Tensorrt: Cannot download ngc models in demo/BERT at release/7.2

[david-macleod]

Description

When following the instructions to reproduce the benchmarks in the demo/BERT readme I get stuck when trying to download models from NGC. For example:

bash scripts/download_model.sh large fp16 384 v2
Error: nvidia/bert_tf_v2_large_fp16_384:2 could not be found.

This error occurs for all configurations I have tried, including the defaults. I have also tried browsing NGC online catalogue to find the models manually but I cannot find the tags the download_model.sh script is searching for.

Thanks

Environment

TensorRT Version: 7.2.1
GPU Type: T4
Nvidia Driver Version: 455.32.00
CUDA Version: 11.1
CUDNN Version: 8.06
Operating System + Version: Ubuntu 18.04
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag): demo/BERT container

[mk-nvidia]

@david-macleod Yes it's a bug on our side, we're working to re-upload these checkpoints soon. Sorry for the inconvenience.

[david-macleod]

Thanks for the quick reply, is there any sort of estimated time frame for this?

[samringer]

I'm also seeing this exact error as well!

[willsq]

Me too - thanks for the reply, is there a blocker on this that might affect timescales?

[rajeevsrao]

Sorry for the inconvenience. Some of our NGC checkpoints were deleted - we will upload the new checkpoints by tomorrow (11/4) morning. I will post an update here as soon as that happens.

[rajeevsrao]

@david-macleod @samringer @willsq FYI fix posted to master and release/7.2 branches today:
https://github.com/NVIDIA/TensorRT/commit/f1d38b5456e15e2683a4513ea3cb14ed6d94cf53