From 52fba48de91af34e3b756858b585ec89f5acee7d Mon Sep 17 00:00:00 2001 From: whs Date: Fri, 24 Jun 2022 15:13:17 +0800 Subject: [PATCH] Refine config API in ACT (#1178) 1. Make config of ACT support path, set, list and tuple 2. Refine the implementation of loading config 3. Add wrapper for dataloader to return data in dict format --- demo/auto_compression/detection/eval.py | 3 +- demo/auto_compression/detection/run.py | 9 +- .../image_classification/run.py | 4 +- demo/auto_compression/nlp/run.py | 4 +- .../semantic_segmentation/README.md | 18 ++- ...seg_dataset.yaml => humanseg_dataset.yaml} | 0 .../configs/pp_humanseg/pp_humanseg_auto.yaml | 3 - .../configs/pp_humanseg/pp_humanseg_qat.yaml | 3 - .../pp_humanseg/pp_humanseg_sparse.yaml | 3 - .../configs/pp_liteseg/pp_liteseg_auto.yaml | 3 - .../configs/pp_liteseg/pp_liteseg_qat.yaml | 3 - .../configs/pp_liteseg/pp_liteseg_sparse.yaml | 3 - .../semantic_segmentation/run.py | 31 +++-- paddleslim/auto_compression/auto_strategy.py | 1 + paddleslim/auto_compression/compressor.py | 32 +++-- paddleslim/auto_compression/config_helpers.py | 84 +++++++++--- .../create_compressed_program.py | 16 ++- .../auto_compression/strategy_config.py | 49 ++++++- paddleslim/auto_compression/utils/__init__.py | 3 + .../auto_compression/utils/dataloader.py | 52 ++++++++ tests/act/qat_dist_train.yaml | 19 +++ tests/act/test_act_api.py | 125 ++++++++++++++++++ 22 files changed, 382 insertions(+), 86 deletions(-) rename demo/auto_compression/semantic_segmentation/configs/dataset/{pp_humanseg_dataset.yaml => humanseg_dataset.yaml} (100%) create mode 100644 paddleslim/auto_compression/utils/dataloader.py create mode 100644 tests/act/qat_dist_train.yaml create mode 100644 tests/act/test_act_api.py diff --git a/demo/auto_compression/detection/eval.py b/demo/auto_compression/detection/eval.py index b91d4300..8cc252cd 100644 --- a/demo/auto_compression/detection/eval.py +++ b/demo/auto_compression/detection/eval.py @@ -129,7 +129,8 @@ def eval(): def main(): global global_config - _, _, global_config = load_slim_config(FLAGS.config_path) + all_config = load_slim_config(FLAGS.config_path) + global_config = all_config["Global"] reader_cfg = load_config(global_config['reader_config']) dataset = reader_cfg['EvalDataset'] diff --git a/demo/auto_compression/detection/run.py b/demo/auto_compression/detection/run.py index 2d6dcc9b..7c6c0fd8 100644 --- a/demo/auto_compression/detection/run.py +++ b/demo/auto_compression/detection/run.py @@ -128,8 +128,9 @@ def eval_function(exe, compiled_test_program, test_feed_names, test_fetch_list): def main(): global global_config - compress_config, train_config, global_config = load_slim_config( - FLAGS.config_path) + all_config = load_slim_config(FLAGS.config_path) + assert "Global" in all_config, f"Key 'Global' not found in config file. \n{all_config}" + global_config = all_config["Global"] reader_cfg = load_config(global_config['reader_config']) train_loader = create('EvalReader')(reader_cfg['TrainDataset'], @@ -167,11 +168,9 @@ def main(): model_filename=global_config["model_filename"], params_filename=global_config["params_filename"], save_dir=FLAGS.save_dir, - strategy_config=compress_config, - train_config=train_config, + config=all_config, train_dataloader=train_loader, eval_callback=eval_func) - ac.compress() diff --git a/demo/auto_compression/image_classification/run.py b/demo/auto_compression/image_classification/run.py index 086a5bfe..817afa1d 100644 --- a/demo/auto_compression/image_classification/run.py +++ b/demo/auto_compression/image_classification/run.py @@ -100,7 +100,6 @@ if __name__ == '__main__': args = parser.parse_args() print_arguments(args) paddle.enable_static() - compress_config, train_config, _ = load_config(args.config_path) data_dir = args.data_dir if args.image_reader_type == 'paddle': @@ -121,8 +120,7 @@ if __name__ == '__main__': model_filename=args.model_filename, params_filename=args.params_filename, save_dir=args.save_dir, - strategy_config=compress_config, - train_config=train_config, + config=args.config_path, train_dataloader=train_dataloader, eval_callback=eval_function, eval_dataloader=reader_wrapper( diff --git a/demo/auto_compression/nlp/run.py b/demo/auto_compression/nlp/run.py index b089d7aa..01b853be 100644 --- a/demo/auto_compression/nlp/run.py +++ b/demo/auto_compression/nlp/run.py @@ -242,7 +242,6 @@ if __name__ == '__main__': print_arguments(args) paddle.enable_static() - compress_config, train_config, _ = load_config(args.config_path) if train_config is not None: train_config.optimizer_builder[ 'apply_decay_param_fun'] = apply_decay_param_fun @@ -256,8 +255,7 @@ if __name__ == '__main__': model_filename=args.model_filename, params_filename=args.params_filename, save_dir=args.save_dir, - strategy_config=compress_config, - train_config=train_config, + config=args.config_path, train_dataloader=train_dataloader, eval_callback=eval_function if compress_config is None or 'HyperParameterOptimization' not in compress_config else diff --git a/demo/auto_compression/semantic_segmentation/README.md b/demo/auto_compression/semantic_segmentation/README.md index 00223417..092fe398 100644 --- a/demo/auto_compression/semantic_segmentation/README.md +++ b/demo/auto_compression/semantic_segmentation/README.md @@ -114,7 +114,8 @@ python run.py \ --model_filename='model.pdmodel' \ --params_filename='model.pdiparams' \ --save_dir='./save_model' \ - --config_path='configs/pp_humanseg/pp_humanseg_auto.yaml' \ + --strategy_config='configs/pp_humanseg/pp_humanseg_auto.yaml' \ + --dataset_config='configs/dataset/humanseg_dataset.yaml' \ --deploy_hardware='SD710' # 多卡启动 @@ -124,7 +125,8 @@ python -m paddle.distributed.launch run.py \ --model_filename='model.pdmodel' \ --params_filename='model.pdiparams' \ --save_dir='./save_model' \ - --config_path='configs/pp_humanseg/pp_humanseg_auto.yaml' \ + --strategy_config='configs/pp_humanseg/pp_humanseg_auto.yaml' \ + --dataset_config='configs/dataset/humanseg_dataset.yaml' \ --deploy_hardware='SD710' ``` - 自行配置稀疏参数进行非结构化稀疏和蒸馏训练，配置参数含义详见[自动压缩超参文档](https://github.com/PaddlePaddle/PaddleSlim/blob/27dafe1c722476f1b16879f7045e9215b6f37559/demo/auto_compression/hyperparameter_tutorial.md)。具体命令如下所示： @@ -136,7 +138,8 @@ python run.py \ --model_filename='model.pdmodel' \ --params_filename='model.pdiparams' \ --save_dir='./save_model' \ - --config_path='configs/pp_humanseg/pp_humanseg_sparse.yaml' + --strategy_config='configs/pp_humanseg/pp_humanseg_sparse.yaml' \ + --dataset_config='configs/dataset/humanseg_dataset.yaml' # 多卡启动 export CUDA_VISIBLE_DEVICES=0,1 @@ -145,7 +148,8 @@ python -m paddle.distributed.launch run.py \ --model_filename='model.pdmodel' \ --params_filename='model.pdiparams' \ --save_dir='./save_model' \ - --config_path='configs/pp_humanseg/pp_humanseg_sparse.yaml' + --strategy_config='configs/pp_humanseg/pp_humanseg_sparse.yaml' \ + --dataset_config='configs/dataset/humanseg_dataset.yaml' ``` - 自行配置量化参数进行量化和蒸馏训练，配置参数含义详见[自动压缩超参文档](https://github.com/PaddlePaddle/PaddleSlim/blob/27dafe1c722476f1b16879f7045e9215b6f37559/demo/auto_compression/hyperparameter_tutorial.md)。具体命令如下所示： @@ -157,7 +161,8 @@ python run.py \ --model_filename='model.pdmodel' \ --params_filename='model.pdiparams' \ --save_dir='./save_model' \ - --config_path='configs/pp_humanseg/pp_humanseg_qat.yaml' + --strategy_config='configs/pp_humanseg/pp_humanseg_qat.yaml' \ + --dataset_config='configs/dataset/humanseg_dataset.yaml' # 多卡启动 export CUDA_VISIBLE_DEVICES=0,1 @@ -166,7 +171,8 @@ python -m paddle.distributed.launch run.py \ --model_filename='model.pdmodel' \ --params_filename='model.pdiparams' \ --save_dir='./save_model' \ - --config_path='configs/pp_humanseg/pp_humanseg_qat.yaml' + --strategy_config='configs/pp_humanseg/pp_humanseg_qat.yaml' \ + --dataset_config='configs/dataset/humanseg_dataset.yaml' ``` 压缩完成后会在`save_dir`中产出压缩好的预测模型，可直接预测部署。 