File size: 11,077 Bytes
7f43c1b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright 2023 Katherine Crowson and The HuggingFace Team. All rights reserved.
#
# 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 dataclasses import dataclass
from typing import Optional, Tuple, Union

import flax
import jax.numpy as jnp
from scipy import integrate

from ..configuration_utils import ConfigMixin, register_to_config
from .scheduling_utils_flax import (
    CommonSchedulerState,
    FlaxKarrasDiffusionSchedulers,
    FlaxSchedulerMixin,
    FlaxSchedulerOutput,
    broadcast_to_shape_from_left,
)


@flax.struct.dataclass
class LMSDiscreteSchedulerState:
    common: CommonSchedulerState

    # setable values
    init_noise_sigma: jnp.ndarray
    timesteps: jnp.ndarray
    sigmas: jnp.ndarray
    num_inference_steps: Optional[int] = None

    # running values
    derivatives: Optional[jnp.ndarray] = None

    @classmethod
    def create(
        cls, common: CommonSchedulerState, init_noise_sigma: jnp.ndarray, timesteps: jnp.ndarray, sigmas: jnp.ndarray
    ):
        return cls(common=common, init_noise_sigma=init_noise_sigma, timesteps=timesteps, sigmas=sigmas)


@dataclass
class FlaxLMSSchedulerOutput(FlaxSchedulerOutput):
    state: LMSDiscreteSchedulerState


class FlaxLMSDiscreteScheduler(FlaxSchedulerMixin, ConfigMixin):
    """
    Linear Multistep Scheduler for discrete beta schedules. Based on the original k-diffusion implementation by
    Katherine Crowson:
    https://github.com/crowsonkb/k-diffusion/blob/481677d114f6ea445aa009cf5bd7a9cdee909e47/k_diffusion/sampling.py#L181

    [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__`
    function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`.
    [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and
    [`~SchedulerMixin.from_pretrained`] functions.

    Args:
        num_train_timesteps (`int`): number of diffusion steps used to train the model.
        beta_start (`float`): the starting `beta` value of inference.
        beta_end (`float`): the final `beta` value.
        beta_schedule (`str`):
            the beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from
            `linear` or `scaled_linear`.
        trained_betas (`jnp.ndarray`, optional):
            option to pass an array of betas directly to the constructor to bypass `beta_start`, `beta_end` etc.
        prediction_type (`str`, default `epsilon`, optional):
            prediction type of the scheduler function, one of `epsilon` (predicting the noise of the diffusion
            process), `sample` (directly predicting the noisy sample`) or `v_prediction` (see section 2.4
            https://imagen.research.google/video/paper.pdf)
        dtype (`jnp.dtype`, *optional*, defaults to `jnp.float32`):
            the `dtype` used for params and computation.
    """

    _compatibles = [e.name for e in FlaxKarrasDiffusionSchedulers]

    dtype: jnp.dtype

    @property
    def has_state(self):
        return True

    @register_to_config
    def __init__(
        self,
        num_train_timesteps: int = 1000,
        beta_start: float = 0.0001,
        beta_end: float = 0.02,
        beta_schedule: str = "linear",
        trained_betas: Optional[jnp.ndarray] = None,
        prediction_type: str = "epsilon",
        dtype: jnp.dtype = jnp.float32,
    ):
        self.dtype = dtype

    def create_state(self, common: Optional[CommonSchedulerState] = None) -> LMSDiscreteSchedulerState:
        if common is None:
            common = CommonSchedulerState.create(self)

        timesteps = jnp.arange(0, self.config.num_train_timesteps).round()[::-1]
        sigmas = ((1 - common.alphas_cumprod) / common.alphas_cumprod) ** 0.5

        # standard deviation of the initial noise distribution
        init_noise_sigma = sigmas.max()

        return LMSDiscreteSchedulerState.create(
            common=common,
            init_noise_sigma=init_noise_sigma,
            timesteps=timesteps,
            sigmas=sigmas,
        )

    def scale_model_input(self, state: LMSDiscreteSchedulerState, sample: jnp.ndarray, timestep: int) -> jnp.ndarray:
        """
        Scales the denoising model input by `(sigma**2 + 1) ** 0.5` to match the K-LMS algorithm.

        Args:
            state (`LMSDiscreteSchedulerState`):
                the `FlaxLMSDiscreteScheduler` state data class instance.
            sample (`jnp.ndarray`):
                current instance of sample being created by diffusion process.
            timestep (`int`):
                current discrete timestep in the diffusion chain.

        Returns:
            `jnp.ndarray`: scaled input sample
        """
        (step_index,) = jnp.where(state.timesteps == timestep, size=1)
        step_index = step_index[0]

        sigma = state.sigmas[step_index]
        sample = sample / ((sigma**2 + 1) ** 0.5)
        return sample

    def get_lms_coefficient(self, state: LMSDiscreteSchedulerState, order, t, current_order):
        """
        Compute a linear multistep coefficient.

