fix: parse_messages_json drops content when key order is content-before-role

.github/actions/stamp-version/action.yml

@@ -0,0 +1,24 @@
+name: Stamp version
+description: Resolve the release version and write python/cactus/_version.py
+runs:
+  using: composite
+  steps:
+    - shell: bash
+      env:
+        TAG: ${{ github.event.release.tag_name }}
+        INPUT_VERSION: ${{ github.event.inputs.version }}
+      run: |
+        if [ -n "$TAG" ]; then
+          raw="${TAG#v}"
+        elif [ -n "$INPUT_VERSION" ]; then
+          raw="${INPUT_VERSION#v}"
+        else
+          raw=$(sed 's/^v//' CACTUS_VERSION)
+        fi
+        version=$(RAW="$raw" python3 -c 'import os; r=os.environ["RAW"]; print(r + ".0" if r.count(".") == 1 else r)')
+        if [ -z "$version" ]; then
+          echo "::error::No version source available (TAG, INPUT_VERSION, CACTUS_VERSION all empty)"
+          exit 1
+        fi
+        printf '__version__ = "%s"\n' "$version" > python/cactus/_version.py
+        echo "Stamped version: $version"

.github/assemble-docs.sh

@@ -0,0 +1,223 @@
+#!/usr/bin/env bash
+# Assembles site_docs/ from scattered source files and rewrites internal links.
+# Used by the release.yml workflow.
+set -euo pipefail
+
+DOCS_VERSION="${1:-}"
+
+if sed --version >/dev/null 2>&1; then
+  sedi() { sed -i "$@"; }
+else
+  sedi() { sed -i '' "$@"; }
+fi
+
+rm -rf site_docs
+mkdir -p site_docs/docs site_docs/python site_docs/apple site_docs/android \
+         site_docs/flutter site_docs/rust site_docs/swift site_docs/kotlin \
+         site_docs/react-native site_docs/blog site_docs/assets
+
+cp -r assets/* site_docs/assets/
+
+echo "docs.cactuscompute.com" > site_docs/CNAME
+
+mkdir -p site_docs/stylesheets site_docs/javascripts
+cp .github/docs-overrides/stylesheets/custom.css site_docs/stylesheets/custom.css
+cp .github/docs-overrides/javascripts/mathjax.js site_docs/javascripts/mathjax.js
+
+[ -f CONTRIBUTING.md ] && cp CONTRIBUTING.md site_docs/CONTRIBUTING.md
+[ -f DCO.md ] && cp DCO.md site_docs/DCO.md
+
+cp docs/*.md site_docs/docs/
+
+cp python/README.md site_docs/python/README.md
+cp apple/README.md site_docs/apple/README.md
+cp android/README.md site_docs/android/README.md
+cp bindings/flutter/README.md site_docs/flutter/README.md
+cp bindings/swift/README.md site_docs/swift/README.md
+cp bindings/kotlin/README.md site_docs/kotlin/README.md
+cp bindings/react-native/README.md site_docs/react-native/README.md
+cp bindings/rust/README.md site_docs/rust/README.md
+
+if [ -d blog ] && ls blog/*.md >/dev/null 2>&1; then
+  cp blog/*.md site_docs/blog/
+fi
+
+{
+  echo '---'
+  echo 'title: "Cactus"'
+  echo 'description: "Energy-efficient AI inference engine for phones, wearables, Macs, and ARM devices."'
+  echo '---'
+  echo ''
+  cat README.md
+} > site_docs/index.md
+
+sedi 's/^# Cactus$//' site_docs/index.md
+
+sedi 's|(/docs/cactus_engine\.md)|(docs/cactus_engine.md)|g' site_docs/index.md
+sedi 's|(/docs/cactus_graph\.md)|(docs/cactus_graph.md)|g' site_docs/index.md
+sedi 's|(/docs/cactus_index\.md)|(docs/cactus_index.md)|g' site_docs/index.md
+sedi 's|(/docs/cactus_kernels\.md)|(docs/cactus_kernels.md)|g' site_docs/index.md
