How it works

There is a root file for the model index: model-index.yml that contains (or links to) all the metadata in a consistent format. An example with a single model:

Models:
  - Name: Inception v3
    Metadata:
      FLOPs: 5731284192
      Parameters: 23834568
      Epochs: 90
      Batch Size: 32
      Training Data: ImageNet  
      Training Techniques: 
        - RMSProp
        - Weight Decay
        - Gradient Clipping
        - Label Smoothing
      Training Resources: 8x V100 GPUs
      Training Time: 24 hours
      Architecture:
        - Auxiliary Classifier
        - Inception-v3 Module
    Results:
      - Task: Image Classification
        Dataset: ImageNet
        Metrics:
          Top 1 Accuracy: 74.67%
          Top 5 Accuracy: 92.1%
    Paper: https://arxiv.org/abs/1512.00567v3
    Code: https://github.com/rwightman/pytorch-image-models/blob/timm/models/inception_v3.py#L442
    Weights: https://download.pytorch.org/models/inception_v3_google-1a9a5a14.pth 
    README: docs/inception-v3-readme.md

Every field except for Name is optional.

The fields you can see above are common fields that are automatically recognized by Papers with Code and enable comparison across different models. You can also add any number of custom fields that are specific to your model or library. For example:

Models:
  - Name: My new model
    Metadata:
      Training Time: 24 hours
      my parameter: 120
      my parameter2: 
        sub parameter1: value 1
        sub parameter2: value 2
      
      

So you can mix-and-match from our set of common field and any other field you want to add.

We recommend putting the model-index.yml file in the root of your repository (so that relative links such as docs/inception-v3-readme.md are easier to write), but you can also put it anywhere else in the repository (e.g. int your docs/ or models/ folder).

Storing metadata in markdown files

Metadata can also be directly stored in model’s README file. For example in this docs/rexnet.md file:

<!--
Type: model-index
Name: RexNet
Metadata: 
  Epochs: 400
  Batch Size: 512
Paper: https://arxiv.org/abs/2007.00992v1
-->

# Summary

Rank Expansion Networks (ReXNets) follow a set of new design 
principles for designing bottlenecks in image classification models.

## Usage

import timm
m = timm.create_model('rexnet_100', pretrained=True)
m.eval()

In this case, you just need to include this markdown file into the global model-index.yml file:

Models:
  - docs/rexnet.md