4 ms·
AFAIK, XLA is the only compiler that can target TPUs, so anything written for TPUs would need to lower to HLO. That said, why would you want to to run non-ML wo
by lordswork 3y ago
AFAIK, XLA is the only compiler that can target TPUs, so anything written for TPUs would need to lower to HLO. That said, why would you want to to run non-ML workloads on such an ML-specialized chip?
- sxp 3y agoI was interested in seeing if raytracing or similar embarrassingly parallel workloads could be ported to a TPU.