Skip to content

Ship current Vulkan prebuilts and restore Volta CUDA support #22

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
danielhanchen merged 3 commits into from
Oct 7, 2026
Merged

Ship current Vulkan prebuilts and restore Volta CUDA support #22

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
35941e9
fix: include Volta kernels in Linux CUDA prebuilts
oobabooga Oct 6, 2026
4791cab
fix: ship current native Vulkan builds for Linux and Windows
oobabooga Oct 6, 2026
51e2e8e
ci: verify Vulkan SDKs and target Ubuntu 22.04
oobabooga Oct 6, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
112 changes: 85 additions & 27 deletions .github/workflows/unsloth-sd-prebuilt.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,26 +1,24 @@
# SPDX-License-Identifier: MIT
# Copyright 2026-present the Unsloth AI Inc. team.

name: Unsloth SD prebuilt (CPU/Apple/CUDA)

# Build and publish OUR OWN stable-diffusion.cpp (sd-cli + sd-server) prebuilts for
# the platforms where the native engine is the FASTER choice: CPU (Linux/WSL/Windows),
# Apple (Metal), and now Linux CUDA. The last one is a correction: this pipeline was
# built on the premise that a GPU host runs diffusers/torch instead, and MiniMax-H3
# breaks that premise by needing more VRAM than a consumer card has, which sends those
# hosts to the GGUF engine after all.
name: Unsloth SD prebuilt (CPU/Apple/CUDA/Vulkan)

# Build and publish our stable-diffusion.cpp (sd-cli + sd-server) prebuilts for
# CPU, Apple, CUDA, and Vulkan. MiniMax-H3 uses the native engine on GPU hosts too.
# Vulkan must carry the same H3 and graph-segment fixes as the other backends,
# rather than sending AMD hosts to an older upstream fallback binary.
#
# Shape mirrors unslothai/llama.cpp's prebuilt pipeline, simplified for a 5-way
# CPU/Apple matrix plus one best-effort CUDA leg:
# Shape mirrors unslothai/llama.cpp's prebuilt pipeline, simplified for a 7-way
# CPU/Apple/Vulkan matrix plus one best-effort CUDA leg:
# resolve -- take THIS repository at the requested ref, merge the pinned PR set
# in scripts/unsloth/pr-set.json, stamp a source tarball, name it
# after the upstream release the tree descends from (supply-chain
# aged), decide if our release already exists.
# build-unix -- macOS arm64 (Metal) + x64, Linux x64 + arm64 (matrix).
# build-windows-- Windows x64 (MSVC + Ninja).
# build-unix -- macOS arm64 (Metal) + x64, Linux x64 + arm64, Linux Vulkan.
# build-windows-- Windows x64 CPU + Vulkan (MSVC + Ninja).
# build-linux-cuda -- Linux x64, CUDA 12.8, best effort (continue-on-error).
# assemble -- fingerprint gate, sha256 + manifest, coverage gate, atomic
# draft->publish. If any of the five CPU/Apple legs fails, assemble
# draft->publish. If any CPU/Apple/Vulkan leg fails, assemble
# is skipped and nothing is published (the Studio needs the full
# asset set). The CUDA leg is exempt: it ships when it built and is
# absent when it did not, so it can never hold back the fallbacks.
Expand Down Expand Up @@ -278,7 +276,17 @@ jobs:
arch: aarch64
defines: ""
gate: "false"
- label: Linux-Ubuntu-22.04-x86_64-vulkan
runner: ubuntu-22.04
arch: x86_64
defines: "-DSD_VULKAN=ON"
gate: "false"
vulkan: "true"
runs-on: ${{ matrix.runner }}
env:
VULKAN_VERSION: 1.4.328.1
# SDK checksums: https://vulkan.lunarg.com/sdk/files.json
VULKAN_SHA256: 241e75b56c91c0d210ed07a7c638ec05a3e5b0e4c66ba9f0ba0f102d823ad6bf
steps:
- name: Checkout mirror (tooling)
uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
Expand All @@ -297,6 +305,22 @@ jobs:
mkdir -p src
tar -xzf "srcpkg/sd-source-${{ needs.resolve.outputs.tag }}.tar.gz" -C src

- name: Install Vulkan build dependencies
if: ${{ matrix.vulkan == 'true' }}
run: |
set -euo pipefail
sudo apt-get update
sudo apt-get install -y libvulkan-dev xz-utils
# Jammy has no glslc package. Use the SDK tools while building against glibc 2.35.
archive="$RUNNER_TEMP/vulkansdk.tar.xz"
sdk_root="$RUNNER_TEMP/vulkansdk"
curl --fail --location --retry 3 --output "$archive" "https://sdk.lunarg.com/sdk/download/${VULKAN_VERSION}/linux/vulkansdk-linux-x86_64-${VULKAN_VERSION}.tar.xz"
echo "$VULKAN_SHA256 $archive" | sha256sum --check --strict
mkdir -p "$sdk_root"
tar -xJf "$archive" -C "$sdk_root" --strip-components=1
echo "VULKAN_SDK=$sdk_root/x86_64" >> "$GITHUB_ENV"
echo "$sdk_root/x86_64/bin" >> "$GITHUB_PATH"

