Run PNN models on available GPUs and CPUs

FP32 · INT8 · INT4 · 1-bit

FP32 · INT8

Models of any size, split into parts and run part by part across machines — distributed and fault-tolerant, so a failed machine does not stop the run.

• • Models larger than one GPU's memory
• • Parts run in sequence, each on the hardware that suits it
• • A failed part is re-run elsewhere