commit | a8d17fa96e445e3b93c70b0e6aaa24a58a7777f0 | [log] [tgz] |
---|---|---|
author | Scott Todd <scott.todd0@gmail.com> | Fri Dec 06 09:56:27 2024 -0800 |
committer | GitHub <noreply@github.com> | Fri Dec 06 17:56:27 2024 +0000 |
tree | 6a9f4ee8ce9e2c0f2435a2fabdd0e0d10b04fe22 | |
parent | 1c73358fdc244e91b4c8e27fa0adc74c7c6c1fe3 [diff] |
Include prereleases from downstream projects in pip release links page. (#19391) This fixes https://github.com/iree-org/iree/issues/19193. Nightly releases of packages from https://github.com/iree-org/iree-turbine will now be available from https://iree.dev/pip-release-links.html with (for example) ```bash pip install --pre --find-links https://iree.dev/pip-release-links.html iree-turbine ``` Previously, this relied on its own index page at https://github.com/iree-org/iree-turbine/releases/expanded_assets/dev-wheels . While this link will continue to work, the consolidated page is more convenient for users. ## Projects included I'm starting with just iree-turbine. See the notes below about other packages. We could still switch to a [PEP 503-compliant package index](https://peps.python.org/pep-0503/) (usable with `--extra-index-url`, _not_ `-f, --find-links`), like what PyTorch [provides](https://pytorch.org/get-started/locally/) at https://download.pytorch.org/whl/cpu. Note the 50+ packages listed there, included mirrors of a few versions of all of PyTorch's dependencies. ## Testing * Publish action: https://github.com/ScottTodd/iree/actions/runs/12188757072/job/34002585332 * Pushed to GitHub pages: https://scotttodd.github.io/iree/pip-release-links.html * Installed using the release links: ```bash python3.11 -m venv 3.11.venv source 3.11.venv/bin/activate pip install --pre --find-links https://scotttodd.github.io/iree/pip-release-links.html iree-turbine --no-deps # Looking in links: https://scotttodd.github.io/iree/pip-release-links.html # Collecting iree-turbine # Downloading https://github.com/iree-org/iree-turbine/releases/download/dev-wheels/iree_turbine-3.1.0rc20241205-py3-none-any.whl (292 kB) # ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 292.3/292.3 kB 8.1 MB/s eta 0:00:00 # Installing collected packages: iree-turbine # Successfully installed iree-turbine-3.1.0rc20241205 ``` ## Other considerations We may want to also include downstream packages like https://github.com/nod-ai/shark-ai (release index page: https://github.com/nod-ai/shark-ai/releases/expanded_assets/dev-wheels). I'm currently trying to simplify workflows for that project ([example here](https://github.com/nod-ai/shark-ai/blob/1763a82a5ebc002d47edb5bc53106f9b312069b9/.github/workflows/ci-shark-ai.yml#L48-L69)). I left some notes in the workflow file too, but I could see us extending this to also include torch-mlir and stablehlo packages. The extra release scraping doesn't cost us much CI time, git storage on the `gh-pages` branch, or network bandwidth on the website. I also don't foresee any issues with making it easier to install nightly versions of those packages. I would pause a bit if the list grows too much beyond that though. ### Note on the `shark-ai` package The `shark-ai` meta package (https://pypi.org/project/shark-ai/, https://github.com/nod-ai/shark-ai/tree/main/shark-ai) pins dependencies in ways that isn't currently compatible with this package installation flow. Specifically, the latest nightly releases pin `shortfin==3.1.0` which can't be resolved by a release candidate like `3.1.0rc20241205`, so this doesn't work as one might expect: ```bash pip install --pre --find-links https://iree.dev/pip-release-links.html shark-ai # Successfully installed shark-ai-3.0.0 # Test with an explicit version to demonstrate the issue: pip install --pre --find-links https://scotttodd.github.io/iree/pip-release-links.html \ shark-ai==3.1.0rc20241205 # ERROR: Could not find a version that satisfies the requirement shortfin==3.1.0 (from shark-ai) (from versions: 2.9.0rc20241108, 2.9.0rc20241109, 2.9.0rc20241110, 2.9.0rc20241111, 2.9.0, 2.9.1rc20241112, 2.9.1rc20241113, 2.9.1rc20241114, 2.9.1, 2.9.2rc20241115, 2.9.2rc20241116, 2.9.2rc20241117, 2.9.2rc20241118, 2.9.2, 3.0.0rc20241118, 3.0.0, 3.1.0rc20241119, 3.1.0rc20241120, 3.1.0rc20241121, 3.1.0rc20241122, 3.1.0rc20241123, 3.1.0rc20241124, 3.1.0rc20241125, 3.1.0rc20241126, 3.1.0rc20241127, 3.1.0rc20241128, 3.1.0rc20241129, 3.1.0rc20241130, 3.1.0rc20241201, 3.1.0rc20241202, 3.1.0rc20241203, 3.1.0rc20241204, 3.1.0rc20241205) # ERROR: No matching distribution found for shortfin==3.1.0 ```
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
Package | Release status |
---|---|
GitHub release (stable) | |
GitHub release (nightly) | |
Python iree-base-compiler | |
Python iree-base-runtime |
Host platform | Build status |
---|---|
Linux | |
macOS | |
Windows |
For the full list of workflows see https://iree.dev/developers/general/github-actions/.
See our website for more information.
Community meeting recordings: IREE YouTube channel
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.