16 KiB

Raw Blame History

OpenVINO Backend for llama.cpp

OpenVINO is an open-source toolkit for optimizing and deploying high-performance AI inference, specifically designed for Intel hardware, including CPUs, GPUs, and NPUs, in the cloud, on-premises, and on the edge. This document describes the OpenVINO backend for llama.cpp, which enables hardware-accelerated inference on Intel® CPUs, GPUs, and NPUs while remaining compatible with the existing GGUF model ecosystem. The backend translates GGML compute graphs into OpenVINO graphs and leverages graph compilation, kernel fusion, and device-specific optimizations to improve inference performance on supported Intel hardware.

The OpenVINO backend is implemented in ggml/src/ggml-openvino and provides a translation layer for core GGML operations. The OpenVINO backend replaces the standard GGML graph execution path with Intel's OpenVINO inference engine. This approach allows the same GGUF model file to run on Intel CPUs, Intel GPUs (integrated and discrete), and Intel NPUs without changes to the model or the rest of the llama.cpp stack. When a ggml_cgraph is dispatched to OpenVINO backend, it:

Walks the GGML graph and identifies inputs, outputs, weights, and KV cache tensors.
Translates the GGML operations into an ov::Model using OpenVINO's frontend API.
Compiles and caches the model for the target device.
Binds GGML tensor memory to OpenVINO inference tensors and runs inference.

Supported Devices

OpenVINO backend supports the following hardware:

Intel CPUs
Intel GPUs (integrated and discrete)
Intel NPUs

Although OpenVINO supports a wide range of Intel hardware, the llama.cpp OpenVINO backend has been validated specifically on AI PCs such as the Intel® Core™ Ultra Series 1 and Series 2.

Supported Model Precisions

FP16
BF16 (on Intel Xeon)
Q8_0
Q4_0
Q4_1
Q4_K
Q4_K_M
Q5_K (converted to Q8_0_C at runtime)
Q6_K (converted to Q8_0_C at runtime)

[!NOTE] Accuracy validation and performance optimizations for quantized models are a work in progress.

Quantization Support Details

CPU and GPU

Q4_0, Q4_1, Q4_K_M, Q6_K models are supported
Q5_K and Q6_K tensors are converted to Q8_0_C

NPU

Primary supported quantization scheme is Q4_0
Q6_K tensors are requantized to Q4_0_128 in general. For embedding weights, Q6_K tensors are requantized to Q8_0_C except for the token embedding matrix which is dequantized to fp16

Additional Notes

Both Q4_0 and Q4_1 models use Q6_K for the token embedding tensor and the final matmul weight tensor (often the same tensor)
Q4_0 models may produce some Q4_1 tensors if an imatrix is provided during quantization using llama-quantize
Q4_K_M models may include both Q6_K and Q5_K tensors (observed in Phi-3)

Validated Models

The following models have been validated for functionality on Intel® Core™ Ultra Series 1 and Series 2:

Build Instructions

Prerequisites

Linux or Windows system with Intel hardware (CPU, GPU, or NPU)
For Intel GPU or NPU Usage: Install the appropriate hardware drivers for your Intel GPU or NPU. For detailed instructions, see: Additional Configurations for Hardware Acceleration.

Linux:

Git, CMake, and Ninja software tools are needed for building.

  sudo apt-get update
  sudo apt-get install -y build-essential libcurl4-openssl-dev libtbb12 cmake ninja-build python3-pip curl wget tar

OpenCL

  sudo apt install ocl-icd-opencl-dev opencl-headers opencl-clhpp-headers intel-opencl-icd

Windows:

Download and install Microsoft Visual Studio 2022 Build Tools. During installation, select the "Desktop development with C++" workload.

Install required tools:

# Windows PowerShell
winget install Git.Git
winget install GNU.Wget
winget install Ninja-build.Ninja

Install OpenCL using vcpkg:

# Windows PowerShell
cd C:\
git clone https://github.com/microsoft/vcpkg
cd vcpkg
.\bootstrap-vcpkg.bat
.\vcpkg install opencl
# Optional but recommended: Integrate vcpkg with Visual Studio / CMake:
.\vcpkg integrate install

1. Install OpenVINO Runtime

Follow the guide to install OpenVINO Runtime from an archive file: Linux | Windows

Linux:

📦 Click to expand OpenVINO installation from an archive file on Ubuntu

wget https://raw.githubusercontent.com/ravi9/misc-scripts/main/openvino/ov-archive-install/install-openvino-from-archive.sh
chmod +x install-openvino-from-archive.sh
./install-openvino-from-archive.sh

Verify OpenVINO is initialized properly:

echo $OpenVINO_DIR

2. Build llama.cpp with OpenVINO Backend

Clone the OpenVINO-enabled llama.cpp fork and build it:

git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp

Linux:

source /opt/intel/openvino/setupvars.sh
cmake -B build/ReleaseOV -G Ninja -DCMAKE_BUILD_TYPE=Release -DGGML_OPENVINO=ON
cmake --build build/ReleaseOV --parallel

Windows:

# x64 Native Tools Command Prompt for VS 2022
"C:\Program Files (x86)\Intel\openvino_2026.0\setupvars.bat"
cmake -B build\ReleaseOV -G Ninja -DCMAKE_BUILD_TYPE=Release -DGGML_OPENVINO=ON -DLLAMA_CURL=OFF -DCMAKE_TOOLCHAIN_FILE=C:\vcpkg\scripts\buildsystems\vcpkg.cmake
cmake --build build\ReleaseOV --parallel

[!NOTE] Use x64 Native Tools Command Prompt for Windows build. After building, you could use either cmd or PowerShell to run the OpenVINO backend.

