class TikiTakaIdealizedPreset(UnitCellRPUConfig):
"""Configuration using Tiki-taka with
:class:`~aihwkit.simulator.presets.devices.IdealizedPresetDevice`.
See :class:`~aihwkit.simulator.configs.devices.TransferCompound`
for details on Tiki-taka-like optimizers.
The default peripheral hardware
(:class:`~aihwkit.simulator.presets.utils.PresetIOParameters`) and
analog update
(:class:`~aihwkit.simulator.presets.utils.PresetUpdateParameters`)
configuration is used otherwise.
"""
device: UnitCell = field(
default_factory=lambda: TransferCompound(
unit_cell_devices=[IdealizedPresetDevice(), IdealizedPresetDevice()],
transfer_forward=PresetIOParameters(),
transfer_update=PresetUpdateParameters(),
transfer_every=1.0,
units_in_mbatch=True,
))
forward: IOParameters = field(default_factory=PresetIOParameters)
backward: IOParameters = field(default_factory=PresetIOParameters)
update: UpdateParameters = field(default_factory=PresetUpdateParameters)
def get_transfer_compound(gamma, **kwargs):
"""Get a Tiki-taka compound with reference cell """
def custom_device(**kwargs):
"""Custom device """
return SoftBoundsDevice(w_max_dtod=0.0,
w_min_dtod=0.0,
w_max=1.0,
w_min=-1.0,
**kwargs)
rpu_config = UnitCellRPUConfig(device=TransferCompound(
# Devices that compose the Tiki-taka compound.
unit_cell_devices=[
# fast "A" matrix
ReferenceUnitCell(
[custom_device(**kwargs),
custom_device(**kwargs)]),
# slow "C" matrix
ReferenceUnitCell(
[custom_device(**kwargs),
custom_device(**kwargs)])
],
gamma=gamma,
))
return rpu_config
def get_rpu_config(self):
return UnitCellRPUConfig(
device=TransferCompound(unit_cell_devices=[
SoftBoundsDevice(w_max_dtod=0, w_min_dtod=0),
SoftBoundsDevice(w_max_dtod=0, w_min_dtod=0)
],
transfer_forward=IOParameters(
is_perfect=True),
transfer_every=1,
gamma=0.1))
from aihwkit.simulator.rpu_base import cuda
# Prepare the datasets (input and expected output).
x = Tensor([[0.1, 0.2, 0.4, 0.3], [0.2, 0.1, 0.1, 0.3]])
y = Tensor([[1.0, 0.5], [0.7, 0.3]])
# The Tiki-taka learning rule can be implemented using the transfer device.
rpu_config = UnitCellRPUConfig(
device=TransferCompound(
# Devices that compose the Tiki-taka compound.
unit_cell_devices=[
SoftBoundsDevice(w_min=-0.3, w_max=0.3),
SoftBoundsDevice(w_min=-0.6, w_max=0.6)
],
# Make some adjustments of the way Tiki-Taka is performed.
units_in_mbatch=True, # batch_size=1 anyway
transfer_every=2, # every 2 batches do a transfer-read
n_cols_per_transfer=1, # one forward read for each transfer
gamma=0.0, # all SGD weight in second device
scale_transfer_lr=True, # in relative terms to SGD LR
transfer_lr=1.0, # same transfer LR as for SGD
)
)
# Make more adjustments (can be made here or above).
rpu_config.forward.inp_res = 1/64. # 6 bit DAC
# same backward pass settings as forward
rpu_config.backward = rpu_config.forward
from aihwkit.simulator.rpu_base import cuda
# Prepare the datasets (input and expected output).
x = Tensor([[0.1, 0.2, 0.4, 0.3], [0.2, 0.1, 0.1, 0.3]])
y = Tensor([[1.0, 0.5], [0.7, 0.3]])
# The Tiki-taka learning rule can be implemented using the transfer device.
rpu_config = UnitCellRPUConfig(device=TransferCompound(
# Devices that compose the Tiki-taka compound.
unit_cell_devices=[
SoftBoundsDevice(w_min=-0.3, w_max=0.3),
SoftBoundsDevice(w_min=-0.6, w_max=0.6)
],
# Make some adjustments of the way Tiki-Taka is performed.
units_in_mbatch=True, # batch_size=1 anyway
transfer_every=2, # every 2 batches do a transfer-read
n_reads_per_transfer=1, # one forward read for each transfer
gamma=0.0, # all SGD weight in second device
scale_transfer_lr=True, # in relative terms to SGD LR
transfer_lr=1.0, # same transfer LR as for SGD
fast_lr=0.1, # SGD update onto first matrix constant
transfer_columns=True # transfer use columns (not rows)
))
# Make more adjustments (can be made here or above).
rpu_config.forward.inp_res = 1 / 64. # 6 bit DAC
# same backward pass settings as forward
rpu_config.backward = rpu_config.forward