50667cd2a4
The float wai.neural.mlicpp decodes correctly only while its encoder evaluates the entropy network bit-identically to the decoder. Its validation run, with the encoder at the runtime's default thread count and the decoder single-threaded, put one scale on the other side of a table edge at symbol 416 of 12,288 and decoded garbage from there. Matching the execution settings fixes that run, but it stays a convention between two implementations. int_mlicpp makes the agreement a property of the arithmetic: the hyper-decoder, the entropy network re-run before each of the 2*K channel-slice x checkerboard passes, and the synthesis are all integer networks, so every decoder derives the encoder's scale bucket for every element, on any platform and at any thread count. - wai-rs/src/int_mlicpp.rs: the decoder (a WIM1 payload; a one-file WMB1 model bundle, so a prior's SHA-256 pins every network and table), built from int_hyper_synth's networks, int_entropy's bucketing and the rANS decoder. - Totality. Before the decode allocates anything, each network's shapes are derived from its layer geometry (the new int_hyper_synth::net_shape, which shares its geometry with the checked run), and a header whose shapes the model does not produce, or on which any network would form a buffer over Header::MAX_TENSOR (2^26) elements, is refused. Latents go through int_entropy::dequantize (sym * 2^frac + base, None where i64 cannot hold it; Kani harness), and every network through int_hyper_synth::run_net_checked (run_net layer by layer, only while no value can leave its type). So a decode returns an image or an error for any payload, the same one with and without overflow checks. - IntModel::from_bytes, the compact network form every integer neural decoder loads, is total on damaged bytes: it no longer reserves for the layer count it reads before checking the bytes hold that many (about 480 GB for 2^32 - 1), and forms layer sizes and read offsets in checked arithmetic (they overflowed a 32-bit usize). - run_net and decode share a per-layer function and to_rgb8; every committed vector still reproduces its hash. - Registry: NeuralIntegerExact, Lossy, royalty-free (the decoder is this crate's; the patent status of the techniques was not examined). A new registry test records the basis of every NeuralIntegerExact capability. - The neural dispatch decodes it and re-checks a pinned prior against the exact bytes it decodes with. - tools/wai_int_mlicpp_export.py writes the committed vector tests/vectors/int_mlicpp/. Its weights are synthetic and seeded (no trained model is available), so the vector shows decode-equivalence and nothing about compression quality. - Tests: tests/int_mlicpp.rs (the pinned image, bucket and symbol hashes under every kernel policy and thread count; one unit in each of four chosen parameters, one per value path, changes the image; damaged payloads never panic), tests/int_mlicpp_preflight.rs (a counting global allocator: refusing a header the model cannot fit, or parsing a damaged network, allocates a few KB at most; the decoder without the pre-flight fails it, and aborts on a 3.4 PB allocation for a synthesis wider than the cap), a byte-exact-conformance golden, a neural dispatch test and unit tests. CI runs them on the conformance matrix in both profiles, on the byte-equality matrix, and in neural-gates. - Evidence: all of the above pass, and the pinned hashes reproduce, on macOS arm64 and on wasm32-wasip2 under wasmtime 49.0.1, in dev and release. x86_64 was not measured here (no x86_64 translation layer is installed); the Linux legs come from the byte-exact matrix, and the Windows runner is offline. - SPEC 5 and 7, the README, determinism-tiers 3, 5 and 6, and the CHANGELOG record it. Co-Authored-By: Claude Opus 5.5 <[email protected]>
