The Model() Class

Model(model_type, model_path)

The base class for a text generation model.

Parameters:

  • model_type (string): Optional. The type of model to use. Defaults to GPT-NEO.

  • model_path (string): Optional. The path to the model to use. Defaults to EleutherAI/gpt-neo-125m, but can be any model from the HuggingFace model hub, or a path to a local folder containing the model's files.

Returns:

A Model object.

Attributes

Model.name

The name of the model.

Type: string

Model.path

The location where the model's files are stored once downloaded. Defaults to [Python package location]/[andromeda folder]/andromeda-latest.

Type: string

Model.train_args

Training arguments for the model. Wrapper for HappyTransformer's GENTrainArgs class.

Properties:

  • Model.train_args.learning_rate: defaults to 5e-05
  • Model.train_args.num_train_epochs: defaults to 3
  • Model.train_args.batch_size: defaults to 1
  • Model.train_args.adam_beta1: defaults to 0.9
  • Model.train_args.adam_beta2: defaults to 0.999
  • Model.train_args.adam_epsilon: defaults to 1e-08
  • Model.train_args.max_grad_norm: defaults to 1.0
  • Model.train_args.save_preprocessed_data: defaults to False
  • Model.train_args.save_preprocessed_data_path: defaults to ''
  • Model.train_args.load_preprocessed_data: defaults to False
  • Model.train_args.load_preprocessed_data_path: defaults to ''
  • Model.train_args.preprocessing_processes: defaults to 1
  • Model.train_args.mlm_probability: defaults to 0.15
  • Model.train_args.fp16: defaults to False

Type: GENTrainArgs object

Model.eval_args

Evaluation arguments for the model. Wrapper for HappyTransformer's GENEvalArgs class.

Properties:

  • Model.eval_args.batch_size: defaults to 1
  • Model.eval_args.save_preprocessed_data: defaults to False
  • Model.eval_args.save_preprocessed_data_path: defaults to ''
  • Model.eval_args.load_preprocessed_data: defaults to False
  • Model.eval_args.load_preprocessed_data_path: defaults to ''
  • Model.eval_args.preprocessing_processes: defaults to 1
  • Model.eval_args.mlm_probability: defaults to 0.15

Type: GENEvalArgs object

Model.config

The model configuration. Wrapper for HappyTransformer's GENSettings class.

Properties:

  • Model.config.min_length: defaults to 10
  • Model.config.max_length: defaults to 50
  • Model.config.do_sample: defaults to False
  • Model.config.early_stopping: defaults to False
  • Model.config.num_beams: defaults to 1
  • Model.config.temperature: defaults to 1
  • Model.config.top_k: defaults to 50
  • Model.config.no_repeat_ngram_size: defaults to 0
  • Model.config.top_p: defaults to 1
  • Model.config.bad_words: defaults to None

Type: GENSettings object

Methods

Model.train()

Trains the model. Training arguments can be configured using the train_args attribute.

Parameters:

  • input_filepath (string): The path to the file containing the training data.

Returns:

None.

Model.evaluate()

Evaluates the model. Evaluation arguments can be configured using the eval_args attribute.

Parameters:

  • input_filepath (string): The path to the file containing the evaluation data.

Returns:

loss (float).

Model.generate()

Uses the model to generate text. Generation arguments can be configured using the config attribute.

Parameters:

  • text (string): The text to use as the prompt for the model.
  • raw (bool): Optional. Whether to return the raw output from the model. Defaults to False.

Returns:

The generated text as a string, or the raw output from the model (an object of type GenerationResult).

Model.save()

Saves the model to disk, overwriting any previously saved files. The model will be saved to the path attribute, along with any configuration changes.

Parameters:

  • None

Returns:

  • None.