Hardware Platforms

xLLM supports multiple accelerator backends for large-scale model inference. This section collects the hardware-specific entry points for environment setup, runtime device selection, launch scripts, and model support.

Platform Guides

• NVIDIA GPU - CUDA backend setup and launch entry points.
• Ascend NPU - Ascend NPU setup, runtime environment, and HCCL launch notes.
• Cambricon MLU - MLU backend setup and launch entry points.
• Hygon DCU - Hygon DCU backend setup and launch entry points.

Common Workflow

1. Prepare the platform-specific container image from the explicit commands in each platform guide.
2. Build xLLM inside the container, or use a release image that already includes xllm.
3. Start the service with the matching device backend in Launch xllm.
4. Check model and modality coverage in the Model Support List.