tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
The PyTorch model with `hardswish` operator and input_dtype=float64 crashed when load to relay. ## Actual behavior ``` data types float64 and float32 do not match in BroadcastRel data types float64...
Implementation of Col2Im operation from ONNX opset on TVM side
onnx_model = onnx.load(path_onnx) model_shape = {"input": [1,3,224,224]} model, params = relay.frontend.from_onnx(onnx_model, model_shape) target = "llvm -mtriple=mipsel-linux-gnu" compiler = "path/to/mips-linux-gnu/bin/mips-linux-gnu-gcc" with tvm.transform.PassContext(opt_level=3): lib = relay.build(model, target=target, params=params) lib.export_library(args.tvm, cc=compiler, options=options) Traceback...
Thanks for participating in the TVM community! We use https://discuss.tvm.ai for any general usage questions and discussions. The issue tracker is used for actionable items such as feature proposals discussion,...
Fixed sync warp being incorrectly treated no-op in cude codegen cc @tqchen
Major changes of this pull request: - Change the fp8-related test `requires_cuda_compute_version` from 9 to 8.9 (since sm_89 ada architecture also supports fp8 tensor cores, which is the platform I...
As reported in #16877, shape inference performed during a Relax transformation may produce different results than shape inference performed during TVMScript parsing. While Relax transformations call `Analyzer::MarkGlobalNonNegValue` for each shape...
> [Last release v0.15.0](https://github.com/apache/tvm/issues/16277) was proposed at the end of Jan. and the release day is 25 Jan, more detail refer [v0.15.0 release schedule](https://github.com/apache/tvm/issues/16277). It has been almost **three months**...
Hi I use conda under windows for a development environment, I encounter a problem with TVM, namely that I have performed all the installation steps the error below  any...
When used on a block with a init statement, blockize creates a separate init block that is not discoverable by any means. This hinders further scheduling, like tensorizing the init...
