亚马逊云科技周一发布了TorchNeuron开发工具，使AI开发者能够将基于PyTorch框架开发的应用快速迁移至亚马逊自研AI芯片Trainium，仅需修改单行配置代码即可运行，无需重写整个应用1。这一举措针对性极强——约85%的AI应用采用PyTorch开发1。
为了进一步打破硬件绑定，亚马逊同时宣布向PyTorch主项目贡献代码，目标是从框架层面移除对特定芯片的依赖1。AWS Annapurna Labs技术副总裁Colin Brace表示，"真正的故事是移除让开发者无法真正选择的锁定"1。实际效果已初步显现：音频处理公司Deepgram使用原生PyTorch迁移到Trainium后，处理性能实现了每美元处理70%更多音频的提升1。
Amazon Web Services announced TorchNeuron on Monday, a new developer tool designed to enable AI startups to port PyTorch-based applications to Amazon's custom Trainium chip with minimal code changes.1 The tool requires developers to modify only a single configuration line to run their existing PyTorch applications on Trainium, eliminating the need for extensive code rewrites.1
Approximately 85% of AI applications are built using PyTorch.1 AWS plans to contribute code directly to the PyTorch project to decouple the framework from hardware-specific vendor dependencies, aimed at giving developers genuine alternatives rather than forcing them into single-supplier arrangements.1 According to Colin Brace, vice president of technology at AWS Annapurna Labs, "The real story is removing the lock-in that prevents developers from having true choice."1
Early adoption has shown promising results. Deepgram, an AI company, migrated to Trainium using native PyTorch and achieved a 70% improvement in audio processing efficiency per dollar spent.1 Trainium itself handles both training and inference workloads while integrating tightly with cloud infrastructure.1
评论
还没有评论，欢迎留下第一条。