Add a new model

Adding a new model to the benchmark harness is very simple. Please perform the following steps:

• Save the model in some space that can be accessed from the internet (e.g. AWS).
  • The models can also be saved in the benchmarking repository, but it is not recommended.
• Write up a json file describing how you plan to run the model. The json files are organized based on the framework the model uses. For caffe2 models, you can use specification/models/caffe2/shufflenet/shufflenet.json as an example.
  • You do not need to worry about the MD5 hash of the files you specify. They will be populated automatically when you run the test locally.
• Test your newly created benchmark file locally.
• Add your json file to specification/models/<framework>/<model name>/<benchmark name>.json
• Submit a PR.