tinygrad
tinygrad.org
- Free plan
- Windows
- macOS
- Linux
- Web
- Self-hosted
What tinygrad is
tinygrad is a deep learning stack for neural network training, combining a tensor library, compiler, JIT, and training tools. It supports forward and backward passes with automatic differentiation, and tensor operations run lazily when a tensor is realized. The library includes neural network classes, optimizers, and state loading and saving. TinyJit captures and replays kernels in a decorated function, and tensors can be sharded across multiple GPUs. Listed accelerators include OpenCL, CPU, Metal, CUDA, AMD, NV, Qualcomm, and WebGPU. tinygrad supports Python and local training; the project says it is used in openpilot to run its driving model on a Snapdragon 845 GPU. It is free and identifies its license as MIT. Installation instructions cover pip, though the project recommends installing from source. The documentation says the API has been stable for a while, but the project is not yet version 1.0. Its FAQ says tinygrad is not faster than PyTorch for most use cases. Development discussion is directed to GitHub and Discord.
Who it suits
It suits Python users working on deep learning and neural network training who want a free, open-source stack with listed accelerator options. The project's maturity and performance caveats are relevant for users evaluating it for production or speed-sensitive workloads.
Good to know, and what to watch
- Includes tensor library, compiler, JIT, and training tools.
- Supports autodiff and multi-GPU tensor sharding.
- Lists CPU and multiple accelerator backends.
- Free with an MIT license.
- Installation documented through pip and source.
- Project is not yet version 1.0.
- FAQ says it is not faster than PyTorch for most use cases.
- Project recommends installing from source.
Plans
Every plan and pricetinygrad
Free plan
The verdict
tinygrad provides a broad deep learning stack at no cost, with multiple accelerator options and training components. Its pre-1.0 status and stated performance caveat should factor into evaluation.
What the maker publishes
- Accelerators
- Listed accelerators include OpenCL, CPU, Metal, CUDA, AMD, NV, Qualcomm, and WebGPU. source
- Autodiff
- The project supports forward and backward passes with autodiff. source
- Community support
- The project directs development discussion to GitHub and Discord. source
- Installation
- The project recommends installing from source and also documents installation with pip. source
- Integration
- The project says tinygrad is used in openpilot to run its driving model on a Snapdragon 845 GPU. source
- Intended users
- The project describes tinygrad as a framework for deep learning and neural network training. source
- JIT
- tinygrad provides TinyJit to capture and replay kernels in a decorated function. source
- Lazy execution
- Tensor operations are lazy and run when the tensor is realized. source
- License
- The GitHub repository identifies the project license as MIT. source
- Maturity
- The documentation says tinygrad is not yet version 1.0, while noting its API has been stable for a while. source
- Multi GPU
- The documentation says tensors can be sharded across multiple GPUs. source
- Neural networks
- The library includes neural network classes, optimizers, and state load/save management. source
- Performance caveat
- The project FAQ says tinygrad is not yet faster than PyTorch for most use cases. source
- Purpose
- tinygrad is an end-to-end deep learning stack with a tensor library, compiler, JIT, and tools for training. source
Checked against the maker's pages on 7 Oct 2026.