[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-VoltaML--voltaML":3,"tool-VoltaML--voltaML":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",140436,2,"2026-04-05T23:32:43",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":77,"owner_location":77,"owner_email":77,"owner_twitter":77,"owner_website":77,"owner_url":78,"languages":79,"stars":96,"forks":97,"last_commit_at":98,"license":99,"difficulty_score":100,"env_os":101,"env_gpu":102,"env_ram":103,"env_deps":104,"category_tags":109,"github_topics":77,"view_count":23,"oss_zip_url":77,"oss_zip_packed_at":77,"status":16,"created_at":110,"updated_at":111,"faqs":112,"releases":142},1327,"VoltaML\u002FvoltaML","voltaML","⚡VoltaML is a lightweight library to convert and run your ML\u002FDL deep learning models in high performance inference runtimes like TensorRT, TorchScript, ONNX and TVM.","voltaML 是一个开源的轻量级 Python 库，帮你把训练好的机器学习或深度学习模型一键转换成 TensorRT、TorchScript、ONNX、TVM 等高性能推理格式，并自动部署到 CPU 或 NVIDIA GPU 上。只需一行代码，就能把推理速度提升 2-10 倍，同时支持 FP16、INT8 量化，显著降低显存占用和延迟。\n\n它解决了“模型训练完却跑不快”的痛点：不用再手动写冗长的转换脚本、调优参数或适配不同硬件。voltaML 内置硬件感知编译，自动针对 RTX、A100、Jetson 等设备做最优优化。\n\n适合算法工程师、深度学习研究员、MLOps 开发者，以及想把 Stable Diffusion、YOLO、BERT 等模型快速落地到生产环境的团队。官方还提供了 ResNet、YOLOv5\u002Fv6、DeepLabv3、BERT 等完整示例 Notebook，拿来即用。","\u003Cp align=\"center\">\n  \u003Cimg width=\"600\" height=\"120\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_ff5c1ae61154.jpg\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n\u003Cimg width=\"141\" alt=\"Screenshot 2022-10-19 at 3 55 14 PM\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_4d17d2295af0.png\">\n\u003C\u002Fp>\n  \n\u003Cp align=\"center\">\n  \u003Cb> Accelerate your machine learning and deep learning models by upto 10X \u003C\u002Fb> \n\u003C\u002Fp>\n\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FpY5SVyHmWm\"> \u003Cimg src=\"https:\u002F\u002Fdcbadge.vercel.app\u002Fapi\u002Fserver\u002FpY5SVyHmWm\" \u002F> \u003C\u002Fa> \n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n\n### **🔥UPDATE: [Stable-Diffusion\u002FDreamBooth Acceleration](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML-fast-stable-diffusion). Upto 2.5X speed up in inference🔥**\n\u003C\u002Fdiv>\n\n\u003Cp style=\"text-align: center;\"> \u003C\u002Fp>\n\n**voltaML** is an open-source lightweight library to accelerate your machine learning and deep learning models. VoltaML can optimize, compile and deploy your models to your target CPU and GPU devices, with just ***one line of code.***\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_83070471d2b6.gif\" alt=\"animated\" \u002F>\n\u003C\u002Fp>\n\n\n#### Out of the box support for \n\n\n✅ FP16 Quantization\u003Cbr\u002F>\n\n✅ Int8 Quantization*\u003Cbr\u002F>\n\n✅ Hardware specific compilation\n\n\u003Cbr>\n\n\u003Cimg width=\"1100\" alt=\"Screenshot 2022-10-17 at 12 06 26 PM\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_303a3e636684.png\">\n\u003Cbr>\n\n**voltaML has compilation support for the following:**\n\n\n\u003Cimg width=\"1102\" alt=\"Screenshot 2022-06-13 at 3 43 03 PM\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_f7e5851e8301.png\">\n\n\n## Installation\n\n### Own setup:\n\nRequirements:\n\n* CUDA Version >11.x \u003Cbr\u002F>\n* TensorRT == 8.4.1.2\u003Cbr\u002F>\n* PyTorch == 1.12 cu11.x\u003Cbr\u002F>\n* NVIDIA Driver version > 510\n\n````\ngit clone https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML.git\ncd voltaML\npython setup.py