fix: remove GPU backends from GitHub runner builds - CPU only #122

Workflow file for this run

.github/workflows/release.yml at 07e31e0

	name: Release

	on:
	push:
	tags:
	- 'v*'
	branches:
	- test-release # Test on specific branch
	workflow_dispatch: # Allow manual testing

	jobs:
	# 🚧 MANDATORY RELEASE GATES - ALL MUST PASS OR ENTIRE RELEASE STOPS
	preflight:
	name: "🚧 Release Gates - MANDATORY VALIDATION"
	runs-on: ubuntu-latest
	env:
	CARGO_NET_GIT_FETCH_WITH_CLI: true
	outputs:
	should_publish: ${{ steps.gates.outputs.should_publish }}
	steps:
	- uses: actions/checkout@v4

	- name: Configure Git for private dependencies
	run: \|
	git config --global url."https://${{ secrets.VISION_PRIVATE_TOKEN }}@github.com/".insteadOf "https://github.com/"

	- name: Install Rust
	uses: dtolnay/rust-toolchain@stable

	- name: "🚧 GATE 1/7: Core Build Validation"
	run: \|
	echo "::group::Gate 1: Core Build"
	cargo build --release --no-default-features --features huggingface
	echo "✅ Core build successful"
	echo "::endgroup::"

	- name: "🚧 GATE 2/7: CUDA Build Validation (No Timeout - Can Take Hours)"
	run: \|
	echo "::group::Gate 2: CUDA Build"
	echo "⏳ CUDA compilation can take 19+ hours - letting it run to natural completion"

	# Check if CUDA Toolkit is available first
	if command -v nvcc >/dev/null 2>&1; then
	echo "✅ CUDA Toolkit found, attempting CUDA build..."
	cargo build --release --no-default-features --features llama-cuda
	echo "✅ CUDA build completed successfully"
	else
	echo "⚠️ CUDA Toolkit not found on runner (nvcc not available)"
	echo "🔄 Validating CPU-only llama build instead..."

	# Validate that CPU-only llama build works
	cargo build --release --no-default-features --features llama
	echo "✅ CPU-only llama build completed successfully"
	echo "📝 Note: CUDA validation skipped - this is expected on standard GitHub runners"
	fi
	echo "::endgroup::"

	- name: "🚧 GATE 3/7: Template Packaging Validation (Issue #60 Protection)"
	run: \|
	echo "::group::Gate 3: Template Packaging"
	# Check for Docker templates with OS-agnostic path handling
	# Use --allow-dirty to handle uncommitted Cargo.lock changes from dependency resolution
	if cargo package --allow-dirty --list \| grep -E "(^\|[/\\\\])templates[/\\\\]docker[/\\\\]Dockerfile$" > /dev/null; then
	echo "✅ Docker templates properly included in package"
	else
	echo "❌ Required Docker template missing from package - Issue #60 regression!"
	echo "Package contents:"
	cargo package --allow-dirty --list \| grep -i docker \|\| echo "No docker files found"
	exit 1
	fi
	echo "::endgroup::"

	- name: "🚧 GATE 4/7: Binary Size Constitutional Limit (20MB)"
	run: \|
	echo "::group::Gate 4: Binary Size"
	size=$(stat -c%s target/release/shimmy 2>/dev/null \|\| echo "0")
	max_size=$((20 * 1024 * 1024))
	if [ "$size" -gt "$max_size" ]; then
	echo "❌ Binary size ${size} exceeds constitutional limit of ${max_size} bytes"
	exit 1
	fi
	echo "✅ Binary size ${size} bytes within constitutional limit"
	echo "::endgroup::"

	- name: "🚧 GATE 5/7: Test Suite Validation"
	run: \|
	echo "::group::Gate 5: Tests"
	# Run tests that don't require model files (exclude integration tests needing .gguf files)
	cargo test --lib --no-default-features --features huggingface -- --skip test_generate_handler --skip test_chat_completions
	echo "✅ Core tests passing"
	echo "::endgroup::"

	- name: "🚧 GATE 5.5/7: Issue Regression Tests"
	run: \|
	echo "::group::Gate 5.5: Issue Regression Prevention"
	echo "🔄 Running issue-specific regression tests to prevent user-reported bug regressions..."

