DAG Scheduler on MLIR
A scheduler for ML computation graphs built on MLIR, exploring how operator ordering and placement affect end-to-end latency.
Source →Graduate student · Systems, ML systems & computer architecture
I build software and hardware that makes machine learning run faster and cheaper. My interests span ML compilers, parallel programming, inference serving, and FPGA design. I like taking a slow system, finding where the time goes, and fixing it.
A scheduler for ML computation graphs built on MLIR, exploring how operator ordering and placement affect end-to-end latency.
Source →An inference-serving system for Mixture-of-Experts models, focused on expert routing, batching and memory-efficient execution.
Source →Recommendation-model inference split across CPU and GPU to balance embedding lookups and dense compute, built for a parallel programming course.
Source →Hardware design project from the FPGA course. Summarize what it accelerates and the speedup or resource numbers you achieved.
Source →A model that predicts NBA player scoring from game and player features.
Source →