8 papers
Shieldstral
Antonia Calvi, Avinash Sooriyarachchi, Giada Pistilli +273
We introduce Shieldstral, a 3B-parameter policy-adaptive multimodal safety classifier that matches or outperforms models nearly 7 its size on text safety benchmarks and set…
PPU-Bench:Real World Benchmark for Personalized Partial Unlearning in Vision Language Models
Jiahui Guang, Zexun Zhan, Zhenlin Xu +5
Multimodal Large Language Models (MLLMs) may memorize sensitive cross-modal information during pretraining. However, existing MLLM unlearning benchmarks rely on synthetic knowledge…
From Storage to Steering: Memory Control Flow Attacks on LLM Agents
Zhenlin Xu, Xiaogang Zhu, Yu Yao +2
Modern agentic systems allow Large Language Model (LLM) agents to tackle complex tasks through extensive tool usage, forming structured control flows of tool selection and executio…
The Appeal and Reality of Recycling LoRAs with Adaptive Merging
Haokun Liu, Gyung Hyun Je, Marco Ciccone +3
The widespread availability of fine-tuned LoRA modules for open pre-trained models has led to an interest in methods that can adaptively merge LoRAs to improve performance. These m…
Voxtral Realtime
Mistral-AI, :, Alexander H. Liu +166
We introduce Voxtral Realtime, a natively streaming automatic speech recognition model that matches offline transcription quality at sub-second latency. Unlike approaches that adap…
Spherical Leech Quantization for Visual Tokenization and Generation
Yue Zhao, Hanwen Jiang, Zhenlin Xu +3
Non-parametric quantization has received much attention due to its efficiency on parameters and scalability to a large codebook. In this paper, we present a unified formulation of…
The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text
Nikhil Kandpal, Brian Lester, Colin Raffel +24
Large language models (LLMs) are typically trained on enormous quantities of unlicensed text, a practice that has led to scrutiny due to possible intellectual property infringement…
Self-Supervised Multi-Object Tracking with Path Consistency
Zijia Lu, Bing Shuai, Yanbei Chen +2
In this paper, we propose a novel concept of path consistency to learn robust object matching without using manual object identity supervision. Our key idea is that, to track a obj…