| commit | 46d934711cd0c3222f74bd7e98a00e8538b5d002 | [log] [tgz] |
|---|---|---|
| author | bjacob <benoitjacob@google.com> | Thu Dec 14 17:44:42 2023 -0500 |
| committer | GitHub <noreply@github.com> | Thu Dec 14 14:44:42 2023 -0800 |
| tree | cac6268ef5030208ef8be48f178c3629a536d4c5 | |
| parent | bc0b7d42bbd04b4af0a86eb56556ad8fcc6985a2 [diff] |
Tweaks to e2e matmul tests (#15930) This restricts the range of random `bf16` test matrix elements, so that we avoid larger-accumulation-size matmul tests getting accumulators >= 256, at which point integers are no longer exactly representable in `bf16` and accumulation becomes very inaccurate. This is what was causing test failures on large tests specifically in https://github.com/openxla/iree/pull/15911 . This also fixes the computation of the window of the test matrices that is displayed on error. The problem arose when diagnosing a matmul test error in cases where M and N are both small, and K is larger. The new window computation is also just simpler.
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
See our website for more information.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.