中间表示#
- A guide on good usage of
non_blocking
andpin_memory()
in PyTorch - Background
- A PyTorch perspective
non_blocking=True
- Synergies
- Other copy directions (GPU -> CPU, CPU -> MPS)
- Practical recommendations
- Additional considerations
- Conclusion
- Additional resources
- 模型剪枝
- 参数化教程
- 使用 FX 构建简单的 CPU 性能分析器
- 在pytorch中通道最后的内存格式
- Forward-mode 自动微分
- (beta) Building a Convolution/Batch Norm fuser in FX