| commit | 8ab68b674c5426baf84085514ad9b3eb238a7d19 | [log] [tgz] |
|---|---|---|
| author | Scott Todd <scott.todd0@gmail.com> | Tue Mar 26 14:04:42 2024 -0700 |
| committer | GitHub <noreply@github.com> | Tue Mar 26 14:04:42 2024 -0700 |
| tree | e162e7d5487773bfeec4acb5c1e9f3ac1acc8c3a | |
| parent | 61a1f2e1c7958915bc54bf2a0c477701d46a895c [diff] |
Start testing real weight models from external test suite. (#16801) See https://github.com/nod-ai/SHARK-TestSuite/pull/103 for the changes to the external test suite itself, which this pulls in. The test suite now has cases like [`pytorch/models/resnet50/`](https://github.com/nod-ai/SHARK-TestSuite/tree/main/iree_tests/pytorch/models/resnet50) which include: * `resnet50.mlirbc` (150 KB, stored in the repo using git LFS) * `splats.irpa` (72 KB, stored in the repo using git LFS) * `test_cases.json` with metadata pointing at `real_weights.irpa` (195 MB) and input/output `.bin` or `.npy` files * flagfiles for splats (running but not checking results, since they are 0s) and real weights (running and checking results) Starting with just resnet50 and opt-125M, but I have SDXL models (unet, clip, vae) just about ready to go in too. Now IREE 1. Runs the test cases in the `onnx/` folder using one config file (with flags to use and XFAIL lists) and set of pytest options (max parallelism) 2. Downloads the remote files from Azure for "real weight" PyTorch tests 3. Runs the test cases in the `pytorch/` folder using a different config file (local-task instead of local-sync, empty XFAIL list) and set of pytest options (no parallelism) ci-exactly: build_packages,regression_test_amdgpu_vulkan,regression_test_cpu
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.