recompute.py 22.2 KB
Newer Older
J
JZ-LIANG 已提交
1
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
2
#
J
JZ-LIANG 已提交
3 4 5
# 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
6
#
J
JZ-LIANG 已提交
7
#     http://www.apache.org/licenses/LICENSE-2.0
8
#
J
JZ-LIANG 已提交
9 10 11 12 13 14 15 16
# 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 paddle
from paddle.fluid import core
17 18
from paddle.autograd import PyLayer
from paddle.autograd.py_layer import LegacyPyLayer
S
ShenLiang 已提交
19

J
JZ-LIANG 已提交
20 21
from paddle.fluid import framework
import contextlib
S
ShenLiang 已提交
22
from paddle.fluid.framework import in_dygraph_mode
J
JZ-LIANG 已提交
23 24

import logging
25

26
logger = logging.getLogger(__name__)
27 28
formatter = logging.Formatter(fmt='%(asctime)s %(levelname)-8s %(message)s',
                              datefmt='%Y-%m-%d %H:%M:%S')
29 30 31
ch = logging.StreamHandler()
ch.setFormatter(formatter)
logger.addHandler(ch)
J
JZ-LIANG 已提交
32

33 34
__all__ = []

J
JZ-LIANG 已提交
35 36 37 38

def detach_variable(inputs):
    out = []
    for inp in inputs:
S
ShenLiang 已提交
39
        if not isinstance(inp, (core.eager.Tensor, core.VarBase)):
J
JZ-LIANG 已提交
40 41 42 43 44 45 46 47 48 49 50
            out.append(inp)
            continue

        x = inp.detach()
        x.stop_gradient = inp.stop_gradient
        out.append(x)
    return tuple(out)


def check_recompute_necessary(inputs):
    if not any(input_.stop_gradient == False for input_ in inputs
S
ShenLiang 已提交
51
               if isinstance(input_, (core.eager.Tensor, paddle.Tensor))):
52
        logger.warn(
J
JZ-LIANG 已提交
53 54 55 56 57
            "[Recompute]: None of the inputs to current recompute block need grad, "
            "therefore there is NO need to recompute this block in backward !")


@contextlib.contextmanager
58 59
def swith_rng_state_tracker(rng_state, tracker):
    from paddle.distributed.fleet.meta_parallel.parallel_layers.random import get_rng_state_tracker
J
JZ-LIANG 已提交
60
    orig_cuda_rng_state = paddle.get_cuda_rng_state()
61 62
    orig_cuda_rng_tracker = get_rng_state_tracker().get_states_tracker()

J
JZ-LIANG 已提交
63
    paddle.set_cuda_rng_state(rng_state)
64
    get_rng_state_tracker().set_states_tracker(tracker)
J
JZ-LIANG 已提交
65 66 67 68
    try:
        yield
    finally:
        paddle.set_cuda_rng_state(orig_cuda_rng_state)
69
        get_rng_state_tracker().set_states_tracker(orig_cuda_rng_tracker)
J
JZ-LIANG 已提交
70 71


72
class LegacyRecomputeFunction(LegacyPyLayer):
73

S
ShenLiang 已提交
74 75
    @staticmethod
    def forward(ctx, run_function, preserve_rng_state, *args):
76
        from paddle.distributed.fleet.meta_parallel.parallel_layers.random import get_rng_state_tracker
S
ShenLiang 已提交
77

78
        # store for recomputing
S
ShenLiang 已提交
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
        ctx.run_function = run_function
        ctx.preserve_rng_state = preserve_rng_state

        # NOTE the number of outputs of backward() should be equal to the number of tensors in forward()'s input
        # the order of tensors in backward()'s output should be the same as tensors in forward()'s input
        # None tensor inputs will be filtered in backward inputs.

