rhysdg/robonnx

An onnx/TensorRT implementation of Cardiff NLP's emotion detection RoBERTa model

Robonnx

An onnx/TensorRT implementation of Cardiff NLP's emotion detection RoBERTa model
Explore the docs »



Report Bug · Request Feature

Table of Contents

• About the Project
  • Built With
• Getting Started
  • Prerequisites
  • The Story so Far
  • Notebooks
  • Scripts and Tools
  • Supplementary Data
• Proposed Updates
• License
• Contact

About The Project

Built With

• Huggingface
• ONNX
• Docker

Getting Started

Prerequisites

Coming Soon

The Story So Far

Notebooks

1. Coming soon

Tools and Scripts

1. Coming soon

Testing

Coming soon

Supplementary Data

Coming soon

Future updates

Coming Soon

License

subject to change following discussion

Distributed under the Apache License. See LICENSE for more information.

Contact

• Project Link: https://github.com/rhysdg/robonnx
• Email: Rhys

robonnx's Issues

Add instance wide verbosity

• Add the ability to flag the verbosity of OnnxSession on and off. Particularly handy with regards to standard unit testing.

Update ONNX providers for multiple contexts

• Add CPUExecutionProvide fallback within roberta_onnx
• Currently passing testing within ubuntu-latest runner via our CI mechanism but fails with a clean CPU backed EC2 instance

Hello!

• For all the new students heading through the door check the issues section here, a one stop shop to debug, assign changes and track enhancements.

Add tensorrt implementation

• onnx IR converts successfully with trtexec as an initial inisight. A pycuda implementation would be great!
• Tracks at about 4ms on a AGX XAvier right now although this is likely not a useful model for an embedded system. RTX 2060, V100 benches on their way.

Populate README.md

• update current readme with appropriate info