←搜索结果

搜索“Hugging Face”

按匹配度 · 共 1193 条结果 清除搜索

星期五

  1. vLLM 官方博客

    vLLM 如何集成 Transformers 建模后端

    vLLM 集成 Transformers 建模后端，让兼容模型在沿用 Transformers 实现的同时获得 vLLM 的吞吐量与延迟优化。该后端现已支持视觉语言模型，并可通过 model_impl="transformers"、OpenAI 兼容 API 或 LLM API 部署；对于 vLLM 尚未原生支持的兼容模型，系统也可自动切换到 Transformers 实现。

    ly 2025); Transformers and vLLM: Inference in Action; vLLM’s Deployment Superpower: OpenAI Compatibility; Why do we need the Transformers modeling backend?; Case Study: Helium The Hugging Face Transformers library offers a flexible, unified interface to a vast ecosystem of model architectures. From research to fine-tuning on custom dataset, Transformers is the go-to toolkit for all. But when it comes to deploying these models at scale, inference speed and efficiency often take center stage. Ente

    匹配字段正文

    #engineering#教程/实践#部署/工程#多模态

星期四

  1. Zyphra Research

    Zyphra 推出 ZR1-1.5B：面向数学与代码的小型推理模型

    Zyphra 发布 ZR1-1.5B，一个基于 DeepSeek-R1-Distill-Qwen-1.5B、用强化学习在代码与数学题上训练的小型推理模型。它在 LCB_Generation 上与 Claude3-Opus、Gemma2-27B 持平，在竞赛数学上超过 Qwen2.5-72B，且推理 token 比 R1-Distill-1.5B 少 60%、比 DeepScaleR 少 53.5%。截断到 2048 token 时，其数学评测平均 pass@1 达 40.39%，高于 Qwen-2.5-Math-7B-Instruct 的 37.01%。

    DeepScaleR. Overall ZR1-1.5B demonstrates strong generalization across disparate domains as well as coherent and efficient reasoning traces compared to models of a similar scale. Hugging Face Introduction We introduce ZR1-1.5B, a small reasoning model trained extensively on both verified coding and mathematics problems with reinforcement learning (RL). ZR1-1.5B outperforms many significantly larger general non-reasoning models on code generation, while maintaining performance close to state-of-

    匹配字段正文

    #lab#模型发布#推理#编码

星期五

  1. Prime Intellect

    Prime Intellect 融资 1500 万美元，打造开放超智能栈

    Prime Intellect 宣布完成 1500 万美元融资，由 Founders Fund 领投，Menlo Ventures、Andrej Karpathy、Clem Delangue、Dylan Patel、Tri Dao、Emad Mostaque 等参投，累计融资超过 2000 万美元。公司称其 Compute Exchange 已连接开发者、创业公司、云厂商和数据中心，并用 PRIME 框架跨三大洲训练了 10B 参数模型，还发布了 INTELLECT-1、METAGENE-1、INTELLECT-MATH 以及基于 DeepSeek-R1 的合成推理数据集等成果。下一步将扩大算力接入、推出托管评测与专用推理，并继续建设 RL 基础设施。

    $15M to Build The Open Superintelligence Stack

    Today, we're announcing that we've raised $15M in funding, led by Founders Fund, with participation from Menlo Ventures, Andrej Karpathy (EurekaAI, Tesla, OpenAI), Clem Delangue (Hugging Face), Dylan Patel (SemiAnalysis), Tri Dao (Together.AI), Emad Mostaque (Stability AI), Jake Medwell (8VC Co-Founder) and many others. This brings our total funding to over $20M, accelerating our work on the open superintelligence stack. Scaling Open + Democratized AI Prime Intellect is already operating at sca

    匹配字段正文

    #company#行业动态#开源生态#数据/训练

星期四

  1. Mistral AI精选

    Mistral AI 发布 24B 参数模型 Mistral Small 3

    Mistral AI 发布延迟优化的 24B 参数模型 Mistral Small 3，并以 Apache 2.0 许可开放预训练和指令微调检查点。官方称其 MMLU 准确率超过 81%、推理速度达 150 tokens/s，与 Llama 3.3 70B instruct 表现相当且在同一硬件上快 3x 以上。量化后可在单张 RTX 4090 或配备 32GB RAM 的 Macbook 上本地运行，并已通过 la Plateforme 及合作平台提供。

