[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-colibri-v1-11-deepseek-v41-flash":3,"topics-all":38,"news-related-8c5bd0e1-551a-405b-b2fc-a67520bf84c5":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"8c5bd0e1-551a-405b-b2fc-a67520bf84c5","Colibri v1.11.0 发布:纯 C 引擎直读 DeepSeek V4.1 Flash,552B 从 SSD 流进 CPU","开源纯 C 推理引擎 Colibri 发布 v1.11.0,新增第九个模型家族:DeepSeek V4.1 Flash 的 552B 官方 checkpoint 零转换直读,510 GB 权重从 SSD 流式进内存,CPU 上一轮对话 25.1 秒。","9 月 10 日 DeepSeek 发布 V4.1 Flash,552B 总参、16B 激活,fp4 专家加 fp8 稠密层;三天后,开源推理引擎 Colibri 发布 v1.11.0,新增的第九个模型家族引擎把这 510 GB 的官方 checkpoint 直接从 SSD 流式读进内存——不做任何格式转换,纯 CPU 跑通。\n\n先说引擎定位。Colibri 是纯 C 写的 MoE 推理引擎,每个模型家族一个 .c 文件,运行时零依赖:不调 BLAS、无 Python、不强制 GPU。核心思路是把 VRAM、RAM、SSD 视作同一套推理存储层级,routed experts 平时躺在磁盘上,路由证明需要时才搬进来。仓库目前 31.1k stars、2230 次提交、Apache 2.0,九个家族从 744B 的 GLM-5.2\u002F5.3 覆盖到 2.8T 的 Kimi K3。\n\n## V4.1 引擎:203 GB 记忆永不进 RAM\n\nRelease notes 的细节密度很高。552B checkpoint 磁盘占用 510 GB,其中 203 GB 是名为 Engram 的双份 n-gram 记忆,共 3.84 亿行,永不进 RAM,每次按需从盘上读几百字节。稠密层是 fp8(32×32 ue8m0 分块),专家权重 fp4,布局与 Kimi K3 引擎已支持的 mxfp4 字节一致——「零转换」的技术底座就在这里。README 给出的对照:V4.1 Flash 每 token 只需读 4.5 GB 专家权重,GLM-5.2 是 12.7 GB。32 层视觉塔、DSML 格式工具调用、DSpark 三段投机解码全部实现;CI 对照团队自写的 torch 纯 CPU 参考实现逐 token 校验——官方 forward 依赖 tilelang GPU kernel,CPU 上根本跑不了——视觉塔对到 5e-06。\n\n## 78.7 秒到 25.1 秒:把失败也写进文档\n\n冷缓存、每轮清缓存、同 prompt 同 seed 实测:一轮对话从 78.7 秒优化到 25.1 秒(0.305 → 0.957 tok\u002Fs),手段是批量专家读取(READ_DEPTH 默认 8,实测拐点)、注意力矩阵按位置块整读、expert-major MoE 布局,且每一步与被替换版本 bit-exact。五轮多轮会话 1.14–1.58 tok\u002Fs。慢吗?慢。但这个团队把失败也写了进去:两种把专家读取藏进矩阵乘的预取方案,实测更差,删掉,连同数字一起写进 docs\u002Fdeepseek-v41.md,给下一个尝试者留起点。\n\n## 开源权重的第二战场\n\n这个 release 的新闻性不在速度,在生态响应:模型发布三天,第三方纯 C 引擎原生直读 checkpoint。这与 8 月底 WARP 把 313B GLM-5.3-Flash 专家搬进 NVMe、Cohere 的单文件 megakernel 是同一条赛道的三种解法——引擎层正成为开放权重真正落地的战场。API 会涨价、限流、停服;能在一台没有 GPU 的机器上持有而不是租用模型,这份选择权是引擎给的。\n\n所以下次看到「重磅开源」,别只看跑分——看看几天内有没有引擎把它跑起来,那才是开放权重含金量的真实刻度。\n\n参考:github.com\u002FJustVugg\u002Fcolibri\u002Freleases\u002Ftag\u002Fv1.11.0(Release Notes)· github.com\u002FJustVugg\u002Fcolibri(README)","https:\u002F\u002Fgithub.com\u002FJustVugg\u002Fcolibri\u002Freleases\u002Ftag\u002Fv1.11.0","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"b52db7e9-7c58-42c3-9536-5132cb2f8f72","deepseek",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"d3b77e6e-cffb-4951-846f-34a6a091178d","en","Colibri v1.11: DeepSeek V4.1 Flash 552B runs off SSD in pure C","Pure-C engine Colibri v1.11.0 adds a ninth family: DeepSeek V4.1 Flash's 552B checkpoint streams off SSD, zero conversion, 25.1s a turn on CPU.","On September 10, DeepSeek released V4.1 Flash — 552B total parameters, 16B active, fp4 experts over an fp8 dense trunk. Three days later, the open-source inference engine Colibri shipped v1.11.0 with its ninth model-family engine, which streams that 510 GB official checkpoint straight off an SSD into memory: zero format conversion, running on pure CPU.\n\nFirst, the engine itself. Colibri is a MoE inference engine written in pure C — one .c file per model family, zero runtime dependencies: no BLAS, no Python at runtime, no GPU required. Its core idea treats VRAM, RAM and SSD as a single inference memory hierarchy: routed experts live on disk and are staged in only when routing proves they are needed. The repository currently sits at 31.1k stars, 2,230 commits, Apache 2.0, with nine families ranging from the 744B GLM-5.2\u002F5.3 up to the 2.8T Kimi K3.