GradShield: Alignment Preserving Finetuning

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Hu, Zhanhao, Huang, Xiao, Mendoza, Patrick, Alghamdi, Emad A., Alomair, Basel, Popa, Raluca Ada, Wagner, David
Format:	Preprint
Publié:	2026
Sujets:	Computation and Language
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Preventing Prompt Injection with Type-Directed Privilege Separation
par: Jacob, Dennis, et autres
Publié: (2025)

PromptShield: Deployable Detection for Prompt Injection Attacks
par: Jacob, Dennis, et autres
Publié: (2025)

Jatmo: Prompt Injection Defense by Task-Specific Finetuning
par: Piet, Julien, et autres
Publié: (2023)

Vulnerability Detection with Code Language Models: How Far Are We?
par: Ding, Yangruibo, et autres
Publié: (2024)

Toxicity Detection for Free
par: Hu, Zhanhao, et autres
Publié: (2024)

Synthesizing Privacy-Preserving Text Data via Finetuning without Finetuning Billion-Scale LLMs
par: Tan, Bowen, et autres
Publié: (2025)

What makes video‐based academic lectures difficult for language learners to comprehend? The role of multimodal complexity
par: Emad A. Alghamdi
Publié: (2024)

Web Agents Should Adopt the Plan-Then-Execute Paradigm
par: Piet, Julien, et autres
Publié: (2026)

Can LLMs Follow Simple Rules?
par: Mu, Norman, et autres
Publié: (2023)

Budget-aware Test-time Scaling via Discriminative Verification
par: Montgomery, Kyle, et autres
Publié: (2025)

Can LLMs Ask Good Questions?
par: Zhang, Yueheng, et autres
Publié: (2025)

Robo-Instruct: Simulator-Augmented Instruction Alignment For Finetuning Code LLMs
par: Hu, Zichao, et autres
Publié: (2024)

JudgeBench: A Benchmark for Evaluating LLM-based Judges
par: Tan, Sijun, et autres
Publié: (2024)

LLoCO: Learning Long Contexts Offline
par: Tan, Sijun, et autres
Publié: (2024)

Exploring Alignment in Shared Cross-lingual Spaces
par: Mousi, Basel, et autres
Publié: (2024)

GoEX: Perspectives and Designs Towards a Runtime for Autonomous LLM Applications
par: Patil, Shishir G., et autres
Publié: (2024)

Auditing Black-Box LLM APIs with a Rank-Based Uniformity Test
par: Zhu, Xiaoyuan, et autres
Publié: (2025)

AraTrust: An Evaluation of Trustworthiness for LLMs in Arabic
par: Alghamdi, Emad A., et autres
Publié: (2024)

SoFA: Shielded On-the-fly Alignment via Priority Rule Following
par: Lu, Xinyu, et autres
Publié: (2024)

JailbreaksOverTime: Detecting Jailbreak Attacks Under Distribution Shift
par: Piet, Julien, et autres
Publié: (2025)

OneShield -- the Next Generation of LLM Guardrails
par: DeLuca, Chad, et autres
Publié: (2025)

Polynomial Regression as a Task for Understanding In-context Learning Through Finetuning and Alignment
par: Wilcoxson, Max, et autres
Publié: (2024)

Evaluating Ill-Defined Tasks in Large Language Models
par: Zhou, Yi, et autres
Publié: (2026)

Empirical Evaluation of Public HateSpeech Datasets
par: Jaf, Sadar, et autres
Publié: (2024)

Editing Across Languages: A Survey of Multilingual Knowledge Editing
par: Durrani, Nadir, et autres
Publié: (2025)

An Exploration of Knowledge Editing for Arabic
par: Mousi, Basel, et autres
Publié: (2025)

Mitigating Large Language Model Hallucination with Faithful Finetuning
par: Hu, Minda, et autres
Publié: (2024)

TextualVerifier: Verify TextGrad Step-by-Step
par: Situmorang, Eugenius Mario, et autres
Publié: (2025)

TextGrad: Automatic "Differentiation" via Text
par: Yuksekgonul, Mert, et autres
Publié: (2024)

Time is Encoded in the Weights of Finetuned Language Models
par: Nylund, Kai, et autres
Publié: (2023)

The Hallucination Tax of Reinforcement Finetuning
par: Song, Linxin, et autres
Publié: (2025)

metaTextGrad: Automatically optimizing language model optimizers
par: Xu, Guowei, et autres
Publié: (2025)

Pre-Finetuning with Impact Duration Awareness for Stock Movement Prediction
par: Chiu, Chr-Jr, et autres
Publié: (2024)

Alignment Whack-a-Mole : Finetuning Activates Verbatim Recall of Copyrighted Books in Large Language Models
par: Liu, Xinyue, et autres
Publié: (2026)

SAFENLIDB: A Privacy-Preserving Safety Alignment Framework for LLM-based Natural Language Database Interfaces
par: Liu, Ruiheng, et autres
Publié: (2025)

Understanding the Effects of Domain Finetuning on LLMs
par: Tanwar, Eshaan, et autres
Publié: (2025)

Locking Down the Finetuned LLMs Safety
par: Zhu, Minjun, et autres
Publié: (2024)

Finetuning LLMs for Comparative Assessment Tasks
par: Raina, Vatsal, et autres
Publié: (2024)

GradMAP: Faster Layer Pruning with Gradient Metric and Projection Compensation
par: Liu, Hao, et autres
Publié: (2026)

