【Prim】remove gelu precision threshold (#52350)

* [prim] promote gelu precision threshold

python/paddle/fluid/tests/unittests/prim_op_test.py

@@ -626,13 +626,12 @@ class PrimForwardChecker:
             )
             raise RuntimeError(msg)
         for i in range(len(ret)):
-            if not np.allclose(
+            np.testing.assert_allclose(
                 ret[i],
                 self.eager_desire[i],
                 rtol=self.fw_comp_rtol,
                 atol=self.fw_comp_atol,
-            ):
-                msg = (
+                err_msg=(
                     'Check static comp forward api out failed. Mismatch between static comp '
                     'and eager on %s, when enable_fw_comp is %s,the forward api out tensor\'s index is : %d \n'
                     'static comp forward api out tensor:\n%s\n eager forward api out tensor:\n%s\n'
@@ -643,8 +642,8 @@ class PrimForwardChecker:
                         ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
         paddle.disable_static()
         core._set_prim_forward_enabled(False)
 
@@ -696,10 +695,12 @@ class PrimForwardChecker:
             )
             raise RuntimeError(msg)
         for i in range(len(ret)):
-            if not np.allclose(
-                ret[i], self.eager_desire[i], rtol=rtol, atol=atol
-            ):
-                msg = (
+            np.testing.assert_allclose(
+                ret[i],
+                self.eager_desire[i],
+                rtol=rtol,
+                atol=atol,
+                err_msg=(
                     'Check jit comp forward api out failed. Mismatch between jit comp '
                     'and eager on %s, when enable_fw_comp is %s,the forward api out tensor\'s index is : %d \n'
                     'jit comp forward api out tensor:\n%s\n eager forward api out tensor:\n%s\n'
@@ -710,8 +711,8 @@ class PrimForwardChecker:
                         ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
         core._set_prim_forward_enabled(False)
         net.forward.program_cache.clear()
 
@@ -781,10 +782,12 @@ class PrimForwardChecker:
             )
             raise RuntimeError(msg)
         for i in range(len(ret)):
-            if not np.allclose(
-                ret[i], self.eager_desire[i], rtol=rtol, atol=atol
-            ):
-                msg = (
+            np.testing.assert_allclose(
+                ret[i],
+                self.eager_desire[i],
+                rtol=rtol,
+                atol=atol,
+                err_msg=(
                     'Check jit comp with cinn forward api out failed. Mismatch between jit comp and eager on %s, '
                     'when enable_fw_comp is %s, enable_cinn is %s, the forward api out tensor\'s index is : %d \n'
                     'jit comp forward api out tensor:\n%s\n eager forward api out tensor:\n%s\n'
@@ -796,8 +799,8 @@ class PrimForwardChecker:
                         ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
         core._set_prim_forward_enabled(False)
         net.forward.program_cache.clear()
 
@@ -975,13 +978,12 @@ class PrimGradChecker(PrimForwardChecker):
             )
             raise RuntimeError(msg)
         for i in range(len(actual_ret)):
-            if not np.allclose(
+            np.testing.assert_allclose(
                 actual_ret[i],
                 self.eager_desire[i],
                 rtol=atol,
                 atol=rtol,
-            ):
-                msg = (
+                err_msg=(
                     'Check eager comp grad out failed. Mismatch between eager comp '
                     'and eager on %s, when enable_rev_comp is %s,the eager comp grad out tensor\'s index is : %d \n'
                     'eager comp grad out tensor:\n%s\n eager grad out tensor:\n%s\n'
@@ -992,8 +994,8 @@ class PrimGradChecker(PrimForwardChecker):
                         actual_ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
         core.set_prim_eager_enabled(False)
 
