解决TensorRT报错：Assertion Error in trtSmToCask: 0 (Unsupported SM.)

最新推荐文章于 2025-11-05 11:03:12 发布
原创 最新推荐文章于 2025-11-05 11:03:12 发布 · 2.8k 阅读 · 1 ·
CC 4.0 BY-SA版权
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
文章标签：

#c++ #linux #gitlab

本文介绍了在使用TensorRT时遇到的错误信息，错误源于TensorRT版本7.2.1与CUDA 11.5版本不兼容。解决办法包括重新安装CUDA或TensorRT。建议选择重装CUDA，因为不同TensorRT版本间可能存在代码不一致性，且重装TensorRT可能需要大量代码修改。同时提供了TensorRT、CUDA和cudnn的版本匹配指南。
部署运行你感兴趣的模型镜像

错误信息：

E20211126 11:55:13.654 140412916315904 tensorrt.cpp:10] …/rtSafe/cuda/caskUtils.cpp (98) - Assertion Error in trtSmToCask: 0 (Unsupported SM.)

原因：TensorRT版本与cuda或者cudnn版本不匹配，如笔者的TensorRT版本为7.2.1，而cuda为最新的11.5版本，其中cuda版本过高无法适配：
在这里插入图片描述

解决：重装cuda或者TensorRT
建议：重装cuda，因为不同版本的TensorRT在使用时存在代码不一致的情况，如TensorRT7.2.1和TensorRT8.2.1要求重写虚函数时添加noexcept，如果代码量不大可以考虑重装TensorRT
附上TensorRT、cuda、cudnn各版本的匹配要求：TensorRT release note

您可能感兴趣的与本文相关的镜像

PyTorch 2.5

PyTorch 2.5

PyTorch
Cuda

PyTorch 是一个开源的 Python 机器学习库，基于 Torch 库，底层由 C++ 实现，应用于人工智能领域，如计算机视觉和自然语言处理

tensorrt报错 [F] [TRT] Assertion failed: Unsupported SM.
qq_30841655的博客
01-06 5648
目录0. 完整报错信息1. 原因 0. 完整报错信息 在安装完tensorRT，运行minist测试用例时，发生如下报错 &&&& RUNNING TensorRT.sample_mnist # ./sample_mnist [01/06/2022-15:14:48] [I] Building and running a GPU inference engine for MNIST [01/06/2022-15:27:59] [F] [TRT] Assertion faile
运行调用TensorRT程序报错 Version tag does not match. Internal Error (Engine deserialization failed.) 参考解决方法
qq_39779233的博客
07-10 2591
运行调用TensorRT程序报错 Version tag does not match. Internal Error (Engine deserialization failed.) 参考解决方法
【笔记】解决 “AssertionError: Torch not compiled with CUDA enabled“ 错误
love530love的博客
11-05 3219
解决PyTorch的CUDA错误"AssertionError: Torch not compiled with CUDA enabled"指南：该错误通常由PyTorch未启用CUDA支持或版本冲突导致。解决方法包括：1.检查当前PyTorch版本；2.从官网获取正确安装命令（含torchvision和torchaudio）；3.强制重装指定版本；4.验证安装结果。其他方案包括完全重装、使用PEP517参数等。预防措施建议使用虚拟环境并注意版本兼容性。最终通过正确安装CUDA支持的Py
ModuleNotFoundError: No module named ‘pycuda‘ | No module named ‘tensorrt‘ | 【TensorRT报错收集】
专注于AI领域前沿技术学习与分享：目标检测、图像修复、超分重建、AI工程化
07-14 9337
No module named ‘tensorrt‘、TensorRT报错收集
fatal error: NvInfer.h: No such file or directory | TensorRT 报错处理 | 【成功解决】
热门推荐
专注于AI领域前沿技术学习与分享：目标检测、图像修复、超分重建、AI工程化
07-16 1万+
遇到报错【 一】： fatal error: NvInfer.h: No such file or directory #include "NvInfer.h" 通过 find -name NvInfer.h 查找发现 该头文件所在目录 可以发现，该头文件 是属于 TensorRT 下的一个 专有头文件，因此我们的C++ 代码在编译时需要能够找到它 find -name NvInfer.h ./project/project21/modelTrans/tensorRT/tensorRT7.
