block.py 1.11 KB
Newer Older
1
"""Token blocks."""
2
from typing import List
Woosuk Kwon's avatar
Woosuk Kwon committed
3

Woosuk Kwon's avatar
Woosuk Kwon committed
4
from vllm.utils import Device
Woosuk Kwon's avatar
Woosuk Kwon committed
5

6
7
DEFAULT_LAST_ACCESSED_TIME = -1

Woosuk Kwon's avatar
Woosuk Kwon committed
8
9

class PhysicalTokenBlock:
10
    """Represents the state of a block in the KV cache."""
Woosuk Kwon's avatar
Woosuk Kwon committed
11
12
13
14
15
16

    def __init__(
        self,
        device: Device,
        block_number: int,
        block_size: int,
17
18
        block_hash: int,
        num_hashed_tokens: int,
Woosuk Kwon's avatar
Woosuk Kwon committed
19
20
21
22
    ) -> None:
        self.device = device
        self.block_number = block_number
        self.block_size = block_size
23
24
        self.block_hash = block_hash
        self.num_hashed_tokens = num_hashed_tokens
Woosuk Kwon's avatar
Woosuk Kwon committed
25
26

        self.ref_count = 0
27
28
29
        self.last_accessed = DEFAULT_LAST_ACCESSED_TIME

        self.computed = False
Woosuk Kwon's avatar
Woosuk Kwon committed
30
31
32
33

    def __repr__(self) -> str:
        return (f'PhysicalTokenBlock(device={self.device}, '
                f'block_number={self.block_number}, '
34
35
36
37
                f'num_hashed_tokens={self.num_hashed_tokens}, '
                f'ref_count={self.ref_count}, '
                f'last_accessed={self.last_accessed}, '
                f'computed={self.computed})')
38
39
40
41


# Mapping: logical block number -> physical block.
BlockTable = List[PhysicalTokenBlock]