Warning: Undefined array key "ruVppY" in /www/wwwroot/www.myshirtai.com/wp-includes/style-engine/class-wp-style-engine-processor.php on line 1
开源项目归档 - 渗透智能

open source project

Open source projects are community-driven programs designed to drive technological innovation and knowledge dissemination through collaboration and sharing.

NextStep-1:自回归图像生成的”终极形态”,14B参数模型开源了!

阶跃星辰(StepFun)团队开源了NextStep-1,一款14B参数的纯自回归图像生成模型。该模型直接在连续视觉空间生成图像,无需依赖扩散模型或离散化处理,由14B参数Transformer骨干和157M参数流匹配头组成。它支持高保真文生图及精准图像编辑(如物体增删、背景修改),在GenEval(0.73)、GenAI-Bench等基准测试中表现优异,接近顶尖扩散模型。但存在生成不稳定、解码延迟等挑战,标志着自回归图像生成新阶段。

NextStep-1:自回归图像生成的”终极形态”,14B参数模型开源了! Read More "

浏览器自动化开源项目,让 AI 真正“上网干活”

Nanobrowser是近期在GitHub爆火的开源AI浏览器自动化框架,上线一周获17,000+星标。其核心采用双智能体协作模式:Planner拆解自然语言指令为操作步骤,Navigator在真实网页中执行、读取等操作。该项目支持本地运行及多模型接入,可实现论文抓取、比价、舆情监控等网页自动化任务,典型案例显示其2分半完成论文数据抓取,成本仅0.1元。

浏览器自动化开源项目,让 AI 真正“上网干活” Read More "

MiniMax M2: Domestic open-source model kills like crazy! 8% price, hit Claude level performance!

MiniMax发布新一代开源大模型M2,性能跻身全球前五,价格仅为Claude 4.5的8%。该模型总参数230B,激活参数仅10B,推理速度超100 tokens/秒。在编程、Agent工作流和多模态任务上表现优异,打破了AI领域高性能、低价格、高速度的"不可能三角"。

MiniMax M2: Domestic open-source model kills like crazy! 8% price, hit Claude level performance! Read More "

DeepAnalyze: let AI become your exclusive data scientist! Open source projects in depth analysis

DeepAnalyze是由中国人民大学与清华大学团队联合开发的开源代理大型语言模型,首个面向自主数据科学的端到端解决方案。其核心能力包括自动完成数据准备、分析、建模、可视化及报告生成全流程,支持CSV、Excel等多格式数据源,无需人工干预。DeepAnalyze-8B(8B参数)在基准测试中表现超越GPT-4o-mini等商业模型,且模型权重、代码与训练数据完全开源,可部署为专属数据科学助手。

DeepAnalyze: let AI become your exclusive data scientist! Open source projects in depth analysis Read More "

DeepAnalyze: let AI become your exclusive data scientist! Open source projects in depth analysis

DeepAnalyze是由中国人民大学与清华大学团队联合开发的开源代理大型语言模型,首个面向自主数据科学的端到端解决方案。其核心能力包括自动完成数据准备、分析、建模、可视化及报告生成全流程,支持CSV、Excel等多格式数据源,无需人工干预。DeepAnalyze-8B(8B参数)在基准测试中表现超越GPT-4o-mini等商业模型,且模型权重、代码与训练数据完全开源,可部署为专属数据科学助手。

DeepAnalyze: let AI become your exclusive data scientist! Open source projects in depth analysis Read More "

OpenMemory MCP: Breaking the Memory Barrier Between AI Tools

Mem0's OpenMemory MCP is a locally-run "memory backpack" solution designed to solve the problem of contextual information loss between different AI tools. The system allows AI applications such as Claude and Cursor to share memories through a standardized protocol, with all data stored locally on the device to ensure privacy and security. Core features include structured memory organization, user permission control, and cross-platform compatibility, supporting seamless workflows in a variety of scenarios from project collaboration to content creation. The project is currently open-sourced on GitHub, with future plans to add features such as memory expiration and cloud backup.OpenMemory MCP significantly improves the efficiency and experience of collaborating with multiple AI tools by maintaining contextual continuity.

OpenMemory MCP: Breaking the Memory Barrier Between AI Tools Read More "