3. Download Sample Model

Download models for testing:

# Linux
mkdir -p ~/models/
wget https://huggingface.co/unsloth/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_0.gguf \
     -O ~/models/Llama-3.2-1B-Instruct-Q4_0.gguf

# Windows PowerShell
mkdir C:\models
Invoke-WebRequest -Uri https://huggingface.co/unsloth/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_0.gguf -OutFile C:\models\Llama-3.2-1B-Instruct-Q4_0.gguf

# Windows Command Line
mkdir C:\models
curl -L https://huggingface.co/unsloth/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_0.gguf -o C:\models\Llama-3.2-1B-Instruct-Q4_0.gguf

4. Run Inference with OpenVINO Backend

When using the OpenVINO backend, the first inference token may have slightly higher latency due to on-the-fly conversion to the OpenVINO graph. Subsequent tokens and runs will be faster.

# If device is unset or unavailable, defaults to CPU.
# If the system has multiple GPUs, use GPU.0 or GPU.1 to explicitly target a specific GPU.

# Linux
export GGML_OPENVINO_DEVICE=GPU
# To run llama-simple:
./build/ReleaseOV/bin/llama-simple -m ~/models/Llama-3.2-1B-Instruct-Q4_0.gguf -n 50 "The story of AI is "
# To run in chat mode:
./build/ReleaseOV/bin/llama-cli -m ~/models/Llama-3.2-1B-Instruct-Q4_0.gguf

# Windows Command Line
set GGML_OPENVINO_DEVICE=GPU
# Windows PowerShell
$env:GGML_OPENVINO_DEVICE = "GPU"

# To run llama-simple
build\ReleaseOV\bin\llama-simple.exe -m "C:\models\Llama-3.2-1B-Instruct-Q4_0.gguf" -n 50 "The story of AI is "
# To run in chat mode:
build\ReleaseOV\bin\llama-cli.exe -m "C:\models\Llama-3.2-1B-Instruct-Q4_0.gguf"

[!NOTE] On systems with multiple GPUs, use GPU.0 or GPU.1 to explicitly target specific GPU. See OpenVINO GPU Device for more details.

Docker Build

You can build and run llama.cpp with OpenVINO backend using Docker.

# Build the base runtime image with compiled shared libraries and minimal dependencies.
docker build -t llama-openvino:base -f .devops/openvino.Dockerfile .

# Build the complete image with all binaries, Python tools, gguf-py library, and model conversion utilities.
docker build --target=full -t llama-openvino:full -f .devops/openvino.Dockerfile .

# Build a minimal CLI-only image containing just the llama-cli executable.
docker build --target=light -t llama-openvino:light -f .devops/openvino.Dockerfile .

# Builds a server-only image with llama-server executable, health check endpoint, and REST API support.
docker build --target=server -t llama-openvino:server -f .devops/openvino.Dockerfile .

# If you are behind a proxy:
docker build --build-arg http_proxy=$http_proxy --build-arg https_proxy=$https_proxy --target=light -t llama-openvino:light -f .devops/openvino.Dockerfile .

Run llama.cpp with OpenVINO backend Docker container. Save sample models in ~/models as shown above. It will be mounted to the container in the examples below.

#  Run Docker container
docker run --rm -it -v ~/models:/models llama-openvino:light --no-warmup -m /models/Llama-3.2-1B-Instruct-Q4_0.gguf

# With Intel GPU access (iGPU or dGPU)
docker run --rm -it -v ~/models:/models \
--device=/dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
llama-openvino:light --no-warmup -m /models/Llama-3.2-1B-Instruct-Q4_0.gguf

# With Intel NPU access
docker run --rm -it --env GGML_OPENVINO_DEVICE=NPU -v ~/models:/models \
--device=/dev/accel --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) \
llama-openvino:light --no-warmup -m /models/Llama-3.2-1B-Instruct-Q4_0.gguf

Run Llama.cpp Server with OpenVINO Backend:

# Run the Server Docker container
docker run --rm -it -p 8080:8080 -v ~/models:/models llama-openvino:server --no-warmup -m /models/Llama-3.2-1B-Instruct-Q4_0.gguf

# In a NEW terminal, test the server with curl

# If you are behind a proxy, make sure to set NO_PROXY to avoid proxy for localhost
export NO_PROXY=localhost,127.0.0.1

# Test health endpoint
curl -f http://localhost:8080/health

# Test with a simple prompt
curl -X POST "http://localhost:8080/v1/chat/completions" -H "Content-Type: application/json" \
 -d '{"messages":[{"role":"user","content":"Write a poem about OpenVINO"}],"max_tokens":100}' | jq .

