Assymetric activations are forced

nncf/experimental/torch_fx/quantization/quantize_model.py

@@ -24,6 +24,7 @@
 import nncf
 from nncf.common.factory import NNCFGraphFactory
 from nncf.common.quantization.structs import QuantizationPreset
+from nncf.common.quantization.structs import QuantizationScheme
 from nncf.data import Dataset
 from nncf.parameters import CompressWeightsMode
 from nncf.parameters import ModelType
@@ -32,6 +33,7 @@
 from nncf.parameters import TargetDevice
 from nncf.quantization.advanced_parameters import AdvancedCompressionParameters
 from nncf.quantization.advanced_parameters import AdvancedQuantizationParameters
+from nncf.quantization.advanced_parameters import QuantizationParameters
 from nncf.quantization.algorithms.post_training.algorithm import PostTrainingQuantization
 from nncf.quantization.algorithms.weight_compression.algorithm import WeightCompression
 from nncf.scopes import IgnoredScope
@@ -66,6 +68,17 @@ def quantize_impl(
     copied_model = deepcopy(model)
     # copied_model = model
 
+    if advanced_parameters is None:
+        advanced_parameters = AdvancedQuantizationParameters()
+    # torch.fx supports only assymetric activations quantization
+    # force to use only this type of quantization
+    activations_quantization_params = advanced_parameters.activations_quantization_params
+    if activations_quantization_params is None:
+        activations_quantization_params = QuantizationParameters()
+
+    activations_quantization_params.mode = QuantizationScheme.ASYMMETRIC
+    advanced_parameters.activations_quantization_params = activations_quantization_params
+
     quantization_algorithm = PostTrainingQuantization(
         preset=preset,
         target_device=target_device,

nncf/experimental/torch_fx/transformations.py

@@ -108,11 +108,8 @@ def insert_one_qdq(
 
     # Quantized functions accepts only uint8 as an input
     if target_point.target_type != TargetType.OPERATION_WITH_WEIGHTS and qparams["_dtype_"] == torch.int8:
-        qparams["_zero_point_"] = qparams["_zero_point_"] - qparams["_quant_min_"]
-        quants_len = qparams["_quant_max_"] - qparams["_quant_min_"]
-        qparams["_quant_min_"] = 0
-        qparams["_quant_max_"] = quants_len
-        qparams["_dtype_"] = torch.uint8
+        raise RuntimeError("Wrong parameters: activations should always be uint8")
+
     # TODO: map FakeQuantizePramaeters to qparams for quantize/dequantize
     # 2. replace activation_post_process node with quantize and dequantize
     graph = model.graph