Spaces:
Sleeping
Sleeping
File size: 5,272 Bytes
0fdcb79 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 |
# Copyright 2021 AlQuraishi Laboratory
# Copyright 2021 DeepMind Technologies Limited
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from functools import partialmethod
from typing import Optional
from abc import ABC, abstractmethod
import torch
import torch.nn as nn
from dockformerpp.model.primitives import Linear, LayerNorm
from dockformerpp.utils.precision_utils import is_fp16_enabled
from dockformerpp.utils.tensor_utils import permute_final_dims
class BaseTriangleMultiplicativeUpdate(nn.Module, ABC):
"""
Implements Algorithms 11 and 12.
"""
@abstractmethod
def __init__(self, c_z, c_hidden, _outgoing):
"""
Args:
c_z:
Input channel dimension
c:
Hidden channel dimension
"""
super(BaseTriangleMultiplicativeUpdate, self).__init__()
self.c_z = c_z
self.c_hidden = c_hidden
self._outgoing = _outgoing
self.linear_g = Linear(self.c_z, self.c_z, init="gating")
self.linear_z = Linear(self.c_hidden, self.c_z, init="final")
self.layer_norm_in = LayerNorm(self.c_z)
self.layer_norm_out = LayerNorm(self.c_hidden)
self.sigmoid = nn.Sigmoid()
def _combine_projections(self,
a: torch.Tensor,
b: torch.Tensor,
) -> torch.Tensor:
if(self._outgoing):
a = permute_final_dims(a, (2, 0, 1))
b = permute_final_dims(b, (2, 1, 0))
else:
a = permute_final_dims(a, (2, 1, 0))
b = permute_final_dims(b, (2, 0, 1))
p = torch.matmul(a, b)
return permute_final_dims(p, (1, 2, 0))
@abstractmethod
def forward(self,
z: torch.Tensor,
mask: Optional[torch.Tensor] = None,
inplace_safe: bool = False,
_add_with_inplace: bool = False
) -> torch.Tensor:
"""
Args:
x:
[*, N_res, N_res, C_z] input tensor
mask:
[*, N_res, N_res] input mask
Returns:
[*, N_res, N_res, C_z] output tensor
"""
pass
class TriangleMultiplicativeUpdate(BaseTriangleMultiplicativeUpdate):
"""
Implements Algorithms 11 and 12.
"""
def __init__(self, c_z, c_hidden, _outgoing=True):
"""
Args:
c_z:
Input channel dimension
c:
Hidden channel dimension
"""
super(TriangleMultiplicativeUpdate, self).__init__(c_z=c_z,
c_hidden=c_hidden,
_outgoing=_outgoing)
self.linear_a_p = Linear(self.c_z, self.c_hidden)
self.linear_a_g = Linear(self.c_z, self.c_hidden, init="gating")
self.linear_b_p = Linear(self.c_z, self.c_hidden)
self.linear_b_g = Linear(self.c_z, self.c_hidden, init="gating")
def forward(self,
z: torch.Tensor,
mask: Optional[torch.Tensor] = None,
inplace_safe: bool = False,
_add_with_inplace: bool = False,
) -> torch.Tensor:
"""
Args:
x:
[*, N_res, N_res, C_z] input tensor
mask:
[*, N_res, N_res] input mask
Returns:
[*, N_res, N_res, C_z] output tensor
"""
if mask is None:
mask = z.new_ones(z.shape[:-1])
mask = mask.unsqueeze(-1)
z = self.layer_norm_in(z)
a = mask
a = a * self.sigmoid(self.linear_a_g(z))
a = a * self.linear_a_p(z)
b = mask
b = b * self.sigmoid(self.linear_b_g(z))
b = b * self.linear_b_p(z)
# Prevents overflow of torch.matmul in combine projections in
# reduced-precision modes
a_std = a.std()
b_std = b.std()
if(is_fp16_enabled() and a_std != 0. and b_std != 0.):
a = a / a.std()
b = b / b.std()
if(is_fp16_enabled()):
with torch.cuda.amp.autocast(enabled=False):
x = self._combine_projections(a.float(), b.float())
else:
x = self._combine_projections(a, b)
del a, b
x = self.layer_norm_out(x)
x = self.linear_z(x)
g = self.sigmoid(self.linear_g(z))
x = x * g
return x
class TriangleMultiplicationOutgoing(TriangleMultiplicativeUpdate):
"""
Implements Algorithm 11.
"""
__init__ = partialmethod(TriangleMultiplicativeUpdate.__init__, _outgoing=True)
class TriangleMultiplicationIncoming(TriangleMultiplicativeUpdate):
"""
Implements Algorithm 12.
"""
__init__ = partialmethod(TriangleMultiplicativeUpdate.__init__, _outgoing=False)
|