    Up and to the Left

    ilable on la Plateforme as mistral-small-latest or mistral-small-2501. Explore our docs to learn how to use our models for text generation. We are also excited to collaborate with Hugging Face, Ollama, Kaggle, Together AI, and Fireworks AI to make the model available on their platforms starting today: The road ahead It’s been exciting days for the open-source community! Mistral Small 3 complements large open-source reasoning models like the recent releases of DeepSeek, and can serve as a strong

    匹配字段正文

    #lab#模型发布#部署/工程#开源生态

星期一

  1. vLLM 官方博客

    Llama Stack 引入 vLLM 推理 Provider

    Llama Stack 引入两种 vLLM 推理 Provider，分别通过 OpenAI 兼容服务器远程调用，以及与 Llama Stack 服务器并行运行。官方以 Llama-3.2-1B-Instruct 演示本地容器和 Kubernetes 部署，并通过 Llama Stack Client 测试推理请求。

    hat runs alongside with Llama Stack server In this article, we will demonstrate the functionality through the remote vLLM inference provider. Prerequisites Linux operating system; Hugging Face CLI if you'd like to download the model via CLI; OCI-compliant container technologies like Podman or Docker (can be specified via the CONTAINER BINARY environment variable when running llama stack CLI commands); Kind for Kubernetes deployment; Conda for managing Python environment Start vLLM Server We firs

    匹配字段正文

    #engineering#教程/实践#部署/工程#开源生态

星期一

  1. Prime Intellect

    Prime Intellect 等发布 7B 宏基因组基础模型 METAGENE-1

    USC、Prime Intellect 与 Nucleic Acid Observatory 合作发布 7B 参数自回归 Transformer 宏基因组基础模型 METAGENE-1，在超过 1.5 万亿碱基对的污水宏基因组 DNA 与 RNA 序列上预训练，用于疫情监测、病原体检测和新兴健康威胁的早期发现。模型采用针对宏基因组序列的 BPE 分词，在病原体检测和宏基因组嵌入基准上取得 SOTA，并演示了病原体与异常检测场景。团队称其 512 token 上下文长度限制了复杂序列设计任务，当前版本风险较低，但未来更大模型开源前需严格安全评估。

    METAGENE-1: Metagenomic Foundation Model

    undation model designed for pandemic monitoring, trained on over 1.5T base pairs of DNA and RNA sequenced from wastewater. Paper: metagene-1-paper; Github: github.com/metagene-ai; Hugging Face: huggingface.co/metagene-ai A collaboration between researchers at USC, Prime Intellect, and the Nucleic Acid Observatory. Overview We pretrain METAGENE-1, a 7-billion-parameter autoregressive transformer language model, which we refer to as a metagenomic foundation model, on a novel corpus of diverse meta

    匹配字段正文

    #company#模型发布#数据/训练#开源生态

星期二

  1. Zyphra Research

    Zyphra 发布 5 万亿 token 数据集 Zyda-2，用 NVIDIA NeMo Curator 将处理时间从 3 周缩短到 2 天

    Zyphra 发布 Zyda-2，一个约 5 万亿 token 的宽松许可数据集，由 DCLM、FineWeb-Edu、Zyda-1 和 Dolma v1.7 的 Common Crawl 部分经跨数据集去重与模型过滤组合而成，已在 HuggingFace 上开放。借助 NVIDIA NeMo Curator 的 GPU 加速，数据处理从 CPU 上的 3 周缩短到 2 天，总拥有成本降低 2 倍。Zyphra 称用 Zyda-2 退火训练的模型在 MMLU、Hellaswag 等聚合评测上优于 DCLM 和 FineWeb-Edu，其 Zamba2 系列模型即基于该数据集的早期版本训练。

