[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-velum-cosyvoice3-cpp-zero-python":3,"topics-all":38,"news-related-bd0a85ff-0a8c-4900-a8fc-65e606a8bce0":67},{"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},"bd0a85ff-0a8c-4900-a8fc-65e606a8bce0","Velum:零 Python 跑 CosyVoice3","HardenedLinux 开源 Velum,把 CosyVoice3 TTS 用 modern C++ 重写,神经网络跑在 GGML+CUDA,运行时不再依赖 Python。四阶段逐级对齐 PyTorch 参考到 GREEN(相对误差 ≤1%),端到端在 CPU 与 CUDA 上均通过校验。","CosyVoice3 是当下表现靠前的开源 TTS 模型,但它的推理栈由 Python 生态堆叠,部署时要把 PyTorch、依赖包、版本冲突一起装好,十几 GB 的运行时是常态。HardenedLinux 在 GitHub 开源的 Velum 直接把这套流水线用 modern C++ 重写,产物只有一个可执行文件,运行时不再依赖 Python。\n\n## 工程取舍:在线零 Python\n\n神经网络(LLM \u002F Flow \u002F HiFT)全部跑在 GGML 之上,带 CUDA 加速;权重转换和 campplus \u002F speech tokenizer \u002F matcha mel 这类声学前处理被挪到离线 Python 一次性导出,C++ 端加载冻结好的二进制文件,部署更新只是拷文件。仓库自带 CMake 工程,默认 `VELUM_ENABLE_CUDA=ON`,CUDA 工具链缺失时退回到纯 CPU 构建;两个后端都链接进同一个二进制,运行时按设备自动选择,可用 `VELUM_BACKEND=cpu` 强制指定。\n\n## 逐级数值对齐 PyTorch 参考\n\n验证策略遵循 PyTorch 参考实现的逐级对齐:每一个 cross-check 都报告 scale-normalized relative error,≤1e-2 标 GREEN(通过),1e-2–1e-1 标 YELLOW(待查),>1e-1 标 RED(结构发散,直接 fail)。仓库用四组 verify_*.py 脚本分别核 DSP 前端(Whisper 128-bin log-mel + Kaldi 80-bin fbank)、Flow 解码器(PreLookaheadLayer + DiT×22 + CFM Euler)、HiFT 声码器、LLM 主干(Qwen2-0.5B + CosyVoice3LM heads + ras_sampling)。端到端 CLI 测试 `tests\u002Fverify_e2e.py` 在 seed 0 上把 LLM → Flow → HiFT 全链路对 PyTorch 做 cross-check,CPU 端 Flow mel 的 max abs 为 1.142e-3、相对误差 1.056e-4,HiFT pcm 相对误差 8.112e-3;CUDA 端分别是 2.220e-3 \u002F 2.052e-4 与 3.495e-3 \u002F 5.036e-3,两端均落在 GREEN 区间。\n\n## 谁写的、边界在哪\n\n这种规模的项目由 AI 直接写代码并不寻常。仓库 README 明示:项目由人类做架构规划,LLM 部分由 deepseek-v4-pro 完成,Coding Assistant 是 Claude Code。需要注意的边界:ONNX 前端(campplus + speech tokenizer)和 matcha 80-bin mel 仍由 Python 预处理生成文件,没有被重写成 C++,属于「暂交给 Python」的部分;LLM 权重 `llm.gguf` 大约 2.6 GB,CUDA 路径需要至少同等显存的卡,文档建议若 `cudaMalloc` 报 OOM 时强制走 CPU。\n\n对希望把 TTS 嵌进产品或 Agent 流水线的团队,Velum 的价值是部署形态的简化:单一可执行、零 Python 运行时、端到端数值对齐参考实现、CUDA\u002FCPU 同二进制可切换。开源协议是 GPL-3.0。","https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85479","d59894d3-308e-4fd8-8865-86dc1eeac4a2",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",{"id":19,"name":20,"slug":20,"description":14,"color":14},"e676a5cf-1f24-472f-a765-86fa21a1bc3c","ai-model",{"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},"6ed8d3f4-8acf-43aa-8768-107fae371949","en","Velum: zero-Python C++ for CosyVoice3","HardenedLinux open-sourced Velum, a modern C++ rewrite of the CosyVoice3 TTS pipeline that runs at runtime without Python. Neural networks run on GGML with optional CUDA; weights and acoustic features are exported once offline in Python. Per-stage cross-checks against PyTorch hit GREEN (\u003C=1e-2) and an end-to-end LLM->Flow->HiFT test passes on both CPU and CUDA.","CosyVoice3 is one of the stronger open-source TTS models, but its inference stack is built on the Python ecosystem: PyTorch plus a long tail of version-pinned dependencies that adds up to several gigabytes and forces every deployer to play package detective. Velum, open-sourced by HardenedLinux on GitHub, rewrites the entire text-to-speech pipeline in modern C++ and ships a single executable that needs no Python at runtime.\n\n## Engineering trade-off: zero Python online\n\nThe neural networks (LLM \u002F Flow \u002F HiFT) run on GGML with CUDA acceleration; weight conversion and acoustic frontend steps (campplus, the speech tokenizer, and the matcha 80-bin mel) are pushed into an offline Python step that runs once and produces frozen files the C++ binary loads verbatim. The repo's CMake project defaults to VELUM_ENABLE_CUDA=ON and falls back to a CPU-only build when no CUDA toolchain is detected; both backends are linked into the same binary, the runtime auto-selects, and VELUM_BACKEND=cpu forces the CPU path.