tlt.models.text_generation.pytorch_hf_text_generation_model.PyTorchHFTextGenerationModel

class tlt.models.text_generation.pytorch_hf_text_generation_model.PyTorchHFTextGenerationModel(model_name: str, model=None, **kwargs)

Class to represent a PyTorch Hugging Face pretrained model that can be used for text generation fine tuning.

__init__(model_name: str, model=None, **kwargs)

Class constructor

Methods

__init__(model_name[, model])	Class constructor
benchmark(dataset[, saved_model_dir, ...])	Use Intel Neural Compressor to benchmark the model with the dataset argument.
evaluate([dataset, enable_auto_mixed_precision])	Evaluates the model on the 'eval_dataset' given in the Trainer arguments
export(output_dir)	Saves the model and tokenizer to the given output_dir directory.
freeze_layer(layer_name)	Freezes the model's layer using a layer name :param layer_name: The layer name that will be frozen in the model :type layer_name: string
generate(input_samples[, temperature, ...])	Generates text completions for the specified input samples.
list_layers([verbose])	Lists all of the named modules (e.g.
load_from_directory(model_dir)	Loads a saved pytorch model from the given model_dir directory.
optimize_graph(output_dir[, overwrite_model])	Performs FP32 graph optimization using the Intel Neural Compressor on the model and writes the inference-optimized model to the output_dir. Graph optimization includes converting variables to constants, removing training-only operations like checkpoint saving, stripping out parts of the graph that are never reached, removing debug operations like CheckNumerics, folding batch normalization ops into the pre-calculated weights, and fusing common operations into unified versions. :param output_dir: Writable output directory to save the optimized model :type output_dir: str :param overwrite_model: Specify whether or not to overwrite the output_dir, if it already exists (default: False) :type overwrite_model: bool.
quantize(output_dir, dataset[, config, ...])	Performs post training quantization using the Intel Neural Compressor on the model using the dataset.
train(dataset, output_dir[, epochs, ...])	Trains the model using the specified text generation dataset.
unfreeze_layer(layer_name)	Unfreezes the model's layer using a layer name :param layer_name: The layer name that will be frozen in the model :type layer_name: string

Attributes

framework	Framework with which the model is compatible
learning_rate	Learning rate for the model
model_name	Name of the model
preprocessor	Preprocessor for the model
use_case	Use case (or category) to which the model belongs