| title: | max.nn.kv_cache |
|---|
| type: | module |
|---|
| lang: | python |
|---|
| wrapper_class: | rst-module-autosummary |
|---|
.. automodule:: max.nn.kv_cache
:no-members:
.. currentmodule:: max.nn.kv_cache
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/class.rst
CacheLeafParamInterface
HostStagingAlloc
KVCacheAssignments
KVCacheBuffer
KVCacheBufferInterface
KVCacheParamInterface
KVCacheParams
KVCacheGroupId
MHAKVCacheParams
MLAKVCacheParams
MSAKVCacheParams
KVCacheQuantizationConfig
KVConnectorType
KVConnectorConfigInterface
NullKVConnectorConfig
KVCacheMemory
MultiKVCacheBuffer
MultiKVCacheParams
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/class.rst
KVCacheInputs
KVCacheInputsPerDevice
BatchCharacteristics
PagedCacheValues
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/class.rst
CacheLeafKind
KVLeafRegion
PagedKVLeafRegion
RecurrentKVLeafRegion
RecurrentLeafInputs
RecurrentStateBuffer
RecurrentStateInputsPerDevice
RecurrentStateParams
RecurrentStateRegion
ScratchRecurrentKVLeafRegion
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/class.rst
AttnKey
AttnKeyInterface
MHAAttnKey
MLAAttnKey
MSAAttnKey
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/class.rst
KVCacheMetrics
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/data.rst
KVHashAlgo
PACKED_PAGE_STRIDE
.. autosummary::
:nosignatures:
:toctree: generated
:template: autosummary/function.rst
build_max_lengths_tensors
compute_max_seq_len_fitting_in_cache
compute_num_device_blocks
estimated_memory_size
max_pages_per_request
packed_page_stride
padded_lut_cols
recurrent_leaf
recurrent_leaves
spec_decode_cache_slack