     def check_static_comp(self):
@@ -1075,10 +1077,12 @@ class PrimGradChecker(PrimForwardChecker):
             )
             raise RuntimeError(msg)
         for i in range(len(actual_ret)):
-            if not np.allclose(
-                actual_ret[i], self.eager_desire[i], rtol=rtol, atol=atol
-            ):
-                msg = (
+            np.testing.assert_allclose(
+                actual_ret[i],
+                self.eager_desire[i],
+                rtol=rtol,
+                atol=atol,
+                err_msg=(
                     'Check static comp grad out failed. Mismatch between static comp '
                     'and eager on %s, when enable_fw_comp is %s,enable_rev_comp is %s,the forward api out tensor\'s index is : %d \n'
                     'static comp grad out tensor:\n%s\n eager grad out tensor:\n%s\n'
@@ -1090,8 +1094,8 @@ class PrimGradChecker(PrimForwardChecker):
                         actual_ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
         core._set_prim_forward_enabled(False)
         core._set_prim_backward_enabled(False)
         paddle.disable_static()
@@ -1179,10 +1183,12 @@ class PrimGradChecker(PrimForwardChecker):
             )
             raise RuntimeError(msg)
         for i in range(len(ret)):
-            if not np.allclose(
-                ret[i], self.eager_desire[i], rtol=rtol, atol=atol
-            ):
-                msg = (
+            np.testing.assert_allclose(
+                ret[i],
+                self.eager_desire[i],
+                rtol=rtol,
+                atol=atol,
+                err_msg=(
                     'Check jit comp grad out failed. Mismatch between jit comp '
                     'and eager on %s, when enable_fw_comp is %s, enable_rev_comp is %s,the grad out tensor\'s index is : %d \n'
                     'jit comp grad out tensor:\n%s\n eager grad out out tensor:\n%s\n'
@@ -1194,8 +1200,8 @@ class PrimGradChecker(PrimForwardChecker):
                         ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
         core._set_prim_forward_enabled(False)
         core._set_prim_backward_enabled(False)
         net.forward.program_cache.clear()
@@ -1297,10 +1303,12 @@ class PrimGradChecker(PrimForwardChecker):
             )
             raise RuntimeError(msg)
         for i in range(len(ret)):
-            if not np.allclose(
-                ret[i], self.eager_desire[i], rtol=rtol, atol=atol
-            ):
-                msg = (
+            np.testing.assert_allclose(
+                ret[i],
+                self.eager_desire[i],
+                rtol=rtol,
+                atol=atol,
+                err_msg=(
                     'Check jit comp with cinn grad out failed. Mismatch between jit comp with cinn '
                     'and eager on %s, when enable_fw_comp is %s, enable_rev_comp is %s, enable_cinn is %s,'
                     'the grad out tensor\'s index is : %d ,jit comp with cinn grad out tensor:\n%s\n eager grad out out tensor:\n%s\n'
@@ -1313,8 +1321,9 @@ class PrimGradChecker(PrimForwardChecker):
                         ret[i],
                         self.eager_desire[i],
                     )
-                )
-                raise RuntimeError(msg)
+                ),
+            )
+
         core._set_prim_forward_enabled(False)
         core._set_prim_backward_enabled(False)
         net.forward.program_cache.clear()

python/paddle/fluid/tests/unittests/test_activation_op.py

@@ -2038,9 +2038,10 @@ class TestGeluApproximate(TestActivation):
         self.outputs = {'Out': out}
         self.attrs = {"approximate": approximate}
 
-        # The backward decomposite of gelu is inconsistent with raw kernel,
-        # lower threshold to support 1e-5 for pass the unittest
-        self.rev_comp_rtol = 1e-5
+        # The backward decomposite of gelu is inconsistent with raw kernel on
+        # cpu device, lower threshold to support 1e-8 for pass the unittest
+        self.rev_comp_rtol = 1e-8
+        self.rev_comp_atol = 1e-8
 
     def test_check_output(self):
         self.check_output(check_prim=True)
@@ -2068,9 +2069,10 @@ class TestGelu(TestActivation):
         self.inputs = {'X': x}
         self.outputs = {'Out': out}
         self.attrs = {"approximate": approximate}
-        # The backward decomposite of gelu is inconsistent with raw kernel,
-        # lower threshold to support 1e-5 for pass the unittest
-        self.rev_comp_rtol = 1e-5
+        # The backward decomposite of gelu is inconsistent with raw kernel on
+        # cpu, lower threshold to support 1e-8 for pass the unittest
+        self.rev_comp_rtol = 1e-8
+        self.rev_comp_atol = 1e-8
 
     def if_enable_cinn(self):
         self.enable_cinn = False
@@ -2104,9 +2106,10 @@ class TestGELUAPI(unittest.TestCase):
         )
         self.enable_cinn = False
 
-        # The backward decomposite of gelu is inconsistent with raw kernel,
-        # lower threshold to support 1e-5 for pass the unittest
-        self.rev_comp_rtol = 1e-5
+        # The backward decomposite of gelu is inconsistent with raw kernel on
+        # cpu, lower threshold to support 1e-8 for pass the unittest
+        self.rev_comp_rtol = 1e-8
+        self.rev_comp_atol = 1e-8
 
     def test_static_api(self):
         with paddle_static_guard():
@@ -3870,11 +3873,17 @@ def create_test_act_fp16_class(
     check_prim=False,
     enable_cinn=True,
     grad_atol=0.80,
+    **kwargs
 ):
     @unittest.skipIf(
         not paddle.is_compiled_with_cuda(), "core is not compiled with CUDA"
     )
     class TestActFp16(parent):
+        def setUp(self):
+            super().setUp()
+            for k, v in kwargs.items():
+                setattr(self, k, v)
+
         def init_dtype(self):
             self.dtype = np.float16
 
@@ -3937,7 +3946,13 @@ create_test_act_fp16_class(TestAsinh, grad_atol=0.85)
 create_test_act_fp16_class(TestAtanh, grad_atol=0.85)
 create_test_act_fp16_class(TestRound, grad_check=False)
 create_test_act_fp16_class(TestRelu, check_prim=True)
-create_test_act_fp16_class(TestGelu, check_prim=True, enable_cinn=False)
+create_test_act_fp16_class(
+    TestGelu,
+    check_prim=True,
+    enable_cinn=False,
+    rev_comp_rtol=1e-3,
+    rev_comp_atol=1e-3,
+)
 create_test_act_fp16_class(TestBRelu)
 create_test_act_fp16_class(TestRelu6)
 create_test_act_fp16_class(TestSoftRelu, check_dygraph=False, grad_atol=0.85)