Model¶

A model represents all the information about a Machine Learning model.

Model is represented by a dict with these common fields:

Name - the only required field, with the name of the model.
Metadata - a dict of Metadata, or a link to a file (yml/json) containing it.
Results - a list of Result dicts.
Paper - URL to the paper, or a dict with paper title, url and
Code - link to code snippet snippet
Weights - link to download the pretrained weights
Config - link to the config file used for training
README - the content of, or a link to the README.md file for the model
Image - path or URL to an image for this model
In Collection- name of the Collection to which this model belongs.

The fields above will be automatically recognized by model-index, but you can also add any number of custom fields to it.

All field names are case-insensitive.

Metadata¶

Metadata is a dict of common and custom metadata. The common fields are:

FLOPs - The number of FLOPs of the model
Parameters - The total number of parameters of the model
Epochs - Number of training epochs
Batch Size - Input batch size
Training Data - Names of dataset on which the models is trained on
Training Techniques - A list of training techniques (for the full list see Methods on Papers with Code)
Training Resources - The hardware used for training
Training Time - How many hours or days it takes to train.
Architecture - A list of architectural features of the model (for the full list see Methods on Papers with Code)

You can also add any other custom field that is specific to your model.

Result¶

Result is a dict capturing the evaluation results of the model. It has these fields:

Task - Name of the task (for full see Benchmarks on Papers with Code)
Dataset - Name of the dataset (for full see Datasets on Papers with Code)
Metrics - a list of dictionaries with metrics. For relevant metrics consult the see Benchmarks on Papers with Code.

A full example¶

An example of the full model dict is shown below:

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 
Config: configs/inception-v3-config.json
README: docs/inception-v3-readme.md