Apply comments

Author: nikita-malininn

optimum/intel/openvino/configuration.py

@@ -658,6 +658,13 @@ def __init__(
         self.overflow_fix = overflow_fix
         self.smooth_quant_alpha = smooth_quant_alpha
         self.activation_format = activation_format
+
+        f8_formats = ["f8e4m3", "f8e5m2"]
+        if self.activation_format in f8_formats and self.weight_format in f8_formats:
+            logger.info(
+                f"{self.activation_format} for activations and {self.weight_format} weights were found. A symmetrical scheme will be used."
+            )
+            self.sym = True
         self.post_init()
 
     def post_init(self):
@@ -674,16 +681,6 @@ def post_init(self):
                 f"SmoothQuant alpha parameter must be in range [0, 1], but found {self.smooth_quant_alpha}"
             )
 
-        if not self.sym:
-            if self.activation_format != "int8":
-                raise ValueError(
-                    f"Asymmetric quantization can not be performed in {self.activation_format} activation format."
-                )
-            if self.weight_format != "int8":
-                raise ValueError(
-                    f"Asymmetric quantization can not be performed in {self.weight_format} weight format."
-                )
-
 
 class OVConfig(BaseConfig):
     CONFIG_NAME = "openvino_config.json"
@@ -708,8 +705,7 @@ def __init__(
             "compression", None
         )  # A field for backward-compatability of training-time compression parameters
         if self.quantization_config is not None:
-            if isinstance(self.quantization_config, OVWeightQuantizationConfig):
-                self.dtype = self.quantization_config.weight_format
+            self.dtype = self.quantization_config.weight_format
         else:
             self.dtype = dtype
 

tests/openvino/test_exporters_cli.py

@@ -123,7 +123,7 @@ class OVCLIExportTestCase(unittest.TestCase):
             "text-generation",
             "llama",
             "f8e4m3",
-            "--dataset wikitext2 --num-samples 1 --smooth-quant-alpha 0.9 --trust-remote-code --sym",
+            "--dataset wikitext2 --num-samples 1 --smooth-quant-alpha 0.9 --trust-remote-code",
             (13,),
             (16,),
         ),
@@ -418,7 +418,7 @@ def test_exporters_cli_full_quantization(
         model_type: str,
         quant_mode: str,
         option: str,
-        expected_num_fq_nodes_per_model: Tuple[int],
+        expected_num_f_nodes_per_model: Tuple[int],
         expected_num_weight_nodes_per_model: Tuple[int],
     ):
         with TemporaryDirectory() as tmpdir:
@@ -432,10 +432,10 @@ def test_exporters_cli_full_quantization(
             models = [model]
             if task == "automatic-speech-recognition":
                 models = [model.encoder, model.decoder, model.decoder_with_past]
-            self.assertEqual(len(expected_num_fq_nodes_per_model), len(models))
+            self.assertEqual(len(expected_num_f_nodes_per_model), len(models))
             for i, model in enumerate(models):
                 actual_num_f_nodes, actual_num_weight_nodes = get_num_quantized_nodes(model)
-                self.assertEqual(expected_num_fq_nodes_per_model[i], actual_num_f_nodes)
+                self.assertEqual(expected_num_f_nodes_per_model[i], actual_num_f_nodes)
                 self.assertEqual(expected_num_weight_nodes_per_model[i], actual_num_weight_nodes[quant_mode])
 
     def test_exporters_cli_int4_with_local_model_and_default_config(self):