Mar 6, 2023 · In this paper, we present FlexBlock, a DNN training accelerator with three BFP modes, possibly different among activation, weight, and gradient tensors.

Mar 13, 2022 · We develop a flexible DNN training accelerator, dubbed FlexBlock, which supports three different BFP precision modes, possibly different among activation, ...

With the hardware support, we empirically demonstrate the possibility of training DNNs even with 4-bit arithmetic (FB12) for computing feature maps and local ...

In this work, we propose a flexible DNN training hardware, i.e., FlexBlock, that supports multiple BFP precision modes (Fig. 1(a)), possibly different among ...

FlexBlock is presented, a DNN training accelerator with three BFP modes, possibly different among activation, weight, and gradient tensors, ...

Oct 22, 2024 · We evaluate the effectiveness of FlexBlock using representative DNNs on CIFAR, ImageNet and WMT14 datasets. As a result, training in FlexBlock ...

As a result, training in FlexBlock significantly improves the training speed by 1.5~5.3x and the energy efficiency by 2.4~7.0x on average compared to other ...

Backed up by this algorithmic opportunity, we develop a flexible DNN training accelerator, dubbed FlexBlock, which supports three different BFP precision modes, ...

FlexBlock is presented, a DNN training accelerator with three BFP modes, possibly different among activation, weight, and gradient tensors, ...

Oct 25, 2023 · Flexblock: A flexible dnn training accelerator with multi-mode block floating point support. IEEE Transactions on Computers (TC),. 72(9):2522 ...