File size: 15,606 Bytes
d711508
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
# Copyright 2023-present the HuggingFace Inc. team.
#
# 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.

import math
from typing import Any, Set, Tuple

import torch
import torch.nn as nn
import torch.nn.functional as F

from peft.tuners.lycoris_utils import LycorisLayer


class LoHaLayer(nn.Module, LycorisLayer):
    # All names of layers that may contain adapter weights
    adapter_layer_names = ("hada_w1_a", "hada_w1_b", "hada_w2_a", "hada_w2_b", "hada_t1", "hada_t2")
    # other_param_names is defined on parent class

    def __init__(self, base_layer: nn.Module):
        super().__init__()
        LycorisLayer.__init__(self, base_layer)

        # LoHa info
        self.hada_w1_a = nn.ParameterDict({})
        self.hada_w1_b = nn.ParameterDict({})
        self.hada_w2_a = nn.ParameterDict({})
        self.hada_w2_b = nn.ParameterDict({})
        self.hada_t1 = nn.ParameterDict({})
        self.hada_t2 = nn.ParameterDict({})

    @property
    def _available_adapters(self) -> Set[str]:
        return {*self.hada_w1_a, *self.hada_w1_b, *self.hada_w2_a, *self.hada_w2_b, *self.hada_t1, *self.hada_t2}

    def create_adapter_parameters(self, adapter_name: str, r: int, shape: Tuple[int, ...]):
        # https://github.com/KohakuBlueleaf/LyCORIS/blob/eb460098187f752a5d66406d3affade6f0a07ece/lycoris/modules/loha.py#L130C9-L143C75
        if len(shape) == 4:
            self.hada_t1[adapter_name] = nn.Parameter(torch.empty(r, r, shape[2], shape[3]))
            self.hada_w1_a[adapter_name] = nn.Parameter(torch.empty(r, shape[0]))  # out_dim, 1-mode
            self.hada_w1_b[adapter_name] = nn.Parameter(torch.empty(r, shape[1]))  # in_dim , 2-mode

            self.hada_t2[adapter_name] = nn.Parameter(torch.empty(r, r, shape[2], shape[3]))
            self.hada_w2_a[adapter_name] = nn.Parameter(torch.empty(r, shape[0]))  # out_dim, 1-mode
            self.hada_w2_b[adapter_name] = nn.Parameter(torch.empty(r, shape[1]))  # in_dim , 2-mode
        else:
            self.hada_w1_a[adapter_name] = nn.Parameter(torch.empty(shape[0], r))
            self.hada_w1_b[adapter_name] = nn.Parameter(torch.empty(r, shape[1]))

            self.hada_w2_a[adapter_name] = nn.Parameter(torch.empty(shape[0], r))
            self.hada_w2_b[adapter_name] = nn.Parameter(torch.empty(r, shape[1]))

    def reset_adapter_parameters(self, adapter_name: str):
        # Original implementation performs initialization with normal distribution
        # https://github.com/KohakuBlueleaf/LyCORIS/blob/3549fdef8f564761d68b695a08ef88b1122fdedc/lycoris/modules/loha.py#L158

        # FedPara paper proposes to perform He initialization, let's stick with it
        # It is enough to initialize only single matrix with zeros to make adapter do nothing after initialization
        if adapter_name in self.hada_w1_a.keys():
            nn.init.kaiming_uniform_(self.hada_w1_a[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_w1_b[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_w2_a[adapter_name], a=math.sqrt(5))
            nn.init.zeros_(self.hada_w2_b[adapter_name])
        if adapter_name in self.hada_t1.keys():
            nn.init.kaiming_uniform_(self.hada_t1[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_t2[adapter_name], a=math.sqrt(5))

    def reset_adapter_parameters_random(self, adapter_name: str):
        # Original implementation performs initialization with normal distribution
        # https://github.com/KohakuBlueleaf/LyCORIS/blob/3549fdef8f564761d68b695a08ef88b1122fdedc/lycoris/modules/loha.py#L158