- name: Build sd-cli + sd-server
working-directory: src
run: |
Expand Down Expand Up @@ -338,8 +362,7 @@ jobs:
if-no-files-found: error
retention-days: 7

# Linux CUDA. The rest of this pipeline is CPU and Apple on the premise that GPU hosts use
# diffusers/torch, and that premise fails for MiniMax-H3: its Diffusers path needs about
# Linux CUDA. GPU hosts also need native inference for MiniMax-H3: its Diffusers path needs about
# 68.5 GB of VRAM, so every consumer card falls back to the GGUF engine, which on Linux had
# no accelerated build at all. Measured on one box: 65 s/step at 320x192 on 96 CPU threads,
# against 21.5 s/step at 960x544 from a CUDA build. That is the difference between a render
Expand All @@ -348,7 +371,7 @@ jobs:
# continue-on-error, and deliberately so. The CPU and Apple assets are what Studio falls back
# to, and a CUDA toolchain failure must never stop them publishing. assemble collects bundles
# by the sd-*-bin-* pattern, so this asset appears when it built and is simply absent when it
# did not; the coverage gate lists only the five CPU/Apple assets and must keep doing so.
# did not; the coverage gate requires CPU/Apple/Vulkan assets but leaves CUDA optional.
build-linux-cuda:
name: Linux-Ubuntu-22.04-x86_64-cuda12
needs: resolve
Expand All @@ -359,7 +382,7 @@ jobs:
# the objects, and these two do.
env:
CUDA_VERSION: "12.8.1"
CUDA_ARCHS: "75;80;86;89;90;100;120"
CUDA_ARCHS: "70;75;80;86;89;90;100;120"
steps:
- name: Checkout mirror (tooling)
uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
Expand Down Expand Up @@ -431,9 +454,9 @@ jobs:
working-directory: src
run: |
set -euo pipefail
# Turing through Blackwell. 12.8 is the first toolkit that can emit sm_100 (B200) and
# sm_120 (RTX 50), and anything older than Turing is not a realistic host for a 20 GB
# video denoiser. build.yml's Windows leg uses the same list plus 61 and 70.
# Volta through Blackwell. V100 can run H3 with quantized weights and CPU offload,
# but a device probe succeeds even when the binary has no sm_70 kernels. Include
# sm_70 so generation does not fail at the first CUDA kernel launch.
#
# The CUDA launcher matters more than the other two here: Jimver installs nvcc outside
# the default search, and nvcc is nearly the whole build, so caching only C/CXX would
Expand Down Expand Up @@ -516,10 +539,22 @@ jobs:
key: ccache-sd-cuda-${{ env.CUDA_VERSION }}-${{ steps.cckey.outputs.archs }}-${{ needs.resolve.outputs.tag }}-

build-windows:
name: win-cpu-x64
name: ${{ matrix.label }}
needs: resolve
if: ${{ needs.resolve.outputs.exists != 'true' || github.event_name == 'workflow_dispatch' }}
runs-on: windows-2022
strategy:
fail-fast: false
matrix:
include:
- label: win-cpu-x64
vulkan: "OFF"
- label: win-vulkan-x64
vulkan: "ON"
env:
VULKAN_VERSION: 1.4.328.1
# SDK checksums: https://vulkan.lunarg.com/sdk/files.json
VULKAN_SHA256: a8675df6d538079c2a719a9373994948091db785b48f142e024254e76348d16c
# Gate on the environment, matching unslothai/unsloth's release-desktop.yml:
# the job that can read the signing credentials is named explicitly.
environment: release-signing
Expand Down Expand Up @@ -551,6 +586,24 @@ jobs:
- name: Setup MSVC
uses: ilammy/msvc-dev-cmd@0b201ec74fa43914dc39ae48a89fd1d8cb592756 # v1.13.0

- name: Install Vulkan SDK
if: ${{ matrix.vulkan == 'ON' }}
shell: pwsh
run: |
$installer = Join-Path $env:RUNNER_TEMP 'VulkanSDK-Installer.exe'
$sdkRoot = Join-Path $env:RUNNER_TEMP 'VulkanSDK'
curl.exe --fail --location --retry 3 --output $installer "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/vulkansdk-windows-X64-${env:VULKAN_VERSION}.exe"
if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
if ((Get-FileHash -LiteralPath $installer -Algorithm SHA256).Hash -ne $env:VULKAN_SHA256) {
throw 'Vulkan SDK installer SHA256 mismatch'
}
$install = Start-Process -FilePath $installer -ArgumentList @(
'--root', "`"$sdkRoot`"", '--accept-licenses', '--default-answer', '--confirm-command', 'install'
) -Wait -PassThru
if ($install.ExitCode -ne 0) { exit $install.ExitCode }
Add-Content $env:GITHUB_ENV "VULKAN_SDK=$sdkRoot"
Add-Content $env:GITHUB_PATH "$sdkRoot\Bin"