diff --git a/demo/auto_compression/semantic_segmentation/configs/dataset/pp_humanseg_dataset.yaml b/demo/auto_compression/semantic_segmentation/configs/dataset/humanseg_dataset.yaml similarity index 100% rename from demo/auto_compression/semantic_segmentation/configs/dataset/pp_humanseg_dataset.yaml rename to demo/auto_compression/semantic_segmentation/configs/dataset/humanseg_dataset.yaml diff --git a/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_auto.yaml b/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_auto.yaml index 01d8ebf9..8adde821 100644 --- a/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_auto.yaml +++ b/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_auto.yaml @@ -1,6 +1,3 @@ -Global: - reader_config: configs/dataset/pp_humanseg_dataset.yaml - TrainConfig: epochs: 14 eval_iter: 400 diff --git a/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_qat.yaml b/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_qat.yaml index 1414c9b5..d11af163 100644 --- a/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_qat.yaml +++ b/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_qat.yaml @@ -1,6 +1,3 @@ -Global: - reader_config: configs/dataset/pp_humanseg_lite.yaml - Distillation: alpha: 1.0 loss: l2 diff --git a/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_sparse.yaml b/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_sparse.yaml index bc8215dd..5e71f396 100644 --- a/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_sparse.yaml +++ b/demo/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_sparse.yaml @@ -1,6 +1,3 @@ -Global: - reader_config: configs/dataset/pp_humanseg_lite.yaml - Distillation: alpha: 1.0 loss: l2 diff --git a/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_auto.yaml b/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_auto.yaml index 41d8f71a..42ff6c5d 100644 --- a/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_auto.yaml +++ b/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_auto.yaml @@ -1,6 +1,3 @@ -Global: - reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml - TrainConfig: epochs: 14 eval_iter: 90 diff --git a/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_qat.yaml b/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_qat.yaml index 663a954a..6bb5b42d 100644 --- a/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_qat.yaml +++ b/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_qat.yaml @@ -1,6 +1,3 @@ -Global: - reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml - Distillation: alpha: 1.0 loss: l2 diff --git a/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_sparse.yaml b/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_sparse.yaml index b755ffb6..f3a6d958 100644 --- a/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_sparse.yaml +++ b/demo/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_sparse.yaml @@ -1,6 +1,3 @@ -Global: - reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml - Distillation: alpha: 1.0 loss: l2 diff --git a/demo/auto_compression/semantic_segmentation/run.py b/demo/auto_compression/semantic_segmentation/run.py index ba46883b..4e4706d2 100644 --- a/demo/auto_compression/semantic_segmentation/run.py +++ b/demo/auto_compression/semantic_segmentation/run.py @@ -3,9 +3,9 @@ import argparse import random import paddle import numpy as np -from paddleseg.cvlibs import Config +from paddleseg.cvlibs import Config as PaddleSegDataConfig from paddleseg.utils import worker_init_fn -from paddleslim.auto_compression.config_helpers import load_config + from paddleslim.auto_compression import AutoCompression from paddleseg.core.infer import reverse_transform from paddleseg.utils import metrics @@ -34,10 +34,15 @@ def parse_args(): default=None, help="directory to save compressed model.") parser.add_argument( - '--config_path', + '--strategy_config', type=str, default=None, help="path of compression strategy config.") + parser.add_argument( + '--dataset_config', + type=str, + default=None, + help="path of dataset config.") parser.add_argument( '--deploy_hardware', type=str, @@ -148,13 +153,15 @@ if __name__ == '__main__': args = parse_args() - compress_config, train_config, global_config = load_config(args.config_path) - cfg = Config(global_config['reader_config']) - - train_dataset = cfg.train_dataset - eval_dataset = cfg.val_dataset + # step1: load dataset config and create dataloader + data_cfg = PaddleSegDataConfig(args.dataset_config) + train_dataset = data_cfg.train_dataset + eval_dataset = data_cfg.val_dataset batch_sampler = paddle.io.DistributedBatchSampler( - train_dataset, batch_size=cfg.batch_size, shuffle=True, drop_last=True) + train_dataset, + batch_size=data_cfg.batch_size, + shuffle=True, + drop_last=True) train_loader = paddle.io.DataLoader( train_dataset, batch_sampler=batch_sampler, @@ -163,16 +170,16 @@ if __name__ == '__main__': worker_init_fn=worker_init_fn) train_dataloader = reader_wrapper(train_loader) - # set auto_compression + # step2: create and instance of AutoCompression ac = AutoCompression( model_dir=args.model_dir, model_filename=args.model_filename, params_filename=args.params_filename, save_dir=args.save_dir, - strategy_config=compress_config, - train_config=train_config, + config=args.strategy_config, train_dataloader=train_dataloader, eval_callback=eval_function, deploy_hardware=args.deploy_hardware) + # step3: start the compression job ac.compress() diff --git a/paddleslim/auto_compression/auto_strategy.py b/paddleslim/auto_compression/auto_strategy.py index a1bb5ad7..a6311a3d 100644 --- a/paddleslim/auto_compression/auto_strategy.py +++ b/paddleslim/auto_compression/auto_strategy.py @@ -18,6 +18,7 @@ import platform from ..common import get_logger from .utils.predict import predict_compressed_model, with_variable_shape from .strategy_config import * +from paddleslim.analysis import TableLatencyPredictor _logger = get_logger(__name__, level=logging.INFO) diff --git a/paddleslim/auto_compression/compressor.py b/paddleslim/auto_compression/compressor.py index 1909482e..7b57e6d9 100644 --- a/paddleslim/auto_compression/compressor.py +++ b/paddleslim/auto_compression/compressor.py @@ -22,6 +22,7 @@ import shutil from time import gmtime, strftime import platform import paddle +import itertools import paddle.distributed.fleet as fleet from ..quant.quanter import convert, quant_post from ..common.recover_program import recover_inference_program @@ -31,7 +32,9 @@ from ..analysis import TableLatencyPredictor from .create_compressed_program import build_distill_program, build_quant_program, build_prune_program, remove_unused_var_nodes from .strategy_config import TrainConfig, ProgramInfo, merge_config from .auto_strategy import prepare_strategy, get_final_quant_config, create_strategy_config, create_train_config +from .config_helpers import load_config, extract_strategy_config, extract_train_config from .utils.predict import with_variable_shape +from .utils import get_feed_vars, wrap_dataloader _logger = get_logger(__name__, level=logging.INFO) @@ -49,9 +52,8 @@ class AutoCompression: params_filename, save_dir, train_dataloader, + config=None, input_shapes=None, - train_config=None, - strategy_config=None, target_speedup=None, eval_callback=None, eval_dataloader=None, @@ -127,16 +129,29 @@ class AutoCompression: self.final_dir = save_dir if not os.path.exists(self.final_dir): os.makedirs(self.final_dir) - self.strategy_config = strategy_config - self.train_dataloader = train_dataloader + + # load config + assert type(config) in [ + dict, str, set, list, tuple + ], f"The type of config should be in [dict, str, set, list, tuple] but got {type(config)}" + if isinstance(config, str): + config = load_config(config) + self.strategy_config = extract_strategy_config(config) + self.train_config = extract_train_config(config) + + # prepare dataloader + self.feed_vars = get_feed_vars(model_dir, model_filename, + params_filename) + self.train_dataloader = wrap_dataloader(train_dataloader, + self.feed_vars) + self.eval_dataloader = wrap_dataloader(eval_dataloader, self.feed_vars) + if eval_dataloader is None: + eval_dataloader = self._get_eval_dataloader(self.train_dataloader) + self.target_speedup = target_speedup self.eval_function = eval_callback self.deploy_hardware = deploy_hardware - if eval_dataloader is None: - eval_dataloader = self._get_eval_dataloader(train_dataloader) - self.eval_dataloader = eval_dataloader - paddle.enable_static() self._exe, self._places = self._prepare_envs() self.model_type = self._get_model_type(self._exe, model_dir, @@ -158,6 +173,7 @@ class AutoCompression: self.model_dir = infer_shape_model self.model_filename = "infered_shape.pdmodel" self.params_filename = "infered_shape.pdiparams" + if self.strategy_config is None: strategy_config = prepare_strategy( self._exe, self._places, self.model_dir, self.model_filename, diff --git a/paddleslim/auto_compression/config_helpers.py b/paddleslim/auto_compression/config_helpers.py index d7d9bcf7..24152277 100644 --- a/paddleslim/auto_compression/config_helpers.py +++ b/paddleslim/auto_compression/config_helpers.py @@ -12,39 +12,81 @@ # See the License for the specific language governing permissions and # limitations under the License. import yaml -from .strategy_config import * +import os +from paddleslim.auto_compression.strategy_config import * __all__ = ['save_config', 'load_config'] -def load_config(config_path): +def load_config(config): + """Load configurations from yaml file into dict. + Fields validation is skipped for loading some custom information. + Args: + config(str): The path of configuration file. + Returns: + dict: A dict storing configuration information. """ - convert yaml to dict config. - """ - f = open(config_path, 'r') - cfg = yaml.load(f, Loader=yaml.FullLoader) - f.close() + if config is None: + return None + assert isinstance( + config, + str), f"config should be str but got type(config)={type(config)}" + assert os.path.exists(config) and os.path.isfile( + config), f"{config} not found or it is not a file." + with open(config) as f: + cfg = yaml.load(f, Loader=yaml.FullLoader) + return cfg - global_config = {} - if 'Global' in cfg: - for g_key, g_value in cfg["Global"].items(): - global_config[g_key] = g_value - cfg.pop('Global') - compress_config = {} - for key, value in cfg.items(): - default_key = eval(key)(**value) if value is not None else eval(key)() - compress_config[key] = default_key +def extract_strategy_config(config): + """Extract configuration items of strategies from file or dict. + And fields validation is enable. + Args: + config(str, dict): The path of configuration file or a dict storing information about strategies. + Returns: + dict: The key is the name of strategy and the value is an instance of paddleslim.auto_compression.BaseStrategy. + """ + if config is None: + return None + if isinstance(config, str): + config = load_config(config) - if compress_config.get('TrainConfig') != None: - train_config = compress_config.pop('TrainConfig') - else: - train_config = None + compress_config = {} + if isinstance(config, dict): + for key, value in config.items(): + if key in SUPPORTED_CONFIG: + compress_config[key] = eval(key)(**value) if isinstance( + value, dict) else eval(key)() + elif type(config) in [set, list, tuple]: + for key in config: + assert isinstance(key, str) + if key in SUPPORTED_CONFIG: + compress_config[key] = eval(key)() if len(compress_config) == 0: compress_config = None + return compress_config + - return compress_config, train_config, global_config +def extract_train_config(config): + """Extract configuration items of training from file or dict. + And fields validation is enable. + Args: + config(str, dict): The path of configuration file or a dict storing information about training. + Returns: + An instance of paddleslim.auto_compression.TrainConfig + """ + if config is None: + return None + if isinstance(config, str): + config = load_config(config) + if isinstance(config, dict): + for key, value in config.items(): + if key == TRAIN_CONFIG_NAME: + return TrainConfig( + **value) if value is not None else TrainConfig() + # return