import tensorrt报错 please make sure you have pycuda installed
qq_36780295的博客
08-26 718
import tensorrt报错 please make sure you have pycuda installed解决方法： 执行import tensorrt时，报错： 通过搜索错误原因，在https://blog.csdn.net/baobei0112/article/details/88761236看到了如下内容： try: import pycuda.driver as cuda import pycuda.gpuarray as gpuarray import pycuda.autoin
tensorRT报错1: [caskUtils.cpp::trtSmToCask::147] Error Code 1: Internal Error (Unsupported SM: 0x809)
m0_56836274的博客
03-05 2049
毕设需要用到TensorRT,小白我第一次下载没经验。把我遇到的坑和大家分享，避免咱调入同一个坑！在下载完TensorRT-8.2.5.1后，我尝试运行TensorRT自带的MNIST数据集的例子，报错了，报错内容如下：在网上搜了一下，说是这个版本的TensorRT不支持SM,也就是不支持该架构的流多处理器。英伟达官方建议的是4080用TensorRT-8.5.1.7。但是我服务器上的GPU是4090，我尝试用TensorRT-8.6.6.1。
[caskBuilderUtils.cpp::trtSmToCaskCCV::548] Error Code 1: Internal Error (Unsupported SM: 0x809)
XCCCCZ的博客
12-19 1348
如果有C++版本的onnx转tensorrt engine文件的程序，则可直接用高版本的tensorrt repo程序安装tensorrt版后再执行C++版的转换程序即可，tensorrt的python包则可根据需要选择是否安装。如果整个环境不需要支持兼容其他模型或程序，可以考虑把cuda和pytorch之类的都先升级再回头安装此程序，如果考虑到兼容，可以从NVIDIA网站上。解压后，设置LD_LIBRARY_PATH指向此tar包解压出来的目录里的so文件所在的lib子目录。
[TensorRT] INTERNAL ERROR: Assertion failed: mem = nullpt
ltochange的博客
11-18 1987
[TensorRT] ERROR: FAILED_EXECUTION: std::exception ^C[TensorRT] INTERNAL ERROR: Assertion failed: mem != nullptr ../rtExt/cuda/cudaPointWiseRunner.cpp:28 Aborting... 使用tensorrt遇到了上述错误，这个错误发生在构建好rt模型之后，错误的原因是： 有一个输出没有拷贝到显存上，造成了非法访问 def setup_bindi...
[11/27/2025-18:31:50] [I] === Model Options === [11/27/2025-18:31:50] [I] Format: ONNX [11/27/2025-18:31:50] [I] Model: D:\Audio2Face\Audio2Face-3D-SDK\_data\audio2emotion-models\audio2emotion-v2.2\network.onnx [11/27/2025-18:31:50] [I] Output: [11/27/2025-18:31:50] [I] === Build Options === [11/27/2025-18:31:50] [I] Memory Pools: workspace: default, dlaSRAM: default, dlaLocalDRAM: default, dlaGlobalDRAM: default, tacticSharedMem: default [11/27/2025-18:31:50] [I] avgTiming: 8 [11/27/2025-18:31:50] [I] Precision: FP32 [11/27/2025-18:31:50] [I] LayerPrecisions: [11/27/2025-18:31:50] [I] Layer Device Types: [11/27/2025-18:31:50] [I] Calibration: [11/27/2025-18:31:50] [I] Refit: Disabled [11/27/2025-18:31:50] [I] Strip weights: Disabled [11/27/2025-18:31:50] [I] Version Compatible: Disabled [11/27/2025-18:31:50] [I] ONNX Plugin InstanceNorm: Disabled [11/27/2025-18:31:50] [I] ONNX kENABLE_UINT8_AND_ASYMMETRIC_QUANTIZATION_DLA flag: Disabled [11/27/2025-18:31:50] [I] TensorRT runtime: full [11/27/2025-18:31:50] [I] Lean DLL Path: [11/27/2025-18:31:50] [I] Tempfile Controls: { in_memory: allow, temporary: allow } [11/27/2025-18:31:50] [I] Exclude Lean Runtime: Disabled [11/27/2025-18:31:50] [I] Sparsity: Disabled [11/27/2025-18:31:50] [I] Safe mode: Disabled [11/27/2025-18:31:50] [I] Build DLA standalone loadable: Disabled [11/27/2025-18:31:50] [I] Allow GPU fallback for DLA: Disabled [11/27/2025-18:31:50] [I] DirectIO mode: Disabled [11/27/2025-18:31:50] [I] Restricted mode: Disabled [11/27/2025-18:31:50] [I] Skip inference: Disabled [11/27/2025-18:31:50] [I] Save engine: C:\Users\Administrator\AppData\Local\Temp\tmp6s8itunu\network.trt [11/27/2025-18:31:50] [I] Load engine: [11/27/2025-18:31:50] [I] Profiling verbosity: 0 [11/27/2025-18:31:50] [I] Tactic sources: Using default tactic sources [11/27/2025-18:31:50] [I] timingCacheMode: local [11/27/2025-18:31:50] [I] timingCacheFile: [11/27/2025-18:31:50] [I] Enable