install\n````\n### Docker Container 🐳\n````\ndocker pull voltaml\u002Fvoltaml:v0.4\ndocker run -it --gpus=all -p \"8888:8888\" voltaml\u002Fvoltaml:v0.4 \\ \n        jupyter lab --port=8888 --no-browser --ip 0.0.0.0 --allow-root\n````\n## Usage\n\n```python\nimport torch\nfrom voltaml.compile import VoltaGPUCompiler, VoltaCPUCompiler, TVMCompiler\nfrom voltaml.inference import gpu_performance\n\nmodel = torch.load(\"path\u002Fto\u002Fmodel\u002Fdir\")\n\n# compile the model by giving paths\ncompiler = VoltaGPUCompiler(\n        model=model,\n        output_dir=\"destination\u002Fpath\u002Fof\u002Fcompiled\u002Fmodel\",\n        input_shape=(1, 3, 224, 224), # example input shape\n        precision=\"fp16\" # specify precision[fp32, fp16, int8] - Only for GPU compiler\n        target=\"llvm\" # specify target device - Only for TVM compiler\n    )\n\n# returns the compiled model\ncompiled_model = compiler.compile()\n\n# compute and compare performance\ngpu_performance(compiled_model, model, input_shape=(1, 3, 224, 224))\ncpu_performance(compiled_model, model, compiler=\"voltaml\", input_shape=(1, 3, 224, 224))\ncpu_performance(compiled_model, model, compiler=\"tvm\", input_shape=(1, 3, 224, 224))\n\n```\n## Notebooks\n\n01. [ResNet-50](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FResNet50%20Classification%20Demo.ipynb) image classification \n02. [DeeplabV3_MobileNet_v3_Large](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FDeeplabV3%20Segmentation%20Demo.ipynb) Segmentation\n03. [YOLOv5](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FYoloV5%20Demo.ipynb) Object Detection YOLOv5\n04. [YOLOv6](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FYoloV6%20Demo.ipynb) Object Detection YOLOv6 \n05. [Bert_Base_Uncased](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FBert_base_uncased_HuggingFace.ipynb) Huggingface\n\n\n## Benchmarks\n### 🖼️ Classification Models Inference Latency (on GPU) ⏱️\nClassification has been done on Imagenet data, `batch size = 1` and `imagesize = 224` on NVIDIA RTX 2080Ti. In terms of top 1% and 5% accuracy for `int8` models, we have not seen an accuracy drop of more than 1%. \n\n![Pytorch (ms), VoltaGPU FP16 (ms) and VoltaGPU int8 (ms)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_d44e6c5a930f.png)\n\n\n| Model         | Pytorch (ms) | VoltaGPU FP16 (ms) | VoltaGPU int8 (ms) | Pytorch vs Int8 Speed |\n|---------------|--------------|--------------------|--------------------|-----------------------|\n| squeezenet1_1 | 1.6          | 0.2                | 0.2                | 8.4x                  |\n| resnet18      | 2.7          | 0.4                | 0.3                | 9.0x                  |\n| resnet34      | 4.5          | 0.7                | 0.5                | 9.0x                  |\n| resnet50      | 6.6          | 0.7                | 0.5                | 13.2x                 |\n| resnet101     | 13.6         | 1.3                | 1.0                | 13.6x                 |\n| densenet121   | 15.7         | 2.4                | 2.0                | 7.9x                  |\n| densenet169   | 22.0         | 4.4                | 3.8                | 5.8x                  |\n| densenet201   | 26.8         | 6.3                | 5.0                | 5.4x                  |\n| vgg11         | 2.0          | 0.9                | 0.5                | 4.0x                  |\n| vgg16         | 3.5          | 1.2                | 0.7                | 5.0x                  |\n\n### 🧐 Object Detection (YOLO) Models Inference Latency (on GPU) ⏱️\nObject Detection inference was done on a dummy data with `imagesize = 640` and `batch size = 1` on NVIDIA RTX 2080Ti.