default training config when it is not set + return TrainConfig() def save_config(config, config_path): diff --git a/paddleslim/auto_compression/create_compressed_program.py b/paddleslim/auto_compression/create_compressed_program.py index 99fce6fa..417eab25 100644 --- a/paddleslim/auto_compression/create_compressed_program.py +++ b/paddleslim/auto_compression/create_compressed_program.py @@ -45,12 +45,13 @@ def _create_lr_scheduler(train_config): def _create_optimizer(train_config): """create optimizer""" - if 'optimizer_builder' not in train_config: train_config['optimizer_builder'] = {'optimizer': {'type': 'SGD'}} optimizer_builder = train_config['optimizer_builder'] - + assert isinstance( + optimizer_builder, dict + ), f"Value of 'optimizer_builder' in train_config should be dict but got {type(optimizer_builder)}" if 'grad_clip' in optimizer_builder: g_clip_params = optimizer_builder['grad_clip'] g_clip_type = g_clip_params.pop('type') @@ -423,11 +424,13 @@ def build_prune_program(executor, from ..prune import Pruner pruner = Pruner(config["criterion"]) params = [] + original_shapes = {} ### TODO(ceci3): set default prune weight for param in train_program_info.program.global_block().all_parameters(): if config['prune_params_name'] is not None and param.name in config[ 'prune_params_name']: params.append(param.name) + original_shapes[param.name] = param.shape pruned_program, _, _ = pruner.prune( train_program_info.program, @@ -435,6 +438,15 @@ def build_prune_program(executor, params=params, ratios=[config['pruned_ratio']] * len(params), place=place) + _logger.info( + "####################channel pruning##########################") + for param in pruned_program.global_block().all_parameters(): + if param.name in original_shapes: + _logger.info( + f"{param.name}, from {original_shapes[param.name]} to {param.shape}" + ) + _logger.info( + "####################channel pruning end##########################") train_program_info.program = pruned_program elif strategy.startswith('asp'): diff --git a/paddleslim/auto_compression/strategy_config.py b/paddleslim/auto_compression/strategy_config.py index dc99a148..19e10a2d 100644 --- a/paddleslim/auto_compression/strategy_config.py +++ b/paddleslim/auto_compression/strategy_config.py @@ -15,13 +15,43 @@ from collections import namedtuple __all__ = [ - "Quantization", "Distillation", "MultiTeacherDistillation", \ - "HyperParameterOptimization", "ChannelPrune", "UnstructurePrune", \ - "TransformerPrune", "ASPPrune", "merge_config", "ProgramInfo", "TrainConfig", + "BaseStrategy", + "Quantization", + "Distillation", + "MultiTeacherDistillation", + "HyperParameterOptimization", + "ChannelPrune", + "UnstructurePrune", + "TransformerPrune", + "ASPPrune", + "merge_config", + "ProgramInfo", + "TrainConfig", + "SUPPORTED_CONFIG", + "TRAIN_CONFIG_NAME", ] +SUPPORTED_CONFIG = [ + "Quantization", + "Distillation", + "MultiTeacherDistillation", + "HyperParameterOptimization", + "ChannelPrune", + "UnstructurePrune", + "TransformerPrune", + "ASPPrune", + "TrainConfig", +] + +TRAIN_CONFIG_NAME = "TrainConfig" + + +class BaseStrategy: + def __init__(self, name): + self.name = name + -class Quantization: +class Quantization(BaseStrategy): def __init__(self, quantize_op_types=[ 'conv2d', 'depthwise_conv2d', 'mul', 'matmul', 'matmul_v2' @@ -53,6 +83,7 @@ class Quantization: for_tensorrt(bool): If True, 'quantize_op_types' will be TENSORRT_OP_TYPES. Default: False. is_full_quantize(bool): If True, 'quantoze_op_types' will be TRANSFORM_PASS_OP_TYPES + QUANT_DEQUANT_PASS_OP_TYPES. Default: False. """ + super(Quantization, self).__init__("Quantization") self.quantize_op_types = quantize_op_types self.weight_bits = weight_bits self.activation_bits = activation_bits @@ -68,7 +99,7 @@ class Quantization: self.is_full_quantize = is_full_quantize -class Distillation: +class Distillation(BaseStrategy): def __init__(self, loss='l2', node=[], @@ -86,6 +117,7 @@ class Distillation: teacher_model_filename(str, optional): The name of teacher model file. If parameters are saved in separate files, set it as 'None'. Default: 'None'. teacher_params_filename(str, optional): The name of teacher params file. When all parameters are saved in a single file, set it as filename. If parameters are saved in separate files, set it as 'None'. Default : 'None'. """ + super(Distillation, self).