        # save input for backward
        ctx.inputs = []
        ctx.tensor_indices = []
        tensor_inputs = []
        for i, arg in enumerate(args):
            if paddle.is_tensor(arg):
                tensor_inputs.append(arg)
                ctx.tensor_indices.append(i)
                ctx.inputs.append(None)
            else:
                ctx.inputs.append(arg)
        ctx.save_for_backward(*tensor_inputs)

        # NOTE recompute with restore RNG only support one senario where one process for one cuda gpu.
        # one process with multiple gpu and mix-gpu-cpu senarios are not support
        if ctx.preserve_rng_state:
            cur_device = paddle.get_device()
            if 'gpu:' not in cur_device:
                raise RuntimeError(
105 106
                    "Recompute with RNG perserve is not support current device: {}."
                    .format(cur_device))
S
ShenLiang 已提交
107
            ctx.fw_cuda_rng_state = paddle.get_cuda_rng_state()
108 109
            ctx.fwd_cuda_rng_state_tracker = get_rng_state_tracker(
            ).get_states_tracker()
S
ShenLiang 已提交
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

        # TODO support AMP
        tracer = framework._dygraph_tracer()
        ctx.is_fw_autocast = False if tracer._amp_level == core.AmpLevel.O0 else True
        if tracer._amp_level == core.AmpLevel.O2:
            ctx.amp_level = 'O2'
        elif tracer._amp_level in (core.AmpLevel.O1, core.AmpLevel.O0):
            ctx.amp_level = 'O1'
        else:
            raise ValueError("unsupported amp level: {}".format(
                tracer._amp_level))

        if tracer._amp_dtype == 'float16':
            ctx.amp_dtype = 'float16'
        elif tracer._amp_dtype in ('bfloat16', 'float32'):
            ctx.amp_dtype = 'bfloat16'
        else:
            raise ValueError("unsupported amp dtype: {}".format(
                tracer._amp_dtype))

        ctx.amp_white_list, ctx.amp_black_list = tracer._get_amp_op_list()

        with paddle.no_grad():
            outputs = run_function(*args)
        return outputs

    @staticmethod
    def backward(ctx, *args):
138
        from paddle.distributed.fleet.meta_parallel.parallel_layers.random import get_rng_state_tracker
S
ShenLiang 已提交
139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155
        with paddle.fluid.dygraph.guard():
            # TODO need to check the recompute calling is vaild or not

            # Restore inputs
            inputs = list(ctx.inputs)
            tensor_indices = ctx.tensor_indices
            tensors = ctx.saved_tensor()
            for i, idx in enumerate(tensor_indices):
                inputs[idx] = tensors[i]

            # paddle.enable_grad()
            tracer = framework._dygraph_tracer()
            tracer._has_grad = True

            # NOTE support AMP
            # need restore auto_cast state as well as w/b list
            if ctx.preserve_rng_state:
156 157
                with swith_rng_state_tracker(ctx.fw_cuda_rng_state,
                                             ctx.fwd_cuda_rng_state_tracker):
S
ShenLiang 已提交
158 159 160 161 162 163 164 165 166
                    with paddle.amp.auto_cast(
                            enable=ctx.is_fw_autocast,
                            custom_white_list=ctx.amp_white_list,
                            custom_black_list=ctx.amp_black_list,
                            level=ctx.amp_level,
                            dtype=ctx.amp_dtype):
                        detached_inputs = detach_variable(tuple(inputs))
                        outputs = ctx.run_function(*detached_inputs)
            else:
167 168 169 170 171
                with paddle.amp.auto_cast(enable=ctx.is_fw_autocast,
                                          custom_white_list=ctx.amp_white_list,
                                          custom_black_list=ctx.amp_black_list,
                                          level=ctx.amp_level,
                                          dtype=ctx.amp_dtype):
S
ShenLiang 已提交
172 173 174
                    detached_inputs = detach_variable(tuple(inputs))
                    outputs = ctx.run_function(*detached_inputs)

