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

Mgpt

Mgpt (mGPT) is a multilingual generative pretrained transformer, described as a GPT-3 variant, with model code and weights for text generation and research. It was pretrained on 61 languages from 25 language families using Wikipedia and the C4 corpus. Its repository provides Transformers-based generation and fine-tuning examples and describes evaluations for language modelling, cross-lingual natural-language understa

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Mgpt — 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 mGPT when an experiment requires a multilingual pretrained generative model and the repository’s language coverage or evaluation work is relevant. The README’s Transformers example is the clearest starting point for prompt-based generation.

Who it is for

  • Researchers comparing language-model behaviour across multiple languages.
  • Developers experimenting with multilingual text generation using Transformers.
  • Teams studying cross-lingual NLU or probing world knowledge across languages.

Documented uses

  • Generate a continuation from a text prompt using the model and tokenizer.
  • Explore language modelling across the model’s listed languages.
  • Evaluate cross-lingual NLU using the research benchmarks described by the project.
  • Probe world knowledge across the languages covered by its evaluation.
Before you choose

Requirements & limits

The repository states that the source code and language models are available under the MIT license. It links to Hugging Face model cards and Colab notebooks for generation and fine-tuning examples.

  • The README lists 61 pretrained languages, but does not claim uniform performance across them.
  • The project provides model-loading and generation examples rather than a documented hosted inference API.
  • Ordinary generation examples use Transformers; DeepSpeed and Megatron-LM are identified in the context of pretraining.
  • The shown generation example uses .cuda(device) without defining the device variable.

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

Generate text from a prompt with Transformers

Load mGPT and produce a continuation from a text prompt using the repository’s Python example.

Before you start

  • Python environment set up with the repository requirements using pip install -r requirements.txt.
  • The Hugging Face Transformers library and access to the sberbank-ai/mGPT tokenizer and model referenced in the README.
  • A prompt in a language listed by the project.

Steps

  1. Load GPT2Tokenizer and GPT2LMHeadModel from sberbank-ai/mGPT.

  2. Encode the prompt with the tokenizer.

  3. Pass the resulting input IDs to model.generate with generation settings such as the README’s listed length and sampling parameters.

  4. Decode the generated output with the tokenizer.

Expected outcome

A decoded text continuation from the supplied prompt.

Guide limitations

  • The README example calls .cuda(device) but does not define device or document hardware requirements; it does not provide a CPU-specific alternative.
  • The example’s generation parameters are examples, not a guarantee of output quality across languages.
  • The repository does not document a hosted API for this task.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
ai-forever/mgpt · Official homepage title · retained in its source language
How we review sourcesSuggest a correction