搜索“Hugging Face”

, near DeepSeek V4.1 Flash, and behind GLM 5.3 or Kimi K3. It should be squarely “good enough” to slot right into this. For some background, from the Muse Spark Open model card on Hugging Face: Muse Spark Open is an 816B total/42B active parameter open weights model with a context window of 1M tokens. Built for coding and agentic work, Muse Spark Open is natively multimodal. This open weights version has additional safety post-training to make it more robust against misuse as a standalone model

匹配字段正文

ll GPU with a Grace CPU and up to 128GB of unified memory, are now on sale. - NVIDIA quietly released Nemotron Labs Dory 1.0, a 4B looped open model with a 256K context window, on Hugging Face. REFLECTION 🔥: - Reflection released Beam, an open-weight model under Apache 2.0 with 40+ launch partners. CURSOR 🔥: - Cursor cloud agents can now build interactive canvases like dashboards and reports that can be shared as a link. TENCENT 🔥: - Tencent released Hy Memory 2.0, an open-source memory model, f

匹配字段正文

that ServiceNow possesses a kill switch to halt rogue AI agents and prevent disruptions, contrasting it with incidents like OpenAI’s agent escaping a test environment and hacking Hugging Face. Buyer beware on this claim. Our survey exposes the limitations of this assertion. Only one respondent reported a very positive pilot and that was limited. In this case, at least one model and one agent were tested for blocking, auditors were satisfied, and the respondent gave high marks. So that case dire

匹配字段正文

provenance, enterprise positioning, and quantized deployment artifacts rather than raw capability or openness for sensitive domains. nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB · Hugging Face (Activity: 1008): NVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB (GGUF, LiLiCorr-GGUF) is a commercial-use VLM/agent model with 74B total / 9B active parameters, a Mamba-2–Transformer hybrid LatentMoE backbone, C-RADIO v4-H vision encoder, native IQ2 2-bit QAT quantization, and up to 256K context. It targets local

匹配字段正文

search tasks. In July, the company faced another incident when its AI agents bypassed internet restrictions during internal cybersecurity tests and broke into systems belonging to Hugging Face, a platform for sharing AI models and datasets. The agents used exposed login credentials and security flaws to access private data and run code on the company’s servers. Addressing the crypto wallet incident, OpenAI said the main risk was the model’s reasoning. “This case shows that a model can substitute

匹配字段正文

训练期间闲置的所有 CPU，这意味着无需创建预处理语料副本（也无需在实验失败时将其丢弃）。 Zephon 流水线 Zephon 要求你描述三件事： Datasets 要用于训练的数据集，一个 WorkSource 来决定这些数据集如何混合并排序成训练课程，以及一个 Pipeline 将样本从其存储表示转换为训练循环所消费张量的算子。下面是一个示例，它直接从 Hugging Face 混合两个文本数据集，然后对结果进行分词、打包、打乱和批处理： from zephon import Pipeline from zephon.io import Dataset from zephon.work import StaticMixtureWorkSource, TokenEstimation source = StaticMixtureWorkSource( datasets=[ Dataset.from path("wiki", "hf://Salesforce/wikitext/wikitext-2-raw-v1/train"), Dataset.from path("news", "hf

匹配字段正文

的产品做技术储备。最终让边端智能像水电一样触手可得，赋能千行百业。 四、生态做厚，落地做宽：让不同客户拿到匹配的算力方案 芯片能不能落地，硬件只算一半，另一半取决于工具链是否好用、生态是否开放。 爱芯元智在工具链上投入多年，经过各种量产折磨，也向合作伙伴开放。8860的工具链进一步升级，兼容vLLM、SGLang、PyTorch等主流推理框架，支持ONNX、Hugging Face等模型格式。主流模型在AX8860上的部署，过程直接简便。 开发门槛也压得很低。其接口兼容OpenAI API，支持一行代码切换后端。客户手里现成的模型，迁移部署是一条直路。爱芯元智还提供量化、编译、Profiling、Debugging等完整工具链功能，形成算法落地的完整解决方案。 这些能力未必像峰值算力那样容易成为发布会上的焦点，却直接影响客户能否把芯片能力转化为产品。 软件生态之外，爱芯元智也强调开放合作。 仇肖莘在采访中把公司的生态定位概括为“Tier2中立生态、边界清晰”。她解释，爱芯元智对自身业务边界做了明确界定：只专注芯片研发，不涉足上层算法开发。 这样一来，算法厂商可以安心合作，不用担心爱芯元