175
            if isinstance(outputs, core.VarBase):
S
ShenLiang 已提交
176 177 178 179 180 181
                outputs = (outputs, )
            assert len(outputs) == len(args)

            # run backward() with only tensor that requires grad
            forward_outputs_with_grad = []
            # NOTE In Transformer-like network, if user put the attention mask into the recompute segment output,
182
            # pylayer will force the stop_gradient of attention mask to be False, which will make the number of
S
ShenLiang 已提交
183 184 185 186
            # tensor that need grad does not match.
            # the following backward_inputs_with_grad is used to avoid this case.
            backward_inputs_with_grad = []
            for i in range(len(outputs)):
187 188
                if isinstance(outputs[i],
                              core.VarBase) and not outputs[i].stop_gradient:
S
ShenLiang 已提交
189 190 191 192 193 194 195 196 197 198 199 200 201
                    forward_outputs_with_grad.append(outputs[i])
                    backward_inputs_with_grad.append(args[i])

            if len(forward_outputs_with_grad) == 0:
                raise RuntimeError(
                    "none of output has requires_grad=True, this recompute() is not necessary"
                )

            # actually backward
            with paddle.amp.auto_cast(enable=False):
                paddle.autograd.backward(forward_outputs_with_grad,
                                         backward_inputs_with_grad)

202 203
            grads = list(inp._grad_ivar() for inp in detached_inputs
                         if isinstance(inp, core.VarBase))
S
ShenLiang 已提交
204 205 206
            return grads


J
JZ-LIANG 已提交
207
class RecomputeFunction(PyLayer):
208

J
JZ-LIANG 已提交
209
    @staticmethod
210
    def forward(ctx, run_function, preserve_rng_state, *args, **kwargs):
211
        from paddle.distributed.fleet.meta_parallel.parallel_layers.random import get_rng_state_tracker
J
JZ-LIANG 已提交
212

213
        # store for recomputing
J
JZ-LIANG 已提交
214 215
        ctx.run_function = run_function
        ctx.preserve_rng_state = preserve_rng_state
216
        ctx.kwargs = kwargs
J
JZ-LIANG 已提交
217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240

        # NOTE the number of outputs of backward() should be equal to the number of tensors in forward()'s input
        # the order of tensors in backward()'s output should be the same as tensors in forward()'s input
        # None tensor inputs will be filtered in backward inputs.

        # save input for backward
        ctx.inputs = []
        ctx.tensor_indices = []
        tensor_inputs = []
        for i, arg in enumerate(args):
            if paddle.is_tensor(arg):
                tensor_inputs.append(arg)
                ctx.tensor_indices.append(i)
                ctx.inputs.append(None)
            else:
                ctx.inputs.append(arg)
        ctx.save_for_backward(*tensor_inputs)

        # NOTE recompute with restore RNG only support one senario where one process for one cuda gpu.
        # one process with multiple gpu and mix-gpu-cpu senarios are not support
        if ctx.preserve_rng_state:
            cur_device = paddle.get_device()
            if 'gpu:' not in cur_device:
                raise RuntimeError(
241 242
                    "Recompute with RNG perserve is not support current device: {}."
                    .format(cur_device))
J
JZ-LIANG 已提交
243
            ctx.fw_cuda_rng_state = paddle.get_cuda_rng_state()
244 245
            ctx.fwd_cuda_rng_state_tracker = get_rng_state_tracker(
            ).get_states_tracker()
J
JZ-LIANG 已提交
246 247

        # TODO support AMP
248
        tracer = framework._dygraph_tracer()
249 250 251 252 253
        ctx.is_fw_autocast = False if tracer._amp_level == core.AmpLevel.O0 else True
        if tracer._amp_level == core.AmpLevel.O2:
            ctx.amp_level = 'O2'
        elif tracer._amp_level in (core.AmpLevel.O1, core.AmpLevel.O0):
            ctx.amp_level = 'O1'
254
        else:
255 256
            raise ValueError("unsupported amp level: {}".format(
                tracer._amp_level))
257 258 259 260 261 262 263 264 265

        if tracer._amp_dtype == 'float16':
            ctx.amp_dtype = 'float16'
        elif tracer._amp_dtype in ('bfloat16', 'float32'):
            ctx.amp_dtype = 'bfloat16'
        else:
            raise ValueError("unsupported amp dtype: {}".format(
                tracer._amp_dtype))