	# Test Issue #111 - GPU metrics endpoint
	cargo test --test regression_tests test_issue_111_gpu_metrics_endpoint --no-default-features --features huggingface
	echo "✅ Issue #111 (GPU metrics): Regression test passed"

	# Test Issue #112 - SafeTensors engine selection
	cargo test --test regression_tests test_issue_112_safetensors_engine_selection --no-default-features --features huggingface
	echo "✅ Issue #112 (SafeTensors): Regression test passed"

	# Test Issue #113 - OpenAI API frontend compatibility
	cargo test --test regression_tests test_issue_113_openai_api_frontend_compatibility --no-default-features --features huggingface
	echo "✅ Issue #113 (OpenAI compatibility): Regression test passed"

	# Test Issue #114 - MLX distribution features
	cargo test --test regression_tests test_issue_114_mlx_distribution_features --no-default-features --features huggingface
	echo "✅ Issue #114 (MLX distribution): Regression test passed"

	# Test Issue #13 - Qwen model template detection
	cargo test --test regression_tests test_qwen_model_template_detection --no-default-features --features huggingface
	echo "✅ Issue #13 (Qwen templates): Regression test passed"

	# Test Issue #12 - Custom model directories
	cargo test --test regression_tests test_custom_model_directory_environment_variables --no-default-features --features huggingface
	echo "✅ Issue #12 (Custom directories): Regression test passed"

	echo "✅ All issue regression tests passed - no user-reported bug regressions detected"
	echo "::endgroup::"

	- name: "🚧 GATE 6/7: Documentation Validation"
	run: \|
	echo "::group::Gate 6: Documentation"

	# Check if CUDA Toolkit is available for documentation build
	if command -v nvcc >/dev/null 2>&1; then
	echo "✅ CUDA Toolkit found, building docs with all features..."
	cargo doc --no-deps --all-features
	echo "✅ Documentation with all features built successfully"
	else
	echo "⚠️ CUDA Toolkit not found on runner (nvcc not available)"
	echo "🔄 Building documentation without CUDA features..."

	# Build docs without CUDA features to avoid build failures
	cargo doc --no-deps --features "huggingface,llama,mlx"
	echo "✅ Documentation built successfully (CUDA features excluded)"
	echo "📝 Note: CUDA documentation skipped - this is expected on standard GitHub runners"
	fi
	echo "::endgroup::"

	- name: "🚧 GATE 7/7: Crates.io Publication Validation"
	run: \|
	echo "::group::Gate 7: Crates.io Validation"
	echo "🧪 Testing crates.io publication readiness..."

	# Handle uncommitted Cargo.lock (common issue)
	if git status --porcelain \| grep -q "Cargo.lock"; then
	echo "⚠️ Cargo.lock has uncommitted changes (expected from dependency resolution)"
	echo "🔄 Using --allow-dirty flag for crates.io validation"
	DIRTY_FLAG="--allow-dirty"
	else
	echo "✅ No uncommitted changes detected"
	DIRTY_FLAG=""
	fi

	# Dry-run validation to catch issues before real release
	echo "🔍 Running cargo publish --dry-run..."
	if cargo publish --dry-run $DIRTY_FLAG; then
	echo "✅ Crates.io dry-run validation PASSED"
	echo "📦 Package builds successfully and is ready for publication"
	else
	echo "❌ Crates.io dry-run validation FAILED"
	echo "🚫 Blocking release until crates.io issues are resolved"
	exit 1
	fi
	echo "::endgroup::"

	- name: "🎯 RELEASE GATES SUMMARY"
	id: gates
	run: \|
	echo "🎉 ALL 8 MANDATORY GATES PASSED!"
	echo "✅ Gate 1: Core Build"
	echo "✅ Gate 2: CUDA Timeout Protection (Issue #59)"
	echo "✅ Gate 3: Template Packaging (Issue #60)"
	echo "✅ Gate 4: Binary Size Constitutional Limit"
	echo "✅ Gate 5: Test Suite"
	echo "✅ Gate 5.5: Issue Regression Prevention"
	echo "✅ Gate 6: Documentation"
	echo "✅ Gate 7: Crates.io Publication Validation"
	echo "should_publish=true" >> $GITHUB_OUTPUT

