Style

Author: AlexKoff88

optimum/intel/openvino/weight_quantization.py

@@ -96,6 +96,7 @@ def post_init(self):
                     ['wikitext2','c4','c4-new','ptb','ptb-new'], but we found {self.dataset}"""
                 )
 
+
 DEFAULT_4BIT_CONFIGS = {
     "dolly-v2-3b": {"mode": nncf.CompressWeightsMode.INT4_ASYM, "group_size": 32, "ratio": 0.5},
     "gpt-j-6b": {"mode": nncf.CompressWeightsMode.INT4_ASYM, "group_size": 64},
@@ -114,6 +115,7 @@ def post_init(self):
     "qwen-7b-chat": {"mode": nncf.CompressWeightsMode.INT4_SYM, "group_size": 128, "ratio": 0.6},
 }
 
+
 def _check_default_4bit_configs(config: PretrainedConfig):
     return DEFAULT_4BIT_CONFIGS.get(config.name_or_path, None)
 
@@ -150,7 +152,7 @@ def compress_decoder_weights(model, quantization_config: Union[OVWeightQuantizat
             group_size=config.group_size,
             all_layers=config.all_layers,
             sensitivity_metric=config.sensitivity_metric,
-            awq = config.awq,
+            awq=config.awq,
             ignored_scope=config.ignored_scope,
             dataset=dataset,
         )

tests/openvino/test_quantization.py

@@ -181,8 +181,31 @@ class OVWeightCompressionTest(unittest.TestCase):
             dict(mode=nncf.CompressWeightsMode.INT4_ASYM, group_size=-1, ratio=0.8, all_layers=True),
             22,
         ),
-        # TODO: uncomment after fix
-        # (OVModelForCausalLM, "hf-internal-testing/tiny-random-gpt2", dict(mode=nncf.CompressWeightsMode.INT4_SYM, group_size=-1, ratio=0.8, sensitivity_metric=nncf.SensitivityMetric.MEAN_ACTIVATION_MAGNITUDE, dataset="ptb"), 16),
+        (
+            OVModelForCausalLM,
+            "hf-internal-testing/tiny-random-gpt2",
+            dict(
+                mode=nncf.CompressWeightsMode.INT4_SYM,
+                group_size=-1,
+                ratio=0.8,
+                sensitivity_metric=nncf.SensitivityMetric.MEAN_ACTIVATION_MAGNITUDE,
+                dataset="ptb",
+            ),
+            16,
+        ),
+        (
+            OVModelForCausalLM,
+            "hf-internal-testing/tiny-random-gpt2",
+            dict(
+                mode=nncf.CompressWeightsMode.INT4_SYM,
+                group_size=-1,
+                ratio=0.8,
+                sensitivity_metric=nncf.SensitivityMetric.MEAN_ACTIVATION_MAGNITUDE,
+                dataset="ptb",
+                awq=True,
+            ),
+            16,
+        ),
     )
 
     SUPPORTED_ARCHITECTURES_WITH_AUTO_COMPRESSION = (