Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
Tensor<FP32> -> Tensor<FP16_or_INT8>
Cast or transform input tensor arrays to lower-precision formats (FP16 or INT8) to align with mobile NPU constraints.
Problem it solves
Mobile NPUs typically do not support FP32 operations and require optimized low-precision arrays to trigger hardware acceleration.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.