探索测试时训练在近似采样中的强大作用，为生成式AI推理难题提供新思路。
arXiv:2606.11437v1 Announce Type: cross Abstract: Efficiently sampling from a complex probability distribution is a fundamental problem which has beco…
无需训练骨干网络，用表示工程让LLM代理稳定应对工具调用变化，突破传统微调瓶颈。
arXiv:2602.04935v3 Announce Type: replace-cross Abstract: Adapting LLM agents to domain-specific tool calling remains notably brittle under evolving i…
揭秘新型注意力机制：动态线性注意力，同时提升效率与精度，ICML 2026录用论文。
arXiv:2606.10650v1 Announce Type: cross Abstract: The scalability of Large Language Models (LLMs) to long contexts is fundamentally constrained by the…
LLM极低比特量化新突破，线性约束向量量化实现2比特数据高效训练，已被ICML 2026收录。
arXiv:2606.10531v1 Announce Type: cross Abstract: Quantization-aware training (QAT) is essential for extremely low-bit large language models (LLMs). C…
稀疏MoE大模型部署新突破：引入联盟感知策略的专家剪枝方法
arXiv:2606.09886v1 Announce Type: cross Abstract: Sparse Mixture-of-Experts (MoE) large language models achieve strong quality with low per-token comp…
探索用LoRA和NEFTune方法高效微调DeepSeek-R1-8B，降低资源消耗同时提升性能。
arXiv:2606.10392v1 Announce Type: new Abstract: Financial named-entity recognition (NER) is essential for translating unstructured financial reports a…
新论文提出WebChallenger，一个可靠高效的通用Web智能体，专为复杂网页任务设计。
arXiv:2606.10423v1 Announce Type: new Abstract: Autonomous web navigation remains challenging for LLM agents, and the strongest generalist systems rel…
ICML 2026 Oral论文，提出通过扩展正交变换实现大模型训练的内存高效方案。
arXiv:2603.05500v2 Announce Type: replace Abstract: Efficient and stable training of large language models (LLMs) remains a core challenge in modern m…
当LLM强化学习遭遇黑盒差异，这篇论文提出重构框架实现更高效训练。
arXiv:2606.08779v1 Announce Type: new Abstract: Reinforcement Learning (RL) has emerged as a pivotal post-training paradigm, yet it frequently suffers…
提出样本高效后训练方法，突破乐高空间物理推理任务的数据瓶颈
arXiv:2606.07602v1 Announce Type: new Abstract: LLM-based LEGO assembly generation requires both semantic grounding and physical feasibility. We ident…
大疆首款140W氮化镓快充，三口输出支持PD3.1，到手价仅209元
IT之家 6 月 8 日消息，大疆今天发布品牌首款高效能氮化镓快充充电器 POWER 140W，新品主打三口输出，支持 PD 3.1 协议，自带 7A 数显数据线， 预售到手价 209 元 。 据介绍，这款充电器的单口最高输出功率可达 140W，能够为手机、平板和笔记本快速补能。具备两个 USB-C…
NVIDIA推出的高性能推理加速器，专为长时间运行AI代理设计，速度更快、效率更高。
Powers faster, efficient reasoning for long-running agents Discussion | Link
边缘设备高效推理新突破，兼顾性能与资源约束，适合部署场景研究者
arXiv:2603.16867v2 Announce Type: replace-cross Abstract: Large language models (LLMs) with chain-of-thought reasoning achieve state-of-the-art perfor…
每月3美元一键部署，30秒获取HTTPS实例，2026年自托管Memos的最省钱方案。
Last updated: June 2026 Memos is the lightweight, open-source notes app a lot of people land on after getting tired of flomo, Google Keep, or a paid N…
提出通过稀疏性演化进行稀疏微调，高效修复稀疏大语言模型，平衡性能与计算开销。
arXiv:2505.24037v3 Announce Type: replace Abstract: Sparse large language models (LLMs) offer an attractive direction toward efficient deployment, but…
微软Edge内嵌Aion-1.0-Instruct小模型与翻译API，支持145种语言，端侧运行无需云服务，赋能开发者打造AI原生Web体验
IT之家 6 月 3 日消息，在今日开幕的 Build 2026 开发者大会上，微软宣布在去年为 Edge 浏览器推出基于 Phi-4-mini 模型的写作辅助 API 基础上扩展了其端侧 AI 能力，新增了模型和 API。本次更新主要包括三项内容： Aion-1.0-Instruct 小语言模型的…
异构边缘/雾环境中大模型高效服务的新方案，解决部署延迟与资源优化难题。
arXiv:2606.03770v1 Announce Type: cross Abstract: Large Language Models (LLMs) have become integral to modern applications, yet their deployment remai…
高效管理Agentic RL后训练资源的新方案Libra，降低训练成本、提升性能。
arXiv:2606.03077v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become a standard post-training paradigm for large language models (…
提出多层潜在原型方法，高效提升LLM内容审核的准确性与速度
arXiv:2502.16174v4 Announce Type: replace Abstract: Although modern LLMs are aligned with human values during post-training, robust moderation remains…
提出可学习的零阶优化器，无需梯度即可高效微调大模型，大幅降低内存开销。
arXiv:2510.00419v2 Announce Type: replace Abstract: Zeroth-order optimizers have recently emerged as an attractive approach for fine-tuning large lang…