Compilation Cache: Enabled [11/27/2025-18:31:50] [I] Enable Monitor Memory: Disabled [11/27/2025-18:31:50] [I] errorOnTimingCacheMiss: Disabled [11/27/2025-18:31:50] [I] Preview Features: Use default preview flags. [11/27/2025-18:31:50] [I] MaxAuxStreams: -1 [11/27/2025-18:31:50] [I] BuilderOptimizationLevel: -1 [11/27/2025-18:31:50] [I] MaxTactics: -1 [11/27/2025-18:31:50] [I] Calibration Profile Index: 0 [11/27/2025-18:31:50] [I] Weight Streaming: Disabled [11/27/2025-18:31:50] [I] Runtime Platform: Same As Build [11/27/2025-18:31:50] [I] Debug Tensors: [11/27/2025-18:31:50] [I] Distributive Independence: Disabled [11/27/2025-18:31:50] [I] Mark Unfused Tensors As Debug Tensors: Disabled [11/27/2025-18:31:50] [I] Input(s)s format: fp32:CHW [11/27/2025-18:31:50] [I] Output(s)s format: fp32:CHW [11/27/2025-18:31:50] [I] Input build shape (profile 0): input_values=1x5000+8x30000+128x60000 [11/27/2025-18:31:50] [I] Input calibration shapes: model [11/27/2025-18:31:50] [I] === System Options === [11/27/2025-18:31:50] [I] Device: 0 [11/27/2025-18:31:50] [I] DLACore: [11/27/2025-18:31:50] [I] Plugins: [11/27/2025-18:31:50] [I] setPluginsToSerialize: [11/27/2025-18:31:50] [I] dynamicPlugins: [11/27/2025-18:31:50] [I] ignoreParsedPluginLibs: 0 [11/27/2025-18:31:50] [I] [11/27/2025-18:31:50] [I] === Inference Options === [11/27/2025-18:31:50] [I] Batch: Explicit [11/27/2025-18:31:50] [I] Input inference shape : input_values=8x30000 [11/27/2025-18:31:50] [I] Iterations: 10 [11/27/2025-18:31:50] [I] Duration: 3s (+ 200ms warm up) [11/27/2025-18:31:50] [I] Sleep time: 0ms [11/27/2025-18:31:50] [I] Idle time: 0ms [11/27/2025-18:31:50] [I] Inference Streams: 1 [11/27/2025-18:31:50] [I] ExposeDMA: Disabled [11/27/2025-18:31:50] [I] Data transfers: Enabled [11/27/2025-18:31:50] [I] Spin-wait: Disabled [11/27/2025-18:31:50] [I] Multithreading: Disabled [11/27/2025-18:31:50] [I] CUDA Graph: Disabled [11/27/2025-18:31:50] [I] Separate profiling: Disabled [11/27/2025-18:31:50] [I] Time Deserialize: Disabled [11/27/2025-18:31:50] [I] Time Refit: Disabled [11/27/2025-18:31:50] [I] NVTX verbosity: 0 [11/27/2025-18:31:50] [I] Persistent Cache Ratio: 0 [11/27/2025-18:31:50] [I] Optimization Profile Index: 0 [11/27/2025-18:31:50] [I] Weight Streaming Budget: 100.000000% [11/27/2025-18:31:50] [I] Inputs: [11/27/2025-18:31:50] [I] Debug Tensor Save Destinations: [11/27/2025-18:31:50] [I] Dump All Debug Tensor in Formats: [11/27/2025-18:31:50] [I] === Reporting Options === [11/27/2025-18:31:50] [I] Verbose: Disabled [11/27/2025-18:31:50] [I] Averages: 10 inferences [11/27/2025-18:31:50] [I] Percentiles: 90,95,99 [11/27/2025-18:31:50] [I] Dump refittable layers:Disabled [11/27/2025-18:31:50] [I] Dump output: Disabled [11/27/2025-18:31:50] [I] Profile: Disabled [11/27/2025-18:31:50] [I] Export timing to JSON file: [11/27/2025-18:31:50] [I] Export output to JSON file: [11/27/2025-18:31:50] [I] Export profile to JSON file: [11/27/2025-18:31:50] [I] [11/27/2025-18:31:50] [I] === Device Information === [11/27/2025-18:31:50] [I] Available Devices: [11/27/2025-18:31:50] [I] Device 0: "NVIDIA GeForce RTX 4070 Laptop GPU" UUID: GPU-df57adcc-000e-4631-7e44-f044e50478c6 [11/27/2025-18:31:50] [I] Selected Device: NVIDIA GeForce RTX 4070 Laptop GPU [11/27/2025-18:31:50] [I] Selected Device ID: 0 [11/27/2025-18:31:50] [I] Selected Device UUID: GPU-df57adcc-000e-4631-7e44-f044e50478c6 [11/27/2025-18:31:50] [I] Compute Capability: 8.9 [11/27/2025-18:31:50] [I] SMs: 36 [11/27/2025-18:31:50] [I] Device Global Memory: 8187 MiB [11/27/2025-18:31:50] [I] Shared Memory per SM: 100 KiB [11/27/2025-18:31:50] [I] Memory Bus Width: 128 bits (ECC disabled) [11/27/2025-18:31:50] [I] Application Compute Clock Rate: 1.695 GHz [11/27/2025-18:31:50] [I] Application Memory Clock Rate: 8.001 GHz [11/27/2025-18:31:50] [I] [11/27/2025-18:31:50] [I] Note: The application clock rates do not reflect the actual clock rates that the GPU is currently running at. [11/27/2025-18:31:50] [I] [11/27/2025-18:31:50] [I] TensorRT version: 10.13.3 [11/27/2025-18:31:50] [I] Loading standard plugins [11/27/2025-18:31:50] [I] [TRT] [MemUsageChange] Init CUDA: CPU +1, GPU +0, now: CPU 18935, GPU 1133 (MiB) [11/27/2025-18:31:52] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +2609, GPU +8, now: CPU 21877, GPU 1141 (MiB) [11/27/2025-18:31:52] [I] Start parsing network model. [11/27/2025-18:31:53] [I] [TRT] ---------------------------------------------------------------- [11/27/2025-18:31:53] [I] [TRT] Input filename: D:\Audio2Face\Audio2Face-3D-SDK\_data\audio2emotion-models\audio2emotion-v2.2\network.onnx [11/27/2025-18:31:53] [I] [TRT] ONNX IR version: 0.0.7 [11/27/2025-18:31:53] [I] [TRT] Opset version: 12 [11/27/2025-18:31:53] [I] [TRT] Producer name: pytorch [11/27/2025-18:31:53] [I] [TRT] Producer version: 2.3.1 [11/27/2025-18:31:53] [I] [TRT] Domain: [11/27/2025-18:31:53] [I] [TRT] Model version: 0 [11/27/2025-18:31:53] [I] [TRT] Doc string: [11/27/2025-18:31:53] [I] [TRT] ---------------------------------------------------------------- [11/27/2025-18:31:54] [I] Finished parsing network model. Parse time: 1.58212 [11/27/2025-18:31:54] [I] Set shape of input tensor input_values for optimization profile 0 to: MIN=1x5000 OPT=8x30000 MAX=128x60000 [11/27/2025-18:31:57] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored. [11/27/2025-18:31:57] [I] [TRT] Compiler backend is used during engine build. [11/27/2025-18:31:59] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:31:59] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:31:59] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:31:59] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:00] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:00] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:00] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:00] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:00] [W] [TRT] Tactic Device request: 12004MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:00] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 0 due to insufficient memory on requested size of 12588006912 detected for tactic 0x0000000000000000. [11/27/2025-18:32:00] [E] Error[10]: IBuilder::buildSerializedNetworkToStream: Error Code 10: Internal Error (Could not find any implementation for node {ForeignNode[ONNXTRT_squeezeTensor + /model/wav2vec2/feature_extractor/conv_layers.0/Transpose...ONNXTRT_unsqueezeTensor_21]}. In nvinfer1::builder::cgraph::LeafCNode::computeCosts at optimizer/common/tactic/optimizer.cpp:4115) [11/27/2025-18:32:00] [I] Created engine with size: 0 MiB [11/27/2025-18:32:00] [I] Engine built in 