| title: | max.experimental.sharding |
|---|---|
| type: | module |
| lang: | python |
| wrapper_class: | rst-module-autosummary |
.. automodule:: max.experimental.sharding :no-members:
.. currentmodule:: max.experimental.sharding
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/class.rst DeviceMesh
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/class.rst Partial Placement Replicated Sharded Unknown
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/class.rst BufferLayout TensorLayout
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/class.rst DeviceMapping NamedMapping
A sharding rule receives the op's arguments with each tensor replaced by its :class:`TensorLayout`, and returns the :class:`AxisAssignment` rows the op accepts: what each tensor operand must hold and what the op produces. These functions build common rules and rows.
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/class.rst AxisAssignment
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/function.rst match_operand_placement replicated_rows
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/function.rst auto_reshard
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/data.rst ALL_TRANSITIONS DEFAULT_TRANSITIONS Transition
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/function.rst visualize
.. autosummary:: :nosignatures: :toctree: generated :template: autosummary/class.rst ConversionError ShardingError