The bert-large-uncased-sparse-90-unstructured-pruneofa Open Source Model - A Sparse Pretraining Tool Supporting Multilingual Task Fine-Tuning

Bert Large Uncased Sparse 90 Unstructured Pruneofa

Developed by Intel

This model is a sparse pre-trained model achieving 90% sparsity through weight pruning techniques, suitable for fine-tuning on various language tasks.

Large Language Model

Transformers

EnglishOpen Source License:Apache-2.0 #High-Sparsity BERT #One-shot Pruning #Language Model Compression

Downloads 13

Release Time : 3/2/2022

Model Overview

A sparse language model based on BERT-Large architecture, employing a one-shot pruning general method to retain important weight information while reducing computational overhead.

Model Features

High-Sparsity Design

Achieves efficient computation through 90% weight pruning while maintaining model performance.

One-shot Pruning General Method

Requires only a single pruning session to adapt to multiple downstream tasks without repeated pruning.

Knowledge Retention Technology

Uses a teacher-student framework to ensure critical knowledge is preserved after pruning.

Model Capabilities

Text Understanding

Language Feature Extraction

Downstream Task Fine-tuning

Use Cases

Natural Language Processing

Question Answering Systems

Can be fine-tuned to build efficient QA systems

Achieves 83.35 EM/90.20 F1 on SQuADv1.1

Text Classification

Suitable for classification tasks like sentiment analysis

Achieves 92.95% accuracy on SST-2

Natural Language Inference

Applicable to genre natural language inference tasks

Achieves 83.74%/84.20% accuracy on MNLI-m/mm respectively

🚀 90% Sparse BERT-Large (uncased) Prune Once for All

This is a sparse pre-trained model that can be fine-tuned for various language tasks. By pruning weights to zero, it reduces computational overhead while retaining important information.

🚀 Quick Start

Here is an example of how to import this model in Python:

import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/bert-large-uncased-sparse-90-unstructured-pruneofa')

For more code examples, refer to the GitHub Repo.

✨ Features

Sparse Pre - training: The model uses weight pruning to create sparse matrices, reducing computational overhead.
Fine - Tunable: Can be fine - tuned for multiple language tasks such as question - answering, natural language inference, and sentiment classification.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/bert-large-uncased-sparse-90-unstructured-pruneofa')

📚 Documentation

Model Details

This model is a sparse pre - trained model. The weight pruning process sets some neural network weights to zero, resulting in sparser matrices and reduced computational overhead. The "sparse" in the model title indicates the sparsity ratio of the weights. For more details, read Zafrir et al. (2021).

Visualization of Prunce Once for All method from Zafrir et al. (2021):

Property	Details
Model Authors - Company	Intel
Date	September 30, 2021
Version	1
Model Type	NLP - General sparse language model
Architecture	"The method consists of two steps, teacher preparation and student pruning. The sparse pre - trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine - tune the sparse pre - trained models for several language tasks while we prune the pre - trained model only once." (Zafrir et al., 2021)
Paper or Other Resources	Zafrir et al. (2021); GitHub Repo
License	Apache 2.0
Questions or Comments	Community Tab and Intel Developers Discord

Intended Use

Intended Use	Description
Primary intended uses	This is a general sparse language model; in its current form, it is not ready for downstream prediction tasks, but it can be fine - tuned for several language tasks including (but not limited to) question - answering, genre natural language inference, and sentiment classification.
Primary intended users	Anyone who needs an efficient general language model for other downstream tasks.
Out - of - scope uses	The model should not be used to intentionally create hostile or alienating environments for people.

Metrics (Model Performance)

Model	Model Size	SQuADv1.1 (EM/F1)	MNLI - m (Acc)	MNLI - mm (Acc)	QQP (Acc/F1)	QNLI (Acc)	SST - 2 (Acc)
80% Sparse BERT - Base uncased fine - tuned on SQuAD1.1	-	81.29/88.47	-	-	-	-	-
85% Sparse BERT - Base uncased	Medium	81.10/88.42	82.71	83.67	91.15/88.00	90.34	91.46
90% Sparse BERT - Base uncased	Medium	79.83/87.25	81.45	82.43	90.93/87.72	89.07	90.88
90% Sparse BERT - Large uncased	Large	83.35/90.20	83.74	84.20	91.48/88.43	91.39	92.95
85% Sparse DistilBERT uncased	Small	78.10/85.82	81.35	82.03	90.29/86.97	88.31	90.60
90% Sparse DistilBERT uncased	Small	76.91/84.82	80.68	81.47	90.05/86.67	87.66	90.02

All the results are the mean of two separate experiments with the same hyper - parameters and different seeds.

Training and Evaluation Data

Property	Details
Datasets	English Wikipedia Dataset (2500M words).
Motivation	To build an efficient and accurate base model for several downstream language tasks.
Preprocessing	"We use the English Wikipedia dataset (2500M words) for training the models on the pre - training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers (Devlin et al., 2019, Sanh et al., 2019). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1."

Ethical Considerations

Ethical Considerations	Description
Data	The training data come from Wikipedia articles
Human life	The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles.
Mitigations	No additional risk mitigation strategies were considered during model development.
Risks and harms	Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al., 2021, and Bender et al., 2021). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.
Use cases	-

Caveats and Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model.

🔧 Technical Details

The weight pruning process forces some neural network weights to zero, creating sparser matrices. This reduces computational overhead during matrix multiplication while retaining enough important information.

📄 License

This model is licensed under the Apache 2.0 license.

BibTeX entry and citation info

@article{zafrir2021prune,
  title={Prune Once for All: Sparse Pre-Trained Language Models},
  author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
  journal={arXiv preprint arXiv:2111.05754},
  year={2021}
}

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご