Update

Author: rk119

nncf/quantization/algorithms/weight_compression/gptq.py

@@ -215,10 +215,10 @@ def _quantize_weights(
         """
         if wc_params.node_with_weight.metatype in self._backend_entity.convolution_metatypes:
             msg = "Convolution metatypes are not supported"
-            raise nncf.UnsupportedModelError(msg)
+            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 nncf.UnsupportedModelError(msg)
+            raise RuntimeError(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/scale_estimation.py

@@ -117,13 +117,6 @@ def apply(
         scales, zero_points = dict(), dict()
 
         for wp in track(all_weight_params, description="Applying Scale Estimation"):
-            if (
-                wp.node_with_weight.metatype in self._backend_entity.matmul_metatypes
-                and not wp.node_with_weight.layer_attributes.constant_attributes[wp.weight_port_id]["transpose"]
-            ):
-                msg = "Transpose is not supported"
-                raise nncf.UnsupportedModelError(msg)
-
             weight_name = wp.weight_name
             node_name = wp.node_with_weight.node_name
             config = wp.compression_config

tests/openvino/native/quantization/test_weights_compression.py

@@ -11,7 +11,6 @@
 
 import inspect
 import os
-from contextlib import nullcontext
 from typing import Callable, Dict, List, Optional
 
 import numpy as np
@@ -1458,16 +1457,6 @@ 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",
     (
@@ -1482,27 +1471,22 @@ def test_compression_with_different_algo_combinations(input_shape, kwargs):
     ),
     ids=["se", "lora", "gptq_se_awq"],
 )
-def test_compression_with_transpose(transpose_a, transpose_b, raises_error, kwargs):
+def test_compression_with_transpose(kwargs):
     dataset_size = 4
-    model = LMLinearModel(transpose_a=transpose_a, transpose_b=transpose_b).ov_model
+    model = LMLinearModel(transpose_a=True, transpose_b=True).ov_model
     input_data = [np.ones(inp.shape) for inp in model.inputs] * dataset_size
     dataset = Dataset(input_data)
 
-    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,
-        )
+    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):