Runtime Configuration

The OpenVINO backend can be configured using the following environment variables at runtime to control device selection, caching, debugging, and profiling behavior.

Configuration Options

Variable	Default	Description
`GGML_OPENVINO_DEVICE`	`CPU`	Specify the target device (CPU, GPU, NPU). On systems with multiple GPUs, use `GPU.0` or `GPU.1` to explicitly target specific GPU. See OpenVINO GPU Device. When set to NPU, static compilation mode is enabled for optimal performance.
`GGML_OPENVINO_CACHE_DIR`	`not set`	Directory for OpenVINO model caching (recommended: `/tmp/ov_cache`). Enables model caching when set. Not supported on NPU devices.
`GGML_OPENVINO_PREFILL_CHUNK_SIZE`	`256`	Token chunk size for NPU prefill.
`GGML_OPENVINO_STATEFUL_EXECUTION`	`0`	Enable stateful KV cache on for better performance. Recommended on CPU, GPU.
`GGML_OPENVINO_PROFILING`	`0`	Enable execution-time profiling.
`GGML_OPENVINO_DUMP_CGRAPH`	`0`	Dump the GGML compute graph to `cgraph_ov.txt`.
`GGML_OPENVINO_DUMP_IR`	`0`	Serialize OpenVINO IR files with timestamps.
`GGML_OPENVINO_DEBUG_INPUT`	`0`	Enable input debugging and print input tensor info.
`GGML_OPENVINO_DEBUG_OUTPUT`	`0`	Enable output debugging and print output tensor info.
`GGML_OPENVINO_PRINT_CGRAPH_TENSOR_ADDRESS`	`0`	Print tensor address map once.

[!NOTE] GGML_OPENVINO_STATEFUL_EXECUTION is an Experimental feature to allow stateful execution for managing the KV cache internally inside the OpenVINO model, improving performance on CPUs and GPUs. Stateful execution is not effective on NPUs, and not all models currently support this feature. This feature is experimental and has been validated only with the llama-simple, llama-cli, llama-bench, and llama-run applications and is recommended to enable for the best performance. Other applications, such as llama-server and llama-perplexity, are not yet supported.

Example Usage

GPU Inference with Profiling

# If the system has multiple GPUs, use GPU.0 or GPU.1 to explicitly target a specific GPU.

# Linux
export GGML_OPENVINO_CACHE_DIR=/tmp/ov_cache
export GGML_OPENVINO_PROFILING=1
export GGML_OPENVINO_DEVICE=GPU

./build/ReleaseOV/bin/llama-simple -m ~/models/Llama-3.2-1B-Instruct-Q4_0.gguf -n 50 "The story of AI is "

# Windows Command Line
set GGML_OPENVINO_CACHE_DIR=C:\tmp\ov_cache
set GGML_OPENVINO_PROFILING=1
set GGML_OPENVINO_DEVICE=GPU

# Windows PowerShell
$env:GGML_OPENVINO_CACHE_DIR = "C:\tmp\ov_cache"
$env:GGML_OPENVINO_PROFILING = "1"
$env:GGML_OPENVINO_DEVICE = "GPU"

build\ReleaseOV\bin\llama-simple.exe -m "C:\models\Llama-3.2-1B-Instruct-Q4_0.gguf" -n 50 "The story of AI is "

llama-bench

# -fa 1 is required when running llama-bench with the OpenVINO backend.
GGML_OPENVINO_DEVICE=GPU ./llama-bench -fa 1

NPU Notes

Model caching is not yet supported
Does not support llama-server -np > 1 (multiple parallel sequences)
Only supports llama-perplexity -b 512 or smaller

Llama.cpp Tools

The following tools work with the OpenVINO backend on CPU, GPU, NPU:

llama-simple
llama-run
llama-cli
llama-server
llama-bench
llama-perplexity

Work in Progress

Performance and memory optimizations
Accuracy validation
Broader quantization coverage
Support for additional model architectures

16 KiB Raw Blame History

OpenVINO Backend for llama.cpp

Supported Devices

Supported Model Precisions

Quantization Support Details

CPU and GPU

NPU

Additional Notes

Validated Models

Build Instructions

Prerequisites

1. Install OpenVINO Runtime

2. Build llama.cpp with OpenVINO Backend

3. Download Sample Model

4. Run Inference with OpenVINO Backend

Docker Build

Runtime Configuration

Configuration Options

Example Usage

GPU Inference with Profiling

llama-bench

NPU Notes

Llama.cpp Tools

Work in Progress

16 KiB

Raw Blame History