        # FedPara paper proposes to perform He initialization, let's stick with it
        # It is enough to initialize only single matrix with zeros to make adapter do nothing after initialization
        if adapter_name in self.hada_w1_a.keys():
            nn.init.kaiming_uniform_(self.hada_w1_a[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_w1_b[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_w2_a[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_w2_b[adapter_name], a=math.sqrt(5))
        if adapter_name in self.hada_t1.keys():
            nn.init.kaiming_uniform_(self.hada_t1[adapter_name], a=math.sqrt(5))
            nn.init.kaiming_uniform_(self.hada_t2[adapter_name], a=math.sqrt(5))

    def update_layer(
        self,
        adapter_name: str,
        r: int,
        alpha: float,
        rank_dropout: float,
        module_dropout: float,
        init_weights: bool,
        use_effective_conv2d: bool = False,
        **kwargs,
    ) -> None:
        """Internal function to create loha adapter

        Args:
            adapter_name (`str`): Name for the adapter to add.
            r (`int`): Rank for the added adapter.
            alpha (`float`): Alpha for the added adapter.
            rank_dropout (`float`): The dropout probability for rank dimension during training.
            module_dropout (`float`): The dropout probability for disabling adapter during training.
            init_weights (`bool`): Whether to initialize weights.
            use_effective_conv2d (`bool`, *optional*, defaults to `False`):
                Use parameter effective decomposition for Conv2d with ksize > 1.
        """
        if r <= 0:
            raise ValueError(f"`r` should be a positive integer value but the value passed is {r}")

        self.r[adapter_name] = r
        self.alpha[adapter_name] = alpha
        self.scaling[adapter_name] = alpha / r
        self.rank_dropout[adapter_name] = rank_dropout
        self.module_dropout[adapter_name] = module_dropout

        # Determine shape of LoHa weights
        base_layer = self.get_base_layer()
        if isinstance(base_layer, nn.Linear):
            shape = tuple(base_layer.weight.shape)
        elif isinstance(base_layer, nn.Conv2d):
            use_effective_conv2d = use_effective_conv2d and base_layer.kernel_size != (1, 1)
            if use_effective_conv2d:
                shape = (base_layer.out_channels, base_layer.in_channels, *base_layer.kernel_size)
            else:
                shape = (
                    base_layer.out_channels,
                    base_layer.in_channels * base_layer.kernel_size[0] * base_layer.kernel_size[1],
                )
        else:
            raise TypeError(f"LoHa is not implemented for base layers of type {type(base_layer).__name__}")

        # Create weights with provided shape
        self.create_adapter_parameters(adapter_name, r, shape)

        # Initialize weights
        if init_weights:
            self.reset_adapter_parameters(adapter_name)
        else:
            self.reset_adapter_parameters_random(adapter_name)

        # Move new weights to device
        weight = getattr(self.get_base_layer(), "weight", None)
        if weight is not None:
            # the layer is already completely initialized, this is an update
            if weight.dtype.is_floating_point or weight.dtype.is_complex:
                self.to(weight.device, dtype=weight.dtype)
            else:
                self.to(weight.device)
        self.set_adapter(self.active_adapters)

    def get_delta_weight(self, adapter_name: str) -> torch.Tensor:
        # https://github.com/KohakuBlueleaf/LyCORIS/blob/eb460098187f752a5d66406d3affade6f0a07ece/lycoris/modules/loha.py#L178
        if adapter_name in self.hada_t1.keys():
            weight = make_weight_cp(
                self.hada_t1[adapter_name],
                self.hada_w1_a[adapter_name],
                self.hada_w1_b[adapter_name],
                self.hada_t2[adapter_name],
                self.hada_w2_a[adapter_name],
                self.hada_w2_b[adapter_name],
                scale=torch.tensor(self.scaling[adapter_name]),
            )
        else:
            weight = make_weight(
                self.hada_w1_a[adapter_name],
                self.hada_w1_b[adapter_name],
                self.hada_w2_a[adapter_name],
                self.hada_w2_b[adapter_name],
                scale=torch.tensor(self.scaling[adapter_name]),
            )

        base_layer = self.get_base_layer()
        weight = weight.reshape(base_layer.weight.shape)