	# 🚀 REUSE GATE BINARIES - NO RECOMPILATION WASTE
	- name: "Upload gate binaries as artifacts"
	uses: actions/upload-artifact@v4
	with:
	name: gate-built-binary
	path: target/release/shimmy

	# 🚀 LINUX BINARY - REUSE FROM GATES (NO RECOMPILATION)
	reuse-gate-binary:
	needs: preflight
	if: needs.preflight.outputs.should_publish == 'true'
	runs-on: ubuntu-latest
	steps:
	- name: Download gate-built binary
	uses: actions/download-artifact@v4
	with:
	name: gate-built-binary
	path: ./

	- name: Upload as Linux x86_64 artifact (Kitchen Sink)
	uses: actions/upload-artifact@v4
	with:
	name: shimmy-linux-x86_64
	path: shimmy

	# 🚀 KITCHEN SINK BUILDS - All GPU backends in one binary per platform
	# Auto-detection at runtime via --gpu-backend auto (default)
	build:
	needs: preflight
	if: needs.preflight.outputs.should_publish == 'true'
	strategy:
	matrix:
	include:
	# Linux x86_64 - CPU only for GitHub runners
	# NOTE: GPU backends excluded - requires system libraries not on runners
	- os: ubuntu-latest
	target: x86_64-unknown-linux-gnu
	binary-name: shimmy
	artifact-name: shimmy-linux-x86_64
	features: huggingface,llama,vision

	# Linux ARM64 - CPU only (GPU rare on ARM servers)
	- os: ubuntu-latest
	target: aarch64-unknown-linux-gnu
	binary-name: shimmy
	artifact-name: shimmy-linux-aarch64
	features: huggingface,llama,vision
	use-cross: true

	# Windows x64 - CPU only for GitHub runners
	# NOTE: GPU backends excluded - requires system libraries not on runners
	- os: windows-latest
	target: x86_64-pc-windows-msvc
	binary-name: shimmy.exe
	artifact-name: shimmy-windows-x86_64.exe
	features: huggingface,llama,vision

	# macOS Intel - CPU only (MLX requires Apple Silicon)
	- os: macos-latest
	target: x86_64-apple-darwin
	binary-name: shimmy
	artifact-name: shimmy-macos-intel
	features: huggingface,llama,vision

	# macOS ARM64 - MLX GPU for Apple Silicon
	- os: macos-latest
	target: aarch64-apple-darwin
	binary-name: shimmy
	artifact-name: shimmy-macos-arm64
	features: huggingface,llama,mlx,vision

	runs-on: ${{ matrix.os }}
	env:
	CARGO_NET_GIT_FETCH_WITH_CLI: true
	GGML_CUDA_NO_GIT_VER: "1" # Disable git version check in llama.cpp CMake
	steps:
	- uses: actions/checkout@v4

	- name: Configure Git for private dependencies
	run: \|
	git config --global url."https://${{ secrets.VISION_PRIVATE_TOKEN }}@github.com/".insteadOf "https://github.com/"

	- name: Install Rust
	uses: dtolnay/rust-toolchain@stable
	with:
	targets: ${{ matrix.target }}

	- name: Install musl-tools (for musl builds)
	if: matrix.target == 'x86_64-unknown-linux-musl'
	run: \|
	sudo apt-get update
	sudo apt-get install -y musl-tools musl-dev

	- name: Install cross for ARM64 cross-compilation
	if: matrix.use-cross == true
	run: \|
	cargo install cross --git https://github.com/cross-rs/cross

	- name: Build binary
	shell: bash
	env:
	CROSS_NO_WARNINGS: 1
	MACOSX_DEPLOYMENT_TARGET: "12.0"
	run: \|
	# Kitchen Sink: Features defined in matrix (all GPU backends per platform)
	FEATURES="${{ matrix.features }}"

	echo "Building ${{ matrix.artifact-name }} with features: $FEATURES"

	# Use cross for ARM64 cross-compilation, cargo for native builds
	if [[ "${{ matrix.use-cross }}" == "true" ]]; then
	cross build --release --target ${{ matrix.target }} --no-default-features --features "$FEATURES"
	else
	cargo build --release --target ${{ matrix.target }} --no-default-features --features "$FEATURES"
	fi

