Feb 28, 2023 · This work proposes a novel dynamic inference scheme, DynaTran, which prunes activations at runtime with low overhead, substantially reducing the number of ...

Oct 20, 2023 · On the other hand,. AccelTran-Server achieves 5.73× higher throughput and 3.69× lower energy consumption compared to the state-of-the-art.

Nov 1, 2023 · One of our proposed accelerators, AccelTran-Edge, achieves 330K × higher throughput with 93K × lower energy requirement when compared to a ...

To effectively implement these methods, we propose AccelTran, a novel accelerator architecture for transformers. Extensive experiments with different models and ...

This work proposes a novel dynamic inference scheme, DynaTran, which prunes activations at runtime with low overhead, substantially reducing the number of ...

Oct 22, 2024 · AccelTran [16] , a stateof-the-art transformer accelerator, executes dynamic inference by skipping all ineffectual multiply-and-accumulate (MAC) ...

Feb 28, 2023 · This improves the throughput of transformer inference. We further propose tiling the matrices in transformer operations along with diverse ...

AccelTran is a tool to simulate a design space of accelerators on diverse flexible and heterogeneous transformer architectures supported by the FlexiBERT ...

These accelerators have specially- designed hardware modules that leverage sparsity in model weights, data reuse, optimized dataflows, and CNN mapping to attain ...

AccelTran: A sparsity-aware accelerator for dynamic inference with transformers. S Tuli, NK Jha. IEEE Transactions on Computer-Aided Design of Integrated ...