精选

如何利用 NVIDIA Warp 与 MjWarp 加速机器人仿真及学习工作流

该技术文章聚焦于如何结合使用 NVIDIA Warp 与 MjWarp 框架，以加速机器人领域的物理仿真与强化学习等工作流。由于提供的原始素材缺少详细摘要与正文内容，具体的技术架构实现、性能评测对比及实操步骤等细节尚无法进一步展开。整体内容旨在探讨通过高性能 GPU 仿真工具提升机器人开发与训练效率的方法。

英国AI安全研究所与EvalEval推动基准评测结果可复现

该内容探讨了英国人工智能安全研究所（UK AISI）与EvalEval如何致力于提升AI模型基准测试结果的可复现性。由于输入信息仅包含标题，具体的技术方案与评测框架细节尚未披露。该工作旨在应对大模型评估中常见的结果波动与透明度不足问题，推动AI评估与安全基准测试走向标准化与可靠化。

Transformers 现已支持直接运行 llama.cpp 量化模型

Hugging Face 旗下核心框架 Transformers 现已支持直接加载和运行 llama.cpp 的量化格式模型。该功能进一步融合了主流开源推理生态，允许开发者在 Transformers 生态内直接使用经过 llama.cpp 极限量化的模型权重，大幅降低本地推理的显存占用并提升部署灵活性。由于输入信息较少，详细支持范围与参数请参阅官方更新说明。

像物理学家一样剪枝大模型：将层块移除转化为伊辛优化问题

该研究提出了一种从物理学视角出发的大语言模型剪枝新方法，将大模型中的层块移除（Block Removal）形式化建模为伊辛优化问题（Ising Optimization Problem），旨在通过统计物理中的经典优化模型探索更高效、系统化的模型压缩与结构化剪枝方案。注：原始素材仅提供标题，详细算法实现与实验结果待论文正文披露。

Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCL

Rebuilding AUTOMATIC1111 with Gradio Workflow

NeoMME: an efficient Multimodal-native and Multilingual Encoder

Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps

Give Your Coding Agents a Memory You Own

Training a coding model to paint watercolours with TRL and OpenEnv

BenchMIRT: What are LLM benchmarks actually measuring?

Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI

The Open ASR Leaderboard Adds Its First Global South Language

Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers

Granite 4.2 LLMs: How They're Built

Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original

Wire It, Run It, Deploy It: AI Workflows in Gradio

How Hugging Face Inference Endpoints, Jobs, and Buckets Power Search on Papers with Code

Measuring benchmark optimization in speech recognition

Up to 3.2x Faster Inference with LFM2.5-DSpark

How Much Memory Does Your Agent Actually Need?

Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers

Same Cluster, 33 Points More Utilization: What Changed Was the Order

State of Open Models: Summer 2026 Observations

Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage Buckets

What We Learned by Reproducing 2,200 papers from ICML

Introducing OlmoEarth embeddings: Custom embedding exports from OlmoEarth Studio for downstream analysis

Thinking of ACE? We Can Do It with Fewer Tokens

Build Low-Latency Multilingual Voice Agents: Open Weights & Full Deployment Control with NVIDIA Magpie TTS

Making Knowledge Distillation Cheap Enough to Run at Scale