Skip to content

vllm.v1.attention.backends.mamba_attn

M module-attribute 

M = TypeVar('M')

BaseMambaAttentionMetadataBuilder

Bases: AttentionMetadataBuilder[M], ABC

Source code in vllm/v1/attention/backends/mamba_attn.py 
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
class BaseMambaAttentionMetadataBuilder(AttentionMetadataBuilder[M], abc.ABC):
    reorder_batch_threshold: int = 1
    cudagraph_support: ClassVar[AttentionCGSupport] = (
        AttentionCGSupport.UNIFORM_SINGLE_TOKEN_DECODE
    )

    def __init__(
        self,
        kv_cache_spec: AttentionSpec,
        layer_names: list[str],
        vllm_config: VllmConfig,
        device: torch.device,
    ):
        super().__init__(kv_cache_spec, layer_names, vllm_config, device)

        assert isinstance(kv_cache_spec, MambaSpec)
        self.compilation_config = vllm_config.compilation_config
        self.decode_cudagraph_max_bs = min(
            self.vllm_config.scheduler_config.max_num_seqs,
            self.compilation_config.max_cudagraph_capture_size,
        )
        self.state_indices_tensor = torch.empty(
            (self.decode_cudagraph_max_bs,),
            dtype=torch.int32,
            device=device,
        )

    def build_for_cudagraph_capture(
        self, common_attn_metadata: CommonAttentionMetadata
    ) -> M:
        """
        This method builds the metadata for full cudagraph capture.
        Currently, only decode is supported for full cudagraphs with Mamba.
        """
        m = common_attn_metadata

        assert m.num_reqs == m.num_actual_tokens, (
            "Mamba only supports decode-only full CUDAGraph capture. "
            "Make sure all cudagraph capture sizes <= max_num_seq."
        )

        m.max_query_len = 1  # decode-only

        return self.build(0, m)

compilation_config instance-attribute 

compilation_config = compilation_config

cudagraph_support class-attribute 

cudagraph_support: AttentionCGSupport = (
    UNIFORM_SINGLE_TOKEN_DECODE
)

decode_cudagraph_max_bs instance-attribute 

decode_cudagraph_max_bs = min(
    max_num_seqs, max_cudagraph_capture_size
)

reorder_batch_threshold class-attribute instance-attribute 

reorder_batch_threshold: int = 1

state_indices_tensor instance-attribute 

state_indices_tensor = empty(
    (decode_cudagraph_max_bs,), dtype=int32, device=device
)

__init__ 

__init__(
    kv_cache_spec: AttentionSpec,
    layer_names: list[str],
    vllm_config: VllmConfig,
    device: device,
)
Source code in vllm/v1/attention/backends/mamba_attn.py 
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
def __init__(
    self,
    kv_cache_spec: AttentionSpec,
    layer_names: list[str],
    vllm_config: VllmConfig,
    device: torch.device,
):
    super().__init__(kv_cache_spec, layer_names, vllm_config, device)

    assert isinstance(kv_cache_spec, MambaSpec)
    self.compilation_config = vllm_config.compilation_config
    self.decode_cudagraph_max_bs = min(
        self.vllm_config.scheduler_config.max_num_seqs,
        self.compilation_config.max_cudagraph_capture_size,
    )
    self.state_indices_tensor = torch.empty(
        (self.decode_cudagraph_max_bs,),
        dtype=torch.int32,
        device=device,
    )

build_for_cudagraph_capture 

build_for_cudagraph_capture(
    common_attn_metadata: CommonAttentionMetadata,
) -> M

This method builds the metadata for full cudagraph capture. Currently, only decode is supported for full cudagraphs with Mamba.

Source code in vllm/v1/attention/backends/mamba_attn.py 
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
def build_for_cudagraph_capture(
    self, common_attn_metadata: CommonAttentionMetadata
) -> M:
    """
    This method builds the metadata for full cudagraph capture.
    Currently, only decode is supported for full cudagraphs with Mamba.
    """
    m = common_attn_metadata

    assert m.num_reqs == m.num_actual_tokens, (
        "Mamba only supports decode-only full CUDAGraph capture. "
        "Make sure all cudagraph capture sizes <= max_num_seq."
    )

    m.max_query_len = 1  # decode-only

    return self.build(0, m)