266
        ctx.amp_white_list, ctx.amp_black_list = tracer._get_amp_op_list()
J
JZ-LIANG 已提交
267 268

        with paddle.no_grad():
269
            outputs = run_function(*args, **kwargs)
J
JZ-LIANG 已提交
270 271 272 273
        return outputs

    @staticmethod
    def backward(ctx, *args):
274
        from paddle.distributed.fleet.meta_parallel.parallel_layers.random import get_rng_state_tracker
J
JZ-LIANG 已提交
275 276 277 278 279 280 281 282 283 284 285 286 287 288
        with paddle.fluid.dygraph.guard():
            # TODO need to check the recompute calling is vaild or not

            # Restore inputs
            inputs = list(ctx.inputs)
            tensor_indices = ctx.tensor_indices
            tensors = ctx.saved_tensor()
            for i, idx in enumerate(tensor_indices):
                inputs[idx] = tensors[i]

            # paddle.enable_grad()
            tracer = framework._dygraph_tracer()
            tracer._has_grad = True

289 290
            # NOTE support AMP
            # need restore auto_cast state as well as w/b list
J
JZ-LIANG 已提交
291
            if ctx.preserve_rng_state:
292 293
                with swith_rng_state_tracker(ctx.fw_cuda_rng_state,
                                             ctx.fwd_cuda_rng_state_tracker):
294 295 296
                    with paddle.amp.auto_cast(
                            enable=ctx.is_fw_autocast,
                            custom_white_list=ctx.amp_white_list,
297
                            custom_black_list=ctx.amp_black_list,
298 299
                            level=ctx.amp_level,
                            dtype=ctx.amp_dtype):
300
                        detached_inputs = detach_variable(tuple(inputs))
301 302
                        outputs = ctx.run_function(*detached_inputs,
                                                   **ctx.kwargs)
303
            else:
304 305 306 307 308
                with paddle.amp.auto_cast(enable=ctx.is_fw_autocast,
                                          custom_white_list=ctx.amp_white_list,
                                          custom_black_list=ctx.amp_black_list,
                                          level=ctx.amp_level,
                                          dtype=ctx.amp_dtype):
J
JZ-LIANG 已提交
309
                    detached_inputs = detach_variable(tuple(inputs))
310
                    outputs = ctx.run_function(*detached_inputs, **ctx.kwargs)
J
JZ-LIANG 已提交
311

312
            if isinstance(outputs, (core.VarBase, core.eager.Tensor)):
J
JZ-LIANG 已提交
313 314 315 316 317
                outputs = (outputs, )
            assert len(outputs) == len(args)

            # run backward() with only tensor that requires grad
            forward_outputs_with_grad = []
318
            # NOTE In Transformer-like network, if user put the attention mask into the recompute segment output,
319
            # pylayer will force the stop_gradient of attention mask to be False, which will make the number of
320 321 322
            # tensor that need grad does not match.
            # the following backward_inputs_with_grad is used to avoid this case.
            backward_inputs_with_grad = []
J
JZ-LIANG 已提交
323
            for i in range(len(outputs)):
324 325 326 327
                if isinstance(
                        outputs[i],
                    (core.VarBase,
                     core.eager.Tensor)) and not outputs[i].stop_gradient:
J
JZ-LIANG 已提交
328
                    forward_outputs_with_grad.append(outputs[i])
329 330
                    backward_inputs_with_grad.append(args[i])