116 lines
5.3 KiB
YAML
116 lines
5.3 KiB
YAML
name: wai cross-architecture byte-equality
|
|
|
|
# The `replicate` determinism contract, checked on real silicon. WAI claims its
|
|
# integer-exact decode paths (rANS entropy coder, integer fixed-point synthesis)
|
|
# produce BYTE-IDENTICAL output on every machine. This matrix proves it: five
|
|
# OS/arch combinations decode the same fixtures and assert the same golden
|
|
# bytes — rANS streams produced by CompressAI's upstream encoder, and an
|
|
# int_synth output hash produced on the reference machine (macOS arm64). One
|
|
# diverging byte anywhere fails the suite.
|
|
#
|
|
# Coverage is still-image AND temporal, classical AND learned: the rANS coder,
|
|
# integer fixed-point synthesis, the JPEG-AI entropy stage, the inter-frame video
|
|
# decoder (`wai.video.int_motion`), and the neural-keyframe variant
|
|
# (`wai.video.int_hyper`) whose I-frame comes out of a learned conditional codec,
|
|
# and the MLIC++-topology codec (`wai.neural.int_mlicpp`), whose entropy network runs
|
|
# before every checkerboard pass and must give every platform the encoder's scale
|
|
# bucket for every element. Video is the strictest leg — inter-frame decode is a
|
|
# feedback loop, so one divergent half-pel rounding does not stay one byte, it
|
|
# predicts the next frame and compounds for the rest of the clip.
|
|
#
|
|
# The conformance crate (`wai/byte-exact-conformance`) #[path]-includes the
|
|
# pure-integer sources from wai-rs — no system codec libraries, no ONNX runtime
|
|
# — so every leg runs in about a minute.
|
|
|
|
on:
|
|
pull_request:
|
|
paths:
|
|
- 'wai/byte-exact-conformance/**'
|
|
- 'wai/wai-rs/src/neural/int_synth.rs'
|
|
- 'wai/wai-rs/src/neural/jpegai_mans.rs'
|
|
# Every source the crate #[path]-includes must be able to fire this
|
|
# workflow, or a change can break cross-arch equality and merge green.
|
|
- 'wai/wai-rs/src/int_synth.rs'
|
|
- 'wai/wai-rs/src/int_entropy.rs'
|
|
- 'wai/wai-rs/src/jpegai_mans.rs'
|
|
- 'wai/wai-rs/src/rans.rs'
|
|
- 'wai/wai-rs/src/int_video.rs'
|
|
- 'wai/wai-rs/src/int_transform.rs'
|
|
- 'wai/wai-rs/src/int_hyper_synth.rs'
|
|
- 'wai/wai-rs/src/int_mlicpp.rs'
|
|
# The fast kernels int_synth, int_transform and int_hyper_synth dispatch to
|
|
# by default: the goldens here decode through them.
|
|
- 'wai/wai-rs/src/int_kernels.rs'
|
|
- 'wai/wai-rs/tests/vectors/int_video/**'
|
|
- 'wai/wai-rs/tests/vectors/int_hyper/**'
|
|
- 'wai/wai-rs/tests/vectors/int_mlicpp/**'
|
|
- '.github/workflows/wai-byte-equality.yml'
|
|
push:
|
|
branches:
|
|
- main
|
|
paths:
|
|
- 'wai/byte-exact-conformance/**'
|
|
- 'wai/wai-rs/src/neural/int_synth.rs'
|
|
- 'wai/wai-rs/src/neural/jpegai_mans.rs'
|
|
# Every source the crate #[path]-includes must be able to fire this
|
|
# workflow, or a change can break cross-arch equality and merge green.
|
|
- 'wai/wai-rs/src/int_synth.rs'
|
|
- 'wai/wai-rs/src/int_entropy.rs'
|
|
- 'wai/wai-rs/src/jpegai_mans.rs'
|
|
- 'wai/wai-rs/src/rans.rs'
|
|
- 'wai/wai-rs/src/int_video.rs'
|
|
- 'wai/wai-rs/src/int_transform.rs'
|
|
- 'wai/wai-rs/src/int_hyper_synth.rs'
|
|
- 'wai/wai-rs/src/int_mlicpp.rs'
|
|
# The fast kernels int_synth, int_transform and int_hyper_synth dispatch to
|
|
# by default: the goldens here decode through them.
|
|
- 'wai/wai-rs/src/int_kernels.rs'
|
|
- 'wai/wai-rs/tests/vectors/int_video/**'
|
|
- 'wai/wai-rs/tests/vectors/int_hyper/**'
|
|
- 'wai/wai-rs/tests/vectors/int_mlicpp/**'
|
|
- '.github/workflows/wai-byte-equality.yml'
|
|
|
|
jobs:
|
|
byte-equality:
|
|
name: ${{ matrix.label }}
|
|
runs-on: ${{ matrix.os }}
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
# Self-hosted fleet labels, not GitHub-hosted images. macos-x86_64 is
|
|
# dropped: no Intel-Mac runner exists in the fleet, and by the same
|
|
# argument the cargo-classes matrix already makes, it adds no arch
|
|
# coverage — x86_64 is held by linux-x86_64 and windows-x86_64, arm64
|
|
# by linux-aarch64 and the macos-arm64 reference.
|
|
include:
|
|
- { os: linux, label: linux-x86_64 }
|
|
- { os: linux-arm64, label: linux-aarch64 }
|
|
- { os: macos, label: macos-arm64 (reference) }
|
|
- { os: windows, label: windows-x86_64 }
|
|
steps:
|
|
- name: Checkout
|
|
uses: actions/checkout@v4
|
|
|
|
# unix: bootstrap rustup (docker image ships no Rust; macOS host has it but
|
|
# launchd's PATH omits ~/.cargo/bin). Windows is a host executor with rustup
|
|
# already present, and must use pwsh — its `bash` is the WSL stub.
|
|
- name: Install Rust 1.98.0 (unix)
|
|
if: matrix.os != 'windows'
|
|
shell: bash
|
|
run: |
|
|
export PATH="$HOME/.cargo/bin:$PATH"
|
|
command -v rustup >/dev/null 2>&1 || \
|
|
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain 1.98.0
|
|
rustup toolchain install 1.98.0 --profile minimal
|
|
rustup default 1.98.0
|
|
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
|
|
- name: Install Rust 1.98.0 (Windows)
|
|
if: matrix.os == 'windows'
|
|
shell: pwsh
|
|
run: |
|
|
rustup toolchain install 1.98.0 --profile minimal
|
|
rustup default 1.98.0
|
|
|
|
- name: Decode fixtures, assert golden bytes
|
|
working-directory: wai/byte-exact-conformance
|
|
run: cargo test --release -- --nocapture
|