    In conclusion, NeMo Curator is a powerful tool that allows us to not only to scale our data processing capabilities, but optimize costs. Download the Zyda-2 dataset directly from Hugging Face and train higher-accuracy models. For more information see our GitHub tutorial. Appendix Analysis of Global Duplicates Figure 7: Distribution of cluster sizes of duplicates in global dataset (log-log scale). Figure 8: Distribution of cluster sizes of duplicates in DCLM (log-log scale). Figure 9: Distributi

    匹配字段正文

    #lab#开源/仓库#数据/训练#部署/工程

星期五

  1. Prime Intellect精选

    Prime Intellect 启动 INTELLECT-1：首个全球分布式训练的 10B 参数模型

    Prime Intellect 发布 INTELLECT-1，称其为首个在全球分布式节点上训练的 10B 参数模型，并开放算力贡献入口。该模型基于 Llama-3 架构，训练数据混合 55% Fineweb-edu、20% DLCM、20% Stack v2 和 5% OpenWebMath，经 Llama-3 tokenizer 处理后超过 6 万亿 token。配套发布的分布式训练框架 Prime 支持容错训练、节点动态上下线和 int8 伪梯度通信，官方称生产训练中每 100 步同步一次，通信仅占总训练时间的 1-2%。

    INTELLECT–1: Launching the First Globally-Distributed Training of a 10B Parameter Model

    ep to ensure AGI will be open-source, transparent, and accessible. Launch Partners and Contributors We are excited and grateful to be joined by leading open-source AI players like Hugging Face, SemiAnalysis, Arcee, Hyperbolic, Olas, Akash, Schelling AI and many others contributing compute to this training run. Anyone can now contribute resources to advance open-source AI through our platform and later on also more easily with their own hardware. Paradigm Shift for Distributed Training. As Jack C

    匹配字段正文

    #company#模型发布#数据/训练#开源生态

星期一

  1. Zyphra Research

    Zyphra 训练手册：如何构建 SSM、Transformer 与 MoE 混合模型

    Zyphra 发布《Zyphra Training Cookbook》，分享构建混合（SSM、Transformer、MoE）模型的工程经验，供其他技术团队参考。Zyphra 已构建 BlackMamba、Zamba-7B、Zamba2-2.7B、Zamba2-1.2B 等混合模型，以及 Zyda 数据集和 Tree Attention 工程优化。手册聚焦 1B-15B 端侧小模型，建议每 1/4 到 1/6 的序列混合层使用全注意力，其余采用 Mamba 等线性混合器以降低 FLOP 和内存开销。

    e recommend using the EleutherAI cookbook by Quentin Anthony, Hailey Schoelkopf, and Stella Biderman. Token Calculation We provide a script at here that tokenizes text data from a Hugging Face dataset, calculates the total number of tokens, and optionally saves the tokenized dataset. Benchmarks Block Benchmarks and Sizing We provide computation benchmarks for hybrid model blocks such as attention, Mamba1, and Mamba2 here. These are useful for comparing hardware performance and for efficiently si

    匹配字段正文

    #lab#教程/实践#开源生态#数据/训练

星期五

  1. Ollama Blog

    如何在编辑器内用 Continue 和 Ollama 搭建全开源 AI 代码助手

    Continue 联合创始人 Ty Dunn 介绍了在 VS Code 和 JetBrains 中结合 Ollama 与开源 LLM 搭建本地代码助手的方法。推荐用 Mistral AI 的 Codestral 22B 做自动补全和对话，或用 DeepSeek Coder 6.7B 补全加 Llama 3 8B 对话；配合 nomic-embed-text 嵌入向量驱动 @codebase，并可用团队接受的补全建议微调 StarCoder 2 后推送到 Ollama 模型库。

    Continue and Ollama

    use accepted autocomplete suggestions from your team to fine-tune a model like StarCoder 2 to give you better suggestions. a. Extract and load the “accepted tab suggestions” into Hugging Face Datasets b. Use Hugging Face Supervised Fine-tuning Trainer to fine-tune StarCoder 2 c. Push the model to the Ollama model library for your team to use and measure how your acceptance rate changes Learn more about Ollama by using @docs to ask questions with the help of Continue Continue also comes with an

