llama.cpp-sycl-bin

LOW

maintainer orphaned 0 votes scanned 2026-10-07 14:05:47.471701

View on AUR

Why flagged

The package repackages an official precompiled binary from the project's GitHub releases, which is a normal and safe AUR practice; the low severity is due to the flagged metadata (few votes, recent upload), not the content or behavior.

Triggered rules

Low Few votes, recently uploaded zero_votes_recent

Uploaded within the last 14 days with 2 or fewer community votes — little peer review so far.

Low AI review llm_review

An AI model (qwen/qwen3-235b-a22b-2507) reviewed this and agrees it is LOW (confidence 95%): The package repackages an official precompiled binary from the project's GitHub releases, which is a normal and safe AUR practice; the low severity is due to the flagged metadata (few votes, recent upload), not the content or behavior.

PKGBUILD

      1# Maintainer: richc <admin+aur@sys.corbet.ch>
    
      2#
    
      3# llama.cpp for the Intel Arc GPU via SYCL — official precompiled binary, NO compilation.
    
      4# Repackages the upstream GitHub release `llama-<ver>-bin-ubuntu-sycl-fp16-x64.tar.gz`.
    
      5# The fast Intel GPU path (~25 tok/s on Lunar Lake Arc 140V); the NPU is handled
    
      6# separately by openvino-genai-bin (llama.cpp's own OpenVINO backend is ~190x slower
    
      7# and intentionally NOT used here).
    
      8#
    
      9# oneAPI (MKL/DNN/SYCL runtime) is an optdepend, sourced at runtime by the
    
      10# `llama-sycl` launcher — so this survives oneAPI version bumps with no rebuild.
    
      11
    
      12pkgname=llama.cpp-sycl-bin
    
      13pkgver=b10545
    
      14pkgrel=1
    
      15pkgdesc="llama.cpp on Intel Arc GPU via SYCL — official precompiled binary (no compilation)"
    
      16arch=('x86_64')
    
      17url='https://github.com/ggml-org/llama.cpp'
    
      18license=('MIT')
    
      19depends=('glibc' 'gcc-libs' 'curl')
    
      20optdepends=('intel-oneapi-basekit: SYCL runtime (MKL/DNN) — required for the GPU backend')
    
      21makedepends=('patchelf')
    
      22provides=('llama.cpp-sycl')
    
      23noextract=("llama-${pkgver}-sycl-fp16.tar.gz")
    
      24source=(
    
      25  "llama-${pkgver}-sycl-fp16.tar.gz::https://github.com/ggml-org/llama.cpp/releases/download/${pkgver}/llama-${pkgver}-bin-ubuntu-sycl-fp16-x64.tar.gz"
    
      26  'llama-sycl'
    
      27)
    
      28sha256sums=(
    
      29  'cdecc02cab2109d240ec37407c77118e638f17ac2f51518abaa805ff954d0e63'
    
      30  'SKIP'
    
      31)
    
      32
    
      33prepare() {
    
      34  mkdir -p sycl
    
      35  bsdtar -xf "llama-${pkgver}-sycl-fp16.tar.gz" -C sycl
    
      36}
    
      37
    
      38build() {
    
      39  local d="$srcdir/sycl/llama-${pkgver}"
    
      40  # Libraries find their siblings via $ORIGIN. The SYCL runtime (MKL/DNN/SYCL) is
    
      41  # provided at runtime by the launcher sourcing oneAPI — not baked in here.
    
      42  find "$d" -maxdepth 1 -name '*.so*' -type f -exec patchelf --set-rpath '$ORIGIN' {} +
    
      43  find "$d" -maxdepth 1 -type f -executable ! -name '*.so*' -exec patchelf --set-rpath '$ORIGIN' {} +
    
      44}
    
      45
    
      46package() {
    
      47  local d="$srcdir/sycl/llama-${pkgver}"
    
      48  install -d "$pkgdir/usr/lib/llama-sycl"
    
      49  find "$d" -maxdepth 1 -name '*.so*'                       -exec cp -dP {} "$pkgdir/usr/lib/llama-sycl/" \;
    
      50  find "$d" -maxdepth 1 -type f -executable ! -name '*.so*' -exec install -m755 {} "$pkgdir/usr/lib/llama-sycl/" \;
    
      51  install -Dm755 "$srcdir/llama-sycl" "$pkgdir/usr/bin/llama-sycl"
    
      52  install -Dm644 "$d/LICENSE" "$pkgdir/usr/share/licenses/$pkgname/LICENSE"
    
      53}
    
      54

Changes since previous scan

--- PKGBUILD @ 2026-08-20 07:10
+++ PKGBUILD @ 2026-10-07 14:05
@@ -10,7 +10,7 @@
 # `llama-sycl` launcher — so this survives oneAPI version bumps with no rebuild.
 
 pkgname=llama.cpp-sycl-bin
-pkgver=b10488
+pkgver=b10545
 pkgrel=1
 pkgdesc="llama.cpp on Intel Arc GPU via SYCL — official precompiled binary (no compilation)"
 arch=('x86_64')
@@ -26,7 +26,7 @@
   'llama-sycl'
 )
 sha256sums=(
-  '59eed8b6b4a87df1ac0febae3e6d926cebfcc1dbf930f4acba8a4ceaeb3f9f70'
+  'cdecc02cab2109d240ec37407c77118e638f17ac2f51518abaa805ff954d0e63'
   'SKIP'
 )
 

Scan history

Scanned at (UTC)	Severity	Rules
2026-10-07 14:05:47	Low	2
2026-08-21 07:14:31	Clean	0
2026-08-20 07:10:50	Clean	0
2026-08-19 07:41:23	Clean	0
2026-08-18 07:39:23	Clean	0
2026-08-17 07:35:18	Clean	0
2026-08-16 09:32:18	Clean	0
2026-08-15 07:30:18	Clean	0
2026-08-14 07:28:04	Clean	0
2026-08-13 07:25:43	Clean	0
2026-08-12 07:23:13	Clean	0
2026-08-01 09:20:18	Clean	0
2026-07-31 09:17:30	Clean	0
2026-07-30 09:14:07	Clean	0
2026-07-29 09:11:06	Clean	0
2026-07-27 09:34:56	Clean	0
2026-07-25 09:29:42	Clean	0
2026-07-24 09:27:42	Clean	0
2026-07-23 09:22:54	Clean	0
2026-07-22 09:20:02	Clean	0