Hi, I was trying to pre-compile some tensorRT engines with <code cla

how to get `trtexec`? about apollo HOT 1 OPEN

josh-wende commented on June 4, 2024

how to get `trtexec`?

from apollo.

Comments (1)

daohu527 commented on June 4, 2024

Apollo use tensorRT API to build engine.trtexec is not required.

apollo/modules/perception/common/inference/tensorrt/rt_net.cc

Lines 912 to 926 in 3859bee

 // serialize trt engine the first time 

 nvinfer1::ICudaEngine *engine = nullptr; 

 auto trt_cache_path = model_root_ + "/TRTengine.cache"; 

 if (!LoadCache(trt_cache_path)) { 

 engine = builder_->buildEngineWithConfig(*network_, *builder_config_); 

 if (engine->serialize()) { 

 nvinfer1::IHostMemory *serialized_model(engine->serialize()); 

 std::ofstream f(trt_cache_path, std::ios::binary); 

 f.write(reinterpret_cast<char *>(serialized_model->data()), 

 serialized_model->size()); 

 serialized_model->destroy(); 

 AERROR << "Saving serialized model file to " << trt_cache_path; 

 }

install tensorRT

If you must use trtexec, you need to install it manually. link

from apollo.

how to get `trtexec`? about apollo HOT 1 OPEN

Comments (1)

install tensorRT

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	// serialize trt engine the first time
	nvinfer1::ICudaEngine *engine = nullptr;
	auto trt_cache_path = model_root_ + "/TRTengine.cache";
	if (!LoadCache(trt_cache_path)) {
	engine = builder_->buildEngineWithConfig(network_, builder_config_);

	if (engine->serialize()) {
	nvinfer1::IHostMemory *serialized_model(engine->serialize());
	std::ofstream f(trt_cache_path, std::ios::binary);

	f.write(reinterpret_cast<char *>(serialized_model->data()),
	serialized_model->size());
	serialized_model->destroy();
	AERROR << "Saving serialized model file to " << trt_cache_path;
	}