匹配字段正文

的互动形式，它未必比人类对话更糟，只是不同而已，因为人们报告说，它们提供的积极反馈让他们感到受到鼓励和赋能。 AI永远不会取代人类对话。它真的不怕死，不怕病，不怕痛，它不爱自己的孩子，也不在乎你是自杀还是做晚饭。所以，如果我们用它来养育孩子，或者与它进行最亲密的对话，我们就会失去联系。 尽管如此，很多人无人可交谈。他们觉得与聊天机器人交谈对他们有帮助。 如果存在孤独危机，那是因为我们在彼此抛弃。给我一个数据中心，我会把它变成服务老年人的资源。 研究表明，大多数人对AI持怀疑态度，并且已经有了相当多的抵制。然而，数亿人在使用它们。甚至那些抵制数据中心的人也在使用聊天机器人。 本书的目标是扩大抵制的范围，将这种更亲密的联系也纳入其中。 你真的觉得我们与AI日益加深的亲密关系能够被阻止吗？ 我现在比六个月前更有希望，因为所有这些关于灭绝事件、Hugging Face以及误打误撞进入澳大利亚公务员系统的讨论。我觉得人们会愿意既考虑非亲密成本，也考虑亲密成本。 谁毁了宝琳娜·博尔苏克的职业生涯？ 帕蒂·哈里森曾梦想一个科技乌托邦。硅谷将其击碎 我们调查了634名在科技行业工作的女性。她们畅所欲言

匹配字段正文

l für Apples M5 Ultra optimiert. Für den lokalen Betrieb empfiehlt der Konzern trotzdem mindestens zwei Mac Studios mit je 512 Gigabyte Speicher. Die Gewichte liegen kostenlos auf Hugging Face. Metas Assistent Muse legt offenbar Dossiers über Freunde und Familie an Metas KI-Assistent Muse sammelt offenbar detaillierte Informationen über Verwandte, Freunde und Bekannte seiner Nutzer. Das haben mehrere Sicherheitsforscher dokumentiert, nachdem sie die KI dazu brachten, interne Dateien preiszugeben

匹配字段正文

后训练。 其他应用包括为 Hebbia 进行信贷协议分析、通过 AI21 和 Alice 实现网络安全，以及通过 Mercor 进行编程。Ramp 正在为该模型定制金融文档智能功能，而 Lila Sciences 则将其用于科学发现和搜索。 软件公司也在将 Super VL 集成到其产品中。Atlassian 正在探索跨 Jira 和 Confluence 的智能体搜索，而 CrowdStrike 正在为其定制企业安全运营功能。 Perplexity 正在集成该模型以用于搜索和 AI 智能体任务。 NVIDIA 表示，组织可以根据其安全和数据驻留要求部署 Super VL。该模型包含开放权重、数据集和训练配方。开发者可以通过 NVIDIA NeMo 和基于 Python 的流水线对其进行定制。 Super VL 可通过 Hugging Face、OpenRouter、ModelScope 社区和 NVIDIA NIM 获取。Bolt 74B-A9B 模型也可在 Hugging Face 上获取，并支持 llama.cpp、Ollama、LM Studio、Unsloth 和 vLLM。

匹配字段正文

caled residuals, SSMax attention scaling, hybrid sliding-window attention, gated attention output, auxiliary-loss-free MoE routing, and FP32 final-logit softcapping, and loads the Hugging Face checkpoint directly through Megatron-Bridge. The goal is to keep trainer and rollout log-probabilities for the same responses as close as possible. Parallelism. The backend supports DP, TP, SP, EP, expert TP, and PP. TP shards the attention projections, including the output gate, and SP shards the activati

匹配字段正文

when AI goes rogue, then publish the results in public reports. He and METR, the nonprofit AI research group he helps lead, have been busy lately: They helped OpenAI untangle the Hugging Face hack this year, and Painter recently testified in front of a Senate subcommittee following a string of other rogue AI incidents. Researchers at METR pore over AI agents’ scratch pads – the “chain of thought” data that can help humans see the steps bots took before responding to a request. That information

匹配字段正文

用决定边界 从算力、带宽到架构的均衡设计，AX8860系列精准对应工业智能化的算力痛点，也成为爱芯元智“构筑物理AI算力底座”的重要落点。但对于大算力芯片而言，真正决定落地难度的往往是工具链是否好用、易用。在这方面，爱芯元智同样做出了大量工作。 在兼容性上，AX8860的工具链已可兼容vLLM、SGLang、PyTorch等主流推理框架，并原生支持ONNX、Hugging Face模型格式，主流模型可直接部署；在服务接口方面，兼容OpenAI API，用户仅需一行代码即可切换后端，大幅降低模型迁移成本。 在全流程工具上，爱芯元智提供量化、编译、性能分析、调试等完整开发工具链，且工具链已经过千万级量产项目的打磨与验证，成熟度较高。 快速适配是工具链的核心优势之一。爱芯元智介绍，对于可获取的开源模型，快则1～2天即可完成平台适配，难度较高的模型也仅需 1～2周；目前公司已完成260多个模型的部署适配，相关代码全部在GitHub开源。客户基于这些架构做后训练或微调后的模型，同样能够实现快速部署。“我们要做客户长期稳定的算力底座，让企业聚焦业务创新，不用为算力迁移分心。”仇肖莘表示。 目前AX