\n\n![Pytorch (ms) and VoltaGPU FP16 (ms)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_50dabbf0ceec.png)\n\n\n| Model        | Pytorch (ms) | VoltaGPU FP16 (ms) | Pytorch vs FP16 Speed |\n|--------------|--------------|--------------------|-----------------------|\n| YOLOv5n      | 5.2          | 1.2                | 4.3x                  |\n| YOLOv5s      | 5.1          | 1.6                | 3.2x                  |\n| YOLOv5m      | 9.1          | 3.2                | 2.8x                  |\n| YOLOv5l      | 15.3         | 5.1                | 3.0x                  |\n| YOLOv5x      | 30.8         | 6.4                | 4.8x                  |\n| YOLOv6s      | 8.8          | 3.0                | 2.9x                  |\n| YOLOv6l_relu | 23.4         | 5.5                | 4.3x                  |\n| YOLOv6l      | 18.1         | 4.1                | 4.4x                  |\n| YOLOv6n      | 9.1          | 1.6                | 5.7x                  |\n| YOLOv6t      | 8.6          | 2.4                | 3.6x                  |\n| YOLOv5m      | 15.5         | 3.5                | 4.4x                  |\n\n\n### 🎨 Segmentation Models Inference Latency (on GPU) ⏱️\nSegmentation inference was done on a dummy data with `imagesize = 224` and `batch size = 1` on NVIDIA RTX 2080Ti.\n\n![Pytorch (ms), VoltaGPU FP16 (ms) and VoltaGPU Int8 (ms)(1)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_8b9b0355ce2e.png)\n\n\n| Model                       | Pytorch (ms) | VoltaGPU FP16 (ms)  | VoltaGPU Int8 (ms) | Speed Up (X) |\n|-----------------------------|--------------|------------------------|------------------------|--------------|\n| FCN_Resnet50                | 8.3          | 2.3                    | 1.8                    | 3.6x         |\n| FCN_Resnet101               | 14.7         | 3.5                    | 2.5                    | 5.9x         |\n| DeeplabV3_Resnet50          | 12.1         | 2.5                    | 1.3                    | 9.3x         |\n| DeeplabV3_Resnet101         | 18.7         | 3.6                    | 2.0                    | 9.4x         |\n| DeeplabV3_MobileNetV3_Large | 6.1          | 1.5                    | 0.8                    | 7.6x         |\n| DeeplabV3Plus_ResNet50      | 6.1          | 1.1                    | 0.8                    | 7.6x         |\n| DeeplabV3Plus_ResNet34      | 4.7          | 0.9                    | 0.8                    | 5.9x         |\n| UNet_ResNet50               | 6.2          | 1.3                    | 1                      | 6.2x         |\n| UNet_ResNet34               | 4.3          | 1.1                    | 0.8                    | 5.4x         |\n| FPN_ResNet50                | 5.5          | 1.2                    | 1                      | 5.5x         |\n| FPN_ResNet34                | 4.2          | 1.1                    | 1                      | 4.2x         |\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_019ad3ad5e8c.png\" \u002F>\n\u003C\u002Fp>\n\n# 🤗 Accelerating Huggingface Models using voltaML \n\nWe're adding support to accelerate Huggingface NLP models with **voltaML**. This work has been inspired from [ELS-RD's](https:\u002F\u002Fgithub.com\u002FELS-RD\u002Ftransformer-deploy) work. This is still in the early stages and only few models listed in the below table are supported. We're working to add more models soon.