Skip to content

vllm.v1.kv_offload.mediums

BlockIDsLoadStoreSpec

Bases: LoadStoreSpec, ABC

Spec for loading/storing KV blocks from given block numbers.

Source code in vllm/v1/kv_offload/mediums.py
class BlockIDsLoadStoreSpec(LoadStoreSpec, ABC):
    """
    Spec for loading/storing KV blocks from given block numbers.
    """

    def __init__(self, block_ids: list[int]):
        self.block_ids = np.array(block_ids, dtype=np.int64)

    def __repr__(self) -> str:
        return repr(self.block_ids)

CPULoadStoreSpec

Bases: BlockIDsLoadStoreSpec

Spec for loading/storing a KV block to CPU memory.

Source code in vllm/v1/kv_offload/mediums.py
class CPULoadStoreSpec(BlockIDsLoadStoreSpec):
    """
    Spec for loading/storing a KV block to CPU memory.
    """

    @staticmethod
    def medium() -> str:
        return "CPU"

GPULoadStoreSpec

Bases: BlockIDsLoadStoreSpec

Spec for loading/storing a KV block to GPU memory.

Source code in vllm/v1/kv_offload/mediums.py
class GPULoadStoreSpec(BlockIDsLoadStoreSpec):
    """
    Spec for loading/storing a KV block to GPU memory.
    """

    @staticmethod
    def medium() -> str:
        return "GPU"