        Args:
            order (TODO):
            t (TODO):
            current_order (TODO):
        """

        def lms_derivative(tau):
            prod = 1.0
            for k in range(order):
                if current_order == k:
                    continue
                prod *= (tau - state.sigmas[t - k]) / (state.sigmas[t - current_order] - state.sigmas[t - k])
            return prod

        integrated_coeff = integrate.quad(lms_derivative, state.sigmas[t], state.sigmas[t + 1], epsrel=1e-4)[0]

        return integrated_coeff

    def set_timesteps(
        self, state: LMSDiscreteSchedulerState, num_inference_steps: int, shape: Tuple = ()
    ) -> LMSDiscreteSchedulerState:
        """
        Sets the timesteps used for the diffusion chain. Supporting function to be run before inference.

        Args:
            state (`LMSDiscreteSchedulerState`):
                the `FlaxLMSDiscreteScheduler` state data class instance.
            num_inference_steps (`int`):
                the number of diffusion steps used when generating samples with a pre-trained model.
        """

        timesteps = jnp.linspace(self.config.num_train_timesteps - 1, 0, num_inference_steps, dtype=self.dtype)

        low_idx = jnp.floor(timesteps).astype(jnp.int32)
        high_idx = jnp.ceil(timesteps).astype(jnp.int32)

        frac = jnp.mod(timesteps, 1.0)

        sigmas = ((1 - state.common.alphas_cumprod) / state.common.alphas_cumprod) ** 0.5
        sigmas = (1 - frac) * sigmas[low_idx] + frac * sigmas[high_idx]
        sigmas = jnp.concatenate([sigmas, jnp.array([0.0], dtype=self.dtype)])

        timesteps = timesteps.astype(jnp.int32)

        # initial running values
        derivatives = jnp.zeros((0,) + shape, dtype=self.dtype)

        return state.replace(
            timesteps=timesteps,
            sigmas=sigmas,
            num_inference_steps=num_inference_steps,
            derivatives=derivatives,
        )

    def step(
        self,
        state: LMSDiscreteSchedulerState,
        model_output: jnp.ndarray,
        timestep: int,
        sample: jnp.ndarray,
        order: int = 4,
        return_dict: bool = True,
    ) -> Union[FlaxLMSSchedulerOutput, Tuple]:
        """
        Predict the sample at the previous timestep by reversing the SDE. Core function to propagate the diffusion
        process from the learned model outputs (most often the predicted noise).

        Args:
            state (`LMSDiscreteSchedulerState`): the `FlaxLMSDiscreteScheduler` state data class instance.
            model_output (`jnp.ndarray`): direct output from learned diffusion model.
            timestep (`int`): current discrete timestep in the diffusion chain.
            sample (`jnp.ndarray`):
                current instance of sample being created by diffusion process.
            order: coefficient for multi-step inference.
            return_dict (`bool`): option for returning tuple rather than FlaxLMSSchedulerOutput class

        Returns:
            [`FlaxLMSSchedulerOutput`] or `tuple`: [`FlaxLMSSchedulerOutput`] if `return_dict` is True, otherwise a
            `tuple`. When returning a tuple, the first element is the sample tensor.

        """
        if state.num_inference_steps is None:
            raise ValueError(
                "Number of inference steps is 'None', you need to run 'set_timesteps' after creating the scheduler"
            )

        sigma = state.sigmas[timestep]

        # 1. compute predicted original sample (x_0) from sigma-scaled predicted noise
        if self.config.prediction_type == "epsilon":
            pred_original_sample = sample - sigma * model_output
        elif self.config.prediction_type == "v_prediction":
            # * c_out + input * c_skip
            pred_original_sample = model_output * (-sigma / (sigma**2 + 1) ** 0.5) + (sample / (sigma**2 + 1))
        else:
            raise ValueError(
                f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, or `v_prediction`"
            )

        # 2. Convert to an ODE derivative
        derivative = (sample - pred_original_sample) / sigma
        state = state.replace(derivatives=jnp.append(state.derivatives, derivative))
        if len(state.derivatives) > order:
            state = state.replace(derivatives=jnp.delete(state.derivatives, 0))

        # 3. Compute linear multistep coefficients
        order = min(timestep + 1, order)
        lms_coeffs = [self.get_lms_coefficient(state, order, timestep, curr_order) for curr_order in range(order)]

        # 4. Compute previous sample based on the derivatives path
        prev_sample = sample + sum(
            coeff * derivative for coeff, derivative in zip(lms_coeffs, reversed(state.derivatives))
        )

        if not return_dict:
            return (prev_sample, state)

        return FlaxLMSSchedulerOutput(prev_sample=prev_sample, state=state)

    def add_noise(
        self,
        state: LMSDiscreteSchedulerState,
        original_samples: jnp.ndarray,
        noise: jnp.ndarray,
        timesteps: jnp.ndarray,
    ) -> jnp.ndarray:
        sigma = state.sigmas[timesteps].flatten()
        sigma = broadcast_to_shape_from_left(sigma, noise.shape)

        noisy_samples = original_samples + noise * sigma

        return noisy_samples

    def __len__(self):
        return self.config.num_train_timesteps