speechbrain.lobes.models.huggingface_transformers.labse module

This lobe enables the integration of huggingface pretrained LaBSE models. Reference: https://arxiv.org/abs/2007.01852

Transformer from HuggingFace needs to be installed: https://huggingface.co/transformers/installation.html

Authors

Ha Nguyen 2023

Summary

Classes:

LaBSE

This lobe enables the integration of HuggingFace and SpeechBrain pretrained LaBSE models.

Reference

class speechbrain.lobes.models.huggingface_transformers.labse.LaBSE(source, save_path, freeze=True, output_norm=True)[source]

Bases: HFTransformersInterface

This lobe enables the integration of HuggingFace and SpeechBrain pretrained LaBSE models.

Source paper LaBSE: https://arxiv.org/abs/2007.01852 Transformer from HuggingFace needs to be installed: https://huggingface.co/transformers/installation.html

The model can be used as a fixed text-based sentence-level embeddings generator or can be finetuned. It will download automatically the model from HuggingFace or use a local path.

Parameters:

source (str) – HuggingFace hub name: e.g “setu4993/LaBSE”
save_path (str) – Path (dir) of the downloaded model.
freeze (bool (default: True)) – If True, the model is frozen. If False, the model will be trained alongside with the rest of the pipeline.
output_norm (bool (default: True)) – If True, normalize the output.

Example

>>> inputs = ["La vie est belle"]
>>> model_hub = "setu4993/smaller-LaBSE"
>>> save_path = "savedir"
>>> model = LaBSE(model_hub, save_path)
>>> outputs = model(inputs)

forward(input_texts)[source]

This method implements a forward of the labse model, which generates sentence-level embeddings from input text.

Parameters:: (translation) (input_texts) – The list of texts (required).