@inproceedings{zuhri-etal-2026-softpick,
title = "Softpick: No Attention Sink, No Massive Activations with Rectified Softmax",
author = "Zuhri, Zayd Muhammad Kawakibi and
Fuadi, Erland Hilman and
Aji, Alham Fikri",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.905/",
doi = "10.18653/v1/2026.findings-acl.905",
pages = "18179--18213",
ISBN = "979-8-89176-395-1",
abstract = "We introduce softpick, a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations. Our experiments with 340M and 1.8B parameter models demonstrate that softpick achieves 0{\%} sink rate consistently. The softpick transformers produce hidden states with significantly lower kurtosis and creates sparse attention maps. Quantized models using softpick outperform softmax on standard benchmarks, with a particularly pronounced advantage at lower bit precisions. Our analysis and discussion shows how softpick has the potential to open new possibilities for quantization, low-precision training, sparsity optimization, pruning, and interpretability. Our code: https://github.com/zaydzuhri/softpick-attention."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zuhri-etal-2026-softpick">
<titleInfo>
<title>Softpick: No Attention Sink, No Massive Activations with Rectified Softmax</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zayd</namePart>
<namePart type="given">Muhammad</namePart>
<namePart type="given">Kawakibi</namePart>
<namePart type="family">Zuhri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Erland</namePart>
<namePart type="given">Hilman</namePart>
<namePart type="family">Fuadi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alham</namePart>
<namePart type="given">Fikri</namePart>
<namePart type="family">Aji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>We introduce softpick, a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations. Our experiments with 340M and 1.8B parameter models demonstrate that softpick achieves 0% sink rate consistently. The softpick transformers produce hidden states with significantly lower kurtosis and creates sparse attention maps. Quantized models using softpick outperform softmax on standard benchmarks, with a particularly pronounced advantage at lower bit precisions. Our analysis and discussion shows how softpick has the potential to open new possibilities for quantization, low-precision training, sparsity optimization, pruning, and interpretability. Our code: https://github.com/zaydzuhri/softpick-attention.</abstract>
<identifier type="citekey">zuhri-etal-2026-softpick</identifier>
<identifier type="doi">10.18653/v1/2026.findings-acl.905</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.905/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>18179</start>
<end>18213</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Softpick: No Attention Sink, No Massive Activations with Rectified Softmax
%A Zuhri, Zayd Muhammad Kawakibi
%A Fuadi, Erland Hilman
%A Aji, Alham Fikri
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F zuhri-etal-2026-softpick
%X We introduce softpick, a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations. Our experiments with 340M and 1.8B parameter models demonstrate that softpick achieves 0% sink rate consistently. The softpick transformers produce hidden states with significantly lower kurtosis and creates sparse attention maps. Quantized models using softpick outperform softmax on standard benchmarks, with a particularly pronounced advantage at lower bit precisions. Our analysis and discussion shows how softpick has the potential to open new possibilities for quantization, low-precision training, sparsity optimization, pruning, and interpretability. Our code: https://github.com/zaydzuhri/softpick-attention.
%R 10.18653/v1/2026.findings-acl.905
%U https://aclanthology.org/2026.findings-acl.905/
%U https://doi.org/10.18653/v1/2026.findings-acl.905
%P 18179-18213
Markdown (Informal)
[Softpick: No Attention Sink, No Massive Activations with Rectified Softmax](https://aclanthology.org/2026.findings-acl.905/) (Zuhri et al., Findings 2026)
ACL