pytorch/glow
A machine learning compiler and execution engine from PyTorch that optimizes and generates code for neural network inference on hardware accelerators.

Velocity · 7d
+1.0
★ / day
Trend
→steady
star history
Glow lowers traditional neural network dataflow graphs into a two-phase strongly-typed intermediate representation, enabling domain-specific optimizations at the high-level IR and memory-related optimizations at the lower-level IR. The compiler performs machine-specific code generation to leverage specialized hardware features of neural network accelerators from partners like Intel, Habana, Bitmain, and others.