普通电脑也能跑744B参数大模型！colibri开源工具让25GB内存畅玩GLM-5.2。
colibri 是一个非常实用的开源项目，它能让普通电脑也能运行超大语言模型（GLM-5.2（744B），并且可以在无显卡的情况下，仅使用 CPU，但需要至少25G 内存。@Appinn 普通电脑跑 GLM-5.2（744B）模型 colibri 使用纯 C 语言，零依赖。可以按需从硬盘加载 Exp
亲测GLM 5.2在慢电脑上的运行体验，能力逼近闭源模型但面临内存瓶颈，本地部署党必看
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are si…
探索参数高效LLM架构新突破，为边缘部署提供解决方案。
arXiv:2604.21254v3 Announce Type: replace Abstract: LLM architecture research generally aims to maximize model quality subject to fixed compute/latenc…
Next.js 16.3 预览版重磅更新：Turbopack 将开发服务器内存占用直降90%，大幅优化开发体验。
Turbopack in Next.js 16.3 adds development memory eviction, persistent file cache for builds, experimental Rust React Compiler, and `import.meta.glob`…
利用Linux系统压力信息动态修剪大模型KV缓存，专为边缘设备优化内存使用
I thought it'd be interesting to use Linux PSI (Pressure Stall Information) for an LLM runtime to trim the KV cache. This is mainly useful imo for edg…
一行import即可加速MoE微调3.7倍，英伟达开源NeMo AutoModel，降低内存占用超30%，轻松提升训练效率。
在Transformers v5的基础上，增加了专家并行、DeepEP和TransformerEngine
融合反向传播与优化器阶段的梯度处理，大幅降低LLM训练的内存峰值，突破显存瓶颈的新方案。
arXiv:2606.22932v1 Announce Type: new Abstract: Reverse-mode differentiation computes every weight gradient, writes it to memory, and only then lets t…
在Linux上省下4GB内存：告别臃肿的Docker Desktop，直接使用原生Docker引擎更高效。
Docker Desktop was using 4GB of RAM on my machine. Sitting there. Doing nothing. No containers running. Just the daemon. I don't know about you, but 4…
揭示LLM服务中内存受限导致的服务诱导拥塞现象，为推理系统优化提供新视角。
arXiv:2606.15555v1 Announce Type: cross Abstract: In large language model (LLM) serving, each request accumulates persistent graphics processing unit …
针对Agent推理场景的跨轮意图感知KV缓存剪枝方法，显著降低内存占用与延迟。
arXiv:2606.09916v1 Announce Type: cross Abstract: Multi-turn LLM agents fan short queries into long trajectories of tool calls, search results, and in…
ICML 2026 Oral论文，提出通过扩展正交变换实现大模型训练的内存高效方案。
arXiv:2603.05500v2 Announce Type: replace Abstract: Efficient and stable training of large language models (LLMs) remains a core challenge in modern m…
从轻松周末项目到性能噩梦，揭秘浏览器音频波形渲染的隐藏难点
A few months ago I thought displaying an audio waveform in the browser would be a weekend project. I was wrong. The first prototype worked fine with s…
提出闭环内存优化方法，革新软件工程中的内存管理，显著提升系统性能与资源利用率
arXiv:2606.05646v1 Announce Type: cross Abstract: Large language models (LLMs) have enabled powerful software engineering (SE) agents capable of navig…
仅26MB内存的Rust编码代理，对比JS代理节省超95%资源，极低功耗运行5种工作模式。
Article URL: https://gi-dellav.github.io/zerostack/ Comments URL: https://news.ycombinator.com/item?id=48401040 Points: 2 # Comments: 0
LocalClaw：本地模型优先的AI agent框架，用图数据库替代扁平存储，内存仅85MB，无云无API成本
# I Replaced My AI Agent's Flat Fact Store with a Graph Database and It Runs in 85MB I've been building LocalClaw, a local-model-first AI agent framew…
ICML2026接收：动态稀疏性突破LLM训练内存瓶颈，兼顾稳定性与实用规模化
arXiv:2606.00888v1 Announce Type: new Abstract: Dynamic Sparse Training (DST) offers a promising paradigm for improving the training and inference eff…
专为大模型设备端微调设计的内存高效结构化反向传播，突破移动设备6-12GB内存限制，实现隐私保护与性能平衡。
arXiv:2602.13069v2 Announce Type: replace Abstract: On-device fine-tuning enables privacy-preserving personalization of large language models, but mob…
提出可学习的零阶优化器，无需梯度即可高效微调大模型，大幅降低内存开销。
arXiv:2510.00419v2 Announce Type: replace Abstract: Zeroth-order optimizers have recently emerged as an attractive approach for fine-tuning large lang…
突破多模态LLM训练中计算与内存的帕累托前沿，BigMac提出高效优化方案。
arXiv:2605.25451v1 Announce Type: new Abstract: Training multimodal large language models (MLLMs) is challenged by both model and data heterogeneity. …
Spring Native + Cloud Run 组合拳，冷启动降至仅18毫秒，内存仅3MB，部署成本直接归零。
Cet article est la version condensée de mon retour d'expérience complet sur ban.ga — config Maven -Pnative détaillée, registrar RuntimeHintsRegistrar …