J
JZ-LIANG 已提交
331 332 333 334 335
            if len(forward_outputs_with_grad) == 0:
                raise RuntimeError(
                    "none of output has requires_grad=True, this recompute() is not necessary"
                )

336 337 338 339
            # actually backward
            with paddle.amp.auto_cast(enable=False):
                paddle.autograd.backward(forward_outputs_with_grad,
                                         backward_inputs_with_grad)
J
JZ-LIANG 已提交
340

341 342 343 344 345 346 347 348
            if in_dygraph_mode():
                grads = tuple(
                    inp._grad_ivar() for inp in detached_inputs
                    if isinstance(inp, (core.VarBase, core.eager.Tensor)))
            else:
                grads = list(
                    inp._grad_ivar() for inp in detached_inputs
                    if isinstance(inp, (core.VarBase, core.eager.Tensor)))
J
JZ-LIANG 已提交
349 350 351 352 353 354 355
            return grads


def recompute(function, *args, **kwargs):
    """
    recompute intermediate activations to save then memory.

356
    Parameters:
357
        function(paddle.nn.Layer): layer of sequence of layers that describes part of forward pass of the model
358 359 360 361 362 363
              whose intermediate activations will be released to save memory in forward stage and will be recomputed
              in backward stage for gradient calculation.
        *args(Tensor): inputs to the function.
        **kwargs(Dict): Kwargs should only contain the key-value pair of preserve_rng_state, which is used to
              indicate whether to save the forward rng. If it is True, then the last forward rng value will be
              restored when the forward recalculation of backpropagation is performed. The default
364
              preserve_rng_state is True.
J
JZ-LIANG 已提交
365 366

    Returns:
367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
        Output of function on args.

    Examples:
        .. code-block:: python

            import numpy as np
            import paddle
            from paddle.distributed.fleet.utils import recompute
            import random

            # required: gpu

            def get_fc_block(block_idx, input_size, is_last=False):
                block_name = "block_" + str(block_idx)
                block = paddle.nn.Sequential(
                    (block_name + "_fc_0", paddle.nn.Linear(input_size, input_size, bias_attr=False)),
                    (block_name + "_dropout", paddle.nn.Dropout(p=0.5)),
                    (block_name + "_relu_1", paddle.nn.ReLU()),
                    (block_name + "_fc_1", paddle.nn.Linear(input_size, input_size, bias_attr=False)),
                    (block_name + "_relu_2", paddle.nn.ReLU()),
                )
                if is_last:
                    block.add_sublayer(
                        block_name + "_fc_2",
                        paddle.nn.Linear(
                            input_size, 1, bias_attr=False
                        )
                    )
                else:
                    block.add_sublayer(
                        block_name + "_fc_2",
                        paddle.nn.Linear(input_size, input_size, bias_attr=False)
                    )

                return block


            class Naive_fc_net(paddle.nn.Layer):
                def __init__(self, input_size=10,
                            recompute_blocks=[1, 3],
                            recompute_kwargs={}):
                    super(Naive_fc_net, self).__init__()
                    self.recompute_blocks = recompute_blocks
                    self.recompute_kwargs = recompute_kwargs
                    self.runfunc0 = get_fc_block(0, input_size, is_last=False)
                    self.runfunc1 = get_fc_block(1, input_size, is_last=False)
                    self.runfunc2 = get_fc_block(2, input_size, is_last=False)
                    self.runfunc3 = get_fc_block(3, input_size, is_last=False)
                    self.runfunc4 = get_fc_block(4, input_size, is_last=True)
                    self.total_func = [self.runfunc0, self.runfunc1, self.runfunc2, self.runfunc3, self.runfunc4]

                def forward(self, inputs):
                    nums = len(self.total_func)
                    for i in range(nums):
                        if i in self.recompute_blocks:
                            inputs = recompute(self.total_func[i], inputs, **{"preserve_rng_state": True})
                        else:
                            inputs = self.total_func[i](inputs)
                    return inputs

