Developers running LLMs locally on Intel CPUs, integrated GPUs, NPUs, or Arc/Flex/Max discrete GPUs.

IPEX-LLM is a PyTorch library that accelerates local LLM inference and fine-tuning on Intel hardware using low-bit quantization and XPU-specific optimizations.

Self-hosted, installed as a Python package on Intel-equipped PCs and servers.

Formerly BigDL-LLM, actively maintained by Intel with continuous updates through 2026.

It exists to make Intel client and server chips competitive for local LLM workloads without dedicated NVIDIA GPUs.

Integrates with llama.cpp, Ollama, Hugging Face Transformers, LangChain, and vLLM.

Open source, Apache-2.0 license, self-hosted; no API key needed; all inference runs on local Intel hardware.