\n\n```python\nfrom voltaml.compile import VoltaNLPCompile\nfrom voltaml.inference import nlp_performance\n\n\nmodel='bert-base-cased'\nbackend=[\"tensorrt\",\"onnx\"] \nseq_len=[1, 1, 1] \ntask=\"classification\"\nbatch_size=[1,1,1]\n\nVoltaNLPCompile(model=model, device='cuda', backend=backend, seq_len=seq_len)\n\nnlp_performance(model=model, device='cuda', backend=backend, seq_len=seq_len)\n\n```\n\n![Pytorch (ms) and VoltaML FP16 (ms)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_0c8663e58c4b.png)\n\n| Model                                           | Pytorch (ms) | VoltaML FP16 (ms) | SpeedUp |\n|-------------------------------------------------|--------------|-------------------|---------|\n| bert-base-uncased                               | 6.4          | 1                 | 6.4x     |\n| Jean-Baptiste\u002Fcamembert-ner                     | 6.3          | 1                 | 6.3x     |\n| gpt2                                            | 6.6          | 1.2               | 5.5x     |\n| xlm-roberta-base                                | 6.4          | 1.08              | 5.9x     |\n| roberta-base                                    | 6.6          | 1.09              | 6.1x     |\n| bert-base-cased                                 | 6.2          | 0.9               | 6.9x     |\n| distilbert-base-uncased                         | 3.5          | 0.6               | 5.8x     |\n| roberta-large                                   | 11.9         | 2.4               | 5.0x     |\n| deepset\u002Fxlm-roberta-base-squad2                 | 6.2          | 1.08              | 5.7x     |\n| cardiffnlp\u002Ftwitter-roberta-base-sentiment       | 6            | 1.07              | 5.6x     |\n| sentence-transformers\u002Fall-MiniLM-L6-v2          | 3.2          | 0.42              | 7.6x     |\n| bert-base-chinese                               | 6.3          | 0.97              | 6.5x     |\n| distilbert-base-uncased-finetuned-sst-2-english | 3.4          | 0.6               | 5.7x     |\n| albert-base-v2                                  | 6.7          | 1                 | 6.7x     |\n\n\n# voltaTrees ⚡🌴 -> [Link](https:\u002F\u002Fgithub.com\u002FVoltaML\u002Fvolta-trees)\n\nA LLVM-based compiler for XGBoost and LightGBM decision trees.\n\n`voltatrees` converts trained XGBoost and LightGBM models to optimized machine code, speeding-up prediction by ≥10x.\n\n## Example\n\n```python\nimport voltatrees as vt\n\nmodel = vt.XGBoostRegressor.Model(model_file=\"NYC_taxi\u002Fmodel.txt\")\nmodel.compile()\nmodel.predict(df)\n```\n\n## Installation\n\n```bash\ngit clone git clone https:\u002F\u002Fgithub.com\u002FVoltaML\u002Fvolta-trees.git\ncd volta-trees\u002F\npip install -e .\n```\n\n## Benchmarks\n\nOn smaller datasets, voltaTrees is 2-3X faster than Treelite by DMLC. Testing on large scale dataset is yet to be conducted.\n\n### Enterpise Platform 🛣️\n\nAny enterprise customers who would like a fully managed solution hosted on your own cloud, please contact us at \u003Charish@voltaml.com>\n\n- [x] Fully managed and cloud-hosted optimization engine.\n- [x] Hardware targeted optimised dockers for maximum performance.\n- [ ] One-click deployment of the compiled models. \n- [ ] Cost-benefit analysis dashboard for optimal deployment.