OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Mt Dnn

Mt Dnn is a Python and PyTorch research implementation of multi-task deep neural networks for natural-language understanding. It provides code and documented workflows for shared-model training and task-specific fine-tuning, alongside separate implementations for adversarial language-model training and commonsense reasoning.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Mt Dnn — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Mt Dnn when you need research code and task-specific training examples for multi-task NLP, and can prepare the datasets, checkpoints and compute required by the chosen workflow.

Who it is for

  • NLP researchers studying multi-task learning and language-model fine-tuning.
  • Developers adapting training pipelines for GLUE, question answering, sequence labeling or domain adaptation.

Documented uses

  • Train a shared natural-language understanding model across multiple tasks.
  • Fine-tune models for individual GLUE tasks, including STS-B and RTE.
  • Prepare and train sequence-labeling or question-answering workflows using the repository's documented data formats.
  • Explore adversarial fine-tuning, language-model training or the included commonsense reasoning model.
Before you choose

Requirements & limits

The repository is public and specifies the MIT license. The README says it no longer provides public storage for model sharing.

  • The documented pip setup specifies Python 3.6; a Docker setup is also described.
  • The README's base-model experiments used four V100 GPUs, while one GLUE refinement run used eight 32 GB V100 GPUs.
  • Training paths require task data and preprocessing; the README provides distinct instructions for GLUE, SciTail/SNLI, NER and SQuAD.
  • The repository's note about discontinued public model storage affects access to pretrained checkpoints.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up the documented Python environment

Prepare the repository environment using its pip-based setup instructions.

Before you start

  • Python 3.6
  • The repository and its requirements.txt file

Steps

  1. Install Python 3.6.

  2. Install the listed dependencies with pip install -r requirements.txt.

Expected outcome

The README's pip-based environment setup steps are complete.

Guide limitations

  • These setup steps do not train a model or verify that the environment runs successfully.
  • The README also documents a Docker option.
Documentation 1 ↗
Source-based practical guide

Prepare and start toy GLUE training

Follow the README's documented data preparation and training sequence for a toy MT-DNN model.

Before you start

  • A prepared project environment
  • GLUE data
  • The repository's download.sh and experiments/glue/prepro.sh scripts

Steps

  1. Run sh download.sh to download data.

  2. Run sh experiments/glue/prepro.sh to preprocess it.

  3. Run python train.py to start training.

Expected outcome

The documented toy-model training command is launched after GLUE data preparation.

Guide limitations

  • The README reports base-model experiments on four V100 GPUs and says batch size may need to be reduced on other GPUs.
  • The repository notes that it no longer provides public storage for model sharing.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
namisan/mt-dnn · Official homepage title · retained in its source language
How we review sourcesSuggest a correction