Dugan et al. (2020)#

Publication#

RoFT: A Tool for Evaluating Human Detection of Machine-Generated Text

Repositories#

https://github.com/kirubarajan/roft

Available Models#

  • GPT2-XL Recipe Generation Model

    • Description: A GPT2-XL model fine-tuned on Recipe1M+ dataset, available at gs://roft_saved_models/gpt2-xl_recipes.tar.gz

    • Name: dugan2020-roft-recipe

    • Usage:

      from repro.models.dugan2020 import RoFTRecipeGenerator
name = "Redwood Room Apple Pie"
ingredients = [
  "1 tablespoon cornstarch",
  "12 cup sugar",
  "14 cup cream",
  "1 tablespoon lemon juice",
  "3 tablespoons butter",
  "20 ounces apples, slices",
  "9 inches pie crusts, baked",
  "8 ounces cream cheese",
  "13 cup sugar",
  "1 egg",
  "12 cup coconut",
  "12 cup walnuts, chopped"
]
model = RoFTRecipeGenerator()
recipe = model.predict(name, ingredients)

Implementation Notes#

  • Although you can set the random seed, the results on the CPU versus the GPU may be different.

Docker Information#

  • Image name: dugan2020

  • Build command:

    repro setup dugan2020 [--silent]

  • Requires network: No

Testing#

Explain how to run the unittests for this model

repro setup dugan2020 
pytest models/dugan2020/tests

Status#

  • [x] Regression unit tests pass
    The model takes up too much memory to run on Github. See here

  • [ ] Correctness unit tests pass
    No expected outputs provided by the original code

  • [ ] Model runs on full test dataset
    Not tested

  • [ ] Predictions approximately replicate results reported in the paper
    Not tested

  • [ ] Predictions exactly replicate results reported in the paper
    Not tested