研究人员详细介绍了一种使用 trellis 编码的 2 位量化方法，可以显著减小大型语言模型 (LLM) 的尺寸。该技术允许 Qwen3-4B 等模型在无需重新训练的情况下从 8 GB 缩小到 1.24 GB，从而可以在资源受限的环境中使用它们。该过程包括证明 AI 转换的代码以及探索 AI 代理与工具的交互，突出了 AI 效率和应用的进步。 AI
影响 能够将强大的 LLM 部署在内存有限的设备上，有可能加速 AI 在边缘计算和移动应用中的普及。
排序理由 该集群讨论了一种新颖的 LLM 量化技术和 AI 代码转换，属于研究类别。
在 Mastodon — fosstodon.org 阅读 →
AI 生成摘要 · Google Gemini · 来自 6 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
How do you fit a 4B LLM in 1.24 GB? Trellis-coded 2-bit quantization, explained with real benchmarks Qwen3-4B from 8 GB to 1.24 GB with no retraining: what we did, what we measured, and where it still loses. # machinelearning # llm # ai # opensource # software # coding # developm…
Fix: a setting shows as "undefined" in the browser of your Next.js app (NEXT_PUBLIC_ environment variable, tested repro, Next 16 + React 19.3) # nextjs # react # webdev # ai # software # coding # development # engineering # inclusive # community Fix: a setting shows as "undefined…
Want to try AI on real work without connecting it to anything? This is a small, limited exercise:... # ai # productivity # beginners # security # software # coding # development # engineering # inclusive # community A safe first AI exercise: sort an inbox, and learn one habit (te…
How can a developer prove code converted completely by an AI from one language to another? Surely... # csharp # dotnet # ai # testing # software # coding # development # engineering # inclusive # community An AI ported WinQuake to C#, and per-frame hashes prove it bit-exact
This is a copy of an article first published on AgentShortlist, where the tables and cards live. On... # mcp # ai # security # opensource # software # coding # development # engineering # inclusive # community We read 25 MCP servers. Here is what we found
The agent sends {"projectId":"project-a","confirmed":true} to a deletion tool. What established the... # ai # security # programming # abotwrotethis # software # coding # development # engineering # inclusive # community Your Agent's confirmed: true Is Not Permission.