ZeroQuant-HERO: Hardware-Enhanced Robust Optimized Post-Training Quantization Framework for W8A8 Transformers
This work addresses the need for efficient quantization in Transformers for deployment on hardware, though it appears incremental by building on existing solutions like ZeroQuant.
The authors tackled the problem of reducing memory and computational demands in deep neural network inference by developing ZeroQuant-HERO, a hardware-enhanced post-training quantization framework for W8A8 Transformers, which integrates memory bandwidth and compute-intensive operators to optimize hardware performance and allows flexible switching between INT8 and FP16/BF16 modes to enhance accuracy.
Quantization techniques are pivotal in reducing the memory and computational demands of deep neural network inference. Existing solutions, such as ZeroQuant, offer dynamic quantization for models like BERT and GPT but overlook crucial memory-bounded operators and the complexities of per-token quantization. Addressing these gaps, we present a novel, fully hardware-enhanced robust optimized post-training W8A8 quantization framework, ZeroQuant-HERO. This framework uniquely integrates both memory bandwidth and compute-intensive operators, aiming for optimal hardware performance. Additionally, it offers flexibility by allowing specific INT8 modules to switch to FP16/BF16 mode, enhancing accuracy.