[TorchFX] Constant folding is removed from nncf.quantize and OpenVINOQuantizer

Author: daniil-lyakhov

nncf/experimental/torch/fx/quantization/quantize_model.py

@@ -86,7 +86,6 @@ def quantize_impl(
         advanced_parameters=advanced_parameters,
     )
 
-    # To make it easier for bias correction algorithms.
     apply_quantization_transformations(copied_model)
 
     nncf_graph = NNCFGraphFactory.create(copied_model)

nncf/experimental/torch/fx/quantization/quantizer/openvino_quantizer.py

@@ -38,7 +38,6 @@
 from nncf.common.utils.api_marker import api
 from nncf.experimental.torch.fx.nncf_graph_builder import GraphConverter
 from nncf.experimental.torch.fx.node_utils import get_graph_node_by_name
-from nncf.experimental.torch.fx.transformations import fold_constant_except_qdq
 from nncf.quantization.advanced_parameters import FP8QuantizationParameters
 from nncf.quantization.advanced_parameters import QuantizationParameters
 from nncf.quantization.algorithms.min_max.algorithm import MinMaxQuantization
@@ -362,5 +361,4 @@ def transform_for_annotation(self, model: torch.fx.GraphModule) -> torch.fx.Grap
         :param model: Given torch.fx.GraphModule to transform before the annotation.
         :return: The transformed torch.fx.GraphModule ready for the annotation.
         """
-        fold_constant_except_qdq(model)
         return model

nncf/experimental/torch/fx/transformations.py

@@ -733,7 +733,6 @@ def apply_quantization_transformations(model: torch.fx.GraphModule) -> None:
     # to make it easier for algorithms to work
     # with the target graph BatchNorm operations
     # are being fused
-    fold_constant_except_qdq(model)
     _fuse_conv_bn_(model)
 
 

nncf/version.py

@@ -9,7 +9,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "2.16.0"
+__version__ = "2.16.0.dev0+d400543fbdirty"
 
 
 BKC_TORCH_SPEC = "==2.6.*"