diff --git a/test/ir/inference/auto_scan_test.py b/test/ir/inference/auto_scan_test.py index 9f21b94027a9b7b75f8692c616494bb7bf25759c..225318b5f8641d6e48f47e5e809cce8e21e729a5 100755 --- a/test/ir/inference/auto_scan_test.py +++ b/test/ir/inference/auto_scan_test.py @@ -14,7 +14,6 @@ import abc import enum -import logging import os import shutil import time @@ -35,9 +34,12 @@ from program_config import ( import paddle import paddle.inference as paddle_infer from paddle.fluid.core import PassVersionChecker +from paddle.static.log_helper import get_logger LOGLEVEL = os.environ.get("PADDLE_TEST_LOGLEVEL", "INFO").upper() -logging.basicConfig(level=LOGLEVEL, format="%(message)s") +logging = get_logger( + __name__, LOGLEVEL, fmt='%(asctime)s-%(levelname)s: %(message)s' +) settings.register_profile( "ci", @@ -662,9 +664,6 @@ class TrtLayerAutoScanTest(AutoScanTest): ) return config - def get_avalible_input_type(self) -> List[np.dtype]: - return [np.float32] - def assert_tensors_near( self, atol: float, @@ -747,10 +746,13 @@ class TrtLayerAutoScanTest(AutoScanTest): if not skip_baseline: # baseline: gpu run, we only test float32 gpu_config = self.create_inference_config(use_trt=False) + prog_config = prog_config.set_input_type( + np.float16 + ).set_input_type(np.float32) baseline_result = self.run_test_config( model, params, - prog_config.set_input_type(np.float32), + prog_config, gpu_config, feed_data, ) @@ -761,82 +763,78 @@ class TrtLayerAutoScanTest(AutoScanTest): nodes_num, threshold, ) in self.sample_predictor_configs(prog_config): - for input_type in self.get_avalible_input_type(): - prog_config = prog_config.set_input_type(input_type) - if os.path.exists(self.cache_dir): - shutil.rmtree(self.cache_dir) + if os.path.exists(self.cache_dir): + shutil.rmtree(self.cache_dir) - if isinstance(threshold, float): - atol = threshold - rtol = 1e-8 - elif isinstance(threshold, list) or isinstance( - threshold, tuple - ): - atol = threshold[0] - rtol = threshold[1] - else: - raise NotImplementedError - - is_fp8 = ( - pred_config.tensorrt_precision_mode() - == paddle_infer.PrecisionType.Int8 - ) - if (not is_fp8 and quant) or (is_fp8 and not quant): - continue - - ignore_flag = False - for teller, reason, note in self.ignore_cases: - if teller(prog_config, pred_config): - ignore_flag = True - if reason == IgnoreReasons.TRT_NOT_IMPLEMENTED: - self.ignore_log( - f"[TRT_NOT_IMPLEMENTED] {note} vs {self.inference_config_str(pred_config)}" - ) - elif reason == IgnoreReasons.TRT_NOT_SUPPORT: - self.ignore_log( - f"[TRT_NOT_SUPPORT] {note} vs {self.inference_config_str(pred_config)}" - ) - else: - raise NotImplementedError - break - - if ignore_flag: - continue - - try: - pred_config_deserialize = paddle_infer.Config( - pred_config - ) - trt_result = self.run_test_config( - model, params, prog_config, pred_config, feed_data - ) - self.assert_tensors_near( - atol, rtol, trt_result, baseline_result - ) - trt_engine_num, paddle_op_num = nodes_num - self.assert_op_size(trt_engine_num, paddle_op_num) - - # deserialize test - if trt_engine_num > 0: - self.run_test_config( - model, - params, - prog_config, - pred_config_deserialize, - feed_data, + if isinstance(threshold, float): + atol = threshold + rtol = 1e-8 + elif isinstance(threshold, list) or isinstance( + threshold, tuple + ): + atol = threshold[0] + rtol = threshold[1] + else: + raise NotImplementedError + + is_fp8 = ( + pred_config.tensorrt_precision_mode() + == paddle_infer.PrecisionType.Int8 + ) + if (not is_fp8 and quant) or (is_fp8 and not quant): + continue + + ignore_flag = False + for teller, reason, note in self.ignore_cases: + if teller(prog_config, pred_config): + ignore_flag = True + if reason == IgnoreReasons.TRT_NOT_IMPLEMENTED: + self.ignore_log( + f"[TRT_NOT_IMPLEMENTED] {note} vs {self.inference_config_str(pred_config)}" ) + elif reason == IgnoreReasons.TRT_NOT_SUPPORT: + self.ignore_log( + f"[TRT_NOT_SUPPORT] {note} vs {self.inference_config_str(pred_config)}" + ) + else: + raise NotImplementedError + break - self.success_log(f"program_config: {prog_config}") - self.success_log( - f"predictor_config: {self.inference_config_str(pred_config)}" - ) - except Exception as e: - self.fail_log(f"program_config: {prog_config}") - self.fail_log( - f"predictor_config: {self.inference_config_str(pred_config)}" + if ignore_flag: + continue + + try: + pred_config_deserialize = paddle_infer.Config(pred_config) + trt_result = self.run_test_config( + model, params, prog_config, pred_config, feed_data + ) + self.assert_tensors_near( + atol, rtol, trt_result, baseline_result + ) + trt_engine_num, paddle_op_num = nodes_num + self.assert_op_size(trt_engine_num, paddle_op_num) + + # deserialize test + if trt_engine_num > 0: + self.run_test_config( + model, + params, + prog_config, + pred_config_deserialize, + feed_data, ) - self.fail_log(f"\033[1;31m ERROR INFO: {e}\033[0m") - all_passes = False + + self.success_log(f"program_config: {prog_config}") + self.success_log( + f"predictor_config: {self.inference_config_str(pred_config)}" + ) + except Exception as e: + self.fail_log(f"program_config: {prog_config}") + self.fail_log( + f"predictor_config: {self.inference_config_str(pred_config)}" + ) + self.fail_log(f"\033[1;31m ERROR INFO: {e}\033[0m") + all_passes = False self.assertTrue(all_passes) diff --git a/test/ir/inference/program_config.py b/test/ir/inference/program_config.py index 3dfd9b3f95877ec5bfeaae00f78c93861fc4c440..601ddfdc2383359b25bba6a1910dd0cde2d54853 100644 --- a/test/ir/inference/program_config.py +++ b/test/ir/inference/program_config.py @@ -67,7 +67,7 @@ class TensorConfig: def __repr__(self): return str({'shape': self.shape, 'lod': self.lod, 'dtype': self.dtype}) - def astype(self, type: np.dtype): + def convert_type_inplace(self, type: np.dtype): self.data = self.data.astype(type) self.dtype = self.data.dtype return self @@ -277,9 +277,9 @@ class ProgramConfig: def set_input_type(self, type: np.dtype): for inp in self.inputs.values(): - inp.astype(type) + inp.convert_type_inplace(type) for weight in self.weights.values(): - weight.astype(type) + weight.convert_type_inplace(type) return self def get_input_type(self) -> np.dtype: diff --git a/test/ir/inference/test_trt_convert_activation.py b/test/ir/inference/test_trt_convert_activation.py index cec7e624b08d820f806938311a939807f8acb1fb..85f2729f3511243d0d1133dfca82992dc56d6829 100644 --- a/test/ir/inference/test_trt_convert_activation.py +++ b/test/ir/inference/test_trt_convert_activation.py @@ -178,10 +178,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -189,10 +191,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_affine_channel.py b/test/ir/inference/test_trt_convert_affine_channel.py index b326f14ed2f376b1b3c00a45121145094bc12b93..b30a07fc1b5c1377a281ed8d9140153ce1aa81fe 100644 --- a/test/ir/inference/test_trt_convert_affine_channel.py +++ b/test/ir/inference/test_trt_convert_affine_channel.py @@ -133,10 +133,12 @@ class TrtConvertAffineChannelTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -144,10 +146,12 @@ class TrtConvertAffineChannelTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_anchor_generator.py b/test/ir/inference/test_trt_convert_anchor_generator.py index 1e7c380b40fa5d63db42863548f769acd33fdea4..fd91a9945b69430c3beca55a7fbf3d331533263d 100644 --- a/test/ir/inference/test_trt_convert_anchor_generator.py +++ b/test/ir/inference/test_trt_convert_anchor_generator.py @@ -107,10 +107,12 @@ class TrtConvertAnchorGeneratorTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -118,10 +120,12 @@ class TrtConvertAnchorGeneratorTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_arg_max.py b/test/ir/inference/test_trt_convert_arg_max.py index 081a58eadf4ca722e38043b033418447b831a367..f7e8185d3adc3b1be0e8452d4cec2443712c786b 100644 --- a/test/ir/inference/test_trt_convert_arg_max.py +++ b/test/ir/inference/test_trt_convert_arg_max.py @@ -118,10 +118,12 @@ class TrtConvertArgMaxTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -129,10 +131,12 @@ class TrtConvertArgMaxTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_arg_min.py b/test/ir/inference/test_trt_convert_arg_min.py index da448142d4f4478eebe3972782ae8b7cd2eba722..c1e330c1c2bada4abd046012f9f898b9a7fda9bc 100644 --- a/test/ir/inference/test_trt_convert_arg_min.py +++ b/test/ir/inference/test_trt_convert_arg_min.py @@ -118,10 +118,12 @@ class TrtConvertArgMinTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -129,10 +131,12 @@ class TrtConvertArgMinTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_assign.py b/test/ir/inference/test_trt_convert_assign.py index 64dbf16064e94fb5eb8b7f04b39585986f332bd4..55939982d5ee0d5fbe366301cdf0f10b61375e64 100644 --- a/test/ir/inference/test_trt_convert_assign.py +++ b/test/ir/inference/test_trt_convert_assign.py @@ -133,10 +133,12 @@ class TrtConvertAssignTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -144,10 +146,12 @@ class TrtConvertAssignTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_batch_norm.py b/test/ir/inference/test_trt_convert_batch_norm.py index 277eb132602807a9bd97b748be8ace5b43da9d5d..2cd98ad69980024fb7dbae40021652c72faf6304 100644 --- a/test/ir/inference/test_trt_convert_batch_norm.py +++ b/test/ir/inference/test_trt_convert_batch_norm.py @@ -232,10 +232,12 @@ class TrtConvertBatchNormTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -243,10 +245,12 @@ class TrtConvertBatchNormTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_bilinear_interp_v2.py b/test/ir/inference/test_trt_convert_bilinear_interp_v2.py index 148776365c0911113ced7b1486827285f7c41ee3..5fb36deec1ab2dd6fac607fe785d10e41da08647 100644 --- a/test/ir/inference/test_trt_convert_bilinear_interp_v2.py +++ b/test/ir/inference/test_trt_convert_bilinear_interp_v2.py @@ -120,10 +120,12 @@ class TrtConvertBilinearInterpV2Test(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -131,10 +133,12 @@ class TrtConvertBilinearInterpV2Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_bitwise_not.py b/test/ir/inference/test_trt_convert_bitwise_not.py index ab8da867362747970acf6e14d4f2861ad961050e..738ef5af23a247d596150ed10e16714972bafca9 100644 --- a/test/ir/inference/test_trt_convert_bitwise_not.py +++ b/test/ir/inference/test_trt_convert_bitwise_not.py @@ -133,10 +133,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -144,10 +146,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_bmm.py b/test/ir/inference/test_trt_convert_bmm.py index ebd25646a18fc939419cd36fd2a3c59082b053b1..0bb85f9abf9b35dcbaf2fd03b1191c341c3983ce 100644 --- a/test/ir/inference/test_trt_convert_bmm.py +++ b/test/ir/inference/test_trt_convert_bmm.py @@ -98,10 +98,12 @@ class TrtConvertBmmTest_dynamic(TrtLayerAutoScanTest): clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -115,10 +117,12 @@ class TrtConvertBmmTest_dynamic(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), tol_fp32 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), tol_half diff --git a/test/ir/inference/test_trt_convert_cast.py b/test/ir/inference/test_trt_convert_cast.py index 026abc571050a2d088026f60c2979b5ffb865ca9..da181e7922556089ab7d6a11f5a0a176a6f70507 100644 --- a/test/ir/inference/test_trt_convert_cast.py +++ b/test/ir/inference/test_trt_convert_cast.py @@ -164,10 +164,12 @@ class TrtConvertCastTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -175,10 +177,12 @@ class TrtConvertCastTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_clip.py b/test/ir/inference/test_trt_convert_clip.py index c6fd638c2f97a015b7a2f2d4add6cb736aa74ec1..dc46c2a775d0cfc70e331302ea020bb1a43c3530 100644 --- a/test/ir/inference/test_trt_convert_clip.py +++ b/test/ir/inference/test_trt_convert_clip.py @@ -143,10 +143,12 @@ class TrtConvertClipTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -154,10 +156,12 @@ class TrtConvertClipTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_compare_and_logical.py b/test/ir/inference/test_trt_convert_compare_and_logical.py index 79d157e9fe7db64bfa1f784bd5ebe046ba73f629..fde4b2b3587308728dce0622f33137954c8f419d 100755 --- a/test/ir/inference/test_trt_convert_compare_and_logical.py +++ b/test/ir/inference/test_trt_convert_compare_and_logical.py @@ -153,10 +153,12 @@ class TrtConvertLogicalTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -164,10 +166,12 @@ class TrtConvertLogicalTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -295,10 +299,12 @@ class TrtConvertCompareTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -306,10 +312,12 @@ class TrtConvertCompareTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -454,10 +462,12 @@ class TrtConvertLessEqualTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -465,10 +475,12 @@ class TrtConvertLessEqualTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -613,10 +625,12 @@ class TrtConvertGreaterEqualTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -624,10 +638,12 @@ class TrtConvertGreaterEqualTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -750,10 +766,12 @@ class TrtConvertCompareSkipTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -761,10 +779,12 @@ class TrtConvertCompareSkipTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_concat.py b/test/ir/inference/test_trt_convert_concat.py index afe1851759e2c1c1baa40a78ae6a4531b32ef7a5..fcf68b7481615d2a97fb40bd791aa58ea5b9d0a7 100644 --- a/test/ir/inference/test_trt_convert_concat.py +++ b/test/ir/inference/test_trt_convert_concat.py @@ -320,10 +320,12 @@ class TrtConvertConcatTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -331,10 +333,12 @@ class TrtConvertConcatTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_conv2d.py b/test/ir/inference/test_trt_convert_conv2d.py index 432ad97f76ad29d22d597cfcc395bb716d932421..8d3ea2e28ea2f7061e3e5aa9437671a0223dbe8c 100644 --- a/test/ir/inference/test_trt_convert_conv2d.py +++ b/test/ir/inference/test_trt_convert_conv2d.py @@ -171,14 +171,17 @@ class TrtConvertConv2dTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-2, 1e-2) @@ -186,14 +189,17 @@ class TrtConvertConv2dTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-2, 1e-2) @@ -356,11 +362,13 @@ class TrtConvertConv2dNotPersistableTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-2, 1e-2) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-2, 1e-2) diff --git a/test/ir/inference/test_trt_convert_conv2d_fusion.py b/test/ir/inference/test_trt_convert_conv2d_fusion.py index dd6eb895b98756796ebc7a8498440e195284ae71..00d6a8678a9ddc9477e6a567687306d2185b4d33 100644 --- a/test/ir/inference/test_trt_convert_conv2d_fusion.py +++ b/test/ir/inference/test_trt_convert_conv2d_fusion.py @@ -179,14 +179,17 @@ class TrtConvertConv2dFusionTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -194,14 +197,17 @@ class TrtConvertConv2dFusionTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_conv2d_transpose.py b/test/ir/inference/test_trt_convert_conv2d_transpose.py index 42057072f70fc2aefecb953b517fafb68c575359..57c973b4667329c702169ca6af09e004ddcead02 100644 --- a/test/ir/inference/test_trt_convert_conv2d_transpose.py +++ b/test/ir/inference/test_trt_convert_conv2d_transpose.py @@ -194,10 +194,12 @@ class TrtConvertConv2dTransposeTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -208,10 +210,12 @@ class TrtConvertConv2dTransposeTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -338,10 +342,12 @@ class TrtConvertConv2dTransposeTest2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-4 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e0, 1e-3) @@ -349,10 +355,12 @@ class TrtConvertConv2dTransposeTest2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-4 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e0, 1e-3) diff --git a/test/ir/inference/test_trt_convert_conv3d_transpose.py b/test/ir/inference/test_trt_convert_conv3d_transpose.py index 4134460d7d38d6e020fa3a9e67a8f23b6a4a0bda..ba545275d805b4e11f34c1ee35855f6b6043f9cb 100644 --- a/test/ir/inference/test_trt_convert_conv3d_transpose.py +++ b/test/ir/inference/test_trt_convert_conv3d_transpose.py @@ -118,6 +118,7 @@ class TrtConvertConv3dTransposeTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -125,6 +126,7 @@ class TrtConvertConv3dTransposeTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_cross_multihead_matmul.py b/test/ir/inference/test_trt_convert_cross_multihead_matmul.py index a8ce9ee0a59e9015906ef9fde208db61c87134db..d50875e9a93a835fa559bec09a82646245423b7c 100644 --- a/test/ir/inference/test_trt_convert_cross_multihead_matmul.py +++ b/test/ir/inference/test_trt_convert_cross_multihead_matmul.py @@ -270,17 +270,21 @@ class TrtConvertCrossMultiHeadMatmulTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 4), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 4), (1e-2, 