Merge pull request #4000 from LDOUBLEV/fix_nonfinite

fix nonfinite and add quant kl

deploy/slim/quantization/quant_kl.py

+# Copyright (c) 2020 PaddlePaddle Authors. 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 __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, '..', '..', '..')))
+sys.path.append(
+    os.path.abspath(os.path.join(__dir__, '..', '..', '..', 'tools')))
+
+import yaml
+import paddle
+import paddle.distributed as dist
+
+paddle.seed(2)
+
+from ppocr.data import build_dataloader
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import init_model
+import tools.program as program
+import paddleslim
+from paddleslim.dygraph.quant import QAT
+import numpy as np
+
+dist.get_world_size()
+
+
+class PACT(paddle.nn.Layer):
+    def __init__(self):
+        super(PACT, self).__init__()
+        alpha_attr = paddle.ParamAttr(
+            name=self.full_name() + ".pact",
+            initializer=paddle.nn.initializer.Constant(value=20),
+            learning_rate=1.0,
+            regularizer=paddle.regularizer.L2Decay(2e-5))
+
+        self.alpha = self.create_parameter(
+            shape=[1], attr=alpha_attr, dtype='float32')
+
+    def forward(self, x):
+        out_left = paddle.nn.functional.relu(x - self.alpha)
+        out_right = paddle.nn.functional.relu(-self.alpha - x)
+        x = x - out_left + out_right
+        return x
+
+
+quant_config = {
+    # weight preprocess type, default is None and no preprocessing is performed. 
+    'weight_preprocess_type': None,
+    # activation preprocess type, default is None and no preprocessing is performed.
+    'activation_preprocess_type': None,
+    # weight quantize type, default is 'channel_wise_abs_max'
+    'weight_quantize_type': 'channel_wise_abs_max',
+    # activation quantize type, default is 'moving_average_abs_max'
+    'activation_quantize_type': 'moving_average_abs_max',
+    # weight quantize bit num, default is 8
+    'weight_bits': 8,
+    # activation quantize bit num, default is 8
+    'activation_bits': 8,
+    # data type after quantization, such as 'uint8', 'int8', etc. default is 'int8'
+    'dtype': 'int8',
+    # window size for 'range_abs_max' quantization. default is 10000
+    'window_size': 10000,
+    # The decay coefficient of moving average, default is 0.9
+    'moving_rate': 0.9,
+    # for dygraph quantization, layers of type in quantizable_layer_type will be quantized
+    'quantizable_layer_type': ['Conv2D', 'Linear'],
+}
+
+
+def sample_generator(loader):
+    def __reader__():
+        for indx, data in enumerate(loader):
+            images = np.array(data[0])
+            yield images
+
+    return __reader__
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config['Global']['distributed']:
+        dist.init_parallel_env()
+
+    global_config = config['Global']
+
+    # build dataloader
+    config['Train']['loader']['num_workers'] = 0
+    train_dataloader = build_dataloader(config, 'Train', device, logger)
+    if config['Eval']:
+        config['Eval']['loader']['num_workers'] = 0
+        valid_dataloader = build_dataloader(config, 'Eval', device, logger)
+    else:
+        valid_dataloader = None
+
+    paddle.enable_static()
+    place = paddle.CPUPlace()
+    exe = paddle.static.Executor(place)
+
+    if 'inference_model' in global_config.keys():  # , 'inference_model'):
+        inference_model_dir = global_config['inference_model']
+    else:
+        inference_model_dir = os.path.dirname(global_config['pretrained_model'])
+        if  not (os.path.exists(os.path.join(inference_model_dir, "inference.pdmodel")) and \
+            os.path.exists(os.path.join(inference_model_dir, "inference.pdiparams")) ):
+            raise ValueError(
+                "Please set inference model dir in Global.inference_model or Global.pretrained_model for post-quantazition"
+            )
+
+    paddleslim.quant.quant_post_static(
+        executor=exe,
+        model_dir=inference_model_dir,
+        model_filename='inference.pdmodel',
+        params_filename='inference.pdiparams',
+        quantize_model_path=global_config['save_inference_dir'],
+        sample_generator=sample_generator(train_dataloader),
+        save_model_filename='inference.pdmodel',
+        save_params_filename='inference.pdiparams',
+        batch_size=1,
+        batch_nums=None)
+
+
+if __name__ == '__main__':
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)

tests/configs/det_mv3_db.yml

@@ -23,10 +23,10 @@ Architecture:
     name: MobileNetV3
     scale: 0.5
     model_name: large
-    disable_se: True
+    disable_se: False
   Neck:
     name: DBFPN
-    out_channels: 96
+    out_channels: 256
   Head:
     name: DBHead
     k: 50
@@ -74,7 +74,7 @@ Train:
           channel_first: False
       - DetLabelEncode: # Class handling label
       - Resize:
-          # size: [640, 640]
+          size: [640, 640]
       - MakeBorderMap:
           shrink_ratio: 0.4
           thresh_min: 0.3

tests/ocr_det_params.txt

@@ -12,7 +12,7 @@ train_model_name:latest
 train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
 null:null
 ##
-trainer:norm_train|pact_train
+trainer:norm_train|pact_train|fpgm_train
 norm_train:tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
 pact_train:deploy/slim/quantization/quant.py -c tests/configs/det_mv3_db.yml -o
 fpgm_train:deploy/slim/prune/sensitivity_anal.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
@@ -21,7 +21,7 @@ null:null
 null:null
 ##
 ===========================eval_params=========================== 
-eval:tools/eval.py -c tests/configs/det_mv3_db.yml -o
+eval:null
 null:null
 ##
 ===========================infer_params===========================
@@ -35,7 +35,7 @@ export1:null
 export2:null
 ##
 train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
-infer_export:tools/export_model.py -c configs/det/det_mv3_db.yml -o
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
 infer_quant:False
 inference:tools/infer/predict_det.py
 --use_gpu:True|False

tests/ocr_kl_quant_params.txt

+===========================train_params===========================
+model_name:ocr_system
+python:python3.7
+gpu_list:null
+Global.use_gpu:null
+Global.auto_cast:null
+Global.epoch_num:null
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:null
+Global.pretrained_model:null
+train_model_name:null
+train_infer_img_dir:null
+null:null
+##
+trainer:
+norm_train:null
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.pretrained_model:
+norm_export:null 
+quant_export:null 
+fpgm_export:null 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+kl_quant:deploy/slim/quantization/quant_kl.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml  -o  
+infer_quant:True
+inference:tools/infer/predict_det.py
+--use_gpu:TrueFalse
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null

tests/test.sh

@@ -433,7 +433,9 @@ if [ ${MODE} = "infer" ]; then
             save_infer_dir=$(dirname $infer_model)
             set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
             set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
-            export_cmd="${python} ${norm_export} ${set_export_weight} ${set_save_infer_key}"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key}"
+            echo ${infer_run_exports[Count]} 
+            echo  $export_cmd
             eval $export_cmd
             status_export=$?
             status_check $status_export "${export_cmd}" "${status_log}"