A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 11:00 UTC
Update deprecated type hinting in `vllm/adapter_commons` (#18073)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
May 13, 2025 at 16:32 UTC
0b217da646fd4cc08cd0dd20d0ea69f81d64ab35
5 files changed
+18
-20
pyproject.toml
-1
@@ -72,7 +72,6 @@ exclude = [
72
"vllm/version.py" = ["F401"]
73
"vllm/_version.py" = ["ALL"]
74
# Python 3.8 typing. TODO: Remove these excludes after v1.0.0
75
-"vllm/adapter_commons/**/*.py" = ["UP006", "UP035"]
75
"vllm/attention/**/*.py" = ["UP006", "UP035"]
76
"vllm/core/**/*.py" = ["UP006", "UP035"]
77
"vllm/device_allocator/**/*.py" = ["UP006", "UP035"]
vllm/adapter_commons/layers.py
+2
-3
@@ -1,15 +1,14 @@
1
# SPDX-License-Identifier: Apache-2.0
2
3
from dataclasses import dataclass
4
-from typing import Tuple
4
5
6
@dataclass
7
class AdapterMapping:
8
# Per every token in input_ids:
10
- index_mapping: Tuple[int, ...]
9
+ index_mapping: tuple[int, ...]
10
# Per sampled token:
12
- prompt_mapping: Tuple[int, ...]
11
+ prompt_mapping: tuple[int, ...]
12
13
def __post_init__(self):
14
self.index_mapping = tuple(self.index_mapping)
vllm/adapter_commons/models.py
+4
-4
@@ -1,7 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
3
from abc import ABC, abstractmethod
4
-from typing import Any, Callable, Dict, Optional, TypeVar
4
+from typing import Any, Callable, Optional, TypeVar
5
6
from torch import nn
7
@@ -49,9 +49,9 @@ class AdapterModelManager(ABC):
49
model: the model to be adapted.
50
"""
51
self.model: nn.Module = model
52
- self._registered_adapters: Dict[int, Any] = {}
52
+ self._registered_adapters: dict[int, Any] = {}
53
# Dict instead of a Set for compatibility with LRUCache.
54
- self._active_adapters: Dict[int, None] = {}
54
+ self._active_adapters: dict[int, None] = {}
55
self.adapter_type = 'Adapter'
56
self._last_mapping = None
57
@@ -97,7 +97,7 @@ class AdapterModelManager(ABC):
97
raise NotImplementedError
98
99
@abstractmethod
100
- def list_adapters(self) -> Dict[int, Any]:
100
+ def list_adapters(self) -> dict[int, Any]:
101
raise NotImplementedError
102
103
@abstractmethod
vllm/adapter_commons/utils.py
+9
-9
@@ -1,10 +1,10 @@
1
# SPDX-License-Identifier: Apache-2.0
2
3
-from typing import Any, Callable, Dict, Optional, Set
3
+from typing import Any, Callable, Optional
4
5
6
## model functions
7
-def deactivate_adapter(adapter_id: int, active_adapters: Dict[int, None],
7
+def deactivate_adapter(adapter_id: int, active_adapters: dict[int, None],
8
deactivate_func: Callable) -> bool:
9
if adapter_id in active_adapters:
10
deactivate_func(adapter_id)
@@ -13,7 +13,7 @@ def deactivate_adapter(adapter_id: int, active_adapters: Dict[int, None],
13
return False
14
15
16
-def add_adapter(adapter: Any, registered_adapters: Dict[int, Any],
16
+def add_adapter(adapter: Any, registered_adapters: dict[int, Any],
17
capacity: int, add_func: Callable) -> bool:
18
if adapter.id not in registered_adapters:
19
if len(registered_adapters) >= capacity:
@@ -32,23 +32,23 @@ def set_adapter_mapping(mapping: Any, last_mapping: Any,
32
return last_mapping
33
34
35
-def remove_adapter(adapter_id: int, registered_adapters: Dict[int, Any],
35
+def remove_adapter(adapter_id: int, registered_adapters: dict[int, Any],
36
deactivate_func: Callable) -> bool:
37
deactivate_func(adapter_id)
38
return bool(registered_adapters.pop(adapter_id, None))
39
40
41
-def list_adapters(registered_adapters: Dict[int, Any]) -> Dict[int, Any]:
41
+def list_adapters(registered_adapters: dict[int, Any]) -> dict[int, Any]:
42
return dict(registered_adapters)
43
44
45
def get_adapter(adapter_id: int,
46
- registered_adapters: Dict[int, Any]) -> Optional[Any]:
46
+ registered_adapters: dict[int, Any]) -> Optional[Any]:
47
return registered_adapters.get(adapter_id)
48
49
50
## worker functions
51
-def set_active_adapters_worker(requests: Set[Any], mapping: Optional[Any],
51
+def set_active_adapters_worker(requests: set[Any], mapping: Optional[Any],
52
apply_adapters_func,
53
set_adapter_mapping_func) -> None:
54
apply_adapters_func(requests)
@@ -66,7 +66,7 @@ def add_adapter_worker(adapter_request: Any, list_adapters_func,
66
return loaded
67
68
69
-def apply_adapters_worker(adapter_requests: Set[Any], list_adapters_func,
69
+def apply_adapters_worker(adapter_requests: set[Any], list_adapters_func,
70
adapter_slots: int, remove_adapter_func,
71
add_adapter_func) -> None:
72
models_that_exist = list_adapters_func()
@@ -88,5 +88,5 @@ def apply_adapters_worker(adapter_requests: Set[Any], list_adapters_func,
88
add_adapter_func(models_map[adapter_id])
89
90
91
-def list_adapters_worker(adapter_manager_list_adapters_func) -> Set[int]:
91
+def list_adapters_worker(adapter_manager_list_adapters_func) -> set[int]:
92
return set(adapter_manager_list_adapters_func())
vllm/adapter_commons/worker_manager.py
+3
-3
@@ -1,7 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
3
from abc import ABC, abstractmethod
4
-from typing import Any, Optional, Set
4
+from typing import Any, Optional
5
6
import torch
7
@@ -17,7 +17,7 @@ class AbstractWorkerManager(ABC):
17
raise NotImplementedError
18
19
@abstractmethod
20
- def set_active_adapters(self, requests: Set[Any],
20
+ def set_active_adapters(self, requests: set[Any],
21
mapping: Optional[Any]) -> None:
22
raise NotImplementedError
23
@@ -34,5 +34,5 @@ class AbstractWorkerManager(ABC):
34
raise NotImplementedError
35
36
@abstractmethod
37
- def list_adapters(self) -> Set[int]:
37
+ def list_adapters(self) -> set[int]:
38
raise NotImplementedError