This repository provides officially supported models and components for the AllenNLP library, targeting NLP researchers and practitioners. It offers pre-trained models for various tasks like classification, coreference resolution, generation, and sequence tagging, simplifying the application of advanced NLP techniques.

How It Works

The library organizes components (DatasetReader, Model, Predictor) by NLP task. It leverages pre-trained models, often based on architectures like BERT, RoBERTa, and ViLBERT, and provides easy access to them via a programmatic interface or pre-built Docker images. This approach allows users to quickly apply state-of-the-art models without extensive setup.

Quick Start & Requirements

• Install: pip install allennlp-models
• Dependencies: Tied to allennlp core package. Docker images are available for CPU and GPU (CUDA 10.2+).
• Resources: Pre-trained models require downloading, size varies.
• Docs: AllenNLP Models documentation

Highlighted Details

• Supports a wide range of NLP tasks including classification, coreference resolution, generation, language modeling, multiple choice, pair classification, reading comprehension, structured prediction, sequence tagging, and text+vision.
• Offers numerous pre-trained models, including BERT, RoBERTa, GPT-2, and ViLBERT variants, for specific datasets and tasks.
• Provides programmatic access to download and load pre-trained models and their corresponding predictors.
• Includes Dockerfiles for building consistent environments with specific AllenNLP and CUDA versions.

Maintenance & Community

The AllenNLP ecosystem is in maintenance mode, with no new features or dependency upgrades planned. Support and bug fixes were available until December 16th, 2022. Issues should be filed on the allenai/allennlp repository and tagged with "Models".

Licensing & Compatibility

The repository is licensed under Apache 2.0. This license is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

The project is in maintenance mode, meaning no new features or dependency updates will be added. Active support concluded in December 2022.