166 lines
4.3 KiB
Python
166 lines
4.3 KiB
Python
from typing import Callable, Optional, Tuple, cast
|
|
|
|
from ..config import registry
|
|
from ..initializers import glorot_uniform_init, zero_init
|
|
from ..model import Model
|
|
from ..types import Floats1d, Floats2d
|
|
from ..util import get_width, partial
|
|
from .chain import chain
|
|
from .dropout import Dropout
|
|
from .layernorm import LayerNorm
|
|
|
|
|
|
@registry.layers("ClippedLinear.v1")
|
|
def ClippedLinear(
|
|
nO: Optional[int] = None,
|
|
nI: Optional[int] = None,
|
|
*,
|
|
init_W: Optional[Callable] = None,
|
|
init_b: Optional[Callable] = None,
|
|
dropout: Optional[float] = None,
|
|
normalize: bool = False,
|
|
slope: float = 1.0,
|
|
offset: float = 0.0,
|
|
min_val: float = 0.0,
|
|
max_val: float = 1.0,
|
|
) -> Model[Floats2d, Floats2d]:
|
|
if init_W is None:
|
|
init_W = glorot_uniform_init
|
|
if init_b is None:
|
|
init_b = zero_init
|
|
model_attrs = {
|
|
"slope": slope,
|
|
"offset": offset,
|
|
"min_val": min_val,
|
|
"max_val": max_val,
|
|
}
|
|
model: Model[Floats2d, Floats2d] = Model(
|
|
"clipped_linear",
|
|
forward=forward,
|
|
init=partial(init, init_W, init_b),
|
|
dims={"nO": nO, "nI": nI},
|
|
params={"W": None, "b": None},
|
|
attrs=model_attrs,
|
|
)
|
|
if normalize:
|
|
model = chain(model, LayerNorm(nI=nO))
|
|
if dropout is not None:
|
|
model = chain(model, cast(Model[Floats2d, Floats2d], Dropout(dropout)))
|
|
return model
|
|
|
|
|
|
def forward(
|
|
model: Model[Floats2d, Floats2d],
|
|
X: Floats2d,
|
|
is_train: bool,
|
|
) -> Tuple[Floats2d, Callable]:
|
|
slope = model.attrs["slope"]
|
|
offset = model.attrs["offset"]
|
|
min_val = model.attrs["min_val"]
|
|
max_val = model.attrs["max_val"]
|
|
W = cast(Floats2d, model.get_param("W"))
|
|
b = cast(Floats1d, model.get_param("b"))
|
|
Y_preact = model.ops.affine(X, W, b)
|
|
Y = model.ops.clipped_linear(Y_preact, slope, offset, min_val, max_val)
|
|
|
|
def backprop(dY: Floats2d) -> Floats2d:
|
|
dY = model.ops.backprop_clipped_linear(
|
|
dY, Y_preact, slope, offset, min_val, max_val, inplace=False
|
|
)
|
|
model.inc_grad("b", dY.sum(axis=0))
|
|
model.inc_grad("W", model.ops.gemm(dY, X, trans1=True))
|
|
return model.ops.gemm(dY, W)
|
|
|
|
return Y, backprop
|
|
|
|
|
|
def init(
|
|
init_W: Callable,
|
|
init_b: Callable,
|
|
model: Model[Floats2d, Floats2d],
|
|
X: Optional[Floats2d] = None,
|
|
Y: Optional[Floats2d] = None,
|
|
) -> None:
|
|
if X is not None:
|
|
model.set_dim("nI", get_width(X))
|
|
if Y is not None:
|
|
model.set_dim("nO", get_width(Y))
|
|
model.set_param("W", init_W(model.ops, (model.get_dim("nO"), model.get_dim("nI"))))
|
|
model.set_param("b", init_b(model.ops, (model.get_dim("nO"),)))
|
|
|
|
|
|
@registry.layers("HardSigmoid.v1")
|
|
def HardSigmoid(
|
|
nO: Optional[int] = None,
|
|
nI: Optional[int] = None,
|
|
*,
|
|
init_W: Optional[Callable] = None,
|
|
init_b: Optional[Callable] = None,
|
|
dropout: Optional[float] = None,
|
|
normalize: bool = False,
|
|
) -> Model[Floats2d, Floats2d]:
|
|
if init_W is None:
|
|
init_W = glorot_uniform_init
|
|
if init_b is None:
|
|
init_b = zero_init
|
|
return ClippedLinear(
|
|
nO=nO,
|
|
nI=nI,
|
|
init_W=init_W,
|
|
dropout=dropout,
|
|
normalize=normalize,
|
|
slope=0.2,
|
|
offset=0.5,
|
|
)
|
|
|
|
|
|
@registry.layers("HardTanh.v1")
|
|
def HardTanh(
|
|
nO: Optional[int] = None,
|
|
nI: Optional[int] = None,
|
|
*,
|
|
init_W: Optional[Callable] = None,
|
|
init_b: Optional[Callable] = None,
|
|
dropout: Optional[float] = None,
|
|
normalize: bool = False,
|
|
) -> Model[Floats2d, Floats2d]:
|
|
if init_W is None:
|
|
init_W = glorot_uniform_init
|
|
if init_b is None:
|
|
init_b = zero_init
|
|
return ClippedLinear(
|
|
nO=nO,
|
|
nI=nI,
|
|
init_W=init_W,
|
|
dropout=dropout,
|
|
normalize=normalize,
|
|
min_val=-1.0,
|
|
max_val=1.0,
|
|
)
|
|
|
|
|
|
@registry.layers("ReluK.v1")
|
|
def ReluK(
|
|
nO: Optional[int] = None,
|
|
nI: Optional[int] = None,
|
|
*,
|
|
init_W: Optional[Callable] = None,
|
|
init_b: Optional[Callable] = None,
|
|
dropout: Optional[float] = None,
|
|
normalize: bool = False,
|
|
k: float = 6.0,
|
|
) -> Model[Floats2d, Floats2d]:
|
|
if init_W is None:
|
|
init_W = glorot_uniform_init
|
|
if init_b is None:
|
|
init_b = zero_init
|
|
return ClippedLinear(
|
|
nO=nO,
|
|
nI=nI,
|
|
init_W=init_W,
|
|
dropout=dropout,
|
|
normalize=normalize,
|
|
min_val=0.0,
|
|
max_val=k,
|
|
)
|