Practical overview
When it fits
Use it when building a PyTorch application that needs supported transformer architectures, reusable model components, or the documented text-generation wrappers.
Who it is for
- Python developers adding transformer models or text generation to applications.
- Developers composing or studying transformer architecture components.
- Teams integrating transformer models into spaCy applications through the separate integration package.
Documented uses
- Load supported transformer models from Hugging Face Hub.
- Generate text with supported generator wrappers.
- Compose custom transformer architectures from reusable components.
- Apply dynamic 4-bit or 8-bit quantization using bitsandbytes.
Requirements & limits
The project is open source under the MIT license and documents installation through pip.
- The library is built for PyTorch and uses Python.
- The supported architecture list distinguishes encoder-only models, decoder-only models and generator wrappers.
- The example uses a CUDA device; Windows users and Linux users with Ada-generation GPUs are advised to install the CUDA 11.8 PyTorch build explicitly.
- spaCy integration is provided by the separate spacy-curated-transformers package.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Curated Transformers
Install the library and, if needed, select the documented CUDA 11.8 PyTorch build or quantization dependencies.
Before you start
- Python environment with pip.
- For Windows, or Linux with an Ada-generation GPU, install the documented CUDA 11.8 PyTorch build.
Steps
Install the library with the command below.
pip install curated-transformersFor the noted CUDA cases, install PyTorch with the command below.
pip install torch --index-url https://download.pytorch.org/whl/cu118To install dependencies for dynamic quantization, use the command below.
pip install curated-transformers[quantization]
Expected outcome
The library is installed, with the documented PyTorch CUDA build or quantization dependencies selected as applicable.
Guide limitations
- These steps cover installation only; they do not verify that a particular model or device will run.
- Quantization uses bitsandbytes.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- explosion/curated-transformers · Official homepage title · retained in its source language