+sedi 's|(/docs/cactus_quants\.md)|(docs/cactus_quants.md)|g' site_docs/index.md
+sedi 's|(/docs/cactus_transpiler\.md)|(docs/cactus_transpiler.md)|g' site_docs/index.md
+sedi 's|(/docs/cactus_hybrid\.md)|(docs/cactus_hybrid.md)|g' site_docs/index.md
+sedi 's|(/docs/finetuning\.md)|(docs/finetuning.md)|g' site_docs/index.md
+sedi 's|(/docs/compatibility\.md)|(docs/compatibility.md)|g' site_docs/index.md
+sedi 's|(/docs/quickstart\.md)|(docs/quickstart.md)|g' site_docs/index.md
+sedi 's|(/docs/choose-bindings\.md)|(docs/choose-bindings.md)|g' site_docs/index.md
+sedi 's|(/CONTRIBUTING\.md)|(CONTRIBUTING.md)|g' site_docs/index.md
+sedi 's|(/bindings/swift/)|(swift/README.md)|g' site_docs/index.md
+sedi 's|(/bindings/kotlin/)|(kotlin/README.md)|g' site_docs/index.md
+sedi 's|(/bindings/python/)|(python/README.md)|g' site_docs/index.md
+sedi 's|(/bindings/react-native/)|(react-native/README.md)|g' site_docs/index.md
+sedi 's|(/bindings/flutter/)|(flutter/README.md)|g' site_docs/index.md
+sedi 's|(/bindings/rust/)|(rust/README.md)|g' site_docs/index.md
+sedi 's|(/python/)|(python/README.md)|g' site_docs/index.md
+sedi 's|(/apple/)|(apple/README.md)|g' site_docs/index.md
+sedi 's|(/android/)|(android/README.md)|g' site_docs/index.md
+sedi 's|(/blog/hybrid_transcription\.md)|(blog/hybrid_transcription.md)|g' site_docs/index.md
+sedi 's|(/blog/lfm2_24b_a2b\.md)|(blog/lfm2_24b_a2b.md)|g' site_docs/index.md
+sedi 's|(/blog/parakeet\.md)|(blog/parakeet.md)|g' site_docs/index.md
+sedi 's|(/blog/lfm2\.5_350m\.md)|(blog/lfm2.5_350m.md)|g' site_docs/index.md
+sedi 's|(/blog/gemma4\.md)|(blog/gemma4.md)|g' site_docs/index.md
+sedi 's|(/blog/turboquant-h\.md)|(blog/turboquant-h.md)|g' site_docs/index.md
+
+for f in site_docs/docs/*.md; do
+  sedi 's|(/docs/cactus_engine\.md)|(cactus_engine.md)|g' "$f"
+  sedi 's|(/docs/cactus_graph\.md)|(cactus_graph.md)|g' "$f"
+  sedi 's|(/docs/cactus_index\.md)|(cactus_index.md)|g' "$f"
+  sedi 's|(/docs/cactus_kernels\.md)|(cactus_kernels.md)|g' "$f"
+  sedi 's|(/docs/cactus_quants\.md)|(cactus_quants.md)|g' "$f"
+  sedi 's|(/docs/cactus_transpiler\.md)|(cactus_transpiler.md)|g' "$f"
+  sedi 's|(/docs/cactus_hybrid\.md)|(cactus_hybrid.md)|g' "$f"
+  sedi 's|(/docs/finetuning\.md)|(finetuning.md)|g' "$f"
+  sedi 's|(/docs/compatibility\.md)|(compatibility.md)|g' "$f"
+  sedi 's|(/docs/quickstart\.md)|(quickstart.md)|g' "$f"
+  sedi 's|(/docs/choose-bindings\.md)|(choose-bindings.md)|g' "$f"
+  sedi 's|(/docs/index\.md)|(../index.md)|g' "$f"
+  sedi 's|(/blog/hybrid_transcription\.md)|(../blog/hybrid_transcription.md)|g' "$f"
+  sedi 's|(/blog/lfm2_24b_a2b\.md)|(../blog/lfm2_24b_a2b.md)|g' "$f"
+  sedi 's|(/blog/parakeet\.md)|(../blog/parakeet.md)|g' "$f"
+  sedi 's|(/blog/lfm2\.5_350m\.md)|(../blog/lfm2.5_350m.md)|g' "$f"
+  sedi 's|(/blog/gemma4\.md)|(../blog/gemma4.md)|g' "$f"
+  sedi 's|(/blog/turboquant-h\.md)|(../blog/turboquant-h.md)|g' "$f"
+  sedi 's|(/CONTRIBUTING\.md)|(../CONTRIBUTING.md)|g' "$f"
+  sedi 's|(/bindings/swift/)|(../swift/README.md)|g' "$f"
+  sedi 's|(/bindings/kotlin/)|(../kotlin/README.md)|g' "$f"
+  sedi 's|(/bindings/python/)|(../python/README.md)|g' "$f"
