Question about dataset and training

#1
by vvvu - opened

The dataset is split into facts and props/proofs. How did was each training example constructed? Was all the facts for a particular file added to context, and then the associated prop/proof? Was the facts even used during training?

vvvu changed discussion title from Dataset used to Question about dataset and training
Owner

Both parts (facts and props/poofs) were used during fine-tuning.
The used format of the data:

### Context: Filename: {filename} Imports {imports}
{proposition}
{proof}

and

### Context: Filename: {filename} Imports {imports}
{fact}

Sign up or log in to comment