# Error Code: 9: Skipping tactic 0x0000000000000000 due to exception No Myelin Error exists **URL:** \ **Category:** TensorRT\ **Tags:** tensorrt, cudnn, ubuntu, python\ **Created:** [September 13, 2024, 9:50am UTC](https://forums.developer.nvidia.com/t/error-code-9-skipping-tactic-0x0000000000000000-due-to-exception-no-myelin-error-exists/306682 "2024-09-13T09:50:30Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![abhinavpavi](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@abhinavpavi](https://forums.developer.nvidia.com/u/abhinavpavi)\ **Post date:** [September 13, 2024, 9:50am UTC](https://forums.developer.nvidia.com/t/error-code-9-skipping-tactic-0x0000000000000000-due-to-exception-no-myelin-error-exists/306682/1 "2024-09-13T09:50:30Z")
## Description Getting this error while trying to convert ONNX model to TensorRT engine. The original model is a table transformer PyTorch model which is converted to ONNX. Initially I converted PyTorch model to ONNX without dynamic axes and the conversion to TensorRT engine worked fine. After adding the dynamic axes parameter I get this error ## Environment **TensorRT Version** : 10.1.0 **GPU Type** : Nvidia T4 **Nvidia Driver Version** : 535 **CUDA Version** : 12.2 **CUDNN Version** : 8.9.7 **Operating System + Version** : Ubuntu 22.04 **Python Version (if applicable)**: 3.10.12 **PyTorch Version (if applicable)**: 2.1.0+cu121 **Baremetal or Container (if container which image + tag)**: ## PyTorch model to ONNX conversion ```auto import torch import sys import os class NomNom: def __init__ (self, torch_model): self.torch_model = torch_model.eval() def tatr_onnx(self): dummy = self.tatr_inputs() dynamic_axes = { "samples": {1: "width", 2: "height"} } torch.onnx.export(self.torch_model, dummy, "TATR_v3.onnx", export_params=True, input_names=['samples'], output_names=['out'], dynamic_axes=dynamic_axes ) # export_options = torch.onnx.ExportOptions(dynamic_shapes=True) # torch.onnx.dynamo_export(self.torch_model, dummy, export_options=export_options).save("TATR.onnx") def tatr_inputs(self, W=800, H=1000): tensor = torch.randn(3, W, H) return [tensor] def change_cwd(): abspath = os.path.abspath( __file__ ) dname = os.path.dirname(abspath) os.chdir(dname) def TATR_optimizer(): sys.path.append("src") from src.model_loader import model torch_model = model() change_cwd() trial = NomNom(torch_model) trial.tatr_onnx() # onnx_model = onnx.load("TATR.onnx") # onnx.checker.check_model(onnx_model) if __name__ ==" __main__": TATR_optimizer() ``` ## ONNX to TensorRT conversion ```auto def build_engine_onnx(model_file): TRT_LOGGER = trt.Logger(trt.Logger.WARNING) builder = trt.Builder(TRT_LOGGER) network = builder.create_network(0) config = builder.create_builder_config() profile = builder.create_optimization_profile(); profile.set_shape("samples", (3, 80, 80), (3, 800, 1000), (3, 1000, 1000)) config.add_optimization_profile(profile) parser = trt.OnnxParser(network, TRT_LOGGER) config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, common.GiB(1)) # Load the Onnx model and parse it in order to populate the TensorRT network. with open(model_file, "rb") as model: if not parser.parse(model.read()): print("ERROR: Failed to parse the ONNX file.") for error in range(parser.num_errors): print(parser.get_error(error)) return None engine_bytes = builder.build_serialized_network(network, config) runtime = trt.Runtime(TRT_LOGGER) return runtime.deserialize_cuda_engine(engine_bytes) ``` ## Output ```auto [09/13/2024-09:50:09] [TRT] [E] Error Code: 9: Skipping tactic 0x0000000000000000 due to exception No Myelin Error exists [09/13/2024-09:50:09] [TRT] [E] Error Code: 10: Could not find any implementation for node {ForeignNode[/Cast_10.../backbone/backbone.0/Cast]}. [09/13/2024-09:50:09] [TRT] [E] IBuilder::buildSerializedNetwork: Error Code 10: Internal Error (Could not find any implementation for node {ForeignNode[/Cast_10.../backbone/backbone.0/Cast]}.) Traceback (most recent call last): File "/home/smartops/Desktop/table-transformer-trt/TRTNvidia.py", line 168, in trt_inference_wrapper = TRTInference(trt_engine_path, File "/home/smartops/Desktop/table-transformer-trt/TRTNvidia.py", line 23, in __init__ engine = self.build_engine_onnx(trt_engine_path) File "/home/smartops/Desktop/table-transformer-trt/TRTNvidia.py", line 91, in build_engine_onnx return runtime.deserialize_cuda_engine(engine_bytes) TypeError: deserialize_cuda_engine(): incompatible function arguments. The following argument types are supported: 1. (self: tensorrt.tensorrt.Runtime, serialized_engine: buffer) -> tensorrt.tensorrt.ICudaEngine 2. (self: tensorrt.tensorrt.Runtime, stream_reader: tensorrt.tensorrt.IStreamReader) -> tensorrt.tensorrt.ICudaEngine Invoked with: , None ------------------------------------------------------------------- PyCUDA ERROR: The context stack was not empty upon module cleanup. ------------------------------------------------------------------- A context was still active when the context stack was being cleaned up. At this point in our execution, CUDA may already have been deinitialized, so there is no way we can finish cleanly. The program will be aborted now. Use Context.pop() to avoid this problem. ------------------------------------------------------------------- Aborted (core dumped) ``` ---
**Author:** ![AakankshaS](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/aakankshas/32/14047_2.png) [@AakankshaS](https://forums.developer.nvidia.com/u/AakankshaS)\ **Post date:** [September 30, 2024, 8:00pm UTC](https://forums.developer.nvidia.com/t/error-code-9-skipping-tactic-0x0000000000000000-due-to-exception-no-myelin-error-exists/306682/2 "2024-09-30T20:00:39Z")
Have you had teh custom plugin implementation in place? ---
**Author:** ![abhinavpavi](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@abhinavpavi](https://forums.developer.nvidia.com/u/abhinavpavi)\ **Post date:** [October 2, 2024, 1:46pm UTC](https://forums.developer.nvidia.com/t/error-code-9-skipping-tactic-0x0000000000000000-due-to-exception-no-myelin-error-exists/306682/3 "2024-10-02T13:46:12Z")
I dont think so. However when I made the image size static the conversion worked fine