Support SVDQuant

auto_round/algorithms/block_runner.py

@@ -232,14 +232,6 @@ def forward(
                 self.last_output_dict["hidden_states"] = result
             return result
         else:
-            if self.batch_size == 1:
-                outputs = [output.unsqueeze(dim=self.batch_dim).to(out_device) for output in outputs]
-                if output_dict:
-                    output_dict = {
-                        key: [value.unsqueeze(dim=self.batch_dim).to(out_device) for value in values]
-                        for key, values in output_dict.items()
-                    }
-
             outputs = torch.cat(outputs, dim=self.batch_dim).to(out_device)
             if output_dict:
                 self.last_output_dict = {

auto_round/algorithms/registry.py

@@ -47,6 +47,7 @@ def _ensure_pipeline_members_registered() -> None:
         "auto_round.algorithms.quantization.sign_roundv2.quantizer",
         "auto_round.algorithms.quantization.adam_round.adam",
         "auto_round.algorithms.transforms.awq.base",
+        "auto_round.algorithms.transforms.svdquant.apply",
     ):
         importlib.import_module(module_name)
     _pipeline_members_registered = True

auto_round/algorithms/transforms/svdquant/__init__.py

@@ -0,0 +1,19 @@
+# Copyright (c) 2026 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from auto_round.algorithms.transforms.svdquant.apply import SVDQuantTransform
+from auto_round.algorithms.transforms.svdquant.config import SVDQuantConfig
+from auto_round.algorithms.transforms.svdquant.wrapper import SVDQuantLinear
+
+__all__ = ["SVDQuantConfig", "SVDQuantLinear", "SVDQuantTransform"]