DeBERTa-v3-base-mnli-fever-anli Open-source Model - Free to Use, a Powerful Tool for Zero-shot Classification and Language Inference

Deberta V3 Base Mnli Fever Anli

Developed by MoritzLaurer

DeBERTa-v3 model trained on MultiNLI, Fever-NLI, and ANLI datasets, excelling in zero-shot classification and natural language inference tasks

Text Classification

Transformers

EnglishOpen Source License:MIT #Zero-shot classification #Natural language inference #Multi-task training

Downloads 613.93k

Release Time : 3/2/2022

Model Overview

This model performs exceptionally well on natural language inference (NLI) tasks, particularly suitable for zero-shot text classification scenarios. Based on Microsoft's DeBERTa-v3-base architecture, it improves performance through enhanced pre-training objectives.

Model Features

Multi-dataset training

Combines three major datasets: MultiNLI, Fever-NLI, and ANLI, totaling 763,913 NLI sample pairs

Excellent adversarial testing performance

Outperforms most large models on the ANLI adversarial benchmark

Improved pre-training architecture

Utilizes an enhanced version of DeBERTa-v3, significantly boosting performance through optimized pre-training objectives

Model Capabilities

Zero-shot text classification

Natural language inference

Textual entailment judgment

Multi-label classification

Use Cases

Content classification

News classification

Automatically categorizes news into predefined categories like politics, economy, etc., without training

Example accuracy approximately 49.5% (ANLI test set)

Semantic analysis

Opinion contradiction detection

Identifies whether statements in text are self-contradictory

🚀 MoritzLaurer/DeBERTa-v3-base-mnli-fever-anli

This model is designed for zero-shot text classification and natural language inference tasks, leveraging multiple datasets for training and achieving good performance on various evaluation metrics.

🚀 Quick Start

This section is not provided in the original document, so it is skipped.

✨ Features

Task Scope: Supports text classification and zero-shot classification tasks.
Datasets Utilized: Trained on datasets such as multi_nli, facebook/anli, and fever.
Metrics: Evaluated using metrics like accuracy, precision, recall, F1-score, and loss.

📚 Documentation

Model Information

Property	Details
Model Type	MoritzLaurer/DeBERTa-v3-base-mnli-fever-anli
Training Data	`multi_nli`, `facebook/anli`, `fever`

Evaluation Results

The model has been evaluated on different datasets and splits, with the following key results:

anli - test_r3

Metric	Value
Accuracy	0.495
Precision Macro	0.4984740618243923
Precision Micro	0.495
Precision Weighted	0.4984357572868885
Recall Macro	0.49461028192371476
Recall Micro	0.495
Recall Weighted	0.495
F1 Macro	0.4942810999491704
F1 Micro	0.495
F1 Weighted	0.4944671868893595
loss	1.8788293600082397

anli - test_r1

Metric	Value
Accuracy	0.712
Precision Macro	0.7134839439315348
Precision Micro	0.712
Precision Weighted	0.7134676028447461
Recall Macro	0.7119814425203647
Recall Micro	0.712
Recall Weighted	0.712
F1 Macro	0.7119226991285647
F1 Micro	0.712
F1 Weighted	0.7119242267218338
loss	1.0105403661727905

multi_nli - validation_mismatched

Metric	Value
Accuracy	0.902766476810415
Precision Macro	0.9023816542652491
Precision Micro	0.902766476810415
Precision Weighted	0.9034597464719761
Recall Macro	0.9024304801555488
Recall Micro	0.902766476810415
Recall Weighted	0.902766476810415
F1 Macro	0.9023086094638595
F1 Micro	0.902766476810415
F1 Weighted	0.9030161011457231
loss	0.3283354640007019

anli - dev_r1

Metric	Value
Accuracy	0.737

📄 License

The model is released under the MIT license.

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