6.63394 sec. [11/27/2025-18:32:00] [E] Assertion failure: false && "Attempting to access an empty engine!" Generating benchmark data... Generating 5 minutes audio 1/10 Generating 5 minutes audio 2/10 Generating 5 minutes audio 3/10 Generating 5 minutes audio 4/10 Generating 5 minutes audio 5/10 Generating 5 minutes audio 6/10 Generating 5 minutes audio 7/10 Generating 5 minutes audio 8/10 Generating 5 minutes audio 9/10 Generating 5 minutes audio 10/10 &&&& RUNNING TensorRT.trtexec [TensorRT v101303] [b9] # trtexec --onnx=D:\Audio2Face\Audio2Face-3D-SDK\_data\audio2emotion-models\audio2emotion-v2.2\network.onnx --saveEngine=C:\Users\Administrator\AppData\Local\Temp\tmp7bwtotg_\network.trt --device=0 --minShapes=input_values:1x5000 --maxShapes=input_values:128x60000 --optShapes=input_values:8x30000 [11/27/2025-18:32:03] [I] TF32 is enabled by default. Add --noTF32 flag to further improve accuracy with some performance cost. [11/27/2025-18:32:03] [I] === Model Options === [11/27/2025-18:32:03] [I] Format: ONNX [11/27/2025-18:32:03] [I] Model: D:\Audio2Face\Audio2Face-3D-SDK\_data\audio2emotion-models\audio2emotion-v2.2\network.onnx [11/27/2025-18:32:03] [I] Output: [11/27/2025-18:32:03] [I] === Build Options === [11/27/2025-18:32:03] [I] Memory Pools: workspace: default, dlaSRAM: default, dlaLocalDRAM: default, dlaGlobalDRAM: default, tacticSharedMem: default [11/27/2025-18:32:03] [I] avgTiming: 8 [11/27/2025-18:32:03] [I] Precision: FP32 [11/27/2025-18:32:03] [I] LayerPrecisions: [11/27/2025-18:32:03] [I] Layer Device Types: [11/27/2025-18:32:03] [I] Calibration: [11/27/2025-18:32:03] [I] Refit: Disabled [11/27/2025-18:32:03] [I] Strip weights: Disabled [11/27/2025-18:32:03] [I] Version Compatible: Disabled [11/27/2025-18:32:03] [I] ONNX Plugin InstanceNorm: Disabled [11/27/2025-18:32:03] [I] ONNX kENABLE_UINT8_AND_ASYMMETRIC_QUANTIZATION_DLA flag: Disabled [11/27/2025-18:32:03] [I] TensorRT runtime: full [11/27/2025-18:32:03] [I] Lean DLL Path: [11/27/2025-18:32:03] [I] Tempfile Controls: { in_memory: allow, temporary: allow } [11/27/2025-18:32:03] [I] Exclude Lean Runtime: Disabled [11/27/2025-18:32:03] [I] Sparsity: Disabled [11/27/2025-18:32:03] [I] Safe mode: Disabled [11/27/2025-18:32:03] [I] Build DLA standalone loadable: Disabled [11/27/2025-18:32:03] [I] Allow GPU fallback for DLA: Disabled [11/27/2025-18:32:03] [I] DirectIO mode: Disabled [11/27/2025-18:32:03] [I] Restricted mode: Disabled [11/27/2025-18:32:03] [I] Skip inference: Disabled [11/27/2025-18:32:03] [I] Save engine: C:\Users\Administrator\AppData\Local\Temp\tmp7bwtotg_\network.trt [11/27/2025-18:32:03] [I] Load engine: [11/27/2025-18:32:03] [I] Profiling verbosity: 0 [11/27/2025-18:32:03] [I] Tactic sources: Using default tactic sources [11/27/2025-18:32:03] [I] timingCacheMode: local [11/27/2025-18:32:03] [I] timingCacheFile: [11/27/2025-18:32:03] [I] Enable Compilation Cache: Enabled [11/27/2025-18:32:03] [I] Enable Monitor Memory: Disabled [11/27/2025-18:32:03] [I] errorOnTimingCacheMiss: Disabled [11/27/2025-18:32:03] [I] Preview Features: Use default preview flags. [11/27/2025-18:32:03] [I] MaxAuxStreams: -1 [11/27/2025-18:32:03] [I] BuilderOptimizationLevel: -1 [11/27/2025-18:32:03] [I] MaxTactics: -1 [11/27/2025-18:32:03] [I] Calibration Profile Index: 0 [11/27/2025-18:32:03] [I] Weight