bert-base-uncased-mnli-sparse-70-unstructured-no-classifier Open Source Model

Bert Base Uncased Mnli Sparse 70 Unstructured No Classifier

Developed by Intel

This model is fine-tuned from bert-base-uncased-sparse-70-unstructured on the MNLI task (GLUE benchmark), with the classifier layer removed for easier loading into other downstream tasks for training.

Large Language Model

Transformers

English#Sparse BERT Fine-tuning #Natural Language Inference #High-precision NLP

Downloads 17

Release Time : 3/2/2022

Model Overview

This is a sparse BERT-based model specifically fine-tuned for the MNLI task, with the classifier layer removed to facilitate transfer to other tasks.

Model Features

Sparse Architecture

Employs 70% unstructured sparsity in BERT architecture, improving model efficiency

Transferability

Classifier layer removed for easier transfer to other downstream tasks

Multi-task Adaptation

Transfer performance validated on multiple GLUE tasks

Model Capabilities

Natural Language Understanding

Text Classification

Sentence Similarity Calculation

Question Answering Systems

Use Cases

Text Analysis

Natural Language Inference

Determine the relationship between two sentences (entailment/contradiction/neutral)

Matched set accuracy 82.5%, mismatched set accuracy 83.3%

Transfer Learning

Downstream Task Adaptation

Can be transferred to QQP, QNLI, SST-2 and other tasks

Achieved 90.2% accuracy on QQP task

Property	Details
QQP (Acc/F1)	90.2/86.7
QNLI (Acc)	90.3
SST - 2 (Acc)	91.5
STS - B (Pears/Spear)	88.9/88.6
SQuADv1.1 (Acc/F1)	80.5/88.2

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

Bert Base Uncased Mnli Sparse 70 Unstructured No Classifier

Model Overview

Model Features

Model Capabilities

Use Cases

🚀 Sparse BERT base model fine tuned to MNLI without classifier layer (uncased)

🚀 Quick Start

📚 Documentation

Evaluation Results