+  sedi 's|(/bindings/react-native/)|(../react-native/README.md)|g' "$f"
+  sedi 's|(/bindings/flutter/)|(../flutter/README.md)|g' "$f"
+  sedi 's|(/bindings/rust/)|(../rust/README.md)|g' "$f"
+  sedi 's|(/python/)|(../python/README.md)|g' "$f"
+  sedi 's|(/apple/)|(../apple/README.md)|g' "$f"
+  sedi 's|(/android/)|(../android/README.md)|g' "$f"
+done
+
+for f in site_docs/python/README.md site_docs/apple/README.md site_docs/android/README.md site_docs/flutter/README.md site_docs/swift/README.md site_docs/kotlin/README.md site_docs/react-native/README.md site_docs/rust/README.md; do
+  sedi 's|(/docs/cactus_engine\.md)|(../docs/cactus_engine.md)|g' "$f"
+  sedi 's|(/docs/cactus_graph\.md)|(../docs/cactus_graph.md)|g' "$f"
+  sedi 's|(/docs/cactus_index\.md)|(../docs/cactus_index.md)|g' "$f"
+  sedi 's|(/docs/cactus_kernels\.md)|(../docs/cactus_kernels.md)|g' "$f"
+  sedi 's|(/docs/cactus_quants\.md)|(../docs/cactus_quants.md)|g' "$f"
+  sedi 's|(/docs/cactus_transpiler\.md)|(../docs/cactus_transpiler.md)|g' "$f"
+  sedi 's|(/docs/cactus_hybrid\.md)|(../docs/cactus_hybrid.md)|g' "$f"
+  sedi 's|(/docs/finetuning\.md)|(../docs/finetuning.md)|g' "$f"
+  sedi 's|(/docs/compatibility\.md)|(../docs/compatibility.md)|g' "$f"
+  sedi 's|(/docs/quickstart\.md)|(../docs/quickstart.md)|g' "$f"
+  sedi 's|(/docs/choose-bindings\.md)|(../docs/choose-bindings.md)|g' "$f"
+  sedi 's|(/docs/index\.md)|(../index.md)|g' "$f"
+  sedi 's|(/blog/hybrid_transcription\.md)|(../blog/hybrid_transcription.md)|g' "$f"
+  sedi 's|(/blog/lfm2_24b_a2b\.md)|(../blog/lfm2_24b_a2b.md)|g' "$f"
+  sedi 's|(/blog/parakeet\.md)|(../blog/parakeet.md)|g' "$f"
+  sedi 's|(/blog/lfm2\.5_350m\.md)|(../blog/lfm2.5_350m.md)|g' "$f"
+  sedi 's|(/blog/gemma4\.md)|(../blog/gemma4.md)|g' "$f"
+  sedi 's|(/blog/turboquant-h\.md)|(../blog/turboquant-h.md)|g' "$f"
+  sedi 's|(/CONTRIBUTING\.md)|(../CONTRIBUTING.md)|g' "$f"
+  sedi 's|(/bindings/swift/)|(../swift/README.md)|g' "$f"
+  sedi 's|(/bindings/kotlin/)|(../kotlin/README.md)|g' "$f"
+  sedi 's|(/bindings/python/)|(../python/README.md)|g' "$f"
+  sedi 's|(/bindings/react-native/)|(../react-native/README.md)|g' "$f"
+  sedi 's|(/bindings/flutter/)|(../flutter/README.md)|g' "$f"
+  sedi 's|(/bindings/rust/)|(../rust/README.md)|g' "$f"
+  sedi 's|(/python/)|(../python/README.md)|g' "$f"
+  sedi 's|(/apple/)|(../apple/README.md)|g' "$f"
+  sedi 's|(/android/)|(../android/README.md)|g' "$f"
+  sedi 's|(\.\.\/README\.md)|(../index.md)|g' "$f"
+done
+
+if ls site_docs/blog/*.md >/dev/null 2>&1; then
+  for f in site_docs/blog/*.md; do
+    sedi 's|(/docs/cactus_engine\.md)|(../docs/cactus_engine.md)|g' "$f"
+    sedi 's|(/docs/cactus_graph\.md)|(../docs/cactus_graph.md)|g' "$f"
+    sedi 's|(/docs/cactus_index\.md)|(../docs/cactus_index.md)|g' "$f"
+    sedi 's|(/docs/cactus_kernels\.md)|(../docs/cactus_kernels.md)|g' "$f"
+    sedi 's|(/docs/cactus_quants\.md)|(../docs/cactus_quants.md)|g' "$f"
+    sedi 's|(/docs/cactus_transpiler\.md)|(../docs/cactus_transpiler.md)|g' "$f"
+    sedi 's|(/docs/cactus_hybrid\.md)|(../docs/cactus_hybrid.md)|g' "$f"
+    sedi 's|(/docs/finetuning\.md)|(../docs/finetuning.md)|g' "$f"
