Unlock the Power of AI/ML including Generative AI and LLMs Using PyTorch
Trusted by Meta's Core Development Team Since 2019
PyTorch is one of the leading open source machine learning frameworks for building and training AI models. Developed primarily by Meta, it has become the hub of innovation within deep learning.
At Quansight, we offer personalized support to help you build and optimize a PyTorch-based AI/ML system. Whether you’re looking to create a new system or optimize an existing one, our team can help.
Quansight’s PyTorch Support Offerings Include:
Start your journey alongside PyTorch experts who will help you create advanced machine learning systems for your specific business needs.
Accelerate your PyTorch projects and cut costs by optimizing critical computation pathways to improve model performance.
Access customized PyTorch training on topics like natural language processing, computer vision, and code acceleration.
Access Open Source Experts to Build Your AI Strategy
- Travis Oliphant, NumPy, SciPy, and Numba Creator
- Tania Allard, PSF board vice chair, Jupyter Distinguished Contributor, conda steering council member
- Dharhas Pothina, Nebari & Conda Store Creator, Holoviz Steering Committee
- Ralf Gommers, SciPy/NumPy Maintainer, Array API Standard Creator
- Jaime Rodríguez-Guerra, Conda-forge Maintainer & Steering Committee Member
- Pearu Peterson, SciPy Creator & f2py Author
- Andrew James, PyTorch Core Developer, torch.sparse Author
- Melissa Weber Mendonça, NumPy/SciPy/napari maintainer & Senior Developer
- Victor Fomin, PyTorch Core Developer, PyTorch Ignite Project Lead
- Nathan Goldbaum, NumPy Steering Council
- Isuru Fernando, Conda-forge Maintainer & Member of Steering Committee
Quansight’s Contributions to PyTorch
For a closer look at the work we’ve done for PyTorch in recent years, take a look at our annual blog posts highlighting contributions and enhancements to PyTorch.
NumPy, TorchVision & torch.func support in torch.compile
Specialized GPU kernels for Block and semi-structured sparse layouts.
Triton enhancements for GPU reductions & mixed-dtype matrix multiply
Inductor improvements for indirect indexing & Welford reductions
AOTAutograd optimizations & decompositions
Contributions include:
2023 & 2024
Vision transforms 2.0
torch.sparse, torch.func, torch.signal.windows, torch.complex32
PrimTorch, inductor and dynamic shapes improvements
PyTorch as backend for SciPy and scikit-learn
NumPy support in torch.compile
Contributions include:
2022
2021
Contributions include:
Python Array API & NumPy Compatibility
torch.linalg, torch.special, torch.fft, torch.testing, torch.sparse
Forward AD, complex numbers, and functorch coverage
Build time improvements
Vision datasets & transforms 1.0
PyTorch Support Service Packages
Team Support
- $25k retainer for project work
- 5 monthly support tickets
- Support via GitHub issues
- 2 monthly office hours
- 2 business-day SLA
Enterprise Support
- $100k retainer for project work
- 20 monthly support tickets
- Support via dedicated Slack channel
- Assigned PyTorch rep PyTorch
- PyTorch Masterclasses
- Monthly office hours
- 2 business-day SLA
Custom Support
Starting at
- PyTorch customization
- PyTorch training
- Proprietary solutions around PyTorch
- Masterclasses
- Monthly office hours
Frequently Asked Questions
Our company is new to PyTorch, can Quansight still help me?▾
Yes! We are well versed in assisting companies at any stage along the PyTorch journey, including those who are just getting started.
We have a number of PyTorch engineers on staff and have a good system going but need expert help from time to time. Is that something Quansight can offer?▾
Absolutely. Our experts have built many of the internal PyTorch systems and have insight that can help even the most advanced users get more out of their systems.
What do the monthly office hours entail?▾
Your team will have a time window with exclusive access to an expert. You can use this time to learn about PyTorch, get advice on how to design your systems, or discuss anything else you like.
What sort of services do you offer around Large Language Models (LLMs)?▾
Thanks to our expertise in LLMs, AI/ML, cloud infrastructure, and the broader scientific, visualization, and data platform Python stack, we can support you at any stage in building an LLM strategy—from getting started to refining what’s in place. See our LLM Services page to learn more.
What is a PyTorch Masterclass?▾
The best way to get started from scratch is to learn from the people who built the platform! Learn the various components of PyTorch and domain libraries from the maintainers who are working on them every day.
Are the support tickets via GitHub?▾
We can easily adapt to whatever systems you are comfortable with. We can use GitHub issues on the PyTorch repository, private issues on your own ticketing system, or any other medium to track support tickets.
About Quansight
Quansight is a data, science, and engineering firm that specializes in solving complex, data-related problems by leveraging the open source software at the foundation of machine learning and artificial intelligence.
Our knowledge in scientific Python and its ecosystem of libraries and tools allows us to offer bespoke solutions that derive true value from data. We empower clients with critical business insights and actionable knowledge while strengthening the open source projects modern technologies depend on.
Get in Touch
For all inquiries about our services, projects, and business, contact us at: connect@quansight.com
