Lluo/modelopt import restructure (#3825)

Author: lanluo-nvidia

examples/apps/flux_demo.py

@@ -34,7 +34,7 @@ def compile_model(
         enabled_precisions = {torch.float4_e2m1fn_x2}
         ptq_config = mtq.NVFP4_DEFAULT_CFG
         if args.fp4_mha:
-            from modelopt.core.torch.quantization.config import NVFP4_FP8_MHA_CONFIG
+            from modelopt.torch.quantization.config import NVFP4_FP8_MHA_CONFIG
 
             ptq_config = NVFP4_FP8_MHA_CONFIG
 

py/torch_tensorrt/dynamo/conversion/impl/dynamic_block_quantize.py

@@ -209,7 +209,7 @@ def _static_double_quantize(
             quantized data tensor in fp4
         """
 
-        import modelopt.core.torch.quantization.qtensor.nvfp4_tensor as nvfp4_tensor
+        import modelopt.torch.quantization.qtensor.nvfp4_tensor as nvfp4_tensor
 
         if weights_tensor.dtype == torch.float16:
             original_dtype = trt.DataType.HALF