+    sedi 's|(/docs/compatibility\.md)|(../docs/compatibility.md)|g' "$f"
+    sedi 's|(/docs/quickstart\.md)|(../docs/quickstart.md)|g' "$f"
+    sedi 's|(/docs/choose-bindings\.md)|(../docs/choose-bindings.md)|g' "$f"
+    sedi 's|(/docs/index\.md)|(../index.md)|g' "$f"
+    sedi 's|(/CONTRIBUTING\.md)|(../CONTRIBUTING.md)|g' "$f"
+    sedi 's|(/bindings/swift/)|(../swift/README.md)|g' "$f"
+    sedi 's|(/bindings/kotlin/)|(../kotlin/README.md)|g' "$f"
+    sedi 's|(/bindings/python/)|(../python/README.md)|g' "$f"
+    sedi 's|(/bindings/react-native/)|(../react-native/README.md)|g' "$f"
+    sedi 's|(/bindings/flutter/)|(../flutter/README.md)|g' "$f"
+    sedi 's|(/bindings/rust/)|(../rust/README.md)|g' "$f"
+    sedi 's|(/python/)|(../python/README.md)|g' "$f"
+    sedi 's|(/apple/)|(../apple/README.md)|g' "$f"
+    sedi 's|(/android/)|(../android/README.md)|g' "$f"
+    sedi 's|(/blog/hybrid_transcription\.md)|(hybrid_transcription.md)|g' "$f"
+    sedi 's|(/blog/lfm2_24b_a2b\.md)|(lfm2_24b_a2b.md)|g' "$f"
+    sedi 's|(/blog/parakeet\.md)|(parakeet.md)|g' "$f"
+    sedi 's|(/blog/lfm2\.5_350m\.md)|(lfm2.5_350m.md)|g' "$f"
+    sedi 's|(/blog/gemma4\.md)|(gemma4.md)|g' "$f"
+    sedi 's|(/blog/turboquant-h\.md)|(turboquant-h.md)|g' "$f"
+  done
+fi
+
+if [ -f site_docs/CONTRIBUTING.md ]; then
+  sedi 's|(/docs/cactus_engine\.md)|(docs/cactus_engine.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/cactus_graph\.md)|(docs/cactus_graph.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/cactus_index\.md)|(docs/cactus_index.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/cactus_kernels\.md)|(docs/cactus_kernels.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/cactus_quants\.md)|(docs/cactus_quants.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/cactus_transpiler\.md)|(docs/cactus_transpiler.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/cactus_hybrid\.md)|(docs/cactus_hybrid.md)|g' site_docs/CONTRIBUTING.md
+  sedi 's|(/docs/index\.md)|(index.md)|g' site_docs/CONTRIBUTING.md
+fi
+
+if [ -n "$DOCS_VERSION" ]; then
+  {
+    echo "!!! note \"Version ${DOCS_VERSION}\""
+    echo "    You're viewing docs for **${DOCS_VERSION}**. If you are cloning the repository, make sure to check out this release: \`git checkout ${DOCS_VERSION}\`"
+    echo ""
+    cat site_docs/docs/quickstart.md
+  } > site_docs/docs/quickstart.tmp && mv site_docs/docs/quickstart.tmp site_docs/docs/quickstart.md
+fi
+
+for nav_path in \
+  "python/README.md" \
+  "swift/README.md" \
+  "kotlin/README.md" \
+  "flutter/README.md" \
+  "react-native/README.md" \
+  "rust/README.md" \
+  "blog/README.md" \
+  "blog/hybrid_transcription.md" \
+  "blog/lfm2_24b_a2b.md" \
+  "blog/parakeet.md" \
+  "blog/lfm2.5_350m.md" \
+  "blog/gemma4.md" \
+  "blog/turboquant-h.md" \
+  "CONTRIBUTING.md" \
+  "docs/compatibility.md"; do
+  if [ ! -f "site_docs/$nav_path" ]; then
+    grep -vF "$nav_path" mkdocs.yml > mkdocs.yml.tmp && mv mkdocs.yml.tmp mkdocs.yml
+  fi
+done
+
+if ! ls site_docs/blog/*.md >/dev/null 2>&1; then
+  grep -v "^  - Blog:" mkdocs.yml > mkdocs.yml.tmp && mv mkdocs.yml.tmp mkdocs.yml
+fi