    匹配字段正文

    #company#教程/实践#编码#部署/工程

星期日

  1. The Gradient

    对齐的虚假性

    The Gradient 刊文《The Artificiality of Alignment》指出，当前所谓“对齐”研究实际在解决的是打造用户愿意付费的产品，而非人类灭绝这一宏大挑战。文章认为 OpenAI 将构建超级智能列为主要目标，Anthropic 也设有产品页面，两家公司的营收目标始终在治理、产品与技术决策中起作用。

    样。但如果博斯特罗姆是对的——对齐是一个技术问题——那么技术挑战究竟是什么？ 我首先应当说明，AI和对齐的思想版图是多元的。许多关注存在风险的人对OpenAI和Anthropic所采取的方法有强烈批评，事实上也对其产品导向提出了类似担忧。尽管如此，聚焦于这些公司在做什么既是必要的也是充分的：它们目前拥有最强大的模型，而且与另外两家大型模型供应商Mosaic或Hugging Face不同，它们在公开传播中最为认真地对待对齐和“超级智能”。 这一版图的一个重要组成部分，是一个由存在风险所驱动的、紧密团结的个体研究者社群。这个社群围绕AI安全和对齐理论发展出了丰富的词汇，其中许多最初是以详细博客文章的形式发表在LessWrong和AI Alignment Forum等论坛上。 其中一个有助于为技术对齐工作提供语境的想法——或许也是博斯特罗姆所指的更为正式的版本——是意图对齐的概念。在2018年一篇引入该术语的Medium文章中，曾任OpenAI对齐团队负责人的保罗·克里斯蒂亚诺将意图对齐定义为“AI（A）正试图做人类（H）希望它做的事”。以这种方式加以界定后，“对齐问题”突然变得容易处理得多

    匹配字段正文

    #ai-vertical#现象/趋势#安全/对齐

星期四

  1. a16z：News

    a16z 发布 AI Canon：现代 AI 必读资源精选清单

    a16z 发布“AI Canon”精选资源清单，汇集过去几年对 AI 领域影响最大的论文、博客、课程与指南。清单从 Transformer 与潜在扩散模型入门讲起，延伸到 LLM 技术学习、实践指南与 AI 市场分析，并收录以 2017 年 Google 论文《Attention is All You Need》为起点的里程碑研究。

    for the full stack and how the pieces fit together Courses LLM Bootcamp: A practical course for building LLM-based applications with Charles Frye, Sergey Karayev, and Josh Tobin; Hugging Face Transformers: Guide to using open-source LLMs in the Hugging Face transformers library LLM benchmarks Chatbot Arena: An Elo-style ranking system of popular LLMs, led by a team at UC Berkeley. Users can also participate by comparing models head to head; Open LLM Leaderboard: A ranking by Hugging Face, compa

    匹配字段正文

    #analysis#其他

星期六

  1. Lilian Weng：Lil'Log

    可控神经文本生成：Lil'Log 综述解码策略、提示设计与微调

    Lil'Log 发布《Controllable Neural Text Generation》综述，梳理用无条件下语言模型实现可控内容生成的三类方法：测试时引导解码与重采样、通过提示设计优化输出、以及微调基座模型或可操控层。文中介绍了 AutoPrompt 的梯度搜索、Prefix-Tuning 与 P-tuning 等提示设计工作，并补充了 unlikelihood training。

    e tokens compared to standard MLE training. Cited as: Or [1] Patrick von Platen. “How to generate text: using different decoding methods for language generation with Transformers” Hugging face blog, March 18, 2020. [2] Angela Fan, et al. “Hierarchical Neural Story Generation/” arXiv preprint arXiv:1805.04833 (2018). [3] Ari Holtzman et al. “The Curious Case of Neural Text Degeneration.” ICLR 2020. [4] Marjan Ghazvininejad et al. “Hafez: an interactive poetry generation system.” ACL 2017. [5] Ari

    匹配字段正文

    #analysis#教程/实践#数据/训练