@inproceedings{behera-etal-2026-sarathi,
title = "{SARATHI}: Decoupled Data-Free Probing and Adaptive Reconstruction for Structured {LLM} Pruning",
author = "Behera, Sankar and
Singh, Dhruv and
Ahlawat, Satyadev and
Prasad, Yamuna",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1415/",
pages = "32314--32328",
ISBN = "979-8-89176-459-0",
abstract = "Structured pruning reduces the inference cost of large language models (LLMs) by permanently removing neurons and weight rows, yielding dense, hardware-agnostic speedup. Existing post-training methods couple neuron selection and weight reconstruction through a shared calibration dataset, introducing domain bias and out-of-memory failures at high sparsity. We propose SARATHI, a framework that decouples these two stages. A data-free \textit{NMF Residual Subspace Probe} scores neurons using intrinsic weight geometry, without accessing any calibration data, and carries a formal connection to Robust PCA. \textit{Adaptive OBS} then reconstructs the pruned weights with a localized calibration batch, operating in $\mathcal{O}(d^2)$ peak memory per layer and eliminating the global memory bottleneck of coupled methods. A \textit{Bias Shift Compensation} step corrects systematic activation drift in post-LayerNorm architectures that would otherwise cause perplexity collapse at high sparsity. Experiments across six models from two architecture families (gated and non-gated FFN) at 15{--}40{\%} sparsity show that SARATHI achieves competitive perplexity and matches or exceeds existing structured pruning baselines on zero-shot accuracy. Code: \url{https://github.com/snkrbhr/SARATHI}"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="behera-etal-2026-sarathi">
<titleInfo>
<title>SARATHI: Decoupled Data-Free Probing and Adaptive Reconstruction for Structured LLM Pruning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sankar</namePart>
<namePart type="family">Behera</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dhruv</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Satyadev</namePart>
<namePart type="family">Ahlawat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yamuna</namePart>
<namePart type="family">Prasad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Structured pruning reduces the inference cost of large language models (LLMs) by permanently removing neurons and weight rows, yielding dense, hardware-agnostic speedup. Existing post-training methods couple neuron selection and weight reconstruction through a shared calibration dataset, introducing domain bias and out-of-memory failures at high sparsity. We propose SARATHI, a framework that decouples these two stages. A data-free NMF Residual Subspace Probe scores neurons using intrinsic weight geometry, without accessing any calibration data, and carries a formal connection to Robust PCA. Adaptive OBS then reconstructs the pruned weights with a localized calibration batch, operating in \mathcalO(d²) peak memory per layer and eliminating the global memory bottleneck of coupled methods. A Bias Shift Compensation step corrects systematic activation drift in post-LayerNorm architectures that would otherwise cause perplexity collapse at high sparsity. Experiments across six models from two architecture families (gated and non-gated FFN) at 15–40% sparsity show that SARATHI achieves competitive perplexity and matches or exceeds existing structured pruning baselines on zero-shot accuracy. Code: https://github.com/snkrbhr/SARATHI</abstract>
<identifier type="citekey">behera-etal-2026-sarathi</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1415/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>32314</start>
<end>32328</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SARATHI: Decoupled Data-Free Probing and Adaptive Reconstruction for Structured LLM Pruning
%A Behera, Sankar
%A Singh, Dhruv
%A Ahlawat, Satyadev
%A Prasad, Yamuna
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F behera-etal-2026-sarathi
%X Structured pruning reduces the inference cost of large language models (LLMs) by permanently removing neurons and weight rows, yielding dense, hardware-agnostic speedup. Existing post-training methods couple neuron selection and weight reconstruction through a shared calibration dataset, introducing domain bias and out-of-memory failures at high sparsity. We propose SARATHI, a framework that decouples these two stages. A data-free NMF Residual Subspace Probe scores neurons using intrinsic weight geometry, without accessing any calibration data, and carries a formal connection to Robust PCA. Adaptive OBS then reconstructs the pruned weights with a localized calibration batch, operating in \mathcalO(d²) peak memory per layer and eliminating the global memory bottleneck of coupled methods. A Bias Shift Compensation step corrects systematic activation drift in post-LayerNorm architectures that would otherwise cause perplexity collapse at high sparsity. Experiments across six models from two architecture families (gated and non-gated FFN) at 15–40% sparsity show that SARATHI achieves competitive perplexity and matches or exceeds existing structured pruning baselines on zero-shot accuracy. Code: https://github.com/snkrbhr/SARATHI
%U https://aclanthology.org/2026.emnlp-main.1415/
%P 32314-32328
Markdown (Informal)
[SARATHI: Decoupled Data-Free Probing and Adaptive Reconstruction for Structured LLM Pruning](https://aclanthology.org/2026.emnlp-main.1415/) (Behera et al., EMNLP 2026)
ACL