\n\n## Stage-by-stage numerical alignment with PyTorch\n\nEach cross-check reports a scale-normalized relative error: \u003C=1e-2 is GREEN (passes), 1e-2 to 1e-1 is YELLOW (needs investigation), >1e-1 is RED (structural divergence, hard fail). The repo ships four verify_*.py scripts covering the DSP frontend (Whisper 128-bin log-mel plus Kaldi 80-bin fbank), the Flow decoder (PreLookaheadLayer + DiT x22 + CFM Euler), the HiFT vocoder, and the LLM backbone (Qwen2-0.5B + CosyVoice3LM heads + ras_sampling). The end-to-end CLI test, tests\u002Fverify_e2e.py, runs the full LLM -> Flow -> HiFT chain on seed 0 against PyTorch. On the CPU backend, Flow mel max abs is 1.142e-3 with relative error 1.056e-4, HiFT pcm relative error 8.112e-3; on CUDA the numbers are 2.220e-3 \u002F 2.052e-4 and 3.495e-3 \u002F 5.036e-3 respectively. Both backends sit inside the GREEN band.\n\n## Who wrote it, and what is still Python\n\nThis is also an unusual example of LLM-authored C++. The README credits deepseek-v4-pro for the LLM work and Claude Code as the coding assistant, with human-driven architecture. Caveats: the ONNX frontend (campplus + speech tokenizer) and the matcha 80-bin mel are still \"temporarily handed to Python\" and produce files the CLI consumes rather than running them in C++; the LLM weight file llm.gguf is around 2.6 GB and the CUDA path needs at least that much free VRAM, so the docs recommend forcing the CPU backend if cudaMalloc reports out of memory.\n\nFor teams trying to embed TTS into products or agent pipelines, Velum's value is the deployment simplification: one binary, no Python runtime, numerical alignment with the PyTorch reference at every stage, and CPU\u002FCUDA selectable from the same executable. The repository is GPL-3.0 licensed.","velum-cosyvoice3-cpp-zero-python","2026-09-26T07:00:00Z","2026-09-26T07:03:21.933141Z","2026-09-26T07:03:21.933152Z",true,"agent",495,[39,49,58],{"slug":40,"tag_slug":41,"title_zh":42,"title_en":43,"intro_zh":44,"intro_en":45,"id":46,"is_active":35,"created_at":47,"modified_at":48},"image-generation","text-to-image","图像生成：从文生图到逐框编辑的模型竞赛","Image Generation: from text-to-image to frame-level editing","文生图已是红海，战场正转向编辑与可控性：FLUX.2\u002F3 用开源和量化把生图搬进消费级显卡，Qwen-Image、Seedream 等国产模型紧咬不放，GPT Image 2 则把推理能力带进像素。本专题追踪图像生成与编辑模型的发布、架构演进、加速量化与商业化数据，看「好图」如何变成「可用的图」。","Text-to-image is a crowded race, and the frontier has moved to editing and control: FLUX.2\u002F3 push open weights and quantization onto consumer GPUs, Qwen-Image and Seedream close in from China, and GPT Image 2 brings reasoning into pixels. This topic tracks model releases, architecture shifts, acceleration work, and the business numbers behind \"pretty\" becoming \"usable\".","52de7c59-7864-4569-8ae7-3d656d26a1a1","2026-11-20T07:20:23.534506Z","2026-11-20T07:20:23.534527Z",{"slug":50,"tag_slug":50,"title_zh":51,"title_en":52,"intro_zh":53,"intro_en":54,"id":55,"is_active":35,"created_at":56,"modified_at":57},"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":59,"tag_slug":59,"title_zh":60,"title_en":61,"intro_zh":62,"intro_en":63,"id":64,"is_active":35,"created_at":65,"modified_at":66},"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":68},[69,74,79,84,89,94],{"id":70,"title":71,"news_slug":72,"published_at":73},"ab1e6bdf-9433-41f0-be4d-cc0f78523ac2","SonicBoom v0.0.3:一个 .so 干掉手工调优","sonicboom-v003-mlir-compiler","2026-10-19T09:03:05+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"b3ce6178-954b-403d-b4c3-6480888a996d","Velum 把 CosyVoice3 重写成单文件 C++ 推理程序","velum-cxx-cosyvoice3-tts","2026-09-26T07:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"6fa1bc74-c98e-476b-bc4c-9ae057105ffb","ParaTempo:免训练并行推理,延迟最高降 32%、token 省三成","paratempo-temporal-confidence-parallel-reasoning","2026-08-24T17:20:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"761043fc-3329-42e4-914d-3f567b665b90","面壁智能联合 OpenBMB 开源 ForgeStencil：用双 Agent 把 Stencil 优化直接打进真实科学计算软件","mianbiai-openbmb-forgestencil-stencil-optimization","2026-08-04T10:00:00+00:00",{"id":90,"title":91,"news_slug":92,"published_at":93},"f9a6c57b-a98e-407c-9751-93907ff314b0","Capacity-Aware Inference：MoE 推理 1.85× 加速","iclr-2026-capacity-aware-moe","2026-07-27T10:00:00+00:00",{"id":95,"title":96,"news_slug":97,"published_at":98},"ba729d8d-d18e-4e77-824e-1db99ec469eb","近零词嵌入暗坑:五成门控检查点训练静默白跑","near-zero-embeddings-gated-training-stall","2026-11-04T15:20:36+00:00"]