tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
This change adds a callback for visualizing the FLOPS during tuning (AutoTVM & AutoScheduler (**NEW**)) over time using Matplotlib. ## Usage ### With Python Create callback (`multi=False`: 1 window per...
When I compile the model with target ```' webgpu -keys=webgpu,gpu -max_num_threads=256 '; runtime = ' llvm --system-lib -mtriple=wasm32-emscripten -mattr=+bulk-memory '```, I get the error: ```Check failed: target->kind->device_type == kDLVulkan (15...
This is a less-than-ideal fix for #12776. It disables reruns for crashes in xdist workers, which gets rid of the pytest internal failure and correctly reports the test name. Ideally...
initial version building TVM on windows with USE_MICRO enabled
See https://github.com/apache/tvm/pull/12785#issuecomment-1252588590 on https://github.com/apache/tvm/pull/12785: This happened while one job was failed (MacOS) and 2 were still running (Android and Windows). The rerun request went to workflow 3091384727 which was the...
Introduces (disabled) unit tests that can gradually be enabled as we make progress on primfunc slicing (a.k.a. "functionalization"). cc @Hzfengsy @junrushao1994
This change reduced the code size of AOT test runner binary for CMSIS-NN flow. Presently, integration tests for CMSIS-NN generate binaries which includes all the CMSIS-NN kernels sources. This fix...
Previously ccache was disabled because of possible issues with hexagon. Re-enabling it to provide a best effort attempt at using it. @driazati @cconvey
This pull request adds support to enable fp16 and uint8 support for depth to space operator. This uses layout transform to support the given layouts and the required test cases...
