Luminous Computing

I maintain a behavioral simulator written in Haskell that simulates our super-computer's architecture and compile various ML IRs to run on our simulator, including an in-house IR and MHLO (MLIR dialect of HLO). I work with hardware and software teams to use this simulation output to inform our hardware designs, software stack, and customer engagements.I also maintain a high level roofline Python model of our system aimed to gain preliminary insights into our architecture's performance, and to explore various target ML workloads. We also use this software to investigate the performance of various parallelization strategies on our hardware.