Document the uv install route, and the prerequisites it cannot cover - #1
Open
rysk-t wants to merge 1 commit into
Open
Document the uv install route, and the prerequisites it cannot cover#1rysk-t wants to merge 1 commit into
rysk-t wants to merge 1 commit into
Conversation
The install guide only described the conda route, so nothing on the site mentioned uv even though the repository ships pyproject.toml and uv.lock, and uv is now the only supported route on macOS. It also asked readers to install the CUDA toolkit, which is not needed: the PyTorch wheels carry their own CUDA runtime, so what has to be present is the driver. The stable install guide is restructured around the two routes with a shared prerequisites section: the Chromium browser Eel needs (with what it looks for on each OS, and why Edge and Safari do not work), the NVIDIA driver, and the Apple Silicon requirements -- macOS 14+, the Xcode Command Line Tools that the imgui and gevent source builds need, and the Camera / Input Monitoring / Screen Recording prompts. A section on choosing the compute device covers YORU_DEVICE and the training-GUI selector added in v1.1.2. home.md mirrors the README changes, and the stable version pointers move to v1.1.2 across the site. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01G3rHpfKzAnmYkxhmvHKwiP
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The site's install guide describes only the conda route, so nothing here mentions
uv— even though the YORU repository ships its ownpyproject.tomlanduv.lock, and as of v1.1.2 uv is the only supported route on macOS. This brings the docs in line with Kamikouchi-lab/YORU#41 and the v1.1.2 release._guides/01_install.md— restructured around two routesInstall with uv (new): how to install uv itself (
wingeton Windows, the install script or Homebrew elsewhere), thengit clone→uv sync→uv run yoru. Notes that no system Python and no conda are needed, since uv fetches the Python 3.10 the project pins, and that YORU has to be run from the repository root — the launcher resolvesweb/andconfig/relative to the working directory.Install with conda: unchanged apart from the CUDA note below. The step numbering skipped 2, which is also fixed.
Prerequisites (new), the pieces neither installer can provide:
chromemode, so the guide now says exactly where Eel looks on each OS (theApp Paths\chrome.exeregistry key on Windows;Google Chrome.app→Chromium.app→mdfindon macOS; four binary names onPATHon Linux) and quotes the error you get when it finds nothing. Microsoft Edge is Chromium-based but registers asmsedge.exe, so Eel does not find it, and Safari is unsupported by Eel — worth stating, since both are reasonable things to try. Only the launcher needs a browser; every other GUI is a native window.nvidia-smi.xcode-select --installbeforeuv syncbecauseimguiandgeventpublish no arm64 wheels and are compiled during the sync, and the Camera / Input Monitoring / Screen Recording prompts that appear on first use.Choosing the compute device (new): the CUDA → MPS → CPU order, the
YORU_DEVICEoverride and the training-GUI selector added in v1.1.2, plus the MPS-vs-CPU realtime inference note.The CUDA toolkit is not required
Both install pages said to install the CUDA toolkit. It is not needed on either route: the PyTorch wheels ship the CUDA runtime themselves (
cudart, cuBLAS and cuDNN live inside the installedtorchpackage). Verified on a Windows box with no toolkit at all — nonvcconPATH,CUDA_PATHunset — wheretorch.version.cudareports 12.4 and CUDA training runs.The pages now ask for the driver, and note that the
CUDA Versionprinted bynvidia-smiis the driver's ceiling rather than the version in use, which is the usual source of the confusion. The toolkit is presented as optional, fornvccand the profilers.Version pointers → v1.1.2
_config.yml, thehome.mdversions table and guides link, the banner in_guides/01_install.md, the stable-release references inbeta.md(plus a v1.1.2 row in its version history), and the stable pointer in_beta_guides/01_install.md.home.mdalso mirrors the README changes: an "Install via uv" section, a "Compute device" section, and Requirements updated with the OS matrix, the driver requirement and a Software list.Note on the beta pages
I deliberately kept these to a minimum, since Beta 2 is a different code line: only the stable-version pointer and the CUDA toolkit correction (which applies to the beta's wheels just the same). The beta guide's existing "Alternative: install with uv" section is left as is. Drop that commit hunk if you would rather the beta pages move separately.