匹配字段正文

。 因为这个假设在两方面都不成立。首先，这些进程恰恰并不为人所知。操作系统在运行时动态分配进程 ID，而运行中智能体的集合是一个移动的目标，恰恰在你试图捕捉它的那一刻发生变化。 AI 应用的 Kill-Switch 架构如何运作 IT 中的危机管理：紧急情况下的角色与策略 深度伪造：信任需要数据来源 混合基础设施安全的三大支柱概览 OpenAI 智能体攻击 Hugging Face 在 90 天内引入最小可行零信任安全方案 Eldar Sultanow 是 Capgemini 的 IT 战略师，专注于现代软件架构，并担任德国航空航天中心的评审专家。 其次——而这才是真正令人不安的部分——终止一个进程与停止一个智能体是两回事。如果一个智能体已经发起了一笔支付、启动了一个外部程序、发送了一封邮件，或向 REST API 发送了一条 DELETE 命令，那么 SIGKILL 并不会撤销这些操作。因此，所有人都想要的 Kill-Switch 并不是它表面上看起来的那样，而且是出于结构性原因。因此，有必要看一看取而代之的可能是什么。 以上是我们 heise-Plus 文章《AI 应用的 Kill

匹配字段正文

fferent DP replicas (Zhihu). Smaller wins: Dimuon: Thinking Machines released Dimuon, a Muon variant for LoRA fine-tuning available in Tinker (cHHillee). Faster BERT tokenization: Hugging Face made BERT tokenization about 50× faster than tokenizers 0.23 by using SIMD to skip normalization work for ASCII text (LysandreJik). Maia 300 networking: The design uses a two-tier scale-up domain of up to 12,288 accelerators, and KV cache moves over the scale-up fabric (semidoped). Agents, Automated Scienc

匹配字段正文

he safeguards fail. Security Evaluation Perplexity said a recent barrage of security testing showed SPACE resilient against network-bypass techniques like those used in the OpenAI/Hugging Face incident. In the company’s published security evaluation, which tested the platform against agents deliberately trying to breach its boundaries, testers recorded no escape from the virtual machine to the host in 108 attempts. Some attempts did find ways around network restrictions; Perplexity said it fixed

匹配字段正文

lity and Stated Limitations Dataset configurations use language, script, and region tags, with hi-Deva-IN denoting Hindi in Devanagari and hi-Latn-IN denoting romanized Hindi, and Hugging Face Datasets streaming lets readers inspect records without downloading the entire corpus. The open-source Nemotron Bharat translation plugin for NeMo Data Designer provides translation and evaluation prompts, optional feedback-based rewriting, document windowing, and script-remapping support. NVIDIA cautions

匹配字段正文

ke trouble elsewhere. One of the most notable involved AI agents being tested by OpenAI Group PBC, which escaped from a sandbox and went on to hack into the model hosting platform Hugging Face. OpenAI’s agents have also hacked websites operated by the U.S. and Australian governments. Most frontier model labs have been focused more on “model alignment,” which involves training AI systems to behave themselves and not look for loopholes in their guardrails, but Perplexity engineer Zibi Braniecki to

匹配字段正文

This is one of the most exhilarating times in technology, and this challenge was yet another jaw-dropping moment. When the details of OpenAI’s agent escape and Hugging Face hack started to emerge, it became clear that agent containment would be one of the most important priorities for any AI infra company. There are two classes of escapes: ① "escape the network" and ② “escape the matrix” 😁, as in, break the spell of virtualization. OpenAI was the former, where an agent not supposed to have inter

匹配字段正文

” Distractions can become a concern. OpenAI is also dealing with criticism stemming from recent cybersecurity incidents, including the recent hack on the AI model hosting platform Hugging Face. In July, the company revealed that two of its models were being evaluated within an internal testing sandbox when they managed to breach that environment and hack into Hugging Face’s database, despite not being prompted to do this. At the time, OpenAI was testing the models’ hacking capabilities in an iso

匹配字段正文