            def run_model(cuda_state, recompute_block=[], recompute_kwargs={}):
                gen = paddle.seed(10)
                gen.manual_seed(10)
                np.random.seed(10)
                random.seed(10)
                if cuda_state:
                    paddle.set_cuda_rng_state(cuda_state)

                batch_size, input_size = 1, 10
                model = Naive_fc_net(
                    input_size,
                    recompute_blocks=recompute_block,
                    recompute_kwargs=recompute_kwargs)
                optimizer = paddle.optimizer.SGD(learning_rate=0.01, parameters=model.parameters())
                loss_ = []
                param_ = []
                grad_ = []
                for _ in range(5):
                    x_data = np.random.randn(batch_size, input_size).astype(np.float32)
                    x = paddle.to_tensor(x_data)
                    y_pred = model(x)
                    loss = y_pred.mean()
                    loss_.append(np.asarray(loss).tolist())
                    loss.backward()
                    optimizer.step()
                    param_.append(np.asarray(model.parameters()[9]).tolist())
                    grad_.append(np.asarray(model.parameters()[3]._grad_ivar()).tolist())
                    optimizer.clear_grad()

                return loss_, param_, grad_

            cuda_state = paddle.get_cuda_rng_state()
            # without recompute
            loss_ref, param_ref, grad_ref = run_model(
                cuda_state, recompute_block=[]
            )

            loss, param, grad = run_model(cuda_state, recompute_block=[1, 2])
            print("normal_loss: {}, recompute_loss: {}".format(loss_ref, loss))
            # The result of the recompute_loss should be the same as the normal_loss.

J
JZ-LIANG 已提交
468 469 470 471
    """
    # Hack to mix *args with **kwargs in a python 2.7-compliant way
    preserve = kwargs.pop('preserve_rng_state', True)

472 473 474
    if framework._dygraph_tracer()._has_grad:
        check_recompute_necessary(args)

475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
    return RecomputeFunction.apply(function, preserve, *args, **kwargs)


def recompute_sequential(ctx, functions, *args, **kwargs):
    """
    recompute intermediate activations to save then memory for 'Sequential' models.

    Parameters:
        ctx(dict): include 'segments' and  'preserve_rng_state' keys, the key 'segments' (int, default 1), represents the number of chunks to create in the model,
                   the key 'preserve_rng_state' (bool, optional, default=True) indicate whether to save the forward rng. If it is True, then the last forward rng value will be
                   restored when the forward recalculation of backpropagation is performed. and some keys such as 'mp_group', 'offload' and 'partition' are invalid here,
                   they are useful in 'recompute_hybrid' API.
        functions(paddle.nn.Sequential): layer of sequence of layers that describes part of forward pass of the model
              whose intermediate activations will be released to save memory in forward stage and will be recomputed
              in backward stage for gradient calculation.
        *args(Tensor): inputs(tuple) to the function.
        **kwargs(Dict): inputs(dict) to the function.

    Returns:
        Output of function on args and kwargs.

    Examples:
        .. code-block:: python

            model = paddle.nn.Sequential(...)
            input = recompute_sequential({'segments' : 1}, model, input)
    """
    segments = ctx.get('segments', 1)
    preserve_rng_state = ctx.get('preserve_rng_state', True)

    def _run_func(begin, end, funcs):

        def do_run(input):
            for i in range(begin, end + 1):
                input = funcs[i](input)
            return input

        return do_run

    if isinstance(functions, paddle.nn.Sequential):
        functions = list(functions.children())

    segment_size = len(functions) // segments

    end = -1
    for begin in range(0, segment_size * (segments - 1), segment_size):
        end = begin + segment_size - 1
        args = recompute(_run_func(begin, end, functions),
                         *args,
                         preserve_rng_state=preserve_rng_state,
                         **kwargs)
    return _run_func(end + 1, len(functions) - 1, functions)(args)