Adapt Neural Coder to INC New API (#858)

Signed-off-by: zehao-intel <[email protected]>
intel · May 11, 2023 · e74a8a0 · e74a8a0
1 parent 113ca1e
commit e74a8a0
Show file tree

Hide file tree

Showing 5 changed files with 17 additions and 35 deletions.
diff --git a/neural_coder/backends/keras_inc.yaml b/neural_coder/backends/keras_inc.yaml
@@ -19,7 +19,6 @@ transformation:
     - |-
       [+] from neural_compressor.quantization import fit
       [+] from neural_compressor.config import PostTrainingQuantConfig
-      [+] from neural_compressor.experimental import common
       [+] config = PostTrainingQuantConfig(backend='itex')
       [+] quantized_model = fit(MODEL_NAME, conf=config, calib_dataloader=DATALOADER_NAME, eval_func=eval_func)
   order:

diff --git a/neural_coder/backends/onnx_inc_dynamic_quant.yaml b/neural_coder/backends/onnx_inc_dynamic_quant.yaml
@@ -17,14 +17,10 @@ transformation:
     - insert_below_model_definition_line
   content:
     - |-
-      [+] from neural_compressor.experimental import Quantization, common
-      [+] from neural_compressor import options, conf
-      [+] conf.model.framework = 'onnxrt_integerops'
-      [+] conf.quantization.approach = 'post_training_dynamic_quant'
-      [+] quantizer = Quantization(conf)
-      [+] quantizer.model = common.Model(MODEL_NAME)
-      [+] quantizer.eval_func = EVAL_FUNCTION_NAME
-      [+] MODEL_NAME = quantizer()
+      [+] from neural_compressor.quantization import fit
+      [+] from neural_compressor.config import PostTrainingQuantConfig
+      [+] config = PostTrainingQuantConfig(approach='dynamic')
+      [+] MODEL_NAME = fit(MODEL_NAME, conf=config, calib_dataloader=DATALOADER_NAME, eval_func=EVAL_FUNCTION_NAME)
   order:
     - below:
       above:
diff --git a/neural_coder/backends/onnx_inc_static_quant_qdq.yaml b/neural_coder/backends/onnx_inc_static_quant_qdq.yaml
@@ -17,15 +17,10 @@ transformation:
     - insert_below_model_definition_line
   content:
     - |-
-      [+] from neural_compressor.experimental import Quantization, common
-      [+] from neural_compressor import options, conf
-      [+] conf.model.framework = 'onnxrt_qdqops'
-      [+] conf.quantization.approach = 'post_training_static_quant'
-      [+] quantizer = Quantization(conf)
-      [+] quantizer.model = common.Model(MODEL_NAME)
-      [+] quantizer.calib_dataloader = DATALOADER_NAME
-      [+] quantizer.eval_func = EVAL_FUNCTION_NAME
-      [+] MODEL_NAME = quantizer()
+      [+] from neural_compressor.quantization import fit
+      [+] from neural_compressor.config import PostTrainingQuantConfig
+      [+] config = PostTrainingQuantConfig(quant_format='QDQ')
+      [+] MODEL_NAME = fit(MODEL_NAME, conf=config, calib_dataloader=DATALOADER_NAME, eval_func=EVAL_FUNCTION_NAME)
   order:
     - below:
       above:
diff --git a/neural_coder/backends/onnx_inc_static_quant_qlinear.yaml b/neural_coder/backends/onnx_inc_static_quant_qlinear.yaml
@@ -17,15 +17,10 @@ transformation:
     - insert_below_model_definition_line
   content:
     - |-
-      [+] from neural_compressor.experimental import Quantization, common
-      [+] from neural_compressor import options, conf
-      [+] conf.model.framework = 'onnxrt_qlinearops'
-      [+] conf.quantization.approach = 'post_training_static_quant'
-      [+] quantizer = Quantization(conf)
-      [+] quantizer.model = common.Model(MODEL_NAME)
-      [+] quantizer.calib_dataloader = DATALOADER_NAME
-      [+] quantizer.eval_func = EVAL_FUNCTION_NAME
-      [+] MODEL_NAME = quantizer()
+      [+] from neural_compressor.quantization import fit
+      [+] from neural_compressor.config import PostTrainingQuantConfig
+      [+] config = PostTrainingQuantConfig(quant_format='default')
+      [+] MODEL_NAME = fit(MODEL_NAME, conf=config, calib_dataloader=DATALOADER_NAME, eval_func=EVAL_FUNCTION_NAME)
   order:
     - below:
       above:
diff --git a/neural_coder/coders/tensorflow/inc.py b/neural_coder/coders/tensorflow/inc.py
@@ -28,14 +28,11 @@ def transform(self):
                 model_name = "model"
                 indent_level = get_line_indent_level(line)
                 self.result.append(line)
-                self.result.append(" " * indent_level + "from neural_compressor.conf.config import QuantConf")
-                self.result.append(" " * indent_level + "from neural_compressor.experimental import Quantization")
-                self.result.append(" " * indent_level + "from neural_compressor.experimental import common")
-                self.result.append(" " * indent_level + "quant_config = QuantConf()")
-                self.result.append(" " * indent_level + "quant_config.usr_cfg.model.framework = 'tensorflow'")
-                self.result.append(" " * indent_level + "quantizer = Quantization(quant_config)")
-                self.result.append(" " * indent_level + "quantizer.model = common.Model(" + model_name + ")")
-                self.result.append(" " * indent_level + model_name + " = quantizer.fit()")
+                self.result.append(" " * indent_level + "from neural_compressor.quantization import fit")
+                self.result.append(" " * indent_level + "from neural_compressor.config import PostTrainingQuantConfig")
+                self.result.append(" " * indent_level + "from neural_compressor import common")
+                self.result.append(" " * indent_level + "config = PostTrainingQuantConfig()")
+                self.result.append(" " * indent_level + model_name + " = fit(" + model_name + ", conf=config)")
             else:
                 self.result.append(line)
         for index, line in enumerate(self.result):