aiqbenchInformații despre AI & calcul cuantic

...

Română · beta. Traducerea a fost făcută automat și este încă în curs de revizuire. Citește pagina în engleză

perspectiva aiqbenchProduse AI

This release introduces VectraYX-Vision-1B, a sub-2B parameter model designed specifically for Spanish and LATAM cybersecurity tasks with integrated visual reasoning and native tool use capabilities. The authors have addressed critical benchmarking integrity issues by retracting inflated scores from the initial version and correcting implementation bugs related to LoRA learning rates and checkpoint loading. For researchers in low-resource language security, this represents a significant step toward robust, region-specific AI agents that can interpret visual data and execute tools within constrained environments.

Sursă care acoperăarXiv

Ce este aceastaLectura unui model asupra articolului și a acoperirii conexe, scrisă pentru aiqbench. Nu sunt cuvintele publicației.

VectraYX-Vision-1B: A Sub-2B Spanish/LATAM Cybersecurity Vision-Language Model with Structured Visual Reasoning and Native Tool Use

arXiv9 septembrie 20261 min de cititArticolul și perspectiva sunt în engleză.

arXiv:2608.08477v2 Announce Type: replace Abstract: 21 pages, 1 figure, 9 tables. v2: retracted the B6 tool-id score of 0.08 (v1) after finding 3 benchmark harness bugs; under the fixed harness every B6/B7 metric is 0.0. Fixed a LoRA LR bug and a checkpoint-load-order bug. Added a 9-field gate + linear probe. Reframed the NoPE ablation as open, confound quantified. Code/checkpoints on HF.

Abstract:21 pages, 1 figure, 9 tables. v2: retracted the B6 tool-id score of 0.08 (v1) after finding 3 benchmark harness bugs; under the fixed harness every B6/B7 metric is 0.0. Fixed a LoRA LR bug and a checkpoint-load-order bug. Added a 9-field gate + linear probe. Reframed the NoPE ablation as open, confound quantified. Code/checkpoints on HF.

20 pages, 1 figure, 9 tables. v2: retracted the B6 tool-id score of 0.08 (v1) after finding 3 benchmark harness bugs; under the fixed harness every B6/B7 metric is 0.0. Fixed a LoRA LR bug and a checkpoint-load-order bug. Added a 9-field gate + linear probe. Reframed the NoPE ablation as open, confound quantified. Code/checkpoints on HF

Computation and Language (cs.CL)

arXiv:2608.08477 [cs.CL]

(or arXiv:2608.08477v2 [cs.CL] for this version)

Text integral sub licența deschisă a publicației · 126 cuvinte

Citește pe arXiv ↗Întreabă despre acest articol