Documentation | Contributors | Community | Release Notes
TVM is a compiler stack for deep learning systems. It is designed to close the gap between the productivity-focused deep learning frameworks, and the performance- and efficiency-focused hardware backends. TVM works with deep learning frameworks to provide end to end compilation to different backends. Checkout the tvm stack homepage for more information.
Neo-AI/TVM is a downstream branch of TVM that includes vendor- and product-specific features on top of the upstream codebase.
- dev
- This is the development branch with most update to date source code.
© Contributors Licensed under an Apache-2.0 license.
TVM adopts apache committer model, we aim to create an open source project that is maintained and owned by the community. Checkout the Contributor Guide
We learned a lot from the following projects when building TVM.
- Halide: TVM uses HalideIR as data structure for arithmetic simplification and low level lowering. We also learned and adapted some part of lowering pipeline from Halide.
- Loopy: use of integer set analysis and its loop transformation primitives.
- Theano: the design inspiration of symbolic scan operator for recurrence.