refine per review comments

Signed-off-by: xin3he <xin3.he@intel.com>

Author: xin3he

neural_compressor/torch/quantization/__init__.py

@@ -25,7 +25,7 @@
 )
 
 # TODO(Yi): move config to config.py
-from .autotune import autotune, TuningConfig, get_default_tune_config
+from neural_compressor.torch.quantization.autotune import autotune, TuningConfig, get_default_tune_config
 
 ### Quantization Function Registration ###
 import neural_compressor.torch.quantization.weight_only

neural_compressor/torch/quantization/config.py

@@ -98,6 +98,7 @@ def __init__(
             use_mse_search (bool): Enables mean squared error (MSE) search, default is False.
             use_layer_wise (bool): Enables quantize model per layer. Defaults to False.
             export_compressed_model (bool): Enables return model in int format or not. Defaults to False.
+            use_double_quant (bool): Enables double quantization, default is False.
             double_quant_dtype (str): Data type for double_quant scale, default is "int".
             double_quant_bits (int): Number of bits used to represent double_quant scale, default is 4.
             double_quant_use_sym (bool): Indicates whether double_quant scale are symmetric, default is True.

test/3x/torch/quantization/weight_only/test_rtn.py

@@ -102,7 +102,7 @@ def test_mse_search(self):
                 atol_false > atol_true
             ), "use_mse_search=True doesn't help accuracy, maybe is reasonable, please double check."
         except:
-            pass  # atol is very close, double checked the logic.
+            assert torch.allclose(atol_false, atol_true, atol=0.012), "atol is very close, double checked the logic."
 
     def test_layer_wise(self):
         model = copy.deepcopy(self.tiny_gptj)