        # Perform rank dropout during training - drop rows of addition weights
        rank_dropout = self.rank_dropout[adapter_name]
        if self.training and rank_dropout:
            drop = (torch.rand(weight.size(0)) > rank_dropout).to(weight.dtype)
            drop = drop.view(-1, *[1] * len(weight.shape[1:])).to(weight.device)
            # TODO: Investigate if there should be a scaler like in normal dropout during training
            # Original implementation doesn't have it
            # https://github.com/KohakuBlueleaf/LyCORIS/blob/eb460098187f752a5d66406d3affade6f0a07ece/lycoris/modules/loha.py#L193
            drop /= drop.mean()
            weight *= drop

        return weight

    def forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
        previous_dtype = x.dtype

        if self.disable_adapters:
            if self.merged:
                self.unmerge()
            result = self.base_layer(x, *args, **kwargs)
        elif self.merged:
            result = self.base_layer(x, *args, **kwargs)
        else:
            result = self.base_layer(x, *args, **kwargs)

            # Execute all the adapters
            for active_adapter in self.active_adapters:
                if active_adapter not in self._available_adapters:
                    continue

                module_dropout = self.module_dropout[active_adapter]

                # Modify current execution weights
                if (not self.training) or (self.training and torch.rand(1) > module_dropout):
                    result = result + self._get_delta_activations(active_adapter, x, *args, **kwargs)

        result = result.to(previous_dtype)
        return result


class Linear(LoHaLayer):
    """LoHa implemented in Linear layer"""

    def __init__(
        self,
        base_layer: nn.Module,
        adapter_name: str = "default",
        r: int = 0,
        alpha: float = 0.0,
        rank_dropout: float = 0.0,
        module_dropout: float = 0.0,
        init_weights: bool = True,
        **kwargs,
    ):
        super().__init__(base_layer)

        # Create adapter and set it active
        self._active_adapter = adapter_name
        self.update_layer(adapter_name, r, alpha, rank_dropout, module_dropout, init_weights, **kwargs)

    def _get_delta_activations(
        self, adapter_name: str, input: torch.Tensor, *args: Any, **kwargs: Any
    ) -> torch.Tensor:
        delta_weight = self.get_delta_weight(adapter_name)
        # don't add bias here, because the bias is already included in the output of the base_layer
        return F.linear(input, delta_weight)

    def __repr__(self) -> str:
        rep = super().__repr__()
        return "loha." + rep


class Conv2d(LoHaLayer):
    """LoHa implemented in Conv2d layer"""

    def __init__(
        self,
        base_layer: nn.Module,
        adapter_name: str = "default",
        r: int = 0,
        alpha: float = 0.0,
        rank_dropout: float = 0.0,
        module_dropout: float = 0.0,
        use_effective_conv2d: bool = False,
        init_weights: bool = True,
        **kwargs,
    ):
        super().__init__(base_layer)

        # Create adapter and set it active
        self._active_adapter = adapter_name
        self.update_layer(
            adapter_name, r, alpha, rank_dropout, module_dropout, init_weights, use_effective_conv2d, **kwargs
        )

    def _get_delta_activations(
        self, adapter_name: str, input: torch.Tensor, *args: Any, **kwargs: Any
    ) -> torch.Tensor:
        delta_weight = self.get_delta_weight(adapter_name)
        # don't add bias here, because the bias is already included in the output of the base_layer
        base_layer = self.get_base_layer()
        return F.conv2d(
            input,
            delta_weight,
            stride=base_layer.stride,
            padding=base_layer.padding,
            dilation=base_layer.dilation,
            groups=base_layer.groups,
        )

    def __repr__(self) -> str:
        rep = super().__repr__()
        return "loha." + rep