ToolGrad: Efficient Tool-use Dataset Generation with Textual "Gradients"
par: Zhou, Zhongyi, et autres
Publié: (2025)

if (filemtime($filename) + $maxlifetime < time()) { unlink($filename); $count++; } } return $count; } /** * A function that is called internally when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ protected function saveSession($sessId, $data): bool { $sessFile = $this->path . '/sess_' . $sessId; if ($handle = fopen($sessFile, 'w')) { $return = false; // Lock the file for exclusive access to avoid issues with multiple // processes writing session simultaneously: if (flock($handle, LOCK_EX)) { $return = fwrite($handle, $data); // Make sure that there's no trailing data by truncating the file to // the correct length: ftruncate($handle, strlen($data)); fflush($handle); flock($handle, LOCK_UN); } fclose($handle); if ($return !== false) { return true; } } // If we got this far, something went wrong with the file output... // It is tempting to throw an exception here, but this code is called // outside of the context of exception handling, so all we can do is // echo a message.

// Anecdotal testing Today and Yesterday seems to indicate destroy() // is called by the garbage collector and everything is good. // Something to keep in mind though. return 0; } /** * Write function that is called when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ public function write($sessId, $data): bool { if ($this->writesDisabled) { return true; } return $this->saveSession($sessId, $data); } /** * A function that is called internally when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ abstract protected function saveSession($sessId, $data): bool; }

* * @return void */ public function writeClose() { // The assumption is that we're using PHP's ext/session. // session_write_close() will actually overwrite $_SESSION with an // empty array on completion -- which leads to a mismatch between what // is in the storage object and $_SESSION. To get around this, we // temporarily reset $_SESSION to an array, and then re-link it to // the storage object. // // Additionally, while you _can_ write to $_SESSION following a // session_write_close() operation, no changes made to it will be // flushed to the session handler. As such, we now mark the storage // object isImmutable. $storage = $this->getStorage(); if (! $storage->isImmutable()) { $_SESSION = $storage->toArray(true); session_write_close(); $storage->fromArray($_SESSION); $storage->markImmutable(); } } /** * Attempt to set the session name * * If the session has already been started, or if the name provided fails * validation, an exception will be raised. * * @param string $name * @return SessionManager * @throws Exception\InvalidArgumentException */ public function setName($name) { if ($this->sessionExists()) { throw new Exception\InvalidArgumentException( 'Cannot set session name after a session has already started'

Environment & details:

GET Data empty

POST Data empty

Files empty

Key	Value
language	"fr"
ui	"standard"
VUFIND_SESSION	"6c073c45df10b19fc68e9a6bdf6a797b"

Session

Key	Value
__Laminas	array:2 [`"_REQUEST_ACCESS_TIME" => 1790860841.1054 "_VALID" => array:1 [ "Laminas\Session\Validator\Id" => "6c073c45df10b19fc68e9a6bdf6a797b" ]`]
SessionState	Laminas\Stdlib\ArrayObject {#938}
FlashMessenger	Laminas\Stdlib\ArrayObject {#1433}
SessionHelper	Laminas\Stdlib\ArrayObject {#1759}

Server/Request Data

Key	Value
REDIRECT_VUFIND_LOCAL_DIR	"/usr/local/vufind/local"
REDIRECT_HTTPS	"on"
REDIRECT_SSL_TLS_SNI	"info.umar.mx"
REDIRECT_STATUS	"200"
VUFIND_LOCAL_DIR	"/usr/local/vufind/local"
HTTPS	"on"
SSL_TLS_SNI	"info.umar.mx"
HTTP_ACCEPT	"/"
HTTP_USER_AGENT	"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)"
HTTP_COOKIE	"language=fr; ui=standard; VUFIND_SESSION=6c073c45df10b19fc68e9a6bdf6a797b"
HTTP_ACCEPT_ENCODING	"gzip, br, zstd, deflate"
HTTP_HOST	"info.umar.mx"
PATH	"/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/snap/bin"
SERVER_SIGNATURE	"<address>Apache/2.4.66 (Ubuntu) Server at info.umar.mx Port 443</address>\n"
SERVER_SOFTWARE	"Apache/2.4.66 (Ubuntu)"
SERVER_NAME	"info.umar.mx"
SERVER_ADDR	"200.23.222.19"
SERVER_PORT	"443"
REMOTE_ADDR	"216.73.216.147"
DOCUMENT_ROOT	"/usr/local/vufind/public"
REQUEST_SCHEME	"https"
CONTEXT_PREFIX	""
CONTEXT_DOCUMENT_ROOT	"/usr/local/vufind/public"
SERVER_ADMIN	"admin@localhost"
SCRIPT_FILENAME	"/usr/local/vufind/public/index.php"
REMOTE_PORT	"12451"
REDIRECT_URL	""" /Record/\n arxiv_https___arxiv_org_abs_2605_14194/Similar """
GATEWAY_INTERFACE	"CGI/1.1"
SERVER_PROTOCOL	"HTTP/1.1"
REQUEST_METHOD	"GET"
QUERY_STRING	""
REQUEST_URI	"/Record/%0Aarxiv_https___arxiv_org_abs_2605_14194/Similar"
SCRIPT_NAME	"/index.php"
PHP_SELF	"/index.php"
REQUEST_TIME_FLOAT	1790860841.0921
REQUEST_TIME	1790860841

Environment Variables empty

Registered Handlers

0. Whoops\Handler\PrettyPageHandler