Skip to content

vllm.v1.worker.ubatch_utils

UBatchSlices module-attribute 

UBatchSlices: TypeAlias = list[UBatchSlice]

UBatchSlice dataclass

Source code in vllm/v1/worker/ubatch_utils.py 
11
12
13
14
15
16
17
18
19
20
21
22
23
24
@dataclass
class UBatchSlice:
    request_slice: slice
    token_slice: slice

    def is_empty(self) -> bool:
        return (
            self.request_slice.start == self.request_slice.stop
            or self.token_slice.start == self.token_slice.stop
        )

    @property
    def num_tokens(self) -> int:
        return self.token_slice.stop - self.token_slice.start

num_tokens property 

num_tokens: int

request_slice instance-attribute 

request_slice: slice

token_slice instance-attribute 

token_slice: slice

__init__ 

__init__(request_slice: slice, token_slice: slice) -> None

is_empty 

is_empty() -> bool
Source code in vllm/v1/worker/ubatch_utils.py 
16
17
18
19
20
def is_empty(self) -> bool:
    return (
        self.request_slice.start == self.request_slice.stop
        or self.token_slice.start == self.token_slice.stop
    )

check_ubatch_thresholds 

check_ubatch_thresholds(
    config: ParallelConfig,
    num_tokens: int,
    uniform_decode: bool,
) -> bool
Source code in vllm/v1/worker/ubatch_utils.py 
34
35
36
37
38
39
40
41
42
def check_ubatch_thresholds(
    config: ParallelConfig, num_tokens: int, uniform_decode: bool
) -> bool:
    if not config.enable_dbo:
        return False
    if uniform_decode:
        return num_tokens >= config.dbo_decode_token_threshold
    else:
        return num_tokens >= config.dbo_prefill_token_threshold

create_ubatch_slices 

create_ubatch_slices(
    num_scheduled_tokens: ndarray, split_point: int
) -> UBatchSlices
Source code in vllm/v1/worker/ubatch_utils.py 
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
def create_ubatch_slices(
    num_scheduled_tokens: np.ndarray, split_point: int
) -> UBatchSlices:
    # TODO(lucas): Refactor the gpu_model_runner.py so we can pass
    # in cu_num_tokens directly (i.e. query_start_loc)
    cu_num_tokens = np.zeros(len(num_scheduled_tokens) + 1, dtype=np.int32)
    np.cumsum(num_scheduled_tokens, dtype=np.int32, out=cu_num_tokens[1:])

    first_ubatch_token_slice = slice(0, split_point)
    second_ubatch_token_slice = slice(split_point, cu_num_tokens[-1])

    # Determine request slices using exclusive stop semantics
    # First ubatch includes requests whose tokens overlap [0, split_point)
    first_ubatch_req_stop = int(
        np.searchsorted(cu_num_tokens, split_point, side="left")
    )
    first_ubatch_req_slice = slice(0, first_ubatch_req_stop)

    # Second ubatch starts at the request that contains the split_point
    # or the request starting exactly at split_point (if on boundary)
    second_ubatch_req_start = int(
        np.searchsorted(cu_num_tokens, split_point, side="right") - 1
    )
    second_ubatch_req_slice = slice(second_ubatch_req_start, len(cu_num_tokens) - 1)

    return [
        UBatchSlice(first_ubatch_req_slice, first_ubatch_token_slice),
        UBatchSlice(second_ubatch_req_slice, second_ubatch_token_slice),
    ]

is_second_ubatch_empty 

is_second_ubatch_empty(
    orig_num_tokens: int, padded_num_tokens: int
) -> bool
Source code in vllm/v1/worker/ubatch_utils.py 
30
31
def is_second_ubatch_empty(orig_num_tokens: int, padded_num_tokens: int) -> bool:
    return (padded_num_tokens // 2) >= orig_num_tokens