Closed vvxt closed 7 months ago
Increase log level to verbose to understand why it's failing.
Log with set severity to kVERBOSE @cyrusbehr
Hi @vvxt Can you send me the model please.
@vvxt please also specify what Options
you are using. Are you using default? Or did you change them?
// Specify our GPU inference configuration options
Options options;
// Specify what precision to use for inference
// FP16 is approximately twice as fast as FP32.
options.precision = Precision::FP16;
// If using INT8 precision, must specify path to directory containing calibration data.
options.calibrationDataDirectoryPath = "";
// If the model does not support dynamic batch size, then the below two parameters must be set to 1.
// Specify the batch size to optimize for.
options.optBatchSize = 1;
// Specify the maximum batch size we plan on running.
options.maxBatchSize = 1;
Engine engine(options);
Closing due to inactivity. Please rep-open once you can provide the above information.
envs: windows 11 NVIDIA GeForce RTX3050 Laptop vs2022
export ONNX: format=onnx dynamic=True simplify=True
Options: precision=Precision::FP16 optBatchSize=32 maxBatchSize=64
but use trtexec.exe can work properly.