tvm
tvm copied to clipboard
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Prior to this commit, an operator's `FCallPacked` attribute, used to specify a 1:1 mapping between a relax operator and a `PackedFunc` that implements it, was only checked in `CodegenVM`. Any...
This is a follow-up commit to https://github.com/apache/tvm/pull/16637, which updated `relax.transform.FuseOps` to provide additional parameters defining symbolic variables required by the fused functions. While this ensures that `relax.transform.FuseOps` produces well-formed Relax...
This is a follow-up to https://github.com/apache/tvm/pull/15864, which added `LegalizeOps` to the default Relax build pipeline. Since legalization may produce additional TIR PrimFuncs that require scheduling, the output of `LegalizeOps` typically...
Bumps [urllib3](https://github.com/urllib3/urllib3) from 1.26.18 to 1.26.19. Release notes Sourced from urllib3's releases. 1.26.19 🚀 urllib3 is fundraising for HTTP/2 support urllib3 is raising ~$40,000 USD to release HTTP/2 support and...
Currently the examples in https://github.com/apache/tvm/tree/main/apps are not continuously tested. Some are broken and have been for a while (e.g. `bundle_deploy`) while others are commonly used but untested (`howto_deploy`). We should...
Description This pull request aims to enhance model optimization by adding post optimization in MetaSchedule. The proposed approach involves the following steps: 1. Execution of MetaSchedule over an end-to-end model...
Bumps [braces](https://github.com/micromatch/braces) to 3.0.3 and updates ancestor dependency [jest](https://github.com/jestjs/jest/tree/HEAD/packages/jest). These dependencies need to be updated together. Updates `braces` from 3.0.2 to 3.0.3 Commits 74b2db2 3.0.3 88f1429 update eslint. lint, fix...
### Expected behavior The docker image (ci_cpu) builds successfully. ### Actual behavior failed with the following error ```sh 40.25 [ 69%] Built target nnpack_reference_layers 40.88 [ 71%] Generating src/x86_64-fma/2d-fourier-16x16.py.o 43.07...
#### Description Here is a single op: `Cast`  In TVM, when it accepts **NaN** value, it outputs **False**. However, in `PyTorch`, it outputs **True**. In Pytorch and ONNX, `Cast`...
There is an issue due to a race in the TVM/TIR optimization passes when several distinct python threads are each compiling some operator. The race occurs in the common subexpression...