HRM的环境配置需要以下关键步骤（以Linux系统为例）：
Preparation of the basic environment
- 安装NVIDIA驱动和CUDA 12.6工具包
- 配置CUDA环境变量：
export CUDA_HOME=/usr/local/cuda-12.6 - 安装PyTorch（适配CUDA 12.6）：
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
必要组件安装
- 构建工具：
pip3 install packaging ninja wheel setuptools setuptools-scm - FlashAttention（根据GPU架构选择安装方式）：
- Hopper架构（如H100）：需源码编译安装
- Ampere/更早架构（如RTX 30/40系）：直接pip安装
Project deployment
- Cloning the official warehouse:
git clone https://github.com/sapientinc/HRM.git - Install the dependencies:
pip install -r requirements.txt - （可选）配置W&B用于训练监控
This answer comes from the articleHRM: Hierarchical Reasoning Model for Complex ReasoningThe
May not be reproduced without permission:AI productivity tools " 如何安装和配置HRM的运行环境？
English 
简体中文 
日本語 
Deutsch 
Português do Brasil