Streaming: Disabled [11/27/2025-18:32:03] [I] Runtime Platform: Same As Build [11/27/2025-18:32:03] [I] Debug Tensors: [11/27/2025-18:32:03] [I] Distributive Independence: Disabled [11/27/2025-18:32:03] [I] Mark Unfused Tensors As Debug Tensors: Disabled [11/27/2025-18:32:03] [I] Input(s)s format: fp32:CHW [11/27/2025-18:32:03] [I] Output(s)s format: fp32:CHW [11/27/2025-18:32:03] [I] Input build shape (profile 0): input_values=1x5000+8x30000+128x60000 [11/27/2025-18:32:03] [I] Input calibration shapes: model [11/27/2025-18:32:03] [I] === System Options === [11/27/2025-18:32:03] [I] Device: 0 [11/27/2025-18:32:03] [I] DLACore: [11/27/2025-18:32:03] [I] Plugins: [11/27/2025-18:32:03] [I] setPluginsToSerialize: [11/27/2025-18:32:03] [I] dynamicPlugins: [11/27/2025-18:32:03] [I] ignoreParsedPluginLibs: 0 [11/27/2025-18:32:03] [I] [11/27/2025-18:32:03] [I] === Inference Options === [11/27/2025-18:32:03] [I] Batch: Explicit [11/27/2025-18:32:03] [I] Input inference shape : input_values=8x30000 [11/27/2025-18:32:03] [I] Iterations: 10 [11/27/2025-18:32:03] [I] Duration: 3s (+ 200ms warm up) [11/27/2025-18:32:03] [I] Sleep time: 0ms [11/27/2025-18:32:03] [I] Idle time: 0ms [11/27/2025-18:32:03] [I] Inference Streams: 1 [11/27/2025-18:32:03] [I] ExposeDMA: Disabled [11/27/2025-18:32:03] [I] Data transfers: Enabled [11/27/2025-18:32:03] [I] Spin-wait: Disabled [11/27/2025-18:32:03] [I] Multithreading: Disabled [11/27/2025-18:32:03] [I] CUDA Graph: Disabled [11/27/2025-18:32:03] [I] Separate profiling: Disabled [11/27/2025-18:32:03] [I] Time Deserialize: Disabled [11/27/2025-18:32:03] [I] Time Refit: Disabled [11/27/2025-18:32:03] [I] NVTX verbosity: 0 [11/27/2025-18:32:03] [I] Persistent Cache Ratio: 0 [11/27/2025-18:32:03] [I] Optimization Profile Index: 0 [11/27/2025-18:32:03] [I] Weight Streaming Budget: 100.000000% [11/27/2025-18:32:03] [I] Inputs: [11/27/2025-18:32:03] [I] Debug Tensor Save Destinations: [11/27/2025-18:32:03] [I] Dump All Debug Tensor in Formats: [11/27/2025-18:32:03] [I] === Reporting Options === [11/27/2025-18:32:03] [I] Verbose: Disabled [11/27/2025-18:32:03] [I] Averages: 10 inferences [11/27/2025-18:32:03] [I] Percentiles: 90,95,99 [11/27/2025-18:32:03] [I] Dump refittable layers:Disabled [11/27/2025-18:32:03] [I] Dump output: Disabled [11/27/2025-18:32:03] [I] Profile: Disabled [11/27/2025-18:32:03] [I] Export timing to JSON file: [11/27/2025-18:32:03] [I] Export output to JSON file: [11/27/2025-18:32:03] [I] Export profile to JSON file: [11/27/2025-18:32:03] [I] [11/27/2025-18:32:03] [I] === Device Information === [11/27/2025-18:32:03] [I] Available Devices: [11/27/2025-18:32:03] [I] Device 0: "NVIDIA GeForce RTX 4070 Laptop GPU" UUID: GPU-df57adcc-000e-4631-7e44-f044e50478c6 [11/27/2025-18:32:03] [I] Selected Device: NVIDIA GeForce RTX 4070 Laptop GPU [11/27/2025-18:32:03] [I] Selected Device ID: 0 [11/27/2025-18:32:03] [I] Selected Device UUID: GPU-df57adcc-000e-4631-7e44-f044e50478c6 [11/27/2025-18:32:03] [I] Compute Capability: 8.9 [11/27/2025-18:32:03] [I] SMs: 36 [11/27/2025-18:32:03] [I] Device Global Memory: 8187 MiB [11/27/2025-18:32:03] [I] Shared Memory per SM: 100 KiB [11/27/2025-18:32:03] [I] Memory Bus Width: 128 bits (ECC disabled) [11/27/2025-18:32:03] [I] Application Compute Clock Rate: 1.695 GHz [11/27/2025-18:32:03] [I] Application