1e-3) # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 3), (1e-5, 1e-4) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-2, 1e-2) def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_cumsum.py b/test/ir/inference/test_trt_convert_cumsum.py index 50c73afcc0647fdd758973f7608ee02ec0e03ddd..8f58f6b2ada60f1751d00cbc1ccc3c1490d91e18 100644 --- a/test/ir/inference/test_trt_convert_cumsum.py +++ b/test/ir/inference/test_trt_convert_cumsum.py @@ -175,10 +175,12 @@ class TrtConvertCumsum(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_deformable_conv.py b/test/ir/inference/test_trt_convert_deformable_conv.py index ff32472a9d0dad17304f0fb3efebf142d3b17672..55682a287c286778047c36e8073b20ddd8216dd5 100644 --- a/test/ir/inference/test_trt_convert_deformable_conv.py +++ b/test/ir/inference/test_trt_convert_deformable_conv.py @@ -219,11 +219,13 @@ class TrtConvertDeformableConvTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) diff --git a/test/ir/inference/test_trt_convert_depthwise_conv2d.py b/test/ir/inference/test_trt_convert_depthwise_conv2d.py index bdad374e5a7a09e7dad89041d08d868bce3f6d9a..066a517d3f1c1ea4be8265dcb681a9fce7926b31 100644 --- a/test/ir/inference/test_trt_convert_depthwise_conv2d.py +++ b/test/ir/inference/test_trt_convert_depthwise_conv2d.py @@ -151,13 +151,16 @@ class TrtConvertDepthwiseConv2dTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-3, 1e-3, ) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-3, 1e-3, @@ -166,13 +169,16 @@ class TrtConvertDepthwiseConv2dTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-3, 1e-3, ) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-3, 1e-3, diff --git a/test/ir/inference/test_trt_convert_depthwise_conv2d_transpose.py b/test/ir/inference/test_trt_convert_depthwise_conv2d_transpose.py index d346e001d09e315e1bbd15864f6ab03f49fe10ff..ce97776d292053647ed2e5aea7e0da467529110e 100644 --- a/test/ir/inference/test_trt_convert_depthwise_conv2d_transpose.py +++ b/test/ir/inference/test_trt_convert_depthwise_conv2d_transpose.py @@ -154,10 +154,12 @@ class TrtConvertDepthwiseConv2dTransposeTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -168,10 +170,12 @@ class TrtConvertDepthwiseConv2dTransposeTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_dropout.py b/test/ir/inference/test_trt_convert_dropout.py index 6d2509cefd706b9c76053474d92bb2e01a46367d..7fcae9e6cfaab122e3e256cecbf1715a154452b0 100644 --- a/test/ir/inference/test_trt_convert_dropout.py +++ b/test/ir/inference/test_trt_convert_dropout.py @@ -137,10 +137,12 @@ class TrtConvertDropoutTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -148,10 +150,12 @@ class TrtConvertDropoutTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_einsum.py b/test/ir/inference/test_trt_convert_einsum.py index 6f1fb5ebdd4bd987c6808651312223fa37a43aba..593d126aa7edda792c57b6cd12666ab84c93d706 100644 --- a/test/ir/inference/test_trt_convert_einsum.py +++ b/test/ir/inference/test_trt_convert_einsum.py @@ -146,10 +146,12 @@ class TrtConvertEinsumTest_SingleOperand(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -157,10 +159,12 @@ class TrtConvertEinsumTest_SingleOperand(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 @@ -326,10 +330,12 @@ class TrtConvertEinsumTest_DoubuleOperand_Vector_Matrix(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -337,10 +343,12 @@ class TrtConvertEinsumTest_DoubuleOperand_Vector_Matrix(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 @@ -456,10 +464,12 @@ class TrtConvertEinsumTest_DoubuleOperand_Matrix_Matrix(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -467,10 +477,12 @@ class TrtConvertEinsumTest_DoubuleOperand_Matrix_Matrix(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_elementwise.py b/test/ir/inference/test_trt_convert_elementwise.py index 0ac4a2ba4620915d59e591d310f4c590021b61ce..b5ca0d05d47c89d0dc505d838984abbf64c1c572 100644 --- a/test/ir/inference/test_trt_convert_elementwise.py +++ b/test/ir/inference/test_trt_convert_elementwise.py @@ -139,10 +139,12 @@ class TrtConvertElementwiseTestOneInputSpecialCase0(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -150,10 +152,12 @@ class TrtConvertElementwiseTestOneInputSpecialCase0(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -270,10 +274,12 @@ class TrtConvertElementwiseTestOneInputSpecialCase1(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -281,10 +287,12 @@ class TrtConvertElementwiseTestOneInputSpecialCase1(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -431,10 +439,12 @@ class TrtConvertElementwiseTestOneInput(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -442,10 +452,12 @@ class TrtConvertElementwiseTestOneInput(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -599,10 +611,12 @@ class TrtConvertElementwiseTestTwoInputWithoutBroadcast(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -610,8 +624,10 @@ class TrtConvertElementwiseTestTwoInputWithoutBroadcast(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-3, 1e-3) def add_skip_trt_case(self): @@ -770,15 +786,19 @@ class TrtConvertElementwiseTestTwoInputWithBroadcast(TrtLayerAutoScanTest): clear_dynamic_shape() if self.shape1[0] == self.shape2[0]: self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-3, 1e-3) # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-3, 1e-3) def add_skip_trt_case(self): @@ -921,15 +941,19 @@ class TrtConvertElementwiseTestOneInputCornerCase(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 3), (1e-3, 1e-3) # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), (1e-3, 1e-3) def add_skip_trt_case(self): @@ -1069,10 +1093,12 @@ class TrtConvertElementwiseTestTwoInputSkipCase(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -1080,8 +1106,10 @@ class TrtConvertElementwiseTestTwoInputSkipCase(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), (1e-3, 1e-3) def add_skip_trt_case(self): @@ -1187,10 +1215,12 @@ class TrtConvertPowOp(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -1198,10 +1228,12 @@ class TrtConvertPowOp(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -1347,10 +1379,12 @@ class TrtConvertElementwise0D(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -1358,10 +1392,12 @@ class TrtConvertElementwise0D(TrtLayerAutoScanTest): # # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_elementwiseadd_transpose.py b/test/ir/inference/test_trt_convert_elementwiseadd_transpose.py index 83d5855464f187611114d57ef1baf4918c9cadda..dd9c495d4977231d0fc384e5cb8ce051a981127c 100644 --- a/test/ir/inference/test_trt_convert_elementwiseadd_transpose.py +++ b/test/ir/inference/test_trt_convert_elementwiseadd_transpose.py @@ -162,10 +162,12 @@ class TrtConvertElementwiseaddTransposeTest(TrtLayerAutoScanTest): # just support dynamic_shape generate_dynamic_shape(attrs, inputs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), ( diff --git a/test/ir/inference/test_trt_convert_emb_eltwise_layernorm.py b/test/ir/inference/test_trt_convert_emb_eltwise_layernorm.py index cfaf3b9843ed15bff2e01b4f2ead2251e0fd1eae..5e619a1f5b8fb81bca8922abcb083de085df003d 100644 --- a/test/ir/inference/test_trt_convert_emb_eltwise_layernorm.py +++ b/test/ir/inference/test_trt_convert_emb_eltwise_layernorm.py @@ -293,15 +293,19 @@ class TrtConvertEmbEltwiseLayernormTest1(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 5), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 5), 2e-2 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 4), 2e-2 def test(self): diff --git a/test/ir/inference/test_trt_convert_equal.py b/test/ir/inference/test_trt_convert_equal.py index 5879a003d9546a91d6a8f491d29f56356b39216a..72e8a1574e72ef3fed70733720096f23c20f9d7b 100644 --- a/test/ir/inference/test_trt_convert_equal.py +++ b/test/ir/inference/test_trt_convert_equal.py @@ -157,10 +157,12 @@ class TrtConvertEqualOneInputCornerCase(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -168,10 +170,12 @@ class TrtConvertEqualOneInputCornerCase(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_expand_as_v2.py b/test/ir/inference/test_trt_convert_expand_as_v2.py index be5458cac07dcc4afe063a7df6e7fac1c680ba32..05b3cade8b4291639e121de7168711cce7d2164c 100644 --- a/test/ir/inference/test_trt_convert_expand_as_v2.py +++ b/test/ir/inference/test_trt_convert_expand_as_v2.py @@ -154,10 +154,12 @@ class TrtConvertExpandASV2Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -244,9 +246,11 @@ class