Datasets:

AIencoder
/

llama-cpp-wheels

@@ -1,189 +1,95 @@
 ---
 license: mit
-task_categories:
-- text-generation
-language:
-- en
 tags:
-- code
-- llama-cpp
-- llama-cpp-python
-- wheels
-- pre-built
-- binary
-- linux
-- windows
-- macos
-pretty_name: llama-cpp-python Pre-Built Wheels
-size_categories:
-- 1K<n<10K
 ---
-If you like this please consider liking and donating (https://buymeacoffee.com/aiencoder)
-# 🏭 llama-cpp-python Mega-Factory Wheels
-> **"Stop waiting for `pip` to compile. Just install and run."**
-The most complete collection of pre-built `llama-cpp-python` wheels in existence — **8,333 wheels** across every platform, Python version, backend, and CPU optimization level.
-No more `cmake`, `gcc`, or compilation hell. No more waiting 10 minutes for a build that might fail. Just find your wheel and `pip install` it directly.
----
-## 🚀 Why These Wheels?
-Standard wheels target the "lowest common denominator" to avoid crashes on old hardware. This collection goes further — the manylinux wheels are built using a massive **Everything Preset** targeting specific CPU instruction sets, maximizing your **Tokens per Second (T/s)**.
-- **Zero Dependencies:** No `cmake`, `gcc`, or `nvcc` required on your target machine.
-- **Every Platform:** Linux (manylinux, aarch64, i686, RISC-V), Windows (amd64, 32-bit), macOS (Intel + Apple Silicon).
-- **Server-Grade Power:** Optimized builds for `Sapphire Rapids`, `Ice Lake`, `Alder Lake`, `Haswell`, and more.
-- **Full Backend Support:** `OpenBLAS`, `MKL`, `Vulkan`, `CLBlast`, `OpenCL`, `RPC`, and plain CPU builds.
-- **Cutting Edge:** Python `3.8` through experimental `3.14`, plus PyPy `pp38`–`pp310`.
-- **GPU Too:** CUDA wheels (cu121–cu124) and macOS Metal wheels included.
----
-## 📊 Collection Stats
-| Platform | Wheels |
-|:---|---:|
-| 🐧 Linux x86_64 (manylinux) | 4,940 |
-| 🍎 macOS Intel (x86\_64) | 1,040 |
-| 🪟 Windows (amd64) | 1,010 |
-| 🪟 Windows (32-bit) | 634 |
-| 🍎 macOS Apple Silicon (arm64) | 289 |
-| 🐧 Linux i686 | 214 |
-| 🐧 Linux aarch64 | 120 |
-| 🐧 Linux x86\_64 (plain) | 81 |
-| 🐧 Linux RISC-V | 5 |
-| **Total** | **8,333** |
-The manylinux builds alone cover **3,600+ combinations** across versions, backends, Python versions, and CPU profiles.
----
-## 🚀 How to Install
-### Quick Install
-Find your wheel filename (see naming convention below), then:
 ```bash
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/YOUR_WHEEL_NAME.whl"
 ```
-### Common Examples
 ```bash
-# Linux x86_64, Python 3.11, OpenBLAS, Haswell CPU (most common Linux setup)
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.18+openblas_haswell-cp311-cp311-manylinux_2_31_x86_64.whl"
-# Linux x86_64, Python 3.12, Basic CPU (maximum compatibility)
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.18+basic_basic-cp312-cp312-manylinux_2_31_x86_64.whl"
-# Windows, Python 3.11
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.18-cp311-cp311-win_amd64.whl"
-# macOS Apple Silicon, Python 3.12
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.18-cp312-cp312-macosx_11_0_arm64.whl"
-# macOS Intel, Python 3.11
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.18-cp311-cp311-macosx_10_9_x86_64.whl"
-# Linux ARM64 (Raspberry Pi, AWS Graviton), Python 3.11
-pip install "https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.18-cp311-cp311-linux_aarch64.whl"
 ```
----
-## 📁 Wheel Naming Convention
-### manylinux wheels (custom-built)
 ```
-llama_cpp_python-{version}+{backend}_{profile}-{pytag}-{pytag}-{platform}.whl
 ```
-**Versions covered:** `0.3.0` through `0.3.18+`
-**Backends:**
-| Backend | Description |
-|:---|:---|
-| `openblas` | OpenBLAS BLAS acceleration — best general-purpose CPU performance |
-| `mkl` | Intel MKL acceleration — best on Intel CPUs |
-| `basic` | No BLAS, maximum compatibility |
-| `vulkan` | Vulkan GPU backend |
-| `clblast` | CLBlast OpenCL GPU backend |