Memory Clock Rate: 8.001 GHz [11/27/2025-18:32:03] [I] [11/27/2025-18:32:03] [I] Note: The application clock rates do not reflect the actual clock rates that the GPU is currently running at. [11/27/2025-18:32:03] [I] [11/27/2025-18:32:03] [I] TensorRT version: 10.13.3 [11/27/2025-18:32:03] [I] Loading standard plugins [11/27/2025-18:32:03] [I] [TRT] [MemUsageChange] Init CUDA: CPU +5, GPU +0, now: CPU 19468, GPU 1133 (MiB) [11/27/2025-18:32:05] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +2273, GPU +8, now: CPU 22044, GPU 1141 (MiB) [11/27/2025-18:32:05] [I] Start parsing network model. [11/27/2025-18:32:07] [I] [TRT] ---------------------------------------------------------------- [11/27/2025-18:32:07] [I] [TRT] Input filename: D:\Audio2Face\Audio2Face-3D-SDK\_data\audio2emotion-models\audio2emotion-v2.2\network.onnx [11/27/2025-18:32:07] [I] [TRT] ONNX IR version: 0.0.7 [11/27/2025-18:32:07] [I] [TRT] Opset version: 12 [11/27/2025-18:32:07] [I] [TRT] Producer name: pytorch [11/27/2025-18:32:07] [I] [TRT] Producer version: 2.3.1 [11/27/2025-18:32:07] [I] [TRT] Domain: [11/27/2025-18:32:07] [I] [TRT] Model version: 0 [11/27/2025-18:32:07] [I] [TRT] Doc string: [11/27/2025-18:32:07] [I] [TRT] ---------------------------------------------------------------- [11/27/2025-18:32:07] [I] Finished parsing network model. Parse time: 1.56135 [11/27/2025-18:32:07] [I] Set shape of input tensor input_values for optimization profile 0 to: MIN=1x5000 OPT=8x30000 MAX=128x60000 [11/27/2025-18:32:10] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored. [11/27/2025-18:32:10] [I] [TRT] Compiler backend is used during engine build. [11/27/2025-18:32:13] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:13] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:13] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:13] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:13] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:13] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:13] [W] [TRT] Tactic Device request: 8999MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:13] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 1 due to insufficient memory on requested size of 9436397568 detected for tactic 0x0000000000000001. [11/27/2025-18:32:14] [W] [TRT] Tactic Device request: 12004MB Available: 8187MB. Device memory is insufficient to use tactic. [11/27/2025-18:32:14] [W] [TRT] UNSUPPORTED_STATE: Skipping tactic 0 due to insufficient memory on requested size of 12588006912 detected for tactic 0x0000000000000000. [11/27/2025-18:32:14] [E] Error[10]: IBuilder::buildSerializedNetworkToStream: Error Code 10: Internal Error (Could not find any implementation for node {ForeignNode[ONNXTRT_squeezeTensor + /model/wav2vec2/feature_extractor/conv_layers.0/Transpose...ONNXTRT_unsqueezeTensor_21]}. In nvinfer1::builder::cgraph::LeafCNode::computeCosts at optimizer/common/tactic/optimizer.cpp:4115) [11/27/2025-18:32:14] [I] Created engine with size: 0 MiB [11/27/2025-18:32:14] [I] Engine built in 6.72844 sec. [11/27/2025-18:32:14] [E] Assertion failure: false && "Attempting to access an empty engine!" 如何解决
最新发布
11-28
dynamic_axes={'input': {0: 'batch'}}, # 仅允许 batch 维度动态 opset_version=13 # 必须 ≥13 ) ``` 3. **替换非常规算子** 对于错误中的 `Conv/Transpose/Squeeze` 组合： - 检查模型中是否有自定义 ...
