Build engine failure of TensorRT 10.0 when running trtexec with fp16 on GPU NVIDIA 3060 series

[roxanacincan]

Description

I tried building a TRT engine using trtexec and fp16 precision, but the process fails without any errors. The TRT model is never generated. For fp32 the TRT model is correctly generated.

Environment

TensorRT Version: 10.0.0.6EA

NVIDIA GPU: NVIDIA GeForce RTX 3060

NVIDIA Driver Version: 551.86

CUDA Version: 11.8

CUDNN Version: 8.9.7

Operating System: Windows10

Relevant Files

Model link: https://drive.google.com/file/d/1JjA_9Ea4oTf-jnn41pYOMolpWOGYoWay/view?usp=drive_link
Build log: build.log

Steps To Reproduce

Commands or scripts: trtexec --onnx=best_model.onnx --saveEngine=best_model_fp16.plan --fp16 --verbose
I also tried using the build_engine.py script from the python samples directory of TensorRT10 but it returns exactly the same output as trtexec. No errors or warnings, just the process stopping abruptly.

Have you tried the latest release?: yes

Can this model run on other frameworks? For example run ONNX model with ONNXRuntime (polygraphy run <model.onnx> --onnxrt): yes it can run with Polygraphy

[RajUpadhyay]

Can you trying running the command with additional option of --useCudaGraph.
trtexec --onnx=model.onnx --saveEngine=engine_fp16.engine --fp16 --useCudaGraph --verbose

[roxanacincan]

i tried using the option --useCudaGraph, but there were no changes.
I was still not able to build the model even though i don't get any errors.
Also, the build log file looks the same as the previous one i uploaded: build_with_useCudaGraph.log

[zerollzeng]

Just requested access to the model.

[zerollzeng]

Thanks for reporting this, filed internal bug 4609232 to track.

[akhilg-nv]

The bug for this issue has been fixed and should be included in the latest minor release, closing this issue.