#FPGA

StreamTensor compiles PyTorch LLM graphs into stream-scheduled FPGA dataflows using an itensor type and LP-based FIFO sizing to reduce DRAM round-trips, lower latency, and improve energy efficiency.