Update

Author: rk119

nncf/quantization/algorithms/weight_compression/gptq.py

@@ -215,10 +215,7 @@ def _quantize_weights(
         """
         if wc_params.node_with_weight.metatype in self._backend_entity.convolution_metatypes:
             msg = "Convolution metatypes are not supported"
-            raise RuntimeError(msg)
-        if not wc_params.node_with_weight.layer_attributes.constant_attributes[wc_params.weight_port_id]["transpose"]:
-            msg = "Transpose is not supported"
-            raise RuntimeError(msg)
+            raise nncf.UnsupportedModelError(msg)
 
         weight_tensor = self._backend_entity.get_weight(
             wc_params.node_with_weight, wc_params.weight_port_id, model, graph

nncf/quantization/algorithms/weight_compression/openvino_backend.py

@@ -130,6 +130,9 @@ def get_weight_names_and_port_ids(node: NNCFNode, graph: NNCFGraph) -> List[Tupl
         return result
 
     def get_weight(self, node_with_weight: NNCFNode, weight_port_id: int, model: ov.Model, graph: NNCFGraph) -> Tensor:
+        if not node_with_weight.layer_attributes.constant_attributes[weight_port_id]["transpose"]:
+            msg = "Only transposed weights are supported"
+            raise nncf.UnsupportedModelError(msg)
         weight_name = node_with_weight.layer_attributes.constant_attributes[weight_port_id]["name"]
         weight_node = self.name_to_node_mapping[weight_name]
         weight_tensor = get_const_value_as_numpy_tensor(weight_node)

tests/openvino/native/quantization/test_weights_compression.py

@@ -11,6 +11,7 @@
 
 import inspect
 import os
+from contextlib import nullcontext
 from typing import Callable, Dict, List, Optional
 
 import numpy as np
@@ -1457,9 +1458,19 @@ def test_compression_with_different_algo_combinations(input_shape, kwargs):
     )
 
 
+@pytest.mark.parametrize(
+    ("transpose_a", "transpose_b", "raises_error"),
+    [
+        (False, True, False),
+        (True, True, False),
+        (False, False, True),
+        (True, False, True),
+    ],
+    ids=["tb_nota", "ta_tb", "nota_notb", "ta_notb"],
+)
 @pytest.mark.parametrize(
     "kwargs",
-    (
+    [
         dict(scale_estimation=True),
         dict(lora_correction=True),
         dict(
@@ -1468,25 +1479,30 @@ def test_compression_with_different_algo_combinations(input_shape, kwargs):
             scale_estimation=True,
             advanced_parameters=CompressionParams(gptq_params=GPTQParams(subset_size=2)),
         ),
-    ),
+    ],
     ids=["se", "lora", "gptq_se_awq"],
 )
-def test_compression_with_transpose(kwargs):
+def test_compression_with_transpose(transpose_a, transpose_b, raises_error, kwargs):
     dataset_size = 4
-    model = LMLinearModel(transpose_a=True, transpose_b=True).ov_model
+    model = LMLinearModel(transpose_a=transpose_a, transpose_b=transpose_b).ov_model
     input_data = [np.ones(inp.shape) for inp in model.inputs] * dataset_size
     dataset = Dataset(input_data)
 
-    compress_weights(
-        model,
-        mode=CompressWeightsMode.INT4_SYM,
-        ratio=1.0,
-        group_size=8,
-        subset_size=2,
-        dataset=dataset,
-        all_layers=True,
-        **kwargs,
-    )
+    with (
+        pytest.raises(nncf.UnsupportedModelError)
+        if raises_error and not kwargs.get("lora_correction", False)
+        else nullcontext()
+    ):
+        compress_weights(
+            model,
+            mode=CompressWeightsMode.INT4_SYM,
+            ratio=1.0,
+            group_size=8,
+            subset_size=2,
+            dataset=dataset,
+            all_layers=True,
+            **kwargs,
+        )
 
 
 class TestOVTemplateWeightCompression(TemplateWeightCompression):