	- name: Upload artifact
	uses: actions/upload-artifact@v4
	with:
	name: ${{ matrix.artifact-name }}
	path: target/${{ matrix.target }}/release/${{ matrix.binary-name }}

	release:
	needs: [preflight, reuse-gate-binary, build]
	runs-on: ubuntu-latest
	if: startsWith(github.ref, 'refs/tags/') && needs.preflight.outputs.should_publish == 'true'
	permissions:
	contents: write
	steps:
	- uses: actions/checkout@v4

	- name: Download all artifacts
	uses: actions/download-artifact@v4
	with:
	path: ./artifacts

	- name: Prepare release files
	run: \|
	mkdir -p release-files

	# Linux x86_64 - Kitchen Sink (all GPU backends)
	cp artifacts/shimmy-linux-x86_64/shimmy release-files/shimmy-linux-x86_64
	cp artifacts/shimmy-linux-x86_64/shimmy release-files/shimmy # Generic = Linux x64

	# Linux ARM64 - CPU only
	cp artifacts/shimmy-linux-aarch64/shimmy release-files/shimmy-linux-aarch64

	# Windows x64 - Kitchen Sink (all GPU backends)
	cp artifacts/shimmy-windows-x86_64.exe/shimmy.exe release-files/shimmy-windows-x86_64.exe
	cp artifacts/shimmy-windows-x86_64.exe/shimmy.exe release-files/shimmy.exe # Generic = Windows

	# macOS Intel - CPU only
	cp artifacts/shimmy-macos-intel/shimmy release-files/shimmy-macos-intel

	# macOS ARM64 - MLX GPU
	cp artifacts/shimmy-macos-arm64/shimmy release-files/shimmy-macos-arm64

	# List what we're releasing
	echo "Release files:"
	ls -lh release-files/

	- name: Create release
	env:
	GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
	run: \|
	gh release create ${{ github.ref_name }} \
	release-files/* \
	--title "Shimmy ${{ github.ref_name }}" \
	--generate-notes

	- name: Install Rust for crates.io publishing
	uses: dtolnay/rust-toolchain@stable

	- name: "📦 Publish to crates.io"
	env:
	CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}
	run: \|
	echo "🚀 Publishing shimmy ${{ github.ref_name }} to crates.io..."
	# Debug: print working tree state (helps diagnose Cargo.lock vs git status mismatch)
	echo "🔍 git status (porcelain):"
	git status --porcelain \|\| true
	echo "🔍 git diff --name-only against HEAD (if any):"
	git diff --name-only \|\| true

	# When publishing we always allow dirty state for Cargo.lock differences that appear
	# during build/packaging steps (this prevents transient lockfile updates from blocking publish)
	echo "⚠️ Publishing with --allow-dirty to tolerate transient changes to Cargo.lock"
	DIRTY_FLAG="--allow-dirty"

	# Publish to crates.io (dry-run already validated in Gate 7)
	cargo publish $DIRTY_FLAG \|\| (echo "❌ cargo publish failed"; exit 1)

	echo "✅ Successfully published shimmy ${{ github.ref_name }} to crates.io!"
	echo "📦 Users can now install with: cargo install shimmy"

	- name: "🐳 Build and Push Docker Image to GHCR"
	if: always()
	env:
	GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
	run: \|
	echo "🐳 Building and pushing Docker image to GitHub Container Registry..."

	# Login to GHCR
	echo ${{ secrets.GITHUB_TOKEN }} \| docker login ghcr.io -u ${{ github.actor }} --password-stdin

	# Build the Docker image
	docker build -f deploy/Dockerfile -t ghcr.io/${{ github.repository }}:${{ github.ref_name }} -t ghcr.io/${{ github.repository }}:latest .

	# Push both tagged and latest versions
	docker push ghcr.io/${{ github.repository }}:${{ github.ref_name }}
	docker push ghcr.io/${{ github.repository }}:latest

	echo "✅ Docker image published successfully!"
	echo "🐳 Users can now run: docker run ghcr.io/${{ github.repository }}:latest"

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: remove GPU backends from GitHub runner builds - CPU only #122

Workflow file

fix: remove GPU backends from GitHub runner builds - CPU only #122

Uh oh!

Workflow file for this run