update full quant PTQ demo (#1512)

example/full_quantization/picodet/configs/picodet_npu.yaml

@@ -7,11 +7,15 @@ Global:
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 
-Distillation:
-  alpha: 1.0
-  loss: l2
+PTQ:   # Post Training Quantization
+  quantizable_op_type: ["conv2d", "depthwise_conv2d"]
+  activation_quantize_type: 'moving_average_abs_max'
+  algo: avg
+  onnx_format: False
+  batch_size: 32
+  batch_nums: 10
 
-Quantization:
+Quantization:   # Auto Compression
   use_pact: true
   activation_quantize_type: 'moving_average_abs_max'
   weight_bits: 8
@@ -20,6 +24,10 @@ Quantization:
   - conv2d
   - depthwise_conv2d
 
+Distillation:
+  alpha: 1.0
+  loss: l2
+
 TrainConfig:
   train_iter: 8000
   eval_iter: 1000

example/full_quantization/picodet/configs/picodet_npu_with_postprocess.yaml

@@ -7,11 +7,15 @@ Global:
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 
-Distillation:
-  alpha: 1.0
-  loss: l2
+PTQ:   # Post Training Quantization
+  quantizable_op_type: ["conv2d", "depthwise_conv2d"]
+  activation_quantize_type: 'moving_average_abs_max'
+  algo: avg
+  onnx_format: False
+  batch_size: 32
+  batch_nums: 10
 
-Quantization:
+Quantization:   # Auto Compression
   use_pact: true
   activation_quantize_type: 'moving_average_abs_max'
   weight_bits: 8
@@ -20,6 +24,10 @@ Quantization:
   - conv2d
   - depthwise_conv2d
 
+Distillation:
+  alpha: 1.0
+  loss: l2
+
 TrainConfig:
   train_iter: 8000
   eval_iter: 1000

example/full_quantization/picodet/post_quant.py

@@ -41,8 +41,6 @@ def argsparser():
         type=str,
         default='gpu',
         help="which device used to compress.")
-    parser.add_argument(
-        '--algo', type=str, default='avg', help="post quant algo.")
 
     return parser
 
@@ -71,6 +69,7 @@ def main():
                                         reader_cfg['worker_num'],
                                         return_list=True)
     train_loader = reader_wrapper(train_loader, global_config['input_list'])
+    ptq_config = all_config['PTQ']
 
     place = paddle.CUDAPlace(0) if FLAGS.devices == 'gpu' else paddle.CPUPlace()
     exe = paddle.static.Executor(place)
@@ -81,13 +80,15 @@ def main():
         data_loader=train_loader,
         model_filename=global_config["model_filename"],
         params_filename=global_config["params_filename"],
-        batch_size=32,
-        batch_nums=10,
-        algo=FLAGS.algo,
+        quantizable_op_type=ptq_config['quantizable_op_type'],
+        activation_quantize_type=ptq_config['activation_quantize_type'],
+        batch_size=ptq_config['batch_size'],
+        batch_nums=ptq_config['batch_nums'],
+        algo=ptq_config['algo'],
         hist_percent=0.999,
         is_full_quantize=False,
         bias_correction=False,
-        onnx_format=True,
+        onnx_format=ptq_config['onnx_format'],
         skip_tensor_list=None)
 
 

example/full_quantization/ppyoloe/configs/ppyoloe_s_416_qat_dis.yaml

@@ -7,9 +7,10 @@ Global:
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 
-PTQ:
+PTQ:  # Post Training Quantization
   quantizable_op_type: ["conv2d", "depthwise_conv2d"]
   activation_quantize_type: 'moving_average_abs_max'
+  algo: avg
   is_full_quantize: True
   onnx_format: False
   batch_size: 10
@@ -19,7 +20,7 @@ Distillation:
   alpha: 1.0
   loss: soft_label
 
-Quantization:
+Quantization:   # Auto Compression
   onnx_format: true
   use_pact: true
   activation_quantize_type: 'moving_average_abs_max'

example/full_quantization/ppyoloe/configs/ppyoloe_s_qat_dis.yaml

@@ -7,19 +7,16 @@ Global:
   model_filename: model.pdmodel
   params_filename: model.pdiparams
 
-PTQ:
+PTQ:   # Post Training Quantization
   quantizable_op_type: ["conv2d", "depthwise_conv2d"]
   activation_quantize_type: 'moving_average_abs_max'
+  algo: avg
   is_full_quantize: True
   onnx_format: False
   batch_size: 10
   batch_nums: 10
 
-Distillation:
-  alpha: 1.0
-  loss: soft_label
-
-Quantization:
+Quantization:   # Auto Compression
   onnx_format: true
   use_pact: true
   activation_quantize_type: 'moving_average_abs_max'
@@ -27,6 +24,10 @@ Quantization:
   - conv2d
   - depthwise_conv2d
 
+Distillation:
+  alpha: 1.0
+  loss: soft_label
+
 TrainConfig:
   train_iter: 5000
   eval_iter: 1000

example/full_quantization/ppyoloe/post_quant.py

@@ -42,8 +42,6 @@ def argsparser():
         type=str,
         default='gpu',
         help="which device used to compress.")
-    parser.add_argument(
-        '--algo', type=str, default='avg', help="post quant algo.")
 
     return parser
 
@@ -89,7 +87,7 @@ def main():
         activation_quantize_type=ptq_config['activation_quantize_type'],
         batch_size=ptq_config['batch_size'],
         batch_nums=ptq_config['batch_nums'],
-        algo=FLAGS.algo,
+        algo=ptq_config['algo'],
         hist_percent=0.999,
         is_full_quantize=ptq_config['is_full_quantize'],
         bias_correction=False,