_key,
|
|
76
|
+
algorithms=[self._config.auth_config.algorithm],
|
|
77
|
+
options={"verify_exp": False}, # TODO: supports token expiration
|
|
78
|
+
)
|
|
79
|
+
username: str = payload.get("sub")
|
|
80
|
+
if username is None:
|
|
81
|
+
raise credentials_exception
|
|
82
|
+
token_scopes = payload.get("scopes", [])
|
|
83
|
+
token_data = TokenData(scopes=token_scopes, username=username)
|
|
84
|
+
except (JWTError, ValidationError):
|
|
85
|
+
raise credentials_exception
|
|
86
|
+
user = self.get_user(token_data.username)
|
|
87
|
+
if user is None:
|
|
88
|
+
raise credentials_exception
|
|
89
|
+
if "admin" in token_data.scopes:
|
|
90
|
+
return user
|
|
91
|
+
for scope in security_scopes.scopes:
|
|
92
|
+
if scope not in token_data.scopes:
|
|
93
|
+
raise HTTPException(
|
|
94
|
+
status_code=status.HTTP_403_FORBIDDEN,
|
|
95
|
+
detail="Not enough permissions",
|
|
96
|
+
headers={"WWW-Authenticate": authenticate_value},
|
|
97
|
+
)
|
|
98
|
+
return user
|
|
99
|
+
|
|
100
|
+
def get_user(self, username: str) -> Optional[User]:
|
|
101
|
+
for user in self._config.user_config:
|
|
102
|
+
if user.username == username:
|
|
103
|
+
return user
|
|
104
|
+
return None
|
|
105
|
+
|
|
106
|
+
def authenticate_user(self, username: str, password: str):
|
|
107
|
+
user = self.get_user(username)
|
|
108
|
+
if not user:
|
|
109
|
+
return False
|
|
110
|
+
if not verify_password(password, user.password):
|
|
111
|
+
return False
|
|
112
|
+
return user
|
|
113
|
+
|
|
114
|
+
def generate_token_for_user(self, username: str, password: str):
|
|
115
|
+
user = self.authenticate_user(username, password)
|
|
116
|
+
if not user:
|
|
117
|
+
raise HTTPException(
|
|
118
|
+
status_code=status.HTTP_401_UNAUTHORIZED,
|
|
119
|
+
detail="Incorrect username or password",
|
|
120
|
+
headers={"WWW-Authenticate": "Bearer"},
|
|
121
|
+
)
|
|
122
|
+
assert user is not None and isinstance(user, User)
|
|
123
|
+
access_token_expires = timedelta(
|
|
124
|
+
minutes=self._config.auth_config.token_expire_in_minutes
|
|
125
|
+
)
|
|
126
|
+
access_token = create_access_token(
|
|
127
|
+
data={"sub": user.username, "scopes": user.permissions},
|
|
128
|
+
secret_key=self._config.auth_config.secret_key,
|
|
129
|
+
algorithm=self._config.auth_config.algorithm,
|
|
130
|
+
expires_delta=access_token_expires,
|
|
131
|
+
)
|
|
132
|
+
return {"access_token": access_token, "token_type": "bearer"}
|