Make quantization_config a part of OVConfig in OVQuantizer

Author: AlexKoff88

optimum/intel/openvino/configuration.py

@@ -15,6 +15,7 @@
 from typing import Dict, List, Optional, Union
 
 import torch
+from transformers.utils.quantization_config import QuantizationConfigMixin
 
 from optimum.configuration_utils import BaseConfig
 
@@ -83,6 +84,7 @@ def __init__(
         compression: Union[List[Dict], Dict, None] = None,
         input_info: Optional[List] = None,
         save_onnx_model: bool = False,
+        quantization_config: Optional[QuantizationConfigMixin] = None,
         **kwargs,
     ):
         super().__init__()
@@ -91,6 +93,7 @@ def __init__(
         self.save_onnx_model = save_onnx_model
         self._enable_standard_onnx_export_option()
         self.optimum_version = kwargs.pop("optimum_version", None)
+        self.quantization_config = quantization_config
 
     def add_input_info(self, model_inputs: Dict, force_batch_one: bool = False):
         self.input_info = [

optimum/intel/openvino/quantization.py

@@ -33,7 +33,6 @@
 from torch.utils.data import DataLoader, RandomSampler
 from transformers import DataCollator, PreTrainedModel, default_data_collator
 from transformers.pytorch_utils import Conv1D
-from transformers.utils.quantization_config import QuantizationConfigMixin
 
 from optimum.exporters.tasks import TasksManager
 from optimum.quantization_base import OptimumQuantizer
@@ -159,7 +158,6 @@ def quantize(
         self,
         calibration_dataset: Dataset = None,
         save_directory: Union[str, Path] = None,
-        quantization_config: QuantizationConfigMixin = None,
         ov_config: OVConfig = None,
         file_name: Optional[str] = None,
         batch_size: int = 1,
@@ -234,7 +232,7 @@ def quantize(
                 data_collator,
                 remove_unused_columns,
                 weights_only,
-                quantization_config,
+                ov_config,
                 **kwargs,
             )
         elif isinstance(self.model, OVBaseModel):
@@ -313,13 +311,14 @@ def _quantize_ovcausallm(
         data_collator: Optional[DataCollator] = None,
         remove_unused_columns: bool = True,
         weights_only: bool = False,
-        quantization_config: QuantizationConfigMixin = None,
+        ov_config: OVConfig = None,
         **kwargs,
     ):
         save_directory = Path(save_directory)
         save_directory.mkdir(parents=True, exist_ok=True)
 
         if weights_only:
+            quantization_config = None if ov_config is None else ov_config.quantization_config
             if quantization_config is None:
                 # Use default 8-bit compression
                 self.model.model = nncf.compress_weights(self.model.model)