.github/docs-overrides/javascripts/mathjax.js

@@ -0,0 +1,12 @@
+window.MathJax = {
+  tex: {
+    inlineMath: [["\\(", "\\)"]],
+    displayMath: [["\\[", "\\]"]],
+    processEscapes: true,
+    processEnvironments: true
+  },
+  options: {
+    ignoreHtmlClass: ".*|",
+    processHtmlClass: "arithmatex"
+  }
+};

.github/docs-overrides/main.html

@@ -0,0 +1,8 @@
+{% extends "base.html" %}
+
+{% block outdated %}
+  You're not viewing the latest version.
+  <a href="{{ '../' ~ base_url }}">
+    <strong>Click here to go to latest.</strong>
+  </a>
+{% endblock %}

.github/docs-overrides/stylesheets/custom.css

@@ -0,0 +1,50 @@
+@font-face {
+  font-family: 'Geist';
+  src: url('https://cdn.jsdelivr.net/npm/geist@1.3.0/dist/fonts/geist-sans/Geist-Regular.woff2') format('woff2');
+  font-weight: 400;
+  font-display: swap;
+}
+@font-face {
+  font-family: 'Geist';
+  src: url('https://cdn.jsdelivr.net/npm/geist@1.3.0/dist/fonts/geist-sans/Geist-Medium.woff2') format('woff2');
+  font-weight: 500;
+  font-display: swap;
+}
+@font-face {
+  font-family: 'Geist';
+  src: url('https://cdn.jsdelivr.net/npm/geist@1.3.0/dist/fonts/geist-sans/Geist-Bold.woff2') format('woff2');
+  font-weight: 700;
+  font-display: swap;
+}
+@font-face {
+  font-family: 'Geist Mono';
+  src: url('https://cdn.jsdelivr.net/npm/geist@1.3.0/dist/fonts/geist-mono/GeistMono-Regular.woff2') format('woff2');
+  font-weight: 400;
+  font-display: swap;
+}
+
+:root {
+  --md-text-font: 'Geist', -apple-system, BlinkMacSystemFont, sans-serif;
+  --md-code-font: 'Geist Mono', 'SFMono-Regular', 'Menlo', monospace;
+}
+
+[data-md-color-scheme="slate"] {
+  --md-default-bg-color: #000000;
+  --md-footer-bg-color: #000000;
+  --md-footer-bg-color--dark: #000000;
+}
+
+[data-md-color-scheme="slate"] .md-header,
+[data-md-color-scheme="slate"] .md-tabs,
+[data-md-color-scheme="slate"] .md-sidebar {
+  background-color: #000000;
+}
+
+.md-banner--warning {
+  background-color: #d4a017;
+  color: #000000;
+}
+
+.md-banner--warning a {
+  color: #000000;
+}

.github/workflows/build.yml

@@ -0,0 +1,46 @@
+name: Build
+
+on:
+  pull_request:
+    paths:
+      - cactus-engine/**
+      - cactus-graph/**
+      - cactus-kernels/**
+      - apple/**
+      - android/**
+      - .github/workflows/build.yml
+  push:
+    branches: [main]
+    paths:
+      - cactus-engine/**
+      - cactus-graph/**
+      - cactus-kernels/**
+      - apple/**
+      - android/**
+      - .github/workflows/build.yml
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  apple:
+    runs-on: macos-15
+    timeout-minutes: 30
+    steps:
+      - uses: actions/checkout@v4
+      - run: bash cactus-engine/build.sh
+      - run: BUILD_STATIC=true BUILD_XCFRAMEWORK=false bash apple/build.sh
+      - run: bash android/build.sh
+
+  linux:
+    runs-on: ubuntu-24.04-arm
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+      - run: sudo apt-get update && sudo apt-get install -y cmake build-essential libcurl4-openssl-dev
+      - run: bash cactus-engine/build.sh