Pith. sign in

REVIEW 4 cited by

Efficient Softmax Approximation for Deep Neural Networks with Attention Mechanism

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2111.10770 v1 pith:QT3A2VVT submitted 2021-11-21 cs.LG cs.AI

classification cs.LGcs.AI
keywords softmaxnetworksneuralacceleratingapproximationattentionbecausedeep
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

There has been a rapid advance of custom hardware (HW) for accelerating the inference speed of deep neural networks (DNNs). Previously, the softmax layer was not a main concern of DNN accelerating HW, because its portion is relatively small in multi-layer perceptron or convolutional neural networks. However, as the attention mechanisms are widely used in various modern DNNs, a cost-efficient implementation of softmax layer is becoming very important. In this paper, we propose two methods to approximate softmax computation, which are based on the usage of LookUp Tables (LUTs). The required size of LUT is quite small (about 700 Bytes) because ranges of numerators and denominators of softmax are stable if normalization is applied to the input. We have validated the proposed technique over different AI tasks (object detection, machine translation, sentiment analysis, and semantic equivalence) and DNN models (DETR, Transformer, BERT) by a variety of benchmarks (COCO17, WMT14, WMT17, GLUE). We showed that 8-bit approximation allows to obtain acceptable accuracy loss below $1.0\%$.

Discussion (0). Sign in to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. A Systolic Array Architecture for Nonlinear Activation Functions and Softmax Computation using Chebyshev Polynomials

    cs.AR 2026-08 conditional novelty 6.0 of 10

    A shared systolic array computes tanh, sigmoid, ReLU, and softmax via Chebyshev polynomial approximation, with reported error and area/power advantages over CORDIC.

  2. CIMple: Standard-cell SRAM-based CIM with LUT-based split softmax for attention acceleration

    cs.AR 2026-04 unverdicted novelty 5.0 of 10

    CIMple delivers a 32 kb digital SRAM-based compute-in-memory accelerator for transformer self-attention that reaches 26.1 TOPS/W at 0.85 V in 28 nm with INT8 precision using dual-banked architecture and LUT-based spli...

  3. Techniques for Peak Memory Reduction for LoRA Fine-tuning of LLMs on Edge Devices

    cs.LG 2026-06 unverdicted novelty 4.0 of 10

    Presents quantization, checkpointing, softmax approximation, and logits masking to achieve substantial peak memory reductions in LoRA fine-tuning of 3B LLMs.

  4. Integer-only Quantized Transformers for Embedded FPGA-based Time-series Forecasting in AIoT

    cs.LG 2024-07 unverdicted novelty 4.0 of 10

    4-bit and 6-bit integer-only quantized Transformers implemented on Spartan-7 FPGA for AIoT time-series forecasting achieve 0.63% higher test loss than 8-bit baselines but up to 132x speedup and 48x lower energy.

Pith tools