\n- [ ] NVIDIA Triton optimzed dockers for large-scale GPU deployment.\n- [ ] Quantization-Aware-Training (QAT) \n","\u003Cp align=\"center\">\n  \u003Cimg width=\"600\" height=\"120\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_ff5c1ae61154.jpg\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n\u003Cimg width=\"141\" alt=\"Screenshot 2022-10-19 at 3 55 14 PM\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_4d17d2295af0.png\">\n\u003C\u002Fp>\n  \n\u003Cp align=\"center\">\n  \u003Cb> 将您的机器学习和深度学习模型加速高达10倍 \u003C\u002Fb> \n\u003C\u002Fp>\n\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FpY5SVyHmWm\"> \u003Cimg src=\"https:\u002F\u002Fdcbadge.vercel.app\u002Fapi\u002Fserver\u002FpY5SVyHmWm\" \u002F> \u003C\u002Fa> \n\u003C\u002Fdiv>\n\n\u003Cdiv align=\"center\">\n\n### **🔥更新：[Stable-Diffusion\u002FDreamBooth加速](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML-fast-stable-diffusion)。推理速度最高可提升2.5倍🔥**\n\u003C\u002Fdiv>\n\n\u003Cp style=\"text-align: center;\"> \u003C\u002Fp>\n\n**voltaML** 是一个开源的轻量级库，用于加速您的机器学习和深度学习模型。voltaML 可以仅通过 ***一行代码*** 就对您的模型进行优化、编译并部署到目标 CPU 和 GPU 设备上。\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_83070471d2b6.gif\" alt=\"animated\" \u002F>\n\u003C\u002Fp>\n\n\n#### 开箱即用支持\n\n\n✅ FP16量化\u003Cbr\u002F>\n\n✅ Int8量化*\u003Cbr\u002F>\n\n✅ 针对特定硬件的编译\n\n\u003Cbr>\n\n\u003Cimg width=\"1100\" alt=\"Screenshot 2022-10-17 at 12 06 26 PM\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_303a3e636684.png\">\n\u003Cbr>\n\n**voltaML 支持以下编译：**\n\n\n\u003Cimg width=\"1102\" alt=\"Screenshot 2022-06-13 at 3 43 03 PM\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_f7e5851e8301.png\">\n\n\n## 安装\n\n### 自行搭建：\n\n要求：\n\n* CUDA 版本 >11.x \u003Cbr\u002F>\n* TensorRT == 8.4.1.2\u003Cbr\u002F>\n* PyTorch == 1.12 cu11.x\u003Cbr\u002F>\n* NVIDIA 驱动版本 > 510\n\n````\ngit clone https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML.git\ncd voltaML\npython setup.py install\n````\n### Docker 容器 🐳\n````\ndocker pull voltaml\u002Fvoltaml:v0.4\ndocker run -it --gpus=all -p \"8888:8888\" voltaml\u002Fvoltaml:v0.4 \\ \n        jupyter lab --port=8888 --no-browser --ip 0.0.0.0 --allow-root\n````\n## 使用方法\n\n```python\nimport torch\nfrom voltaml.compile import VoltaGPUCompiler, VoltaCPUCompiler, TVMCompiler\nfrom voltaml.inference import gpu_performance\n\nmodel = torch.load(\"path\u002Fto\u002Fmodel\u002Fdir\")\n\n# 通过提供路径来编译模型\ncompiler = VoltaGPUCompiler(\n        model=model,\n        output_dir=\"destination\u002Fpath\u002Fof\u002Fcompiled\u002Fmodel\",\n        input_shape=(1, 3, 224, 224), # 示例输入形状\n        precision=\"fp16\" # 指定精度[fp32, fp16, int8] - 仅适用于 GPU 编译器\n        target=\"llvm\" # 指定目标设备 - 仅适用于 TVM 编译器\n    )\n\n# 返回编译后的模型\ncompiled_model = compiler.compile()\n\n# 计算并比较性能\ngpu_performance(compiled_model, model, input_shape=(1, 3, 224, 224))\ncpu_performance(compiled_model, model, compiler=\"voltaml\", input_shape=(1, 3, 224, 224))\ncpu_performance(compiled_model, model, compiler=\"tvm\", input_shape=(1, 3, 224, 224))\n\n```\n## 笔记本\n\n01. [ResNet-50](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FResNet50%20Classification%20Demo.ipynb) 图像分类 \n02. [DeeplabV3_MobileNet_v3_Large](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FDeeplabV3%20Segmentation%20Demo.ipynb) 分割\n03. [YOLOv5](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FYoloV5%20Demo.ipynb) 目标检测 YOLOv5\n04. [YOLOv6](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FYoloV6%20Demo.ipynb) 目标检测 YOLOv6 \n05. [Bert_Base_Uncased](https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fblob\u002Fmain\u002Fdemo%20notebooks\u002FBert_base_uncased_HuggingFace.ipynb) Huggingface\n\n\n## 基准测试\n### 🖼️ 分类模型推理延迟（在GPU上）⏱️\n分类任务在Imagenet数据集上进行，`batch size = 1`且`imagesize = 224`，使用NVIDIA RTX 2080Ti。