# Below code is a direct copy from https://github.com/KohakuBlueleaf/LyCORIS/blob/eb460098187f752a5d66406d3affade6f0a07ece/lycoris/modules/loha.py#L9


class HadaWeight(torch.autograd.Function):
    @staticmethod
    def forward(ctx, w1a, w1b, w2a, w2b, scale=torch.tensor(1)):
        ctx.save_for_backward(w1a, w1b, w2a, w2b, scale)
        diff_weight = ((w1a @ w1b) * (w2a @ w2b)) * scale
        return diff_weight

    @staticmethod
    def backward(ctx, grad_out):
        (w1a, w1b, w2a, w2b, scale) = ctx.saved_tensors
        grad_out = grad_out * scale
        temp = grad_out * (w2a @ w2b)
        grad_w1a = temp @ w1b.T
        grad_w1b = w1a.T @ temp

        temp = grad_out * (w1a @ w1b)
        grad_w2a = temp @ w2b.T
        grad_w2b = w2a.T @ temp

        del temp
        return grad_w1a, grad_w1b, grad_w2a, grad_w2b, None


class HadaWeightCP(torch.autograd.Function):
    @staticmethod
    def forward(ctx, t1, w1a, w1b, t2, w2a, w2b, scale=torch.tensor(1)):
        ctx.save_for_backward(t1, w1a, w1b, t2, w2a, w2b, scale)

        rebuild1 = torch.einsum("i j k l, j r, i p -> p r k l", t1, w1b, w1a)
        rebuild2 = torch.einsum("i j k l, j r, i p -> p r k l", t2, w2b, w2a)

        return rebuild1 * rebuild2 * scale

    @staticmethod
    def backward(ctx, grad_out):
        (t1, w1a, w1b, t2, w2a, w2b, scale) = ctx.saved_tensors
        grad_out = grad_out * scale

        temp = torch.einsum("i j k l, j r -> i r k l", t2, w2b)
        rebuild = torch.einsum("i j k l, i r -> r j k l", temp, w2a)

        grad_w = rebuild * grad_out
        del rebuild

        grad_w1a = torch.einsum("r j k l, i j k l -> r i", temp, grad_w)
        grad_temp = torch.einsum("i j k l, i r -> r j k l", grad_w, w1a.T)
        del grad_w, temp

        grad_w1b = torch.einsum("i r k l, i j k l -> r j", t1, grad_temp)
        grad_t1 = torch.einsum("i j k l, j r -> i r k l", grad_temp, w1b.T)
        del grad_temp

        temp = torch.einsum("i j k l, j r -> i r k l", t1, w1b)
        rebuild = torch.einsum("i j k l, i r -> r j k l", temp, w1a)

        grad_w = rebuild * grad_out
        del rebuild

        grad_w2a = torch.einsum("r j k l, i j k l -> r i", temp, grad_w)
        grad_temp = torch.einsum("i j k l, i r -> r j k l", grad_w, w2a.T)
        del grad_w, temp

        grad_w2b = torch.einsum("i r k l, i j k l -> r j", t2, grad_temp)
        grad_t2 = torch.einsum("i j k l, j r -> i r k l", grad_temp, w2b.T)
        del grad_temp
        return grad_t1, grad_w1a, grad_w1b, grad_t2, grad_w2a, grad_w2b, None


def make_weight(w1a, w1b, w2a, w2b, scale):
    return HadaWeight.apply(w1a, w1b, w2a, w2b, scale)


def make_weight_cp(t1, w1a, w1b, t2, w2a, w2b, scale):
    return HadaWeightCP.apply(t1, w1a, w1b, t2, w2a, w2b, scale)