allenai / allennlp

An NLP library for developing models on a variety of linguistic tasks.

AllenNLP

AllenNLP is an NLP research library, built on PyTorch, for developing state-of-the-art deep learning models on a wide variety of linguistic tasks. It consists of:

24+ available models for a variety of NLP tasks.
Data processing modules for loading NLP datasets.
A variety of PyTorch modules for use with NLP datasets.

Tasks

Choose a task to see what models are available:

Question Answering

Named Entity Recognition (NER)

Visual Question Answering (VQA)

Sentiment Analysis

Language Modelling

Natural Language Inference

Common Sense Reasoning

Open Information Extraction

Dependency Parsing

Coreference Resolution

Semantic Role Labeling

Constituency Parsing

Show all

Models

Viewing Models for Question Answering:

MODEL	TRAINED ON	EM	PARAMETERS	YEAR
RoBERTa SWAG	SWAG		356 Million	2019
Physical Interaction Question Answering	PIQA		356 Million	2019
Transformer QA	SQuAD	84	355 Million	2019
RoBERTa Common Sense QA	CommonsenseQA		356 Million	2019
Numerically Augmented QA Net	DROP		14 Million	2019
GPT2-based Next Token Language Model	WebText		163 Million	2019
ELMo-BiDAF	SQuAD	71	113 Million	2016
BiDAF	SQuAD	66	12 Million	2016

Models

Viewing Models for Sentiment Analysis:

	MODEL	TRAINED ON	ACCURACY	PARAMETERS	PAPER	YEAR
	RoBERTa large SST	SST	95.11	355 Million		2019
	GLoVe-LSTM	SST	87%	10 Million

Models

Viewing Models for Named Entity Recognition:

MODEL	TRAINED ON	F1	PARAMETERS	YEAR
ELMo-based Named Entity Recognition	CoNLL-2003		98 Million	2017
Fine Grained Named Entity Recognition	OntoNotes 5.0	88	101 Million	2016
Fine Grained Named Entity Recognition with Transformer	OntoNotes 5.0	88	125 Million

Models

Viewing Models for Visual Question Answering:

	MODEL	TRAINED ON	PARAMETERS	PAPER	YEAR
	ViLBERT - Visual Question Answering	Visual Question Answering	245 Million		2019

Models

Viewing Models for Natural Language Inference:

MODEL	TRAINED ON	PARAMETERS	YEAR
RoBERTa SNLI	SNLI	356 Million	2019
RoBERTa MNLI	MultiNLI	356 Million	2019
Enhanced LSTM for Natural Language Inference	SNLI	100 Million	2016
ELMo-based Decomposable Attention	SNLI	94 Million	2016

Models

Viewing Models for Language Modelling:

	MODEL	TRAINED ON	PARAMETERS	PAPER	YEAR
	GPT2-based Next Token Language Model	WebText	163 Million		2019
	BERT-based Masked Language Model	Wikipedia, BookCorpus	131 Million		2018

Models

Viewing Models for Common Sense Reasoning:

MODEL	TRAINED ON	PARAMETERS	YEAR
RoBERTa SWAG	SWAG	356 Million	2019
Physical Interaction Question Answering	PIQA	356 Million	2019
RoBERTa Common Sense QA	CommonsenseQA	356 Million	2019

Models

Viewing Models for Dependency Parsing:

	MODEL	TRAINED ON	LAS	PARAMETERS	PAPER	YEAR
	Deep Biaffine Attention for Neural Dependency Parsing	Penn Treebank	94.44	20 Million		2016

Models

Viewing Models for Open Information Extraction:

	MODEL	TRAINED ON	PARAMETERS	PAPER	YEAR
	Open Information Extraction	AW-OIE	15 Million		2018

Models

Viewing Models for Coreference Resolution:

	MODEL	TRAINED ON	PARAMETERS	PAPER	YEAR
	Coreference Resolution	OntoNotes 5.0	366 Million		2018

Models

Viewing Models for Semantic Role Labeling:

	MODEL	TRAINED ON	F1	PARAMETERS	PAPER	YEAR
	SRL BERT	OntoNotes 5.0	86.49	110 Million		2019

Models

Viewing Models for Constituency Parsing:

	MODEL	TRAINED ON	F1 SCORE	PARAMETERS	PAPER	YEAR
	Constituency Parser with ELMo embeddings	Penn Treebank	94.11	98 Million		2018