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

Ru Gpts

Ru Gpts is a public Python repository containing five autoregressive Russian-language transformer models: ruGPT3XL, ruGPT3Large, ruGPT3Medium, ruGPT3Small and ruGPT2Large. Its documentation includes text-generation and fine-tuning examples, including Google Colab notebooks.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Ru Gpts — 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 Ru Gpts when you want model files and developer examples for Russian-language generation or fine-tuning, rather than a hosted chat or generation service.

Who it is for

  • Developers building Russian-language text generation or completion experiments.
  • Researchers and practitioners comparing or fine-tuning the repository’s model variants.

Documented uses

  • Generate a Russian-language continuation from a supplied prompt.
  • Explore model variants with different documented sequence lengths.
  • Follow the repository’s examples for fine-tuning a model.
Before you choose

Requirements & limits

The repository is public and lists an Apache-2.0 license.

  • The README describes 2048-token sequence lengths for the GPT-3 models and 1024 for ruGPT2Large.
  • The smaller models and ruGPT2Large use Hugging Face Transformers, with Transformers 4.24.0 specified; ruGPT3XL has a more involved setup involving DeepSpeed, Triton and Apex.
  • The repository’s examples include Colab notebooks for generation and fine-tuning.

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 with a smaller ruGPT model using Transformers

Load ruGPT3Large and generate text from a Russian-language prompt.

Before you start

  • Python environment with PyTorch and CUDA available for the documented .cuda() calls.
  • Hugging Face Transformers version 4.24.0.

Steps

  1. Install Transformers: the command below.

    Documented command
    pip install transformers==4.24.0

  2. Load the tokenizer and model sberbank-ai/rugpt3large_based_on_gpt2 with GPT2Tokenizer.from_pretrained and GPT2LMHeadModel.from_pretrained, then move the model to CUDA.

  3. Set a Russian-language prompt, tokenize it with return_tensors="pt", and move the input IDs to CUDA.

  4. Call model.generate(input_ids) and decode the output with the tokenizer.

Expected outcome

The model generates a continuation of the supplied prompt; the README includes an illustrative Russian continuation for a prompt about Alexander Pushkin’s birth.

Guide limitations

  • The documented example uses CUDA; CPU execution is not shown.
  • The README describes its sample output as illustrative, not as a guaranteed result.
  • This recipe covers ruGPT3Large; other models may use different model identifiers or examples.
Documentation 1 ↗
Source-based practical guide

Set up ruGPT3XL in Colab

Install the dependencies and repository files needed for the documented ruGPT3XL usage example in Colab.

Before you start

  • A Colab environment; the repository recommends these installation instructions for Colab.
  • The commands use apt-get, pip, git, and fixed package paths including /usr/local/lib/python3.8/dist-packages/.

Steps

  1. Set LD_LIBRARY_PATH=/usr/lib/ and install clang-9, llvm-9, llvm-9-dev, and llvm-9-tools.

  2. Clone the qywu/apex repository, enter its directory, and install it with the documented C++ and CUDA extension options; then install triton, deepspeed with DS_BUILD_CPU_ADAM=1 DS_BUILD_SPARSE_ATTN=1, transformers, huggingface_hub, and timm==0.3.2.

  3. Clone ai-forever/ru-gpts and copy its trainer_pt_utils.py and _amp_state.py files to the documented Transformers and Apex package paths.

  4. Restart the Colab environment. The README suggests checking with ds_report and importing deepspeed.ops.sparse_attention.sparse_attn_op.

Expected outcome

The Colab setup reaches the documented dependency-installation and sparse-attention check stage for ruGPT3XL.

Guide limitations

  • The README explicitly scopes these installation instructions to Colab.
  • This recipe ends at setup and environment checks; it does not verify a successful generation run.
  • The commands assume the package locations and Python 3.8 path stated in the README.
Documentation 1 ↗
Evidence and transparency

Sources & review

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