hardware-aware-transformers
Explore HAT's ability to leverage Hardware-Aware Transformers to boost natural language processing efficiency. The project offers PyTorch code and includes 50 pre-trained models that aid in locating optimized solutions for distinct hardware, cutting search costs by over 10000 times. HAT provides up to triple the speed and a 3.7-fold reduction in model size with no performance detriment. Featuring latency feedback for hardware like Raspberry Pi and Intel Xeon, HAT presents a cutting-edge method for optimizing machine translation tasks, delivering superior performance across various devices.
