T5 Xxl Lm Adapt

🚀 Google's T5 Version 1.1 - LM-Adapted

Google's T5 Version 1.1 - LM-Adapted is an improved version of the original T5 model, offering enhanced performance and capabilities for various NLP tasks.

🚀 Quick Start

This section provides an overview of the key features and improvements of Google's T5 Version 1.1 - LM-Adapted.

✨ Features

Improvements over the Original T5 Model

• Activation Function: Utilizes GEGLU activation in the feed-forward hidden layer instead of ReLU. Refer to here for more details.
• Dropout: Dropout is turned off during pre-training for better quality. It should be re-enabled during fine-tuning.
• Pre-training Data: Pre-trained solely on the C4 dataset without incorporating downstream tasks.
• Parameter Sharing: There is no parameter sharing between the embedding and classifier layers.
• Model Sizes: "xl" and "xxl" replace "3B" and "11B", with different model shapes (larger d_model, smaller num_heads, and d_ff).

Pretraining Objectives

The model is pre-trained on both denoising and language modeling objectives. This specific checkpoint is initialized from T5 Version 1.1 - XXL and then trained for an additional 100K steps on the LM objective described in the T5 paper. This adaptation enhances the model's suitability for prompt tuning.

Popular Fine-tuned Version

A well - known fine-tuned version of the T5 Version 1.1 - LM Adapted model is BigScience's T0pp.

📚 Documentation

Pretraining Dataset

The model is pre-trained on the C4 dataset.

Other Community Checkpoints

You can find other community checkpoints here.

Paper

The related paper is Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer.

Authors

The authors of the paper are Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. Liu.

Abstract

Transfer learning, which involves pre - training a model on a data - rich task before fine - tuning it on a downstream task, has become a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has led to a variety of approaches, methodologies, and practices. In this paper, the authors explore the transfer learning techniques for NLP by introducing a unified framework that converts every language problem into a text - to - text format. Their systematic study compares pre - training objectives, architectures, unlabeled datasets, transfer approaches, and other factors across dozens of language understanding tasks. By combining insights from their exploration with scale and the new “Colossal Clean Crawled Corpus”, they achieve state - of - the - art results on many benchmarks, including summarization, question answering, text classification, etc. To facilitate future work on transfer learning for NLP, they release their dataset, pre - trained models, and code.

Model Image

📄 License

The project is licensed under the Apache 2.0 license.