-| `opencl` | Generic OpenCL GPU backend |
-| `rpc` | Distributed inference over network |
-**CPU Profiles:**
-| Profile | Instruction Sets | Era | Notes |
-|:---|:---|:---|:---|
-| `basic` | x86-64 baseline | Any | Maximum compatibility |
-| `sse42` | SSE 4.2 | 2008+ | Nehalem |
-| `sandybridge` | AVX | 2011+ | |
-| `ivybridge` | AVX + F16C | 2012+ | |
-| `haswell` | AVX2 + FMA + BMI2 | 2013+ | **Most common** |
-| `skylakex` | AVX-512 | 2017+ | |
-| `icelake` | AVX-512 + VNNI + VBMI | 2019+ | |
-| `alderlake` | AVX-VNNI | 2021+ | |
-| `sapphirerapids` | AVX-512 BF16 + AMX | 2023+ | Highest performance |
-**Python tags:** `cp38`, `cp39`, `cp310`, `cp311`, `cp312`, `cp313`, `cp314`, `pp38`, `pp39`, `pp310`
-**Platform:** `manylinux_2_31_x86_64` (glibc 2.31+, compatible with Ubuntu 20.04+, Debian 11+)
-### Windows / macOS / Linux ARM wheels (from abetlen)
-```
-llama_cpp_python-{version}-{pytag}-{pytag}-{platform}.whl
-```
-These are the official pre-built wheels from the upstream maintainer, covering versions `0.2.82` through `0.3.18+`.
----
-## 🔍 How to Find Your Wheel
-1. **Identify your Python version:** `python --version` → e.g. `3.11` → tag `cp311`
-2. **Identify your platform:**
-   - Linux x86\_64 → `manylinux_2_31_x86_64`
-   - Windows 64-bit → `win_amd64`
-   - macOS Apple Silicon → `macosx_11_0_arm64`
-   - macOS Intel → `macosx_10_9_x86_64`
-3. **Pick a backend** (manylinux only): `openblas` for most use cases
-4. **Pick a CPU profile** (manylinux only): `haswell` works on virtually all modern CPUs
-5. **Browse the files** in this repo or construct the filename directly
----
-## 🏗️ Sources & Credits
-### manylinux Wheels — Built by AIencoder
-The 4,940 manylinux x86\_64 wheels were built by a distributed **4-worker HuggingFace Space factory** system (`AIencoder/wheel-factory-*`) — a custom-built automated pipeline covering every possible llama.cpp cmake option on manylinux:
-- Every backend: OpenBLAS, MKL, Basic, Vulkan, CLBlast, OpenCL, RPC
-- Every CPU hardware profile from baseline x86-64 up to Sapphire Rapids AMX
-- Python 3.8 through 3.14
-- llama-cpp-python versions 0.3.0 through 0.3.18+
-### Windows / macOS / Linux ARM Wheels — abetlen
-The remaining 3,393 wheels (Windows, macOS, Linux aarch64/i686/riscv64, PyPy) were sourced from the official releases by **Andrei Betlen ([@abetlen](https://github.com/abetlen))**, the original author and maintainer of `llama-cpp-python`. These include:
-- CPU wheels for all platforms via `https://abetlen.github.io/llama-cpp-python/whl/cpu/`
-- Metal wheels for macOS GPU acceleration
-- CUDA wheels (cu121–cu124) for Windows and Linux
-> All credit for the underlying library goes to **Georgi Gerganov ([@ggerganov](https://github.com/ggerganov))** and the [llama.cpp](https://github.com/ggml-org/llama.cpp) team, and to **Andrei Betlen** for the Python bindings.
 ---
-## 📝 Notes
-- All wheels are **MIT licensed** (same as llama-cpp-python upstream)
-- manylinux wheels require **glibc 2.31+** (Ubuntu 20.04+, Debian 11+)
-- `manylinux` and `linux_x86_64` are **not the same thing** — manylinux wheels have broad distro compatibility, plain linux wheels do not
-- CUDA wheels require the matching CUDA toolkit to be installed
-- Metal wheels require macOS 11.0+ and an Apple Silicon or AMD GPU
-- This collection is updated periodically as new versions are released

 ---
 license: mit
+pretty_name: llama-cpp-python Wheels — Complete Collection
 tags:
+  - llama-cpp-python
+  - llama.cpp
+  - wheels
+  - cuda
+  - metal
+  - prebuilt
+viewer: false
 ---
+# llama-cpp-python Wheels — the complete collection
+Prebuilt [`llama-cpp-python`](https://github.com/abetlen/llama-cpp-python) wheels, assembled to be
+the most complete single place to grab one — **10,000+ wheels** spanning every published upstream
+version and every backend this project can build, for **Linux, Windows, and macOS**.
