BladeDISC深度學(xué)習(xí)編譯器
BladeDISC 是阿里巴巴集團(tuán)自主研發(fā)并開源的深度學(xué)習(xí)編譯器,旨在為用戶提供通用、透明、易用的深度學(xué)習(xí)性能優(yōu)化能力。BladeDISC支持主流的機(jī)器學(xué)習(xí)框架,如TensorFlow、PyTorch,支持主流硬件,如GPGPU、CPU 等。BladeDISC在架構(gòu)層面根本地解決了深度學(xué)習(xí)領(lǐng)域具有動(dòng)態(tài)尺寸Tensor的優(yōu)化難題,大大提升了深度學(xué)習(xí)模型的性能,降低了部署的難度。同時(shí),BladeDISC 還提供了多種靈活的部署方案,包括插件模式集成到Pytorch和TensorFlow運(yùn)行時(shí),也包括獨(dú)立運(yùn)行與AOT編譯的支持。BladeDISC源于阿里巴巴內(nèi)部諸多的深度學(xué)習(xí)加速場景,也同時(shí)服務(wù)于阿里云上的客戶,在生產(chǎn)中經(jīng)受了廣泛的考驗(yàn)。
功能與支持
前端框架支持情況
| TensorFlow | PyTorch | |
|---|---|---|
| 推理 | Yes | Yes |
| 訓(xùn)練 | Yes | Ongoing |
后端硬件支持情況
| Status | |
|---|---|
| Nvidia GPU | Yes |
| AMD GPU | Yes |
| Hygon DCU | Yes |
| X86 | Yes |
| AArch64 | Yes |
典型模型的加速效果
入門與示例
- How to Setup and Build from Source
- Use Case of TensorFlow Inference and Training
- Use Case of PyTorch Inference
論文發(fā)表
開發(fā)入門
- Tutorial: A Walkthough of the BladeDISC Pass Pipeline
- Introduction on Runtime Abstraction Layer
- TorchBlade Overview
- Tutorial: How to Add a New Torch Operator Converter
演講與文章
- DISC: A Dynamic Shape Compiler for Machine Learning Workload
- Performance optimization practice for dynamic shape AI workloads via a compiler-based approach
- 2022/07/31 BladeDISC: A Practice of Dynamic Shape Deep Learning Compiler(Chinese)
- 2022/07/07 BladeDISC and Torch-MLIR Roadmap Talk on Torch-MLIR Community
- GTC22-S41073, Generalized and Transparent AI Optimization Solutions with AI Compilers from Cloud Service
- GTC22-S41395, Easier-to-use and More Robust TensorRT via PAI-Blade
評(píng)論
圖片
表情
