Add is_autoround_available to improve package install (#1338)

intel_extension_for_transformers/llm/quantization/utils.py

@@ -24,15 +24,16 @@
 from datasets import load_dataset
 from neural_compressor import quantization
 from neural_compressor.adaptor.torch_utils.model_wrapper import WeightOnlyLinear
-from auto_round.export.export_to_itrex.model_wrapper import WeightOnlyLinear as auto_round_woqlinear
 from neural_compressor.utils.utility import LazyImport
 from neural_compressor.config import PostTrainingQuantConfig
-from ...utils.utils import is_ipex_available
+from ...utils.utils import is_ipex_available, is_autoround_available
 from transformers import AutoTokenizer
 
 if is_ipex_available():
     import intel_extension_for_pytorch as ipex
 
+if is_autoround_available():
+    from auto_round.export.export_to_itrex.model_wrapper import WeightOnlyLinear as auto_round_woqlinear # pylint: disable=E0401
 
 torch = LazyImport("torch")
 
@@ -106,7 +107,7 @@ def _replace_linear(
         is_removed = False
 
         if (isinstance(module, torch.nn.Linear) or isinstance(module, WeightOnlyLinear)
-            or isinstance(module, auto_round_woqlinear) or (is_ipex_available()
+            or (is_autoround_available() and isinstance(module, auto_round_woqlinear)) or (is_ipex_available()
             and isinstance(module, ipex.nn.utils._weight_prepack._IPEXLinear))) \
             and (name not in modules_to_not_convert):
             # Check if the current key is not in the `modules_to_not_convert`

intel_extension_for_transformers/utils/utils.py

@@ -72,3 +72,14 @@ def get_gpu_family():
 
 def is_ipex_available():
     return _ipex_available
+
+_autoround_available = importlib.util.find_spec("auto_round") is not None
+_autoround_version = "N/A"
+if _autoround_available:
+    try:
+        _autoround_version = importlib_metadata.version("auto_round")
+    except importlib_metadata.PackageNotFoundError:
+        _autoround_available = False
+
+def is_autoround_available():
+    return _autoround_available

requirements.txt

@@ -2,7 +2,6 @@
 accelerate
 cmake>=3.16
 gguf
-git+https://github.com/intel/auto-round.git@b65830f3f6cb32d92a5c8ba5f80ace12d517357b
 ninja
 optimum-intel
 py-cpuinfo