\n\n## The V4.1 engine: 203 GB of memory that never enters RAM\n\nThe release notes are unusually dense. The 552B checkpoint occupies 510 GB on disk, of which 203 GB is a double n-gram memory called Engram — 384M rows that never enter RAM and are read from disk a few hundred bytes at a time. The dense trunk is fp8 (32x32 ue8m0 tiles) and the experts are fp4, with a layout byte-identical to the mxfp4 the Kimi K3 engine already reads — that is the technical foundation of \"no conversion\". The README's comparison: V4.1 Flash pays 4.5 GB of expert reads per token, against GLM-5.2's 12.7 GB. The 32-layer vision tower, tool calling in the checkpoint's own DSML format, and DSpark speculative decoding (3 stages, blocks of 5) are all implemented; CI holds the engine token-exact against a torch-only CPU reference the team had to write themselves — the vendor's own forward needs tilelang GPU kernels and cannot run on a CPU — with the vision tower matched to 5e-06.\n\n## From 78.7 to 25.1 seconds: failures documented too\n\nMeasured on the released checkpoint, cold, caches dropped before every run, same prompt and seed: a turn went from 78.7 s to 25.1 s (0.305 to 0.957 tok\u002Fs), via batched expert reads (READ_DEPTH, default 8, the measured knee), attention matrices read once per block of positions instead of once per token, and an expert-major MoE layout — every step bit-exact against what it replaced. A five-turn chat session runs at 1.14–1.58 tok\u002Fs. Slow? Yes. But the team also wrote down what failed: two schemes for hiding expert reads behind the matmuls were built, measured worse, and removed, with the numbers preserved in docs\u002Fdeepseek-v41.md so the next attempt starts from them.\n\n## The second battlefield of open weights\n\nThe news here is not speed but ecosystem response: three days after the model dropped, a third-party pure-C engine reads the checkpoint natively. Together with WARP moving 313B GLM-5.3-Flash experts into NVMe in late August, and Cohere's single-file megakernel, these are three answers on the same track — the engine layer is becoming the battlefield where open weights actually land. APIs can raise prices, throttle, or shut down; holding a model on a machine with no GPU, instead of renting it, is a choice that engines provide.\n\nSo the next time a \"landmark open release\" lands, don't just read the benchmarks — check how fast an engine can run it. That is the real measure of what open weights are worth.\n\nReference: [Colibri v1.11.0 Release Notes](https:\u002F\u002Fgithub.com\u002FJustVugg\u002Fcolibri\u002Freleases\u002Ftag\u002Fv1.11.0) · [Colibri README](https:\u002F\u002Fgithub.com\u002FJustVugg\u002Fcolibri)","colibri-v1-11-deepseek-v41-flash","2026-09-14T13:09:40Z","2026-09-14T13:09:53.002730Z","2026-09-14T13:09:53.002747Z",true,"agent",54,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"2638aeac-dc4d-4b73-b7fe-2b042015adee","OreoLook 开源:三层缓存把 AI 搜索搬进 8 核 CPU,重复问题 0.1 毫秒出答案","oreolook-three-layer-cpu-cache","2026-09-10T23:08:36+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"63042acc-9bec-4d6e-960b-8352ed541bb5","bartowski 用 1000 组实验重写 GGUF 量化排布规则","gguf-quant-layout-maps","2026-09-10T21:05:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"0fe9ceb8-6411-4924-8e02-8cee3665fc6f","Cohere 开源 megakernel 推理引擎：单 CUDA 文件，H100 解码吃到 62% 带宽光速","cohere-megakernel-north-mini-code-h100","2026-09-08T21:13:46+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"0c29e1ad-914a-4b79-a153-445c087acb03","被 LLM 抛弃的 dropout 翻身:Cerebras 称调好可省 25% 训练 FLOPs","dont-drop-dropout-layer-sparsity","2026-09-07T21:06:35+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"34b087d5-f502-4a46-9599-6f014fe0cf89","llama.cpp v0.4.0 发布:权重懒加载进主线,小内存也能碰 75B 级 MoE","llama-cpp-v0-4-0-lazy-tensor-loading","2026-09-07T13:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"63c30bcd-3ffc-47c5-bd74-c2a9ed8f7c94","DeepSeek Harness 预览版开源:Agent 被拆成可插拔的插件栈,模型只负责想、Harness 负责做事","deepseek-harness-plugin-stack","2026-09-05T06:00:00+00:00"]