我的收藏提交工具
FOXAI SEARCH

全站搜索

新闻报道 220

training-alignment

Adaptive Scheduling Pipeline for Multi-Instance Asynchronous Reinforcement Learning

arXiv:2611.08864v1 Announce Type: new Abstract: We present the Adaptive Scheduling Pipeline (ASP), an offline rollout scheduler for multi-instance asynchronous reinforcement learning (RL) of large language models (LLMs). ASP jointly reduces three sources of rollout underutilization - padding bubbles from intra-minibatch length heterogeneity, skewness bubbles from long-tail outputs, and sorting bubbles from fluctuations in training-queue occupancy induced by length-sorted dispatch - through a single offline preprocessing pass. This enables efficient multi-instance rollout that overcomes the early scaling plateau of single-instance inference while smoothing the arrival of generated samples at the trainer. We integrate the scheduler into an asynchronous architecture that decouples rollout from training while keeping policy lag observable and bounded. Complete ASP improves end-to-end throughput by up to 3.22 times and reduces iteration time by up to 76.8% relative to MindSpeed-RL, while reducing iteration time by up to 62.5% relative to verl. Within ASP, scheduling reduces mean iteration time by 10.9%. ASP also scales more robustly with additional generation resources than these baselines and shows no visible convergence degradation in the measured bounded-lag regime. ASP has since been used in an internal telecom-domain RL post-training workflow, reducing the reported training period from more than three months to about two months and producing a domain-enhanced model for knowledge answering and tool calling.

翻译服务暂未运行，当前显示原文。
arXiv cs.DC
ai-safety

AttentionSwitch: Jailbreak Defense for LLMs via Joint Learning of Sparse Attention-Head Selection and Scaling

arXiv:2611.08752v1 Announce Type: new Abstract: Large language models have achieved remarkable performance across a wide range of tasks, yet they remain susceptible to jailbreak attacks that circumvent safety mechanisms and elicit harmful or unintended outputs. Although safety can be improved through model fine-tuning, such approaches can be computationally expensive and may degrade previously acquired capabilities through catastrophic forgetting. Recent jailbreak defenses have therefore explored targeted interventions on specific model components associated with unsafe behavior. However, existing approaches typically decouple two inherently related decisions: where to intervene and how to modify the selected components. Such sequential optimization may be suboptimal because the optimal intervention location can depend on the intervention applied there, and vice versa. In this work, we introduce AttentionSwitch, the first jailbreak defense to jointly optimize intervention sites and strengths within a frozen LLM. Specifically, AttentionSwitch learns a sparse set of attention heads together with query-conditioned multiplicative scaling factors, allowing the outputs of the selected heads to be either attenuated or amplified without altering the directions of their representations. Across popular language models and jailbreak attacks, AttentionSwitch outperforms existing defenses, with particularly strong results when training data is scarce.

翻译服务暂未运行，当前显示原文。
arXiv cs.AI
coding-engineering

Claude Fable 5.2 now available on AI Gateway

Claude Fable 5.2, designed for open-ended technical tasks, is now available on AI Gateway with improved coding, reasoning, and vision capabilities. It operates as an orchestrator or advisor, supports model fallbacks for cybersecurity classifier rejections, requires data retention (no Zero Data Retention), and can be accessed via the AI SDK, APIs, or integrated tools like Claude Code and Cursor. AI Gateway offers unified API access, cost tracking, failover options, and pricing transparency without markup.

翻译服务暂未运行，当前显示原文。
Vercel
datasets

Compression of Emotion in Language Model Representations

arXiv:2611.08683v1 Announce Type: new Abstract: Large language models (LLMs) have been evaluated on their ability to identify emotions expressed in text and produce fluent emotional language. However, output predictions alone do not reveal how fine-grained emotion distinctions are organized internally. We study seven LLMs on GoEmotions, an English Reddit dataset annotated with 27 fine-grained emotion categories plus neutral. We compare model predictions and emotion-prompted final-token representations with the 28 GoEmotions categories and with coarser emotion-family and sentiment groupings. We find a consistent pattern of $\textit{overcompression}$ relative to the fine-grained GoEmotions taxonomy. Under the tested KMeans protocol, when we cluster their hidden representations without using emotion labels, the clearest split is into two groups that largely separate positive from negative emotion. Coarser emotion groupings are also recovered with fewer clusters than the full 28-category taxonomy. Yet linear probes retain substantial fine-grained predictive information. These results suggest that fine-grained emotion information remains accessible even though broader affective structure more strongly organizes the unsupervised geometry.

翻译服务暂未运行，当前显示原文。
arXiv cs.CL
ma-ipo

IWIBench: Evaluating Gradient-Free Model Driven Intra-weight Injection

arXiv:2611.08647v1 Announce Type: new Abstract: Intra-weight injection lets one model change another model's behaviour by editing its weights directly, without gradients. Such edits can range from altering lexical associations to enforcing output formats or updating factual knowledge. Despite growing interest in these approaches, there has been no systematic way to evaluate how well models perform them. We introduce IWIBench, which measures a large editor model's ability to modify the weights of a smaller target model (typically Gemma 3 4B IT) to achieve one of ten classical text-editing objectives while preserving its other capabilities. Editors are limited to four tools for reading and writing weights, with bounded tensor access and a fixed budget for generating and inspecting rollouts. A validator applies each proposed patch in a clean environment and compares the edited model against the unmodified target. Evaluating frontier closed- and open-weight models reveals sharply diverging capabilities, both across models and across tasks. A small number of models, most notably Astra and GPT-6.1 Sol, perform far ahead of the rest. The recency of this ability (little of it is evident in GPT-5.6 Sol, for instance), together with qualitative analysis, suggests it has been acquired through recent training. Tasks also vary widely: structured response resisted every attempt, and answer-length control yielded only one successful solution, whereas alias acquisition and recoding proved most amenable. Success in one task family correlated poorly with success in others, indicating an early and uneven capability frontier. We discuss strategies that editors repeatedly adopted and conclude with an outlook on intra-weight injection as an emerging but already potent approach to gradient-free behaviour modification.

翻译服务暂未运行，当前显示原文。
arXiv cs.AI
open-models

Nemotron-Personas-Finance-Japan: A Synthetic Dataset Capturing Japan's Demographics and Financial Context

NVIDIA and Elith have released Nemotron-Personas-Finance-Japan, a large-scale open-source synthetic dataset featuring 9 million personas based on Japan’s official demographic, labor, and financial statistics. Built using NVIDIA’s NeMo Data Designer and SDG-PGMs framework, the dataset includes 50 fields per record with natural Japanese text, covering demographics, household finances, occupations, financial literacy, and personality traits. Designed for commercial and non-commercial use under CC BY 4.0, it enables privacy-safe development of financial AI applications such as conversational agents, customer simulations, and bias testing, while aligning with Japan’s cultural and regulatory context.

翻译服务暂未运行，当前显示原文。
Hugging Face Blog