TrtConvertExpandV2Test2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) # fill_constant will be folded by constnt folding pass! yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_expand_v2.py b/test/ir/inference/test_trt_convert_expand_v2.py index 49330b1fffc2baf8b80b4c5588bbec50d452faf0..5b6cc1c885b7776f80001aae3d645e9ad36c6ff5 100644 --- a/test/ir/inference/test_trt_convert_expand_v2.py +++ b/test/ir/inference/test_trt_convert_expand_v2.py @@ -147,10 +147,12 @@ class TrtConvertExpandV2Test(TrtLayerAutoScanTest): clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -158,10 +160,12 @@ class TrtConvertExpandV2Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -248,9 +252,11 @@ class TrtConvertExpandV2Test2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) # fill_constant will be folded by constnt folding pass! yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def add_skip_trt_case(self): @@ -392,9 +398,11 @@ class TrtConvertExpandV2Test3(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) # fill_constant will be folded by constnt folding pass! yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_fill_any_like.py b/test/ir/inference/test_trt_convert_fill_any_like.py index 33871a58676b14a08ee677dc6434e99c95f276ce..1c3dddf2c36e1c062f933e2bf07fb365a4754474 100644 --- a/test/ir/inference/test_trt_convert_fill_any_like.py +++ b/test/ir/inference/test_trt_convert_fill_any_like.py @@ -167,10 +167,12 @@ class TrtConvertExpandV2Test(TrtLayerAutoScanTest): clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -178,10 +180,12 @@ class TrtConvertExpandV2Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_fill_constant.py b/test/ir/inference/test_trt_convert_fill_constant.py index 767f460d4caae603f08d319c832578a6a050b0ed..73d15d26ab82f90181d141c833cd95b14da8c640 100644 --- a/test/ir/inference/test_trt_convert_fill_constant.py +++ b/test/ir/inference/test_trt_convert_fill_constant.py @@ -142,10 +142,12 @@ class TrtConvertFillConstantTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_flash_multihead_matmul.py b/test/ir/inference/test_trt_convert_flash_multihead_matmul.py index ebede2dce89f0e8c8a382e7a4be66a5481438caf..4d06e4764f2c004d7928eaecdf167931010f433e 100644 --- a/test/ir/inference/test_trt_convert_flash_multihead_matmul.py +++ b/test/ir/inference/test_trt_convert_flash_multihead_matmul.py @@ -265,17 +265,21 @@ class TrtConvertFlashMultiHeadMatmulTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 2), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), (1e-3, 1e-3) # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 2), (1e-5, 1e-4) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), (1e-2, 1e-3) def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_flatten.py b/test/ir/inference/test_trt_convert_flatten.py index 40546a7fbd3bea51cb0e0ee76d7e364a8388d190..636502aa231138bdcc52084259eca5ec7800560e 100644 --- a/test/ir/inference/test_trt_convert_flatten.py +++ b/test/ir/inference/test_trt_convert_flatten.py @@ -101,10 +101,12 @@ class TrtConvertFlattenTest_dim_2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -112,10 +114,12 @@ class TrtConvertFlattenTest_dim_2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -202,10 +206,12 @@ class TrtConvertFlattenTest_dim_3(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -213,10 +219,12 @@ class TrtConvertFlattenTest_dim_3(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -303,10 +311,12 @@ class TrtConvertFlattenTest_dim_4(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -314,10 +324,12 @@ class TrtConvertFlattenTest_dim_4(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) @@ -404,10 +416,12 @@ class TrtConvertFlattenTest_dim_5(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -415,10 +429,12 @@ class TrtConvertFlattenTest_dim_5(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_flatten_contiguous_range.py b/test/ir/inference/test_trt_convert_flatten_contiguous_range.py index f6d5130ed070c212793f64a59d47c25b56cdec11..df7d6102710e6ad74eec6ec81f9c7695898fd52e 100644 --- a/test/ir/inference/test_trt_convert_flatten_contiguous_range.py +++ b/test/ir/inference/test_trt_convert_flatten_contiguous_range.py @@ -130,10 +130,12 @@ class TrtConvertFlattenContiguousRangeTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -141,10 +143,12 @@ class TrtConvertFlattenContiguousRangeTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_flip.py b/test/ir/inference/test_trt_convert_flip.py index 442a3381e38439adcee0b8f54efddb2eb4ef36d0..d56cf50f7c96feab2c4e5389e3c37fa51d3d5125 100644 --- a/test/ir/inference/test_trt_convert_flip.py +++ b/test/ir/inference/test_trt_convert_flip.py @@ -124,10 +124,12 @@ class TrtConvertFlipTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_fused_token_prune.py b/test/ir/inference/test_trt_convert_fused_token_prune.py index 93a5253aad7b0248e51f3ace90a1e038325d285a..672812edb1d3dc95b141537ed52e7e2f29e89636 100644 --- a/test/ir/inference/test_trt_convert_fused_token_prune.py +++ b/test/ir/inference/test_trt_convert_fused_token_prune.py @@ -121,10 +121,12 @@ class TrtConvertFusedTokenPruneTest(TrtLayerAutoScanTest): generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-2, 1e-2) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-1, 1e-2) diff --git a/test/ir/inference/test_trt_convert_gather.py b/test/ir/inference/test_trt_convert_gather.py index 69a2624b77e092eb116ca1fd913329dde74ae8bf..fbf5c818b01bbcff68291bd719c165dae238abab 100644 --- a/test/ir/inference/test_trt_convert_gather.py +++ b/test/ir/inference/test_trt_convert_gather.py @@ -194,10 +194,12 @@ class TrtConvertGatherTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-3 @@ -205,8 +207,10 @@ class TrtConvertGatherTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-3 def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_gather_nd.py b/test/ir/inference/test_trt_convert_gather_nd.py index 8b32e5516b97e6976dd5eeb9f8b651e34809d5c6..ee0d0917c722fa6006062d7cedcff70f3fab18f8 100644 --- a/test/ir/inference/test_trt_convert_gather_nd.py +++ b/test/ir/inference/test_trt_convert_gather_nd.py @@ -93,15 +93,19 @@ class TrtConvertGatherNdTest_dim_4_1(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): @@ -183,15 +187,19 @@ class TrtConvertGatherNdTest_dim_4_1_2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): @@ -273,15 +281,19 @@ class TrtConvertGatherNdTest_dim_4_2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): @@ -363,15 +375,19 @@ class TrtConvertGatherNdTest_dim_4_3(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): @@ -453,15 +469,19 @@ class TrtConvertGatherNdTest_dim_2_2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): @@ -545,15 +565,19 @@ class TrtConvertGatherNdTest_dim_3_3(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 4), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_gelu.py b/test/ir/inference/test_trt_convert_gelu.py index 1f3847ff207c0e84c41c4fec1c166a8da30e5266..aa64994ad4cf7790cf03e04eafb64b7f8468d372 100644 --- a/test/ir/inference/test_trt_convert_gelu.py +++ b/test/ir/inference/test_trt_convert_gelu.py @@ -127,10 +127,12 @@ class TrtConvertGeluTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -138,10 +140,12 @@ class TrtConvertGeluTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_grid_sampler.py b/test/ir/inference/test_trt_convert_grid_sampler.py index 49d2248a5021272c41f840982fa4cb4731d5d432..5c40eb6e5f779980f180e2bd5618ddda18a173fa 100644 --- a/test/ir/inference/test_trt_convert_grid_sampler.py +++ b/test/ir/inference/test_trt_convert_grid_sampler.py @@ -138,8 +138,10 @@ class TrtConvertGridSampler(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_group_norm.py b/test/ir/inference/test_trt_convert_group_norm.py index bd4652c4bbd67fcd7a228c1e2afd491d046e9891..49993c43f149db4f6d911f30bf08aa76de1e5848 100644 --- a/test/ir/inference/test_trt_convert_group_norm.py +++ b/test/ir/inference/test_trt_convert_group_norm.py @@ -125,11 +125,13 @@ class TrtConvertGroupNormTest(TrtLayerAutoScanTest): clear_dynamic_shape() self.trt_param.workspace_size = 2013265920 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -138,11 +140,13 @@ class TrtConvertGroupNormTest(TrtLayerAutoScanTest): self.trt_param.workspace_size = 2013265920 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_hard_sigmoid.py b/test/ir/inference/test_trt_convert_hard_sigmoid.py index