Debian 13 (Trixie) packages for llama-cpp-python 0.3.36, with separately installable accelerator backends.

Fixes

• Align the vendored llama.cpp revision with upstream 0.3.36 (0c1e57098bba43ac29e6e3b677cdceebdd22334f), fixing the Python/native API mismatch and the missing llama_adapter_lora_init_from_file_ptr symbol.
• Replace the obsolete ROCm 7.14 backend configuration with ROCm Core SDK 10.1.
• Include the HIP patch for compatible gfx1250 kernels.

Available backends

• CPU
• Vulkan
• NVIDIA CUDA 12.4, 12.9 and 13.3
• AMD ROCm Core SDK 10.1 / HIP
• Intel SYCL

Install the common runtime and the backend packages appropriate for your hardware. Vendor runtime dependencies are defined by the individual packages.

Packaging maintenance

• Keep upstream sources clean and Debian modifications in quilt patches.
• Remove the stale, tracked PKG-INFO.
• Move Debian build documentation to debian/README.Debian.md.
• Add API and staged-import checks before the accelerator builds.

Validation

The complete Debian build succeeded, Python bindings imported successfully against freshly built CPU libraries, and the resulting packages were verified in TLlama.

The repaired repository also passed source-package validation with both quilt patches applied.

Assets

The attached .deb files are the built Debian packages. For source builds, use the Debian source archives including the vendored llama.cpp sources; GitHub’s automatically generated source archives do not include submodule contents.

DEB packages 0.3.35-3.

Add modular CPU, Vulkan, CUDA, ROCm and SYCL backend packages

• split llama-cpp-python into core and dynamically loadable backend packages
• add CPU backend with multiple optimized CPU variants and OpenBLAS
• add Vulkan backend
• add co-installable CUDA 12.4, 12.9 and 13.3 backends using update-alternatives
• add ROCm 7.14 HIP backend with support for current AMD GPU architectures
• add Intel oneAPI SYCL backend with oneMKL, oneDNN and Level Zero support
• add transitional packages for legacy package names
• add backend alternatives handling and private shared-library dependency resolution
• add CUDA 12.9 compatibility handling for Debian Trixie/glibc 2.41
• isolate external CUDA, ROCm and oneAPI toolchains from Debian system libraries
• update package dependencies, shlibs handling, cleanup rules and runtime RPATHs

Fixed dependencies in CUDA 12.9 backend package

Full Changelog: v0.3.35...v0.3.35-3-deb

deb packages with actual version 0.3.21 for:

generic CPUs
AVX CPUs
AVX2 CPUs
AVX512 CPUs
CUDA GPUs
Vulkan GPUs

PS: ROCm package will be later, CUDA packages will be CUDA12 and CUDA13 at least

deb packages with actual version 0.3.21 for:

• generic CPUs
• AVX CPUs
• AVX2 CPUs
• AVX512 CPUs
• CUDA GPUs
• Vulkan GPUs

PS: ROCm package will be later