fix model export issues (#11397)

Summary:

the upstream GPTQ D75899296 break our model export

Reviewed By: larryliu0820

Differential Revision: D76049137

Author: billmguo

examples/models/llama/source_transformation/lora.py

@@ -13,7 +13,7 @@
 
 import torch
 from torch import nn
-from torchao.quantization.GPTQ import Int8DynActInt4WeightLinear
+from torchao.quantization.linear_quant_modules import Int8DynActInt4WeightLinear
 from torchao.quantization.quant_api import _replace_with_custom_fn_if_matches_filter
 
 

examples/models/llama/source_transformation/pre_quantization.py

@@ -13,7 +13,10 @@
 import torch
 from torch import nn
 
-from torchao.quantization.GPTQ import _check_linear_int4_k, Int8DynActInt4WeightLinear
+from torchao.quantization.linear_quant_modules import (
+    _check_linear_int4_k,
+    Int8DynActInt4WeightLinear,
+)
 from torchao.quantization.quant_api import _replace_with_custom_fn_if_matches_filter
 
 from .quantize import Int8DynActInt8WeightLinear, QuantizedGroupEmbedding

examples/models/llama/source_transformation/quantize.py

@@ -847,7 +847,7 @@ def set_8da4w_computation_dtype(
     module: nn.Module, computation_dtype: torch.dtype
 ) -> nn.Module:
 
-    from torchao.quantization.GPTQ import Int8DynActInt4WeightLinear
+    from torchao.quantization.linear_quant_modules import Int8DynActInt4WeightLinear
 
     def _set_8da4w_computation_dtype(module: nn.Module, dtype: torch.dtype) -> None:
         """