Investigate using smaller context length models for model card creation

[j2whiting]

Now that we have gathered some baseline results from using GPT4 to create model cards for ~775 different models, we should look into using a cheaper model for doing this.

TODO:

Scope techniques for parsing out model card fields from a research paper using smaller context lengths (i.e., 4096).