e65b1e1e32b6ffbb470e430a55cc6eb10c933284..69d7594fd4373918564ac2f2d8c9ff0841993885 100644 --- a/test/ir/inference/test_trt_convert_hard_sigmoid.py +++ b/test/ir/inference/test_trt_convert_hard_sigmoid.py @@ -97,15 +97,19 @@ class TrtConvertHardSigmoidTest_dim_2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_hard_swish.py b/test/ir/inference/test_trt_convert_hard_swish.py index 3c3a98ee1e6aea352bd84b81e1bca5457e4fe9e6..381712bb59636b29798ff253885c8df60725b2b9 100644 --- a/test/ir/inference/test_trt_convert_hard_swish.py +++ b/test/ir/inference/test_trt_convert_hard_swish.py @@ -97,10 +97,12 @@ class TrtConvertHardSwishTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -108,10 +110,12 @@ class TrtConvertHardSwishTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_index_select.py b/test/ir/inference/test_trt_convert_index_select.py index 9d1d9fb26ba3e96ee1da48f59192139a5416f0af..c43347ba19abcc0b92d4e3e028467a745d030e97 100644 --- a/test/ir/inference/test_trt_convert_index_select.py +++ b/test/ir/inference/test_trt_convert_index_select.py @@ -171,10 +171,12 @@ class TrtConvertIndexSelectTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-3 @@ -182,8 +184,10 @@ class TrtConvertIndexSelectTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_instance_norm.py b/test/ir/inference/test_trt_convert_instance_norm.py index 72b728d5cc34bc508daa0f1556fc3058314af882..3200003821d36ca2abc18ffb2b530339a72f703e 100644 --- a/test/ir/inference/test_trt_convert_instance_norm.py +++ b/test/ir/inference/test_trt_convert_instance_norm.py @@ -145,10 +145,12 @@ class TrtConvertInstanceNormTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -156,10 +158,12 @@ class TrtConvertInstanceNormTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_inverse.py b/test/ir/inference/test_trt_convert_inverse.py index 87adffe106d05f102db5d78dfbaf986b73de96de..e36d74294092664d03253b6ba784120eb6ee192b 100644 --- a/test/ir/inference/test_trt_convert_inverse.py +++ b/test/ir/inference/test_trt_convert_inverse.py @@ -82,15 +82,19 @@ class TrtConvertInverse(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 3), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_layer_norm.py b/test/ir/inference/test_trt_convert_layer_norm.py index f3c2cbfeff31dd3d8c448dcc42778dbe1f779c5c..63a7c9ac6279388d5b3ec0693d822919430b7de4 100644 --- a/test/ir/inference/test_trt_convert_layer_norm.py +++ b/test/ir/inference/test_trt_convert_layer_norm.py @@ -125,10 +125,12 @@ class TrtConvertLayerNormTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -136,10 +138,12 @@ class TrtConvertLayerNormTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 @@ -249,10 +253,12 @@ class TrtConvertLayerNormTest_2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -260,10 +266,12 @@ class TrtConvertLayerNormTest_2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_leaky_relu.py b/test/ir/inference/test_trt_convert_leaky_relu.py index 683bace7b0002a3f2f0d3974d7405d39484030e4..4fac037674da1a45c0c5d72ab2f414f1c3300c1c 100644 --- a/test/ir/inference/test_trt_convert_leaky_relu.py +++ b/test/ir/inference/test_trt_convert_leaky_relu.py @@ -103,14 +103,17 @@ class TrtConvertLeakyReluTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -118,14 +121,17 @@ class TrtConvertLeakyReluTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_lookup_table.py b/test/ir/inference/test_trt_convert_lookup_table.py index dd5ee6d8fce41d36d2935fb5d1779dfa45652c47..e1fb64bcdf545f8d29ebeedfff14ed4ec87d0ca4 100644 --- a/test/ir/inference/test_trt_convert_lookup_table.py +++ b/test/ir/inference/test_trt_convert_lookup_table.py @@ -138,10 +138,12 @@ class TrtConvertLookupTableV2Test(TrtLayerAutoScanTest): # for dynamic_shape mode generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_lookup_table_v2.py b/test/ir/inference/test_trt_convert_lookup_table_v2.py index ecc97fc3b94d860a6759d15cdd51bfc159564861..4f2722a27f94587ecac4bea7d01951dd403b2107 100644 --- a/test/ir/inference/test_trt_convert_lookup_table_v2.py +++ b/test/ir/inference/test_trt_convert_lookup_table_v2.py @@ -128,10 +128,12 @@ class TrtConvertLookupTableV2Test(TrtLayerAutoScanTest): # for dynamic_shape mode generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_matmul.py b/test/ir/inference/test_trt_convert_matmul.py index 516542827299c95c05b5edc7266bc9af17c00770..a529648c65cec3dd7a9987fe92fd8e46e00fcc6a 100644 --- a/test/ir/inference/test_trt_convert_matmul.py +++ b/test/ir/inference/test_trt_convert_matmul.py @@ -101,8 +101,10 @@ class TrtConvertMatmulTest_static(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def test(self): @@ -192,8 +194,10 @@ class TrtConvertMatmulTest_dynamic(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_matmul_v2.py b/test/ir/inference/test_trt_convert_matmul_v2.py index f3dae2b3160d7163ea98e83cdfc8a8f4781ad6a8..b511d4a0abf4fec8194cbd0f035257aba3db0010 100644 --- a/test/ir/inference/test_trt_convert_matmul_v2.py +++ b/test/ir/inference/test_trt_convert_matmul_v2.py @@ -96,8 +96,10 @@ class TrtConvertMatmulTest_dynamic(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), (tol_fp32, tol_fp32) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (tol_half, tol_half) def add_skip_trt_case(self): @@ -182,8 +184,10 @@ class TrtConvertMatmulTest_dynamic2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), (tol_fp32, tol_fp32) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (tol_half, tol_half) def add_skip_trt_case(self): @@ -311,8 +315,10 @@ class TrtConvertMatmulTest_dynamic3(TrtLayerAutoScanTest): generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), 1e-3 def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_mish.py b/test/ir/inference/test_trt_convert_mish.py index 36375b67031e0b288bec9445ca5af974d2f9b34b..8206b089c6690b03f06951b411cab9a1040be7f7 100644 --- a/test/ir/inference/test_trt_convert_mish.py +++ b/test/ir/inference/test_trt_convert_mish.py @@ -139,10 +139,12 @@ class TrtConvertMishTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -150,10 +152,12 @@ class TrtConvertMishTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_multiclass_nms.py b/test/ir/inference/test_trt_convert_multiclass_nms.py index 20950d2564b9bc5b7dac88e8b7e49f6bf1f58291..3e40c66993553d827f0e136f40b30f42c3e23ec2 100644 --- a/test/ir/inference/test_trt_convert_multiclass_nms.py +++ b/test/ir/inference/test_trt_convert_multiclass_nms.py @@ -160,10 +160,12 @@ class TrtConvertMulticlassNMSTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -171,6 +173,7 @@ class TrtConvertMulticlassNMSTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_multiclass_nms3.py b/test/ir/inference/test_trt_convert_multiclass_nms3.py index 89f40072986d62d1d817645d05ca9c9ece4a50a9..60f2b0a68a41a46d0443ce072f205bc894cdaebf 100644 --- a/test/ir/inference/test_trt_convert_multiclass_nms3.py +++ b/test/ir/inference/test_trt_convert_multiclass_nms3.py @@ -167,10 +167,12 @@ class TrtConvertMulticlassNMS3Test(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -178,6 +180,7 @@ class TrtConvertMulticlassNMS3Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_multihead_matmul.py b/test/ir/inference/test_trt_convert_multihead_matmul.py index 4a29253f2c8b8acde6db17e325269f3d73e6a6fc..3d3100a70e7808af7773f2bffa7571f4495c911b 100644 --- a/test/ir/inference/test_trt_convert_multihead_matmul.py +++ b/test/ir/inference/test_trt_convert_multihead_matmul.py @@ -385,9 +385,11 @@ class TrtConvertMultiHeadMatmulTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-2, 1e-2) def test(self): @@ -983,16 +985,19 @@ class TrtConvertVitToMultiHeadMatmulTest(TrtLayerAutoScanTest): generate_dynamic_shape(attrs) self.trt_param.workspace_size = 2013265920 self.trt_param.precision = paddle_infer.PrecisionType.Int8 + program_config.set_input_type(np.int8) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-3, 1e-3, ) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-3, 2e-2, ) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(), ( 1e-5, 1e-5, @@ -1364,9 +1369,11 @@ class TrtConvertMultiHeadMatmulTest_biasqk_seqseq(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-2, 1e-2) def test(self): diff --git a/test/ir/inference/test_trt_convert_multihead_matmul_roformer.py b/test/ir/inference/test_trt_convert_multihead_matmul_roformer.py index 2de959a6ee71690e8ffe73654040f881cad74106..f1f8ecdc374993fa8b3043566717bd277bfde09a 100644 --- a/test/ir/inference/test_trt_convert_multihead_matmul_roformer.py +++ b/test/ir/inference/test_trt_convert_multihead_matmul_roformer.py @@ -540,9 +540,11 @@ class TrtConvertMultiHeadMatmulRoformerTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 5), (1e-3, 1e-3) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 5), (1e-3, 1e-3) def test(self): diff --git a/test/ir/inference/test_trt_convert_nearest_interp.py b/test/ir/inference/test_trt_convert_nearest_interp.py index e3dd3c7474080adcd131dbd41c67cc27d8e5d114..a96cbb9f81a9a316ecfcfbea2e7b7cc27eff2cb5 100644 --- a/test/ir/inference/test_trt_convert_nearest_interp.py +++ b/test/ir/inference/test_trt_convert_nearest_interp.py @@ -113,10 +113,12 @@ class TrtConvertNearestInterpTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -124,10 +126,12 @@ class TrtConvertNearestInterpTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_nearest_interp_v2.py b/test/ir/inference/test_trt_convert_nearest_interp_v2.py index 7a9421681768c8eb8f87329c189455cc2e4dd722..d83ec7ecbe986d1cd6ce3c89783b41842ad72e11 100644 --- a/test/ir/inference/test_trt_convert_nearest_interp_v2.py +++ b/test/ir/inference/test_trt_convert_nearest_interp_v2.py @@ -83,10 +83,12 @@ class TrtConvertNearestInterpV2Test(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -94,10 +96,12 @@ class TrtConvertNearestInterpV2Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 @@ -174,10 +178,12 @@ class TrtConvertNearestInterpV2ShapeTensorTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -185,10 +191,12 @@ class TrtConvertNearestInterpV2ShapeTensorTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_one_hot.py b/test/ir/inference/test_trt_convert_one_hot.py index 5332cb950ae88a6a72ed1886e2fc5d79a7e3b746..72ca495ef8bea982fb76be973fc565d9bfd8c439 100644 --- a/test/ir/inference/test_trt_convert_one_hot.py +++ b/test/ir/inference/test_trt_convert_one_hot.py @@ -141,10 +141,12 @@ class TrtConvertOneHotTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -152,10 +154,12 @@ class TrtConvertOneHotTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_p_norm.py b/test/ir/inference/test_trt_convert_p_norm.py index 6fc8e5ec87cb0da85370f0715eb8698e73d0aefa..bf5adf9715ba04c97a741421c595454ecd341033 100644 --- a/test/ir/inference/test_trt_convert_p_norm.py +++ b/test/ir/inference/test_trt_convert_p_norm.py @@ -118,10 +118,12 @@ class TrtConvertPNormTest(TrtLayerAutoScanTest): # for dynamic_shape mode generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_pad.py b/test/ir/inference/test_trt_convert_pad.py index ecfc09090b7f233780b4111f69c8af3e5edb30e1..f20b915c2a467b9a0adf26f447d9ff105f9c0c38 100644 --- a/test/ir/inference/test_trt_convert_pad.py +++ b/test/ir/inference/test_trt_convert_pad.py @@ -104,10 +104,12 @@ class TrtConvertPadTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 @@ -115,10 +117,12 @@ class TrtConvertPadTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_pad3d.py b/test/ir/inference/test_trt_convert_pad3d.py index df0426b23493fb82ee6e0777008b1b2988b16b2b..1237c26c4745d25b90b89ee152822508a77e338d 100644 --- a/test/ir/inference/test_trt_convert_pad3d.py +++ b/test/ir/inference/test_trt_convert_pad3d.py @@ -120,10 +120,12 @@ class TrtConvertPad3dTensorPadding(TrtLayerAutoScanTest): clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -131,10 +133,12 @@ class TrtConvertPad3dTensorPadding(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -230,10 +234,12 @@ class TrtConvertPad3dListPadding(TrtLayerAutoScanTest): clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -241,10 +247,12 @@ class TrtConvertPad3dListPadding(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_pool2d.py b/test/ir/inference/test_trt_convert_pool2d.py index a44273c069b0c34ac5abeb6b6bbca6e8c368d43d..669a83b7f2da0a3dc63897b4ae9221ae5a7f2e01 100644 --- a/test/ir/inference/test_trt_convert_pool2d.py +++ b/test/ir/inference/test_trt_convert_pool2d.py @@ -153,10 +153,12 @@ class TrtConvertPool2dTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -164,10 +166,12 @@ class TrtConvertPool2dTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_preln_residual_bias.py b/test/ir/inference/test_trt_convert_preln_residual_bias.py index a45ddfcae189e2b7814ee363902c16cfb1535268..f9431e07511c3434ba5510cd4b1eab21f54d4342 100644 --- a/test/ir/inference/test_trt_convert_preln_residual_bias.py +++ b/test/ir/inference/test_trt_convert_preln_residual_bias.py @@ -169,10 +169,12 @@ class TrtConvertSkipLayernormTest(TrtLayerAutoScanTest): # for static_shape, fall back to fluid fused op clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 # atol=1e-2 while rtol is 1e-8 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 # atol=1e-2 while rtol is 1e-8 @@ -180,10 +182,12 @@ class TrtConvertSkipLayernormTest(TrtLayerAutoScanTest): # just support dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 # atol=1e-2 while rtol is 1e-8 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 # atol=1e-2 while rtol is 1e-8 diff --git a/test/ir/inference/test_trt_convert_preln_residual_no_bias.py b/test/ir/inference/test_trt_convert_preln_residual_no_bias.py index fd3bdb64c7ede23db826ed60cb329bd9199f6668..7be194a953a2e3dadfae242832ece1710d1d9f57 100644 --- a/test/ir/inference/test_trt_convert_preln_residual_no_bias.py +++ b/test/ir/inference/test_trt_convert_preln_residual_no_bias.py @@ -158,10 +158,12 @@ class TrtConvertSkipLayernormTest(TrtLayerAutoScanTest): # for static_shape, fall back to fluid fused op clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 # atol=1e-2 while rtol is 1e-8 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 # atol=1e-2 while rtol is 1e-8 @@ -169,10 +171,12 @@ class TrtConvertSkipLayernormTest(TrtLayerAutoScanTest): # just support dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 # atol=1e-2 while rtol is 1e-8 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 # atol=1e-2 while rtol is 1e-8 diff --git a/test/ir/inference/test_trt_convert_prelu.py b/test/ir/inference/test_trt_convert_prelu.py index ae53f10ad576107a6fea334da355c1ddf8179cb3..7c627f957ab0c072e04cc63e68d3e3da6dcfd776 100644 --- a/test/ir/inference/test_trt_convert_prelu.py +++ b/test/ir/inference/test_trt_convert_prelu.py @@ -205,10 +205,12 @@ class TrtConvertPreluTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -216,10 +218,12 @@ class TrtConvertPreluTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_qk_multihead_matmul.py b/test/ir/inference/test_trt_convert_qk_multihead_matmul.py index 548f0486e12f68bfae0ab018511f926ace495635..ece654687fe7a4c7fff39ac4f2dcaf3cd27caa61 100644 --- a/test/ir/inference/test_trt_convert_qk_multihead_matmul.py +++ b/test/ir/inference/test_trt_convert_qk_multihead_matmul.py @@ -329,17 +329,21 @@ class TrtConvertQkAttentionTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 3), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-3, 1e-3) # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), (1, 3), (1e-5, 1e-4) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 3), (1e-2, 1e-3) def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_range.py b/test/ir/inference/test_trt_convert_range.py index 5ef6b8b90524d860b6f36547278606d3cabd89bd..d8b139925b0004f6efdb0e64607a5a75e3538796 100644 --- a/test/ir/inference/test_trt_convert_range.py +++ b/test/ir/inference/test_trt_convert_range.py @@ -130,10 +130,12 @@ class TrtConvertRangeDynamicTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-2 @@ -214,10 +216,12 @@ class TrtConvertRangeStaticTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-2 diff --git a/test/ir/inference/test_trt_convert_reduce.py b/test/ir/inference/test_trt_convert_reduce.py index 59acef179644fda85fc5c2402eba97ab0cc7de5f..57de3c5603b375b980dc353c67fce9c3b0f2522f 100644 --- a/test/ir/inference/test_trt_convert_reduce.py +++ b/test/ir/inference/test_trt_convert_reduce.py @@ -164,10 +164,12 @@ class TrtConvertReduceTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -175,10 +177,12 @@ class TrtConvertReduceTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-5) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_reshape.py b/test/ir/inference/test_trt_convert_reshape.py index c30d973651bad125911aad42dc0bc43749d32c03..5762d51c3480f78919234ec5dfb96c22550b2ea4 100644 --- a/test/ir/inference/test_trt_convert_reshape.py +++ b/test/ir/inference/test_trt_convert_reshape.py @@ -172,10 +172,12 @@ class TrtConvertReshapeTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -183,10 +185,12 @@ class TrtConvertReshapeTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -308,8 +312,10 @@ class TrtConvertReshapeTest2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def add_skip_trt_case(self): @@ -419,8 +425,10 @@ class TrtConvertReshapeTest3(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def add_skip_trt_case(self): @@ -509,10 +517,12 @@ class TrtConvertReshapeZeroDimsTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_rnn.py b/test/ir/inference/test_trt_convert_rnn.py index 296da4db148bc619c1e2fa487fc484a47be1dd33..b00ac38ac0e2c466408bbb2990c60354ac7b6042 100644 --- a/test/ir/inference/test_trt_convert_rnn.py +++ b/test/ir/inference/test_trt_convert_rnn.py @@ -264,10 +264,12 @@ class TrtConvertSliceTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), tol_fp32 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), tol_half diff --git a/test/ir/inference/test_trt_convert_roi_align.py b/test/ir/inference/test_trt_convert_roi_align.py index 1e1a83a40e48ac4be8b2dd5c280db4ec4660f973..4a3558ac4615fde7b1a63298f41d38efe09a68bd 100644 --- a/test/ir/inference/test_trt_convert_roi_align.py +++ b/test/ir/inference/test_trt_convert_roi_align.py @@ -190,10 +190,12 @@ class TrtConvertRoiAlignTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -201,10 +203,12 @@ class TrtConvertRoiAlignTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_roll.py b/test/ir/inference/test_trt_convert_roll.py index e5267de3e592897e63dd5b2cbb4d75e18968e478..21e265a4679dc8e62a95246446cf4135d208d66f 100644 --- a/test/ir/inference/test_trt_convert_roll.py +++ b/test/ir/inference/test_trt_convert_roll.py @@ -107,10 +107,12 @@ class TrtConvertRollTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -118,10 +120,12 @@ class TrtConvertRollTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_scale.py b/test/ir/inference/test_trt_convert_scale.py index 55d1752d499037436acadd57b69eafdf0ff22ed1..c3916a65aa627508291c4de69e998c0a1c549ca1 100644 --- a/test/ir/inference/test_trt_convert_scale.py +++ b/test/ir/inference/test_trt_convert_scale.py @@ -165,10 +165,12 @@ class TrtConvertScaleTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -176,10 +178,12 @@ class TrtConvertScaleTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_scatter_nd_add.py b/test/ir/inference/test_trt_convert_scatter_nd_add.py index 66f25781e6beeec8f5979206446f35488945c09f..7667b40331ef4120d94a1f448d445276327e9c8d 100644 --- a/test/ir/inference/test_trt_convert_scatter_nd_add.py +++ b/test/ir/inference/test_trt_convert_scatter_nd_add.py @@ -102,15 +102,19 @@ class TrtConvertScatterNd(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 5), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 5), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 4), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 4), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_set_value.py b/test/ir/inference/test_trt_convert_set_value.py index 569a515e9ada28936373372479a7830d392c3d72..39f2d1c228bb38e406f9a7d8ac2ff5f4416e3f96 100644 --- a/test/ir/inference/test_trt_convert_set_value.py +++ b/test/ir/inference/test_trt_convert_set_value.py @@ -154,11 +154,13 @@ class TrtConvertSetValue(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) self.trt_param.workspace_size = 2013265920 yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-5, 1e-4) self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_shape.py b/test/ir/inference/test_trt_convert_shape.py index 1f2a97018e2c0484b87c19666ecd8071dd384bab..f29473d9ded2effb819f4b87b78e5ee9abc49896 100644 --- a/test/ir/inference/test_trt_convert_shape.py +++ b/test/ir/inference/test_trt_convert_shape.py @@ -99,10 +99,12 @@ class TrtConvertSumTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-3 @@ -110,8 +112,10 @@ class TrtConvertSumTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_shuffle_channel.py b/test/ir/inference/test_trt_convert_shuffle_channel.py index f61ecfbeea0cddfbefca8dc40206d0c8d7b99945..725a6c67c373f47268b653f0e550e6c967681b4c 100644 --- a/test/ir/inference/test_trt_convert_shuffle_channel.py +++ b/test/ir/inference/test_trt_convert_shuffle_channel.py @@ -92,10 +92,12 @@ class TrtConvertShuffleChannelTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -103,10 +105,12 @@ class TrtConvertShuffleChannelTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_slice.py b/test/ir/inference/test_trt_convert_slice.py index ac5b6439030dd3bcc1cf77c3591a1ac6cb0b2b4c..66458d53a125630c30f13f3f8236f8b4502b1fcc 100644 --- a/test/ir/inference/test_trt_convert_slice.py +++ b/test/ir/inference/test_trt_convert_slice.py @@ -131,10 +131,12 @@ class TrtConvertSliceTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -142,10 +144,12 @@ class TrtConvertSliceTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_softmax.py b/test/ir/inference/test_trt_convert_softmax.py index 7fb9a69bb63ee3637d50e2ebe0cf4744e52265c1..713340946865429d0209b4cd91af5e082b32c053 100644 --- a/test/ir/inference/test_trt_convert_softmax.py +++ b/test/ir/inference/test_trt_convert_softmax.py @@ -135,10 +135,12 @@ class TrtConvertSoftmaxTest(TrtLayerAutoScanTest): pass else: self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -146,10 +148,12 @@ class TrtConvertSoftmaxTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_split.py b/test/ir/inference/test_trt_convert_split.py index bfb7099f16f4463af62030599ec4ca7479974d1d..e29cb2ddef1e4b73c7a040db34322970ea2fcf63 100644 --- a/test/ir/inference/test_trt_convert_split.py +++ b/test/ir/inference/test_trt_convert_split.py @@ -250,10 +250,12 @@ class TrtConvertSplitTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -261,10 +263,12 @@ class TrtConvertSplitTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -400,10 +404,12 @@ class TrtConvertSplitTest2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_square.py b/test/ir/inference/test_trt_convert_square.py index 4dd30b4016f9fd15969762b0e31b00bdd7ba11d3..a082726998b03d71f71b95894e055c989b0688ff 100644 --- a/test/ir/inference/test_trt_convert_square.py +++ b/test/ir/inference/test_trt_convert_square.py @@ -117,10 +117,12 @@ class TrtConvertSquareTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -128,10 +130,12 @@ class TrtConvertSquareTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_squeeze2.py b/test/ir/inference/test_trt_convert_squeeze2.py index a24465428f632e8d2f8a95699870c32fe2902938..3b73040c881c9ea87693e6209bd7edb92899df14 100644 --- a/test/ir/inference/test_trt_convert_squeeze2.py +++ b/test/ir/inference/test_trt_convert_squeeze2.py @@ -117,10 +117,12 @@ class TrtConvertSplitTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -128,10 +130,12 @@ class TrtConvertSplitTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_stack.py b/test/ir/inference/test_trt_convert_stack.py index 39f729e2be71283b943758d9592369540ccb2569..f83368e325f46e58b77dc46ec1896c8885eec7a5 100644 --- a/test/ir/inference/test_trt_convert_stack.py +++ b/test/ir/inference/test_trt_convert_stack.py @@ -218,10 +218,12 @@ class TrtConvertStackTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -229,10 +231,12 @@ class TrtConvertStackTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_strided_slice.py b/test/ir/inference/test_trt_convert_strided_slice.py index b814c9f9b5797213362c8f6e22a98901cd867edc..e0e6198a16f4a8b72c7f522c46f8c88b2876b95e 100644 --- a/test/ir/inference/test_trt_convert_strided_slice.py +++ b/test/ir/inference/test_trt_convert_strided_slice.py @@ -122,6 +122,7 @@ class TrtConvertStridedSliceTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -129,6 +130,7 @@ class TrtConvertStridedSliceTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 @@ -217,11 +219,13 @@ class TrtConvertStridedSliceTest2(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 def test(self): diff --git a/test/ir/inference/test_trt_convert_sum.py b/test/ir/inference/test_trt_convert_sum.py index 231921d641b7398d3a89e162ddf0e84185c35e5e..abcd0f8ed663ba4d33ab06955c0c43c0b6e0d67c 100644 --- a/test/ir/inference/test_trt_convert_sum.py +++ b/test/ir/inference/test_trt_convert_sum.py @@ -193,10 +193,12 @@ class TrtConvertSumTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-3 @@ -204,8 +206,10 @@ class TrtConvertSumTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-3 def test(self): @@ -311,10 +315,12 @@ class TrtConvertSumTest1(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-3 @@ -322,8 +328,10 @@ class TrtConvertSumTest1(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_swish.py b/test/ir/inference/test_trt_convert_swish.py index 3db16d47cdabb66e9ec61894695684666a8ced2d..623cada1d460020822c73d242da26e3aa25a9ef7 100755 --- a/test/ir/inference/test_trt_convert_swish.py +++ b/test/ir/inference/test_trt_convert_swish.py @@ -121,10 +121,12 @@ class TrtConvertSwishTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -132,10 +134,12 @@ class TrtConvertSwishTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_take_along_axis.py b/test/ir/inference/test_trt_convert_take_along_axis.py index 63b340f8807a7ebd4d68104fcb0a85cac2373ea2..3d269ea6da406bae1317329c9db37c249d56331e 100644 --- a/test/ir/inference/test_trt_convert_take_along_axis.py +++ b/test/ir/inference/test_trt_convert_take_along_axis.py @@ -160,10 +160,12 @@ class TrtConvertTakeAlongAxisTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( False ), 1e-5 @@ -171,8 +173,10 @@ class TrtConvertTakeAlongAxisTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num(True), 1e-3 def add_skip_trt_case(self): diff --git a/test/ir/inference/test_trt_convert_temporal_shift.py b/test/ir/inference/test_trt_convert_temporal_shift.py index b0b2ce5106213c749227107c4afde0e961e0ff86..9ac5369264aa9ad20b8c0aeb661e9f284ecb3b6e 100755 --- a/test/ir/inference/test_trt_convert_temporal_shift.py +++ b/test/ir/inference/test_trt_convert_temporal_shift.py @@ -108,10 +108,12 @@ class TrtConvertTemporalShiftTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -119,10 +121,12 @@ class TrtConvertTemporalShiftTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_tile.py b/test/ir/inference/test_trt_convert_tile.py index 240fd65b83960686ee4b4ba8f7d8cc5b34d78c4d..ba512438e24e0ab2781e9085e8c21c7bd68ed80c 100644 --- a/test/ir/inference/test_trt_convert_tile.py +++ b/test/ir/inference/test_trt_convert_tile.py @@ -93,10 +93,12 @@ class TrtConvertTileTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -104,10 +106,12 @@ class TrtConvertTileTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -184,10 +188,12 @@ class TrtConvertTileTest2(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -283,10 +289,12 @@ class TrtConvertTileTest3(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_top_k.py b/test/ir/inference/test_trt_convert_top_k.py index 8c8bcbde9935fbaa4ddb67f6db16acdfa5d7ff7a..239e361ebd301e5b47316a813e8a117ded3978da 100644 --- a/test/ir/inference/test_trt_convert_top_k.py +++ b/test/ir/inference/test_trt_convert_top_k.py @@ -118,10 +118,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -129,10 +131,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_top_k_v2.py b/test/ir/inference/test_trt_convert_top_k_v2.py index 29a04c5d04943af058e5a13b8662c15244e6162d..9f20b6260ac4703408fd4dbea1ec21877b898e4b 100644 --- a/test/ir/inference/test_trt_convert_top_k_v2.py +++ b/test/ir/inference/test_trt_convert_top_k_v2.py @@ -147,10 +147,12 @@ class TrtConvertTopKV2Test(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -158,10 +160,12 @@ class TrtConvertTopKV2Test(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_trans_layernorm.py b/test/ir/inference/test_trt_convert_trans_layernorm.py index 3d7236d2bba7dfe32d7dda25f89bc7daa2a7668c..e8623b771dbd83012cf12e9933bc71b9a3779524 100644 --- a/test/ir/inference/test_trt_convert_trans_layernorm.py +++ b/test/ir/inference/test_trt_convert_trans_layernorm.py @@ -226,10 +226,12 @@ class TrtConvertTransLayernormTest(TrtLayerAutoScanTest): # just support dynamic_shape generate_dynamic_shape(attrs, inputs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), ( diff --git a/test/ir/inference/test_trt_convert_transpose.py b/test/ir/inference/test_trt_convert_transpose.py index 9a583537569538a6e4192d8964f7a8fcb7af1646..d0d3aa98313cbf5f2b42987372c4b198ff3d9b56 100644 --- a/test/ir/inference/test_trt_convert_transpose.py +++ b/test/ir/inference/test_trt_convert_transpose.py @@ -139,10 +139,12 @@ class TrtConvertTransposeTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -150,10 +152,12 @@ class TrtConvertTransposeTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_unary.py b/test/ir/inference/test_trt_convert_unary.py index 1d83978f2f06ca34bd5cab68c1ff7d566be74f7a..ba70dfb085d6122fe990529a05bff4d9219cf9a5 100644 --- a/test/ir/inference/test_trt_convert_unary.py +++ b/test/ir/inference/test_trt_convert_unary.py @@ -208,10 +208,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-4 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -219,10 +221,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-4 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 @@ -342,10 +346,12 @@ class TrtConvertLogicalNotTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), (1e-3, 1e-3) @@ -353,10 +359,12 @@ class TrtConvertLogicalNotTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), (1e-3, 1e-3) diff --git a/test/ir/inference/test_trt_convert_unbind.py b/test/ir/inference/test_trt_convert_unbind.py index 875f37ba54a4d79f4a5d73ae79229ddf40cd3490..91464aabeefe95a10d9cb83ef5117371e77146c2 100644 --- a/test/ir/inference/test_trt_convert_unbind.py +++ b/test/ir/inference/test_trt_convert_unbind.py @@ -106,10 +106,12 @@ class TrtConvertUnbind(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_unfold.py b/test/ir/inference/test_trt_convert_unfold.py index 1f63a2eb97e78a052f2a9eff7299ca38692ed0c1..ea18d50e1a446357457281f3be72518197049815 100644 --- a/test/ir/inference/test_trt_convert_unfold.py +++ b/test/ir/inference/test_trt_convert_unfold.py @@ -87,15 +87,19 @@ class TrtConvertUnfold(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (0, 3), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (0, 3), 1e-3 # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), (1, 2), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), (1, 2), 1e-3 def test(self): diff --git a/test/ir/inference/test_trt_convert_unsqueeze2.py b/test/ir/inference/test_trt_convert_unsqueeze2.py index 7e92c52b553875d08023fe5045023c1dc94c1324..d1f27ace1473d0c6e720c3768c0852587272eafe 100644 --- a/test/ir/inference/test_trt_convert_unsqueeze2.py +++ b/test/ir/inference/test_trt_convert_unsqueeze2.py @@ -100,10 +100,12 @@ class TrtConvertSplitTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -111,10 +113,12 @@ class TrtConvertSplitTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_where.py b/test/ir/inference/test_trt_convert_where.py index 17c20e56de229c1a4451916fdeea1250374f72e8..fafa1a137fa9b5195f9e74d4e6ccad91c0a2d7cf 100644 --- a/test/ir/inference/test_trt_convert_where.py +++ b/test/ir/inference/test_trt_convert_where.py @@ -206,10 +206,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 @@ -217,10 +219,12 @@ class TrtConvertActivationTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 diff --git a/test/ir/inference/test_trt_convert_yolo_box.py b/test/ir/inference/test_trt_convert_yolo_box.py index 0bc410cc81f2ba0bb52968311fb2bdfec79a0c8c..dc112ba02d7d28f0ee609b3b66e81bb349015714 100644 --- a/test/ir/inference/test_trt_convert_yolo_box.py +++ b/test/ir/inference/test_trt_convert_yolo_box.py @@ -153,10 +153,12 @@ class TrtConvertYoloBoxTest(TrtLayerAutoScanTest): # for static_shape clear_dynamic_shape() self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, False ), 1e-3 @@ -164,10 +166,12 @@ class TrtConvertYoloBoxTest(TrtLayerAutoScanTest): # for dynamic_shape generate_dynamic_shape(attrs) self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-5 self.trt_param.precision = paddle_infer.PrecisionType.Half + program_config.set_input_type(np.float16) yield self.create_inference_config(), generate_trt_nodes_num( attrs, True ), 1e-3 diff --git a/test/ir/inference/test_trt_convert_yolo_box_head.py b/test/ir/inference/test_trt_convert_yolo_box_head.py index cf31681b1ab9b584d7b7ce03421af2718c89e442..df53e2fec654f71ee6f6c2610bf01868cbd884b9 100644 --- a/test/ir/inference/test_trt_convert_yolo_box_head.py +++ b/test/ir/inference/test_trt_convert_yolo_box_head.py @@ -79,6 +79,7 @@ class TrtConvertYoloBoxHeadTest(TrtLayerAutoScanTest): ) -> (paddle_infer.Config, List[int], float): # for static_shape self.trt_param.precision = paddle_infer.PrecisionType.Float32 + program_config.set_input_type(np.float32) yield self.create_inference_config(), [1, 2], 1e-5 def test(self):