TorchInductor: a PyTorch-native Compiler with Define-by-Run IR and Symbolic Shapes - compiler - PyTorch Dev Discussions
dev-discuss.pytorch.org
lost in translation TorchDynamo must be paired with a compiler backend that converts the captured graphs into fast machine code graph capture problem of PyTorch with dynamic Python bytecode transformation built a lightweight autotuner 260 to select the best backend for each subgraph. exporting to di
1 Users
0 Comments
13 Highlights
1 Notes
Top Highlights
- lost in translation
- TorchDynamo must be paired with a compiler backend that converts the captured graphs into fast machine code
- graph capture problem of PyTorch with dynamic Python bytecode transformation
- built a lightweight autotuner 260 to select the best backend for each subgraph.
- exporting to different backends that differ greatly from PyTorch.
Domain
Ready to highlight and find good content?
Glasp is a social web highlighter that people can highlight and organize quotes and thoughts from the web, and access other like-minded people’s learning.