就`int8`模型的前1%和前5%准确率而言，我们未观察到超过1%的准确率下降。\n\n![Pytorch (ms), VoltaGPU FP16 (ms) and VoltaGPU int8 (ms)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_d44e6c5a930f.png)\n\n\n| 模型         | Pytorch (ms) | VoltaGPU FP16 (ms) | VoltaGPU int8 (ms) | Pytorch vs Int8 速度 |\n|---------------|--------------|--------------------|--------------------|-----------------------|\n| squeezenet1_1 | 1.6          | 0.2                | 0.2                | 8.4x                  |\n| resnet18      | 2.7          | 0.4                | 0.3                | 9.0x                  |\n| resnet34      | 4.5          | 0.7                | 0.5                | 9.0x                  |\n| resnet50      | 6.6          | 0.7                | 0.5                | 13.2x                 |\n| resnet101     | 13.6         | 1.3                | 1.0                | 13.6x                 |\n| densenet121   | 15.7         | 2.4                | 2.0                | 7.9x                  |\n| densenet169   | 22.0         | 4.4                | 3.8                | 5.8x                  |\n| densenet201   | 26.8         | 6.3                | 5.0                | 5.4x                  |\n| vgg11         | 2.0          | 0.9                | 0.5                | 4.0x                  |\n| vgg16         | 3.5          | 1.2                | 0.7                | 5.0x                  |\n\n### 🧐 目标检测（YOLO）模型推理延迟（在GPU上）⏱️\n目标检测推理是在NVIDIA RTX 2080Ti上使用`imagesize = 640`和`batch size = 1`的虚拟数据进行的。\n\n![Pytorch (ms) and VoltaGPU FP16 (ms)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_50dabbf0ceec.png)\n\n\n| 模型        | Pytorch (ms) | VoltaGPU FP16 (ms) | Pytorch vs FP16 速度 |\n|--------------|--------------|--------------------|-----------------------|\n| YOLOv5n      | 5.2          | 1.2                | 4.3x                  |\n| YOLOv5s      | 5.1          | 1.6                | 3.2x                  |\n| YOLOv5m      | 9.1          | 3.2                | 2.8x                  |\n| YOLOv5l      | 15.3         | 5.1                | 3.0x                  |\n| YOLOv5x      | 30.8         | 6.4                | 4.8x                  |\n| YOLOv6s      | 8.8          | 3.0                | 2.9x                  |\n| YOLOv6l_relu | 23.4         | 5.5                | 4.3x                  |\n| YOLOv6l      | 18.1         | 4.1                | 4.4x                  |\n| YOLOv6n      | 9.1          | 1.6                | 5.7x                  |\n| YOLOv6t      | 8.6          | 2.4                | 3.6x                  |\n| YOLOv5m      | 15.5         | 3.5                | 4.4x                  |\n\n### 🎨 分割模型推理延迟（在GPU上）⏱️\n分割推理是在NVIDIA RTX 2080Ti上，使用`imagesize = 224`和`batch size = 1`的虚拟数据完成的。