📄 中文摘要
自2025年初以来，许多硬件供应商和客户对我们的模型管理平台表示了浓厚的兴趣。该平台界面友好，以模型为中心，用户可以一键启动，降低了使用门槛。一些客户甚至愿意支付费用让我们为他们部署类似的设置。然而，该平台最初是为特定硬件设计的，如NVIDIA GPU、华为Ascend、海光DCU和AMD ROCm等。每种硬件都有不同的驱动程序、设备挂载机制、环境变量和安全上下文，适应每种新硬件类型需要编写大量代码，过程相当痛苦。更糟糕的是，后续的维护工作也十分繁琐。
为什么我们构建了 AIMA：一个用于管理 AI 推理的开源项目
📄 English Summary
Why We Built AIMA: An Open-Source Project for Managing AI Inference with AI
Since early 2025, numerous hardware vendors and clients have shown significant interest in our model management platform. The platform features a user-friendly interface, is model-centric, and allows users to get started with a single click, which lowers the barrier to entry. Some clients were even willing to pay for us to deploy a similar setup for them. However, the platform was originally designed for specific hardware types, including NVIDIA GPUs, Huawei Ascend, Hygon DCU, and AMD ROCm. Each hardware type has different drivers, device mounting mechanisms, environment variables, and security contexts, making it painful to adapt to each new hardware type due to the massive amounts of code required. The aftermath of these adaptations has proven to be equally cumbersome.