+It contains two things:
+1. **A full mirror of the official upstream wheels** — abetlen's `cpu`, `metal`, and `cu121`–`cu124`
+   releases, across all historical versions (`0.1.x` → `0.3.x`) and all three OSes.
+2. **Thousands of additional builds that upstream never shipped** — produced here by an automated
+   GitHub Actions "wheel factory": extra CUDA versions, extra CPU instruction-set variants, and
+   extra backends (see the table below).
+> These are self-contained builds. CUDA wheels bundle the CUDA runtime, so you only need an NVIDIA
+> **driver** installed — not the full CUDA toolkit.
+## Install a wheel
+Every wheel is a direct download. Install one by pointing `pip` at its URL:
 ```bash
+pip install https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/<path-to-wheel>
 ```
+For example, llama-cpp-python 0.3.19 with CUDA 12.4 on Python 3.11 / Windows:
 ```bash
+pip install https://huggingface.co/datasets/AIencoder/llama-cpp-wheels/resolve/main/llama_cpp_python-0.3.19+cu124-cp311-cp311-win_amd64.whl
 ```
+## Reading a filename
 ```
+llama_cpp_python-0.3.19+cu124-cp311-cp311-win_amd64.whl
+                 │      │     │     │     └── platform  (win_amd64 / manylinux… / macosx…)
+                 │      │     └─────┴──────── Python tag (cp311 = CPython 3.11)
+                 │      └── build variant / backend  (see table)
+                 └── llama-cpp-python version
 ```
+The part after `+` (the "local version") is the **backend / build variant**. Upstream base wheels
+have no `+` tag (plain CPU). Factory builds encode the backend there, e.g. `+cu126`,
+`+basic_avx2_fma_f16c`, `+openblas_avx512_fma_f16c_vnni_vbmi`, `+vulkan`, `+metal`.
+## What's covered
+| OS | Backends | Notes |
+|----|----------|-------|
+| **Linux** (`manylinux`) | CPU (plain + AVX / AVX2 / AVX-512 up to VNNI·VBMI·BF16·AMX), OpenBLAS, MKL, CLBlast, OpenCL, Vulkan, SYCL, RPC, and **CUDA 12.4 / 12.6 / 12.8 / 12.9** | The broadest coverage — every CPU instruction-set tier so you can match your exact chip |
+| **Windows** (`win_amd64`) | Full upstream mirror + **CUDA 12.4 / 12.6 / 12.8** | CUDA wheels self-contained (driver only) |
+| **macOS** (`macosx`) | Metal + CPU (upstream mirror) | Apple-Silicon and Intel |
+- **Python:** cp38 – cp314 for factory builds; upstream mirror covers whatever upstream shipped (back to cp37 / PyPy on old versions).
+- **Versions:** factory builds target the latest releases (0.3.16 – 0.3.19); the mirror covers everything from 0.1.x onward.
+## Repository layout
+Because HuggingFace caps any single folder at 10,000 files, wheels are organised into **per-version
+subfolders** (`0.3.19/`, `0.1.30/`, …). Earlier wheels live in the repository root; both are equally
+valid — use the Hub file browser or the filename to find what you need.
+## How these are built
+The additional wheels are compiled automatically by a GitHub Actions factory (no GPU required — `nvcc`
+cross-compiles the CUDA kernels). Each wheel is built from the official `abetlen/llama-cpp-python`
+source at the tagged version, repaired to be self-contained, import/structure-checked, and uploaded.
+The queue is derived by diffing this dataset against the target matrix, so the collection fills in
+and stays current on its own.
+## Caveats
+- **CUDA wheels need a matching NVIDIA driver** on the machine that runs them (the CUDA runtime is
+  bundled; the driver is not).
+- **CUDA wheels are large** (~0.6–1.3 GB) because they bundle the runtime and fat CUDA kernels — this
+  is expected and is what makes them work without a separate CUDA install.
+- These are community builds, provided as-is, under the same MIT license as llama-cpp-python.
 ---
+*Collection maintained by [AIencoder](https://huggingface.co/AIencoder). Wheels are builds of
+[abetlen/llama-cpp-python](https://github.com/abetlen/llama-cpp-python).*