First, follow instructions on Installation page to install deeppavlov package for Python 3.6/3.7/3.8/3.9.

DeepPavlov contains a bunch of great pre-trained NLP models. Each model is determined by its config file. List of models is available on the doc page or in the deeppavlov.configs:

from deeppavlov import configs

When you’ve decided on the model (+ config file), there are two ways to train, evaluate and infer it:

Before making choice of an interface, install model’s package requirements (CLI):

python -m deeppavlov install <config_path>
  • where <config_path> is path to the chosen model’s config file (e.g. deeppavlov/configs/classifiers/insults_kaggle_bert.json) or just name without .json extension (e.g. insults_kaggle_bert)

Command line interface (CLI)

To get predictions from a model interactively through CLI, run

python -m deeppavlov interact <config_path> [-d]
  • -d downloads required data – pretrained model files and embeddings (optional).

You can train it in the same simple way:

python -m deeppavlov train <config_path> [-d]

Dataset will be downloaded regardless of whether there was -d flag or not.

To train on your own data, you need to modify dataset reader path in the train section doc. The data format is specified in the corresponding model doc page.

There are even more actions you can perform with configs:

python -m deeppavlov <action> <config_path> [-d]
  • <action> can be
    • download to download model’s data (same as -d),

    • train to train the model on the data specified in the config file,

    • evaluate to calculate metrics on the same dataset,

    • interact to interact via CLI,

    • riseapi to run a REST API server (see docs),

    • risesocket to run a socket API server (see docs),

    • predict to get prediction for samples from stdin or from <file_path> if -f <file_path> is specified.

  • <config_path> specifies path (or name) of model’s config file

  • -d downloads required data


To get predictions from a model interactively through Python, run

from deeppavlov import build_model

model = build_model(<config_path>, download=True)

# get predictions for 'input_text1', 'input_text2'
model(['input_text1', 'input_text2'])
  • where download=True downloads required data from web – pretrained model files and embeddings (optional),

  • <config_path> is path to the chosen model’s config file (e.g. "deeppavlov/configs/ner/ner_ontonotes_bert_mult.json") or deeppavlov.configs attribute (e.g. deeppavlov.configs.ner.ner_ontonotes_bert_mult without quotation marks).

You can train it in the same simple way:

from deeppavlov import train_model

model = train_model(<config_path>, download=True)
  • download=True downloads pretrained model, therefore the pretrained model will be, first, loaded and then trained (optional).

Dataset will be downloaded regardless of whether there was -d flag or not.

To train on your own data, you need to modify dataset reader path in the train section doc. The data format is specified in the corresponding model doc page.

You can also calculate metrics on the dataset specified in your config file:

from deeppavlov import evaluate_model

model = evaluate_model(<config_path>, download=True)

Using GPU

To run or train PyTorch-based DeepPavlov models on GPU you should have CUDA installed on your host machine, and install model’s package requirements. CUDA version should be compatible with DeepPavlov required PyTorch version.


If you use latest NVIDIA architecture, PyTorch installed from PyPI using DeepPavlov could not support your device CUDA capability. You will receive incompatible device warning after model initialization. You can install compatible package from download.pytorch.org. For example:

pip3 install torch==1.8.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html

If you want to run the code on GPU, just make the device visible for the script. If you want to use a particular device, you may set it in command line:

export CUDA_VISIBLE_DEVICES=3; python -m deeppavlov train <config_path>

or in Python script:

import os


In case one wants to left the GPU device visible but use CPU, one can set directly in the configuration file in dictionary with model parameters “device”: “cpu”.

Pretrained models

DeepPavlov provides a wide range of pretrained models and skills. See features overview for more info. Please note that most of our models are trained on specific datasets for specific tasks and may require further training on your data. You can find a list of our out-of-the-box models below.

Docker images

You can run DeepPavlov models in riseapi mode via Docker without installing DP. Both your CPU and GPU (we support NVIDIA graphic processors) can be utilised, please refer our CPU and GPU Docker images run instructions.

Out-of-the-box pretrained models

While the best way to solve most of the NLP tasks lies through collecting datasets and training models according to the domain and an actual task itself, DeepPavlov offers several pretrained models, which can be strong baselines for a wide range of tasks.

You can run these models via Docker or in riseapi/risesocket mode to use in solutions. See riseapi and risesocket modes documentation for API details.

Text Question Answering

Text Question Answering component answers a question based on a given context (e.g, a paragraph of text), where the answer to the question is a segment of the context.


DeepPavlov config








Name Entity Recognition

Named Entity Recognition (NER) classifies tokens in text into predefined categories (tags), such as person names, quantity expressions, percentage expressions, names of locations, organizations, as well as expression of time, currency and others.


DeepPavlov config











Insult Detection

Insult detection predicts whether a text (e.g, post or speech in some public discussion) is considered insulting to one of the persons it is related to.


DeepPavlov config





Paraphrase Detection

Detect if two given texts have the same meaning.


DeepPavlov config