__init__("Distillation") self.loss = loss self.node = node self.alpha = alpha @@ -120,7 +152,7 @@ class MultiTeacherDistillation: self.teacher_params_filename = teacher_params_filename -class HyperParameterOptimization: +class HyperParameterOptimization(BaseStrategy): def __init__(self, ptq_algo=["KL", "hist", "avg", "mse"], bias_correct=[True, False], @@ -138,6 +170,7 @@ class HyperParameterOptimization: batch_num(list(int)): The upper and lower bounds of batch number, the real batch number is uniform sampling in this bounds. max_quant_count(int): Max number of model quantization. Default: 20. """ + super(HyperParameterOptimization, self).__init__("HPO_PTQ") self.ptq_algo = ptq_algo self.bias_correct = bias_correct self.weight_quantize_type = weight_quantize_type @@ -224,7 +257,9 @@ class TrainConfig: epochs=None, train_iter=None, learning_rate=0.02, - optimizer_builder={'optimizer': 'SGD'}, + optimizer_builder={'optimizer': { + 'type': 'SGD' + }}, eval_iter=1000, logging_iter=10, origin_metric=None, diff --git a/paddleslim/auto_compression/utils/__init__.py b/paddleslim/auto_compression/utils/__init__.py index e3c3a49d..7c7236f9 100644 --- a/paddleslim/auto_compression/utils/__init__.py +++ b/paddleslim/auto_compression/utils/__init__.py @@ -14,5 +14,8 @@ from __future__ import absolute_import from .predict import predict_compressed_model +from .dataloader import * +from . import dataloader __all__ = ["predict_compressed_model"] +__all__ += dataloader.__all__ diff --git a/paddleslim/auto_compression/utils/dataloader.py b/paddleslim/auto_compression/utils/dataloader.py new file mode 100644 index 00000000..184f2602 --- /dev/null +++ b/paddleslim/auto_compression/utils/dataloader.py @@ -0,0 +1,52 @@ +import os +import time +import numpy as np +import paddle +from collections.abc import Iterable + +__all__ = ["wrap_dataloader", "get_feed_vars"] + + +def get_feed_vars(model_dir, model_filename, params_filename): + """Get feed vars of model. + """ + paddle.enable_static() + exe = paddle.static.Executor(paddle.CPUPlace()) + [inference_program, feed_target_names, fetch_targets] = ( + paddle.static.load_inference_model( + model_dir, + exe, + model_filename=model_filename, + params_filename=params_filename)) + return feed_target_names + + +def wrap_dataloader(dataloader, names): + """Create a wrapper of dataloader if the data returned by the dataloader is not a dict. + And the names will be the keys of dict returned by the wrapper. + """ + if dataloader is None: + return dataloader + assert isinstance(dataloader, paddle.io.DataLoader) + assert len(dataloader) > 0 + data = next(dataloader()) + if isinstance(data, dict): + return dataloader + + if isinstance(data, Iterable): + assert len(data) == len( + names + ), f"len(data) == len(names), but got len(data): {len(data)} and len(names): {len(names)}" + else: + assert len( + names + ) == 1, f"The length of name should 1 when data is not Iterable but got {len(names)}" + + def gen(): + for i, data in enumerate(dataloader()): + if not isinstance(data, Iterable): + data = [data] + yield dict((name_, np.array(data_)) + for name_, data_ in zip(names, data)) + + return gen diff --git a/tests/act/qat_dist_train.yaml b/tests/act/qat_dist_train.yaml new file mode 100644 index 00000000..d96344ed --- /dev/null +++ b/tests/act/qat_dist_train.yaml @@ -0,0 +1,19 @@ +# For unittests +Quantization: + quantize_op_types: + - conv2d + - depthwise_conv2d + +Distillation: + alpha: 1.0 + loss: l2 + +TrainConfig: + epochs: 1 + eval_iter: 1070 + learning_rate: 2.0e-5 + optimizer_builder: + optimizer: + type: AdamW + weight_decay: 0.01 + origin_metric: 0.7403 diff --git a/tests/act/test_act_api.py b/tests/act/test_act_api.py new file mode 100644 index 00000000..0d7e80fe --- /dev/null +++ b/tests/act/test_act_api.py @@ -0,0 +1,125 @@ +import sys +import os +sys.path.append("../") +import unittest +import tempfile +import paddle +import unittest +import numpy as np +from static_case import StaticCase +from paddle.io import Dataset +from paddleslim.auto_compression import AutoCompression +from paddleslim.auto_compression.config_helpers import load_config + + +class RandomEvalDataset(Dataset): + def __init__(self, num_samples, image_shape=[3, 32, 32], class_num=10): + self.num_samples = num_samples + self.image_shape = image_shape + self.class_num = class_num + + def __getitem__(self, idx): + image = np.random.random(self.image_shape).astype('float32') + return image + + def __len__(self): + return self.num_samples + + +class ACTBase(unittest.TestCase): + def __init__(self, *args, **kwargs): + super(ACTBase, self).__init__(*args, **kwargs) + paddle.enable_static() + self.tmpdir = tempfile.TemporaryDirectory(prefix="test_") + self.infer_model_dir = os.path.join(self.tmpdir.name, "infer") + self.create_program() + self.create_dataloader() + + def create_program(self): + main_program = paddle.static.Program() + startup_program = paddle.static.Program() + with paddle.static.program_guard(main_program, startup_program): + data = paddle.static.data( + name='data', shape=[-1, 3, 32, 32], dtype='float32') + tmp = paddle.static.nn.conv2d( + input=data, num_filters=2, filter_size=3) + out = paddle.static.nn.conv2d( + input=tmp, num_filters=2, filter_size=3) + exe = paddle.static.Executor(paddle.CPUPlace()) + exe.run(startup_program) + + paddle.static.save_inference_model( + self.infer_model_dir, [data], [out], exe, program=main_program) + print(f"saved infer model to [{self.infer_model_dir}]") + + def create_dataloader(self): + # define a random dataset + self.eval_dataset = RandomEvalDataset(32) + + def __del__(self): + self.tmpdir.cleanup() + + +class TestYamlQATDistTrain(ACTBase): + def __init__(self, *args, **kwargs): + super(TestYamlQATDistTrain, self).__init__(*args, **kwargs) + + def test_compress(self): + image = paddle.static.data( + name='data', shape=[-1, 3, 32, 32], dtype='float32') + train_loader = paddle.io.DataLoader( + self.eval_dataset, feed_list=[image], batch_size=4) + ac = AutoCompression( + model_dir=self.tmpdir.name, + model_filename="infer.pdmodel", + params_filename="infer.pdiparams", + save_dir="output", + config="./qat_dist_train.yaml", + train_dataloader=train_loader, + eval_dataloader=train_loader) # eval_function to verify accuracy + ac.compress() + + +class TestSetQATDist(ACTBase): + def __init__(self, *args, **kwargs): + super(TestSetQATDist, self).__init__(*args, **kwargs) + + def test_compress(self): + image = paddle.static.data( + name='data', shape=[-1, 3, 32, 32], dtype='float32') + train_loader = paddle.io.DataLoader( + self.eval_dataset, feed_list=[image], batch_size=4) + ac = AutoCompression( + model_dir=self.tmpdir.name, + model_filename="infer.pdmodel", + params_filename="infer.pdiparams", + save_dir="output", + config={"QAT", "Distillation"}, + train_dataloader=train_loader, + eval_dataloader=train_loader) # eval_function to verify accuracy + ac.compress() + + +class TestDictQATDist(ACTBase): + def __init__(self, *args, **kwargs): + super(TestDictQATDist, self).__init__(*args, **kwargs) + + def test_compress(self): + config = load_config("./qat_dist_train.yaml") + image = paddle.static.data( + name='data', shape=[-1, 3, 32, 32], dtype='float32') + train_loader = paddle.io.DataLoader( + self.eval_dataset, feed_list=[image], batch_size=4) + ac = AutoCompression( + model_dir=self.tmpdir.name, + model_filename="infer.pdmodel", + params_filename="infer.pdiparams", + save_dir="output", + config=config, + train_dataloader=train_loader, + eval_dataloader=train_loader) # eval_function to verify accuracy + ac.compress() + + +if __name__ == '__main__': + unittest.main() -- GitLab