本地大模型机器人玩法汇总站https://spark.tlec.top/zh-cnhttps://spark.tlec.top/plays/agent-content-publishing/https://spark.tlec.top/plays/agent-content-publishing/从链接和部署笔记开始，通过内容接口保存草稿、审核、发布，并自动更新网页和搜索。Sat, 31 Oct 2026 00:00:00 GMTplaysAgent内容管理自动化https://spark.tlec.top/blog/hello/https://spark.tlec.top/blog/hello/本地模型越来越好用，但玩法散落在各个论坛和群里，想找个地方把它们攒起来。Sat, 31 Oct 2026 00:00:00 GMTblog随笔https://spark.tlec.top/news/news-section-launch/https://spark.tlec.top/news/news-section-launch/本栏目由 AI 每天从 Reddit、Hacker News、GitHub Releases 等来源收集本地模型部署相关资讯，筛选、摘要、翻译后经人工审核发布。Sat, 31 Oct 2026 00:00:00 GMTnewshttps://spark.tlec.top/plays/ollama-open-webui/https://spark.tlec.top/plays/ollama-open-webui/十分钟在自己电脑上跑起带网页界面、多用户、可切换模型的聊天助手，局域网里的手机也能用。Fri, 30 Oct 2026 00:00:00 GMTplays入门网页界面https://spark.tlec.top/news/gemma-4-31b-humanlike-chat-released-63cb1ecb/https://spark.tlec.top/news/gemma-4-31b-humanlike-chat-released-63cb1ecb/基于 Google Gemma 4 31B 官方模型微调的 Humanlike-Chat 版本发布，改进了人类语音与聊天风格，支持 GGUF、Safetensors、FP8 及 GPTQ-Int4 本地部署权重。Fri, 30 Oct 2026 00:00:00 GMTnewsGemmaLLMGGUFvLLMFine-tunehttps://spark.tlec.top/news/introducing-deepseek-v41-flash-1017-dynamic-v31-unsloth-gguf-6b1222de/https://spark.tlec.top/news/introducing-deepseek-v41-flash-1017-dynamic-v31-unsloth-gguf-6b1222de/Unsloth 推出 DeepSeek-V4.1-1017 动态 v3.1 预览版 GGUF，在多项基准测试中展现出高精度保留能力，并更新了 Unsloth Desktop 功能。Fri, 30 Oct 2026 00:00:00 GMTnewsUnslothDeepSeekGGUFQuantizationLocal LLMhttps://spark.tlec.top/news/ista-deepseek-v41-flash-1017-gsqrco-quants-comparison-bcca3978/https://spark.tlec.top/news/ista-deepseek-v41-flash-1017-gsqrco-quants-comparison-bcca3978/ISTA 针对与 Unsloth 测试结果的差异做出说明，指出推理预算限制（32K vs 128K）是导致跑分差异的主因。Fri, 30 Oct 2026 00:00:00 GMTnewsDeepSeekQuantizationLLMBenchmarkhttps://spark.tlec.top/news/local-ai-made-it-to-space-gemma-tpu-66b99d03/https://spark.tlec.top/news/local-ai-made-it-to-space-gemma-tpu-66b99d03/Google 确认其 Project Suncatcher 卫星成功在板载 TPU 上运行 Gemma 模型，并向地球传回了生成的首个响应（一首诗）。Fri, 30 Oct 2026 00:00:00 GMTnewsGoogleGemmaEdge AITPUhttps://spark.tlec.top/news/moss-team-ama-local-llama-2b1877ce/https://spark.tlec.top/news/moss-team-ama-local-llama-2b1877ce/复旦大学及上海创新研究院背景的 MOSS 团队在 Reddit 进行 AMA，集中介绍了涵盖图像生成、语音合成、音频与视觉理解的开源模型家族（如 MOSS-Image、MOSS-TTS 系列等）。Fri, 30 Oct 2026 00:00:00 GMTnewsMOSS开源模型文生图语音合成多模态https://spark.tlec.top/news/sensenova-loop-9b-preview-40479274/https://spark.tlec.top/news/sensenova-loop-9b-preview-40479274/SenseNova-Loop-9B-Preview 是一款面向长文本和 Agent 工作流的多模态语言模型，采用共享循环解码器和双流流形约束超连接（mHC）架构，支持 256K 超长上下文。Fri, 30 Oct 2026 00:00:00 GMTnews多模态开源模型长文本Agenthttps://spark.tlec.top/news/sketch-stadium-ai-agent-pictionary-ec56c328/https://spark.tlec.top/news/sketch-stadium-ai-agent-pictionary-ec56c328/开发者推出 Sketch Stadium 演示项目，让 AI Agent 通过 SVG 绘制和猜测来玩 Pictionary（你画我猜），同时展示了 Wakeshift 推送通知 SDK 与 MCP 的结合应用。Fri, 30 Oct 2026 00:00:00 GMTnewsAgentMCPWakeshiftSVGPush Notificationshttps://spark.tlec.top/news/upgrade-flash-next-to-dflash2-6ac0e32f/https://spark.tlec.top/news/upgrade-flash-next-to-dflash2-6ac0e32f/随着 llama.cpp 对 DFlash2 的支持，Qwen3.8-Flash-Next 模型在本地运行的解码速度平均提升 17%，草稿头体积大幅缩减至约 350MB。Fri, 30 Oct 2026 00:00:00 GMTnewsllama.cppQwenDFlash2Apple Silicon量化https://spark.tlec.top/plays/local-coding-assistant/https://spark.tlec.top/plays/local-coding-assistant/用 Continue 插件连接本地 Ollama 模型，补全和对话都不出本机，适合公司代码不能外传的场景。Wed, 28 Oct 2026 00:00:00 GMTplays离线隐私https://spark.tlec.top/plays/anythingllm-knowledge-base/https://spark.tlec.top/plays/anythingllm-knowledge-base/桌面版一键安装，接上本地模型和本地向量库，拖进 PDF、网页、笔记就能问。Sun, 25 Oct 2026 00:00:00 GMTplaysRAG知识库