Repository navigation
Lower explicit runtime slice sizes in CoreML - #11
Draft
matthargett wants to merge 3 commits into
Draft
matthargett wants to merge 3 commits into
matthargett wants to merge 3 commits into
Conversation
Documentation Build ✅Documentation builds successfully! 🎉 Preview will be available after merge. |
This was referenced Sep 24, 2026
Documentation Build: passedMkDocs pages and Rust API docs build successfully. Preview will be available after merge. |
Documentation Build: passedMkDocs pages and Rust API docs build successfully. Preview will be available after merge. |
matthargett
pushed a commit
that referenced
this pull request
Oct 19, 2026
## Summary Release 0.6.0: the first release of the WebNN `MLContext` API from `main`. The 0.5.x releases on crates.io were cut from the `v0.5` branch and are the earlier converter and loader crate. - `Cargo.toml`: version `0.5.11` -> `0.6.0` in the package and in both cargo-c metadata sections; `Cargo.lock` updated for the version bump only. - `CHANGELOG.md`: the Unreleased section becomes `[0.6.0]`, with the entries missing since rustnn#243 added (C/C++ API and Doxygen reference, CoreML, LiteRT, CANN, WPT harness, TensorRT), six **Breaking** lines (rustnn#185/rustnn#245, rustnn#246, rustnn#260, rustnn#264, rustnn#266) and three stale statements corrected (PASS snapshots, the removed spec-mapping page, the `conv2_with_options` alias that rustnn#264 removed). - `README.md`, `docs/index.md`, `docs/user-guide/getting-started.md`: the "not yet published, use the git dependency" text now points at `rustnn = { version = "0.6", features = [...] }`, with the git line kept for following `main`. - `publish.yml`: the release tag must match the `Cargo.toml` version before publishing. `ci.yml`: the `version-check` job is removed; it grepped `pyproject.toml` (deleted in #11) and had no tag trigger, so it never ran. `setup.md` and the workflows README updated accordingly, including the stale `snapshot-sync.yml` name. - Removed `.github/workflows/rustnn-gate.yml.orig`, a committed merge leftover. ## Before tagging - [x] **webnn-graph release.** `cargo publish --dry-run` fails: "dependency `webnn-graph` does not specify a version". webnn-graph `main` is 5 commits past 0.3.0 (rustnn needs the packed 4-bit SafeTensors commit) and still declares 0.3.0. Bump it (0.4.0; ONNX support became optional), give it a versioned `webnn-onnx-utils` dependency (`0.1.1` is published and equals `main`), publish, then switch both git dependencies in `Cargo.toml` to versions (see the TODO comment there). - [ ] Set the changelog date to the tag date, then: merge, tag `v0.6.0`, create the GitHub release. `publish.yml` publishes; docs.rs builds with `onnx-runtime`, `trtx-runtime`, `cann-runtime`, `dynamic-inputs` (trtx-sys and trtx build on docs.rs today). Merge and publish on the same day, since the docs now describe the crates.io dependency. ## Validation - `cargo check --features onnx-runtime` on the branch: clean. - `cargo publish --dry-run`: fails on the unversioned `webnn-graph` dependency (expected, see above). - Docs changes are text only; `make test` not run (no Rust code changed besides the version). ## Documentation - [x] README, getting started and the landing page describe the crates.io release. - [x] `docs/development/setup.md` and `.github/workflows/README.md` match the workflow changes. - [x] Operator support unchanged; no rustdoc changes.
This branch has not been deployed
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
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.
Summary
Lower explicit, input-bound named Slice sizes to runtime MIL
slice_by_sizesizes instead of allocation maxima. Check active bounds before prediction, preserve fixed sizes, and reject unknown/derived labels, unsupported strides and overflowing bounds. Other backends reject named sizes unless they have a runtime lowering.This stacks on #10. Refreshed against current main, preserving the GraphRecorder importer and the newer shape/gather fixes.
Scope
Names are opaque equality labels, not expressions. This cannot repair exports that have already frozen a shape dependency into a literal maximum. webnn-graph #20 tracks producer-based Slice/Range provenance; the broader API is in upstream #225. Neither is a prerequisite for this deliberately limited lowering.
Validation
make test, strict CoreML/dynamic-input Clippy, andmake test-coremlpass on the refreshed revision.Still a draft: the refreshed source has not yet completed a new phone/full-model acceptance run. Earlier device probes and fixed-cache SmolLM runs in #222 do not establish that growing-cache SmolLM is fixed.