TensorRT onnx转engine报Assertion failed: dims.nbDims == 4 || dims.nbDims == 5
修炼之路
08-27 2719
错误信息 在使用TensorRT将onnx转为engine的时候报错，错误信息如下 [08/27/2021-15:27:08] [I] === Model Options === [08/27/2021-15:27:08] [I] Format: ONNX [08/27/2021-15:27:08] [I] Model: glintr100.onnx [08/27/2021-15:27:08] [I] Output: [08/27/2021-15:27:08] [I] === Build Options =
使用TensorRT遇到的小问题
qq_41151162的博客
07-14 2858
1、使用TensorRT生成.engine文件时报错： TensorRT was linked against cuBLAS/cuBLAS LT 11.3.0 but loaded cuBLAS/cuBLAS LT 11.2.0 问题原因：CUDA版本不对 解决：使用的TensorRT版本为7.2.1.6支持的CUDA版本为11.0 update1即11.0.3，安装11.0.3即可 2、本机GPU为gtx 960M，目标机器GPU为P2000，程序在目标机器运行报错： [E] [TRT] ../
TensorRT报错的一百种姿势 | 【❤️TensorRT抱错砖家❤️】
专注于AI领域前沿技术学习与分享：目标检测、图像修复、超分重建、AI工程化
09-28 3969
本博文致力于收集❤️打怪升级过程中❤️遇到的一百种 tensorRT 环境报错 这里的一些报错处理，有的可能没有一步到位 但是，在初次遇到时，必然是当时视角下，最高效排bug的一种方式 文章目录❤️ ModuleNotFoundError: No module named 'tensorrt'???? AttributeError: ‘tensorrt.tensorrt.Builder‘ object has no attribute ‘max_workspace_size‘???? Import
NVIDIA Xavier TensorRT TensorRTX 生成engine时报错
qq_44942565的博客
04-14 5133
NVIDIA Xavier TensorRT TensorRTX 生成engine时报错
TensorRT生成engine文件异常（[TRT] Network::addScale::434, condition: shift.count ＞ 0 ?)
恐龙先生的博客
05-10 5520
Loading weights: ./best.wts [05/10/2022-08:41:14] [E] [TRT] Parameter check failed at: Network.cpp::nvinfer1::Network::addScale::434, condition: shift.count > 0 ? (shift.values != nullptr) : (shift.values == nullptr) 这个错误是wts转engine文件时的报错，在网上找了好久也没有找到.
TensorRT报错1 [caskUtils.cpptrtSmToCask147] Error Code 1 Internal Error (Unsupported SM 0x809)
qq_51366133的博客
04-08 707
TensorRT报错1 [caskUtils.cpptrtSmToCask147] Error Code 1 Internal Error (Unsupported SM 0x809)
AssertionError: Gather function not implemented for CPU tensors 错误解决
qq_41018465的博客
07-04 2599
pytorch模型训练时的一些错误解决
suma++ TensorRT Not Found
weixin_58045467的博客
08-28 1316
TensorRT-8.2.1.8Driver Version: 470.141.03cuda_11.4 cudnn8.2.4ubuntu20.04在復現過程中，這個問題折磨我好久，在github上的issue也有兩個這樣的問題，但是對我來說都不能work，而且在csdn，google中搜索時發現suma++相關介紹挺多的，但是復現或者出現問題想搜索來解決卻沒有什麼參考的，所以寫這篇文章提供一些信息，可能會節約困擾中的你許多時間。1.剛開始，在師兄的幫助下，我們把cmakelist中的TensorRT Fou
TensorRT开发环境搭建
hello_dear_you的博客
08-22 1342
TensorRT的C++开发环境搭建
解决occcont.cpp报错：缺失OCX控件的解决方案
在分析给定文件信息的过程中，我们首先应关注标题中提到的“occcont.cpp报错”，这通常表示在C++开发环境中的一个源代码文件（occcont.cpp）在编译或运行时遇到了问题。具体错误信息是“Debug Assertion Failed”，...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值