Skip to content

HyperDex Installation Guide

This guide will walk you through installing HyperDex-Toolchain to run large language models on HyperAccel's LPU hardware.

Overview

The LPU is built on AMD's Alveo FPGA platform and requires two main components:

  1. Xilinx Runtime (XRT): Low-level drivers and runtime for FPGA hardware
  2. HyperDex-Toolchain: HyperAccel's optimized software stack for LLM inference

The HyperDex-Toolchain includes a runtime package that allows the LPU to fully utilize its hardware capabilities, providing enhanced performance for large language models.

Prerequisites

System Requirements

  • Operating System: RHEL-8/8 or Ubuntu-22.04-LTS
  • Hardware: AMD Alveo FPGA card (U55C)
  • Xilinx Runtime (XRT): Required for FPGA operations

XRT Installation

Before proceeding, make sure XRT is properly installed. Follow our XRT install guide for detailed instructions.

Software Compatibility

The table below shows the supported Python, CUDA, and PyTorch versions. Ensure your environment matches one of these configurations:

Python Version CUDA Version PyTorch Version
3.10, 3.11, 3.12 CPU-only, 12.6 2.7.0+cpu, 2.7.0+cu126

Step 1: Set Up Python Environment

We recommend using uv, a fast Python package and virtual environment manager, to set up your environment quickly and efficiently.

Install uv and Create Virtual Environment

# Install uv package manager
$ curl -LsSf https://astral.sh/uv/install.sh | sh

# Create a new virtual environment with Python 3.10
$ uv venv -p python==3.10 --no-project --seed .hdex-env

# Activate the virtual environment
$ source .hdex-env/bin/activate

Step 2: Install HyperDex Python Package

HyperDex-Toolchain is distributed through HyperAccel's private PyPI server, which requires authentication credentials.

Set Up PyPI Authentication

Environment Variables Required

To run the following commands, you'll need your HyperAccel PyPI credentials. Please contact our support team if you need help obtaining credentials. 

export PYPI_ID="your_pypi_username"
export PYPI_PW="your_pypi_password"
These credentials will be used to authenticate with HyperAccel's private PyPI server.

Choose Your Installation

Select the appropriate installation option based on your hardware setup:

Option A: LPU-Only Environment (CPU PyTorch)

# Install PyTorch CPU version
$ uv pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cpu

# Install HyperDex-Toolchain for CPU
$ uv pip install -i https://pypi.hyperaccel.ai/simple hyperdex-toolchain==1.5.2+cpu

Option B: LPU+GPU Environment (CUDA PyTorch)

# Install PyTorch CUDA version
$ uv pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cu126

# Install HyperDex-Toolchain for CUDA
$ uv pip install -i https://${PYPI_ID}:${PYPI_PW}@pypi.hyperaccel.ai/simple hyperdex-toolchain==1.5.2+cu126

Step 3: Verification and Next Steps

To operate the LPU successfully, ensure both components are properly installed:

  1. ✅ Xilinx Runtime (XRT) - Hardware drivers and runtime
  2. ✅ HyperDex-Toolchain - Python package and dependencies

Verify Installation

Use the following commands to confirm your installation is working correctly:

Check XRT Installation: 

# Verify XRT installation and hardware detection
$ xbutil examine

Check HyperDex-Toolchain Installation: 

# Verify HyperDex-Toolchain package is installed
$ uv pip list | grep hyperdex-toolchain

Ready to Start?

Once installation is complete and verified, you can begin using HyperDex for LLM inference. Proceed to our Quick Start Guide to run your first model on the LPU.