[Paddle] Replace paddle.fluid imports with paddle.base

qa/L0_paddle_lint/test.sh

@@ -22,5 +22,5 @@ then
   cp $TE_PATH/qa/L0_paddle_lint/pylintrc $TE_PATH
   cd $TE_PATH
   echo "Checking Python files"
-  pylint --recursive=y transformer_engine/common transformer_engine/paddle
+  python -m pylint --recursive=y transformer_engine/common transformer_engine/paddle
 fi

tests/paddle/dist_launcher.py

@@ -10,7 +10,10 @@
 import time
 import unittest
 
-from paddle import fluid
+try:
+    from paddle.base import core
+except ImportError:
+    from paddle.fluid import core
 from paddle.distributed.utils.launch_utils import (
     TrainerProc,
     find_free_ports,
@@ -114,7 +117,7 @@ def run_2gpu(
         allocator_strategy="auto_growth",
     ):
         """Run target file in subprocesses"""
-        if (not fluid.core.is_compiled_with_cuda() or fluid.core.get_cuda_device_count() == 0):
+        if (not core.is_compiled_with_cuda() or core.get_cuda_device_count() == 0):
             return
 
         selected_gpus = get_gpus('0,1')

transformer_engine/paddle/layer/base.py

@@ -12,8 +12,12 @@
 import numpy as np
 
 import paddle
-from paddle.fluid import core
-from paddle.fluid.framework import _dygraph_tracer
+try:
+    from paddle.base import core
+    from paddle.base.framework import _dygraph_tracer
+except ImportError:
+    from paddle.fluid import core
+    from paddle.fluid.framework import _dygraph_tracer
 
 from ..constants import FP8BwdTensors, dist_group_type
 from ..cpp_extensions import cast_transpose, cast_transpose_bgrad, cast_to_fp8, transpose

transformer_engine/paddle/profile.py

@@ -5,7 +5,10 @@
 
 from contextlib import contextmanager
 
-from paddle.fluid import core
+try:
+    from paddle.base import core
+except ImportError:
+    from paddle.fluid import core
 
 
 @contextmanager