interfaces.py 2.47 KB
Newer Older
1
2
from abc import ABC, abstractmethod
from typing import Dict, FrozenSet, List, Optional, Protocol
3
4
5
6
7
8
9
10
11
12

from vllm.utils import Device


class Block(ABC):

    @abstractmethod
    def append_token_ids(self, token_ids: List[int]) -> None:
        pass

13
14
    @property
    @abstractmethod
15
16
17
    def block_id(self) -> Optional[int]:
        pass

18
19
    @property
    @abstractmethod
20
21
22
    def token_ids(self) -> List[int]:
        pass

23
24
    @property
    @abstractmethod
25
26
27
    def num_empty_slots(self) -> int:
        pass

28
29
    @property
    @abstractmethod
30
31
32
    def is_full(self) -> bool:
        pass

33
34
    @property
    @abstractmethod
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
    def prev_block(self) -> Optional["Block"]:
        pass

    class Factory(Protocol):

        @abstractmethod
        def __call__(
            self,
            prev_block: Optional["Block"],
            token_ids: List[int],
            block_size: int,
            allocator: "BlockAllocator",
            block_id: Optional[int] = None,
        ) -> "Block":
            pass


class BlockAllocator(ABC):

    @abstractmethod
55
56
    def allocate_mutable(self, prev_block: Optional[Block],
                         device: Device) -> Block:
57
58
59
60
        pass

    @abstractmethod
    def allocate_immutable(self, prev_block: Optional[Block],
61
                           token_ids: List[int], device: Device) -> Block:
62
63
64
65
66
67
68
69
70
71
72
        pass

    @abstractmethod
    def free(self, block: Block) -> None:
        pass

    @abstractmethod
    def fork(self, last_block: Block) -> List[Block]:
        pass

    @abstractmethod
73
    def get_num_free_blocks(self, device: Device) -> int:
74
75
        pass

76
77
78
    @property
    @abstractmethod
    def all_block_ids(self) -> FrozenSet[int]:
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
        pass

    @abstractmethod
    def clear_copy_on_writes(self) -> Dict[int, List[int]]:
        pass

    @abstractmethod
    def mark_blocks_as_computed(self) -> None:
        pass

    @abstractmethod
    def get_common_computed_block_ids(
            self, seq_block_ids: List[List[int]]) -> List[int]:
        pass

    class NoFreeBlocksError(ValueError):
        pass


class DeviceAwareBlockAllocator(BlockAllocator):

    @abstractmethod
    def allocate_mutable(self, prev_block: Optional[Block],
                         device: Device) -> Block:
        pass

    @abstractmethod
    def allocate_immutable(self, prev_block: Optional[Block],
                           token_ids: List[int], device: Device) -> Block:
        pass

    @abstractmethod
    def get_num_free_blocks(self, device: Device) -> int:
        pass