DeepPavlov

DeepPavlov is an open-source natural language processing framework with pretrained models for question answering, named entity recognition, text classification and spelling correction. Developers and researchers use it to train and serve NLP pipelines from Python or the command line.

Every model is defined by a JSON configuration file that wires together dataset readers, tokenizers, embedders and the model itself, so pipelines can be swapped or nested without rewriting code. Pretrained configs cover context and open-domain question answering, knowledge base QA, entity extraction, ranking and sentiment or topic classification.

It installs with pip or runs from Docker images, supports GPUs, and includes BERT, ELMo and fastText embeddings plus multitask BERT. Hyperparameter search and cross-validation tools help when training on your own data.