\n\n![Pytorch (毫秒)、VoltaGPU FP16 (毫秒)和VoltaGPU Int8 (毫秒)(1)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_8b9b0355ce2e.png)\n\n\n| 模型                       | Pytorch (毫秒) | VoltaGPU FP16 (毫秒)  | VoltaGPU Int8 (毫秒) | 加速倍数 (X) |\n|-----------------------------|--------------|------------------------|------------------------|--------------|\n| FCN_Resnet50                | 8.3          | 2.3                    | 1.8                    | 3.6x         |\n| FCN_Resnet101               | 14.7         | 3.5                    | 2.5                    | 5.9x         |\n| DeeplabV3_Resnet50          | 12.1         | 2.5                    | 1.3                    | 9.3x         |\n| DeeplabV3_Resnet101         | 18.7         | 3.6                    | 2.0                    | 9.4x         |\n| DeeplabV3_MobileNetV3_Large | 6.1          | 1.5                    | 0.8                    | 7.6x         |\n| DeeplabV3Plus_ResNet50      | 6.1          | 1.1                    | 0.8                    | 7.6x         |\n| DeeplabV3Plus_ResNet34      | 4.7          | 0.9                    | 0.8                    | 5.9x         |\n| UNet_ResNet50               | 6.2          | 1.3                    | 1                      | 6.2x         |\n| UNet_ResNet34               | 4.3          | 1.1                    | 0.8                    | 5.4x         |\n| FPN_ResNet50                | 5.5          | 1.2                    | 1                      | 5.5x         |\n| FPN_ResNet34                | 4.2          | 1.1                    | 1                      | 4.2x         |\n\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_019ad3ad5e8c.png\" \u002F>\n\u003C\u002Fp>\n\n# 🤗 使用voltaML加速Huggingface模型\n\n我们正在添加对Huggingface NLP模型进行加速的支持，使用**voltaML**。这项工作受到了[ELS-RD](https:\u002F\u002Fgithub.com\u002FELS-RD\u002Ftransformer-deploy)工作的启发。目前仍处于早期阶段，仅支持下表中列出的少数几个模型。我们正努力尽快增加更多模型。\n\n```python\nfrom voltaml.compile import VoltaNLPCompile\nfrom voltaml.inference import nlp_performance\n\n\nmodel='bert-base-cased'\nbackend=[\"tensorrt\",\"onnx\"] \nseq_len=[1, 1, 1] \ntask=\"classification\"\nbatch_size=[1,1,1]\n\nVoltaNLPCompile(model=model, device='cuda', backend=backend, seq_len=seq_len)\n\nnlp_performance(model=model, device='cuda', backend=backend, seq_len=seq_len)\n\n```\n\n![Pytorch (毫秒)和VoltaML FP16 (毫秒)](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_readme_0c8663e58c4b.png)\n\n| 模型                                           | Pytorch (毫秒) | VoltaML FP16 (毫秒) | 加速倍数 |\n|-------------------------------------------------|--------------|-------------------|---------|\n| bert-base-uncased                               | 6.4          | 1                 | 6.4x     |\n| Jean-Baptiste\u002Fcamembert-ner                     | 6.3          | 1                 | 6.3x     |\n| gpt2                                            | 6.6          | 1.2               | 5.5x     |\n| xlm-roberta-base                                | 6.4          | 1.08              | 5.9x     |\n| roberta-base                                    | 6.6          | 1.09              | 6.1x     |\n| bert-base-cased                                 | 6.2          | 0.9               | 6.9x     |\n| distilbert-base-uncased                         | 3.5          | 0.6               | 5.8x     |\n| roberta-large                                   | 11.9         | 2.4               | 5.0x     |\n| deepset\u002Fxlm-roberta-base-squad2                 | 6.2          | 1.08              | 5.7x     |\n| cardiffnlp\u002Ftwitter-roberta-base-sentiment       | 6            | 1.07              | 5.6x     |\n| sentence-transformers\u002Fall-MiniLM-L6-v2          | 3.2          | 0.42              | 7.6x     |\n| bert-base-chinese                               | 6.3          | 0.97              | 6.5x     |\n| distilbert-base-uncased-finetuned-sst-2-english | 3.4          | 0.6               | 5.7x     |\n| albert-base-v2                                  | 6.7          | 1                 | 6.7x     |\n\n\n# voltaTrees ⚡🌴 -> [链接](https:\u002F\u002Fgithub.com\u002FVoltaML\u002Fvolta-trees)\n\n一个基于LLVM的XGBoost和LightGBM决策树编译器。