Merge pull request #1421 from cattidea/fallback-to-old-ir-in-paddle3b2

Fallback to old IR in paddlepaddle 3.0.0b2

Author: risemeup1

paddle2onnx/convert.py

@@ -16,6 +16,21 @@
 import paddle
 import paddle2onnx.paddle2onnx_cpp2py_export as c_p2o
 from paddle2onnx.utils import logging, paddle_jit_save_configs
+from contextlib import contextmanager
+
+
+def get_old_ir_guard():
+    # For old version of PaddlePaddle, donothing guard is returned.
+    @contextmanager
+    def dummy_guard():
+        yield
+
+    if not hasattr(paddle, "pir_utils"):
+        return dummy_guard
+    pir_utils = paddle.pir_utils
+    if not hasattr(pir_utils, "DygraphOldIrGuard"):
+        return dummy_guard
+    return pir_utils.DygraphOldIrGuard
 
 def export(model_filename,
            params_filename,
@@ -66,9 +81,13 @@ def dygraph2onnx(layer, save_file, input_spec=None, opset_version=9, **configs):
         if os.path.isfile(params_file):
             os.remove(params_file)
     save_configs = paddle_jit_save_configs(configs)
-    paddle.jit.save(
-        layer, os.path.join(paddle_model_dir, "model"), input_spec, **save_configs
-    )
+    with get_old_ir_guard()():
+        # In PaddlePaddle 3.0.0b2, PIR becomes the default IR, but PIR export still in development.
+        # So we need to use the old IR to export the model, avoid make users confused.
+        # In the future, we will remove this guard and recommend users to use PIR.
+        paddle.jit.save(
+            layer, os.path.join(paddle_model_dir, "model"), input_spec, **save_configs
+        )
     logging.info("Static PaddlePaddle model saved in {}.".format(paddle_model_dir))
     if not os.path.isfile(params_file):
         params_file = ""