- name: Build sd-cli + sd-server
shell: pwsh
working-directory: src
Expand All @@ -561,6 +614,7 @@ jobs:
-DSD_BUILD_EXAMPLES=ON `
-DSD_SERVER_BUILD_FRONTEND=OFF `
-DSD_WEBP=OFF -DSD_WEBM=OFF `
-DSD_VULKAN=${{ matrix.vulkan }} `
-DGGML_NATIVE=OFF
cmake --build build --config Release -j 3 --target sd-cli sd-server

Expand All @@ -585,7 +639,7 @@ jobs:
BIN_DIR: ${{ github.workspace }}/src/build/bin
OUT_DIR: ${{ github.workspace }}/dist
TAG: ${{ needs.resolve.outputs.tag }}
LABEL: win-cpu-x64
LABEL: ${{ matrix.label }}
COMMIT: ${{ needs.resolve.outputs.commit }}
SOURCE_REPO: ${{ github.repository }}
LICENSE_FILE: ${{ github.workspace }}/src/LICENSE
Expand All @@ -610,8 +664,8 @@ jobs:
- name: Upload bundle
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
with:
name: sd-${{ needs.resolve.outputs.tag }}-bin-win-cpu-x64
path: dist/sd-${{ needs.resolve.outputs.tag }}-bin-win-cpu-x64.zip
name: sd-${{ needs.resolve.outputs.tag }}-bin-${{ matrix.label }}
path: dist/sd-${{ needs.resolve.outputs.tag }}-bin-${{ matrix.label }}.zip
if-no-files-found: error
retention-days: 7

Expand Down Expand Up @@ -671,7 +725,9 @@ jobs:
--publish-repo "$GITHUB_REPOSITORY"
ls -la dist

- name: Coverage gate (all 5 CPU/Apple assets present)
# Vulkan is required: the pinned upstream fallback lacks our H3 fixes.
# Blocking publication on a Vulkan failure prevents promoting an incomplete fix.
- name: Coverage gate (all 7 CPU/Apple/Vulkan assets present)
run: |
set -eu
TAG='${{ needs.resolve.outputs.tag }}'
Expand All @@ -681,7 +737,9 @@ jobs:
"sd-${TAG}-bin-Darwin-macOS-x86_64.zip" \
"sd-${TAG}-bin-Linux-Ubuntu-22.04-x86_64.zip" \
"sd-${TAG}-bin-Linux-Ubuntu-24.04-aarch64.zip" \
"sd-${TAG}-bin-win-cpu-x64.zip"; do
"sd-${TAG}-bin-Linux-Ubuntu-22.04-x86_64-vulkan.zip" \
"sd-${TAG}-bin-win-cpu-x64.zip" \
"sd-${TAG}-bin-win-vulkan-x64.zip"; do
[ -s "dist/$f" ] || { echo "ERROR: missing $f" >&2; fail=1; }
done
[ "$fail" = 0 ] || { echo "ERROR: refusing to publish a partial release" >&2; exit 1; }
Expand Down Expand Up @@ -718,7 +776,7 @@ jobs:
AHEAD='${{ needs.resolve.outputs.ahead }}'
COMMIT='${{ needs.resolve.outputs.commit }}'
REPO="$GITHUB_REPOSITORY"
NOTES="Automated Unsloth stable-diffusion.cpp CPU + Apple prebuild (sd-cli + sd-server), built from [\`${COMMIT}\`](https://github.com/${REPO}/commit/${COMMIT}) in this repository. GPU hosts use diffusers/torch; this native engine targets CPU (Linux/WSL/Windows) and Apple (Metal)."
NOTES="Automated Unsloth stable-diffusion.cpp prebuild (sd-cli + sd-server), built from [\`${COMMIT}\`](https://github.com/${REPO}/commit/${COMMIT}) in this repository. Includes CPU, Apple Metal, and Linux/Windows Vulkan builds, plus Linux CUDA when its build succeeds."
PINS='${{ needs.resolve.outputs.pins }}'
if [ "${AHEAD:-0}" -gt 0 ]; then
NOTES="${NOTES}"$'\n\n'"Not a stock build: the tree is ${AHEAD} commits past the ${UPSTREAM_TAG} release it descends from, which is what the \`-u\` suffix on the tag marks. \`sd-prebuilt-manifest.json\` records the commit."
Expand Down