\n\n`voltatrees`将训练好的XGBoost和LightGBM模型转换为优化后的机器代码，使预测速度提升≥10倍。\n\n## 示例\n\n```python\nimport voltatrees as vt\n\nmodel = vt.XGBoostRegressor.Model(model_file=\"NYC_taxi\u002Fmodel.txt\")\nmodel.compile()\nmodel.predict(df)\n```\n\n## 安装\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FVoltaML\u002Fvolta-trees.git\ncd volta-trees\u002F\npip install -e .\n```\n\n## 基准测试\n\n在较小的数据集上，voltaTrees比DMLC的Treelite快2-3倍。大规模数据集上的测试尚未进行。\n\n### 企业平台 🛣️\n\n任何希望在您自己的云上托管完全管理解决方案的企业客户，请联系我们：\u003Charish@voltaml.com>\n\n- [x] 完全管理和云端托管的优化引擎。\n- [x] 针对硬件优化的Docker镜像，以实现最大性能。\n- [ ] 一键部署编译后的模型。\n- [ ] 最优部署的成本效益分析仪表板。\n- [ ] 针对大规模GPU部署优化的NVIDIA Triton Docker镜像。\n- [ ] 量化感知训练（QAT）","# voltaML 快速上手指南\n\n## 环境准备\n- **操作系统**：Linux（推荐 Ubuntu 20.04+）\n- **CUDA**：11.x（>11.0）\n- **NVIDIA 驱动**：≥ 510\n- **Python**：3.8+\n- **依赖库**：\n  - TensorRT == 8.4.1.2\n  - PyTorch == 1.12+cu11.x\n\n## 安装步骤\n\n### 1. 本地安装\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML.git\ncd voltaML\npython setup.py install\n```\n\n### 2. Docker 一键启动（推荐）\n```bash\ndocker pull voltaml\u002Fvoltaml:v0.4\ndocker run -it --gpus=all -p \"8888:8888\" voltaml\u002Fvoltaml:v0.4 \\\n        jupyter lab --port=8888 --no-browser --ip 0.0.0.0 --allow-root\n```\n浏览器访问 `http:\u002F\u002Flocalhost:8888` 即可使用。\n\n## 基本使用\n\n### 1. 编译并加速 PyTorch 模型\n```python\nimport torch\nfrom voltaml.compile import VoltaGPUCompiler\nfrom voltaml.inference import gpu_performance\n\n# 加载模型\nmodel = torch.load(\"path\u002Fto\u002Fmodel\u002Fdir\")\n\n# 一键编译\ncompiler = VoltaGPUCompiler(\n    model=model,\n    output_dir=\"compiled_model\",\n    input_shape=(1, 3, 224, 224),  # 根据实际输入调整\n    precision=\"fp16\"               # fp32 \u002F fp16 \u002F int8\n)\ncompiled_model = compiler.compile()\n\n# 性能对比\ngpu_performance(compiled_model, model, input_shape=(1, 3, 224, 224))\n```\n\n### 2. 加速 Hugging Face 模型\n```python\nfrom voltaml.compile import VoltaNLPCompile\nfrom voltaml.inference import nlp_performance\n\n# 一行代码加速 BERT\nVoltaNLPCompile(\n    model='bert-base-chinese',\n    device='cuda',\n    backend=['tensorrt', 'onnx'],\n    seq_len=[1, 1, 1],\n    batch_size=[1, 1, 1]\n)\n\n# 查看加速效果\nnlp_performance(\n    model='bert-base-chinese',\n    device='cuda',\n    backend=['tensorrt'],\n    seq_len=[1, 1, 1]\n)\n```\n\n### 3. 决策树加速（voltatrees）\n```bash\npip install voltatrees\n```\n```python\nimport voltatrees as vt\nmodel = vt.XGBoostRegressor.Model(model_file=\"model.txt\")\nmodel.compile()\npred = model.predict(df)   # ≥10× 提速\n```\n\n完成！现在你可以用 **一行代码** 把任何 PyTorch \u002F Hugging Face \u002F 决策树模型提速 5-10 倍。","一家 20 人的 AI 初创公司正在做“实时菜品识别”小程序，用户上传手机照片，后端需在 200 ms 内返回菜品名称与热量信息，模型是 ResNet-50，部署在 2 张 RTX 3080 的云服务器上。\n\n### 没有 voltaML 时\n- 直接用 PyTorch 推理，单张图片平均 480 ms，远超 200 ms 目标，高峰期排队严重，用户投诉“转圈圈”。  \n- 为了压延时，工程师手写 TensorRT 插件，调 FP16、INT8、算子融合，折腾两周，代码 600+ 行，仍因版本不匹配崩溃两次。  \n- GPU 利用率只有 45%，被迫再开 2 台实例，月账单从 800 美元涨到 2000 美元，投资人开始皱眉。  \n- 模型更新后（新增 30 种菜品），又得重新走一遍 TensorRT 编译流程，上线周期被拉长到 5 天。\n\n### 使用 voltaML 后\n- 三行代码把 ResNet-50 编译成 TensorRT FP16，单图延迟降到 42 ms，直接满足 200 ms SLA，高峰期零排队。  \n- 无需手写插件，voltaML 自动完成算子融合与 INT8 校准，工程师把两周压缩成 30 分钟，专注业务逻辑。  \n- GPU 利用率飙到 85%，两台服务器直接下线，月账单回到 800 美元，投资人重新露出笑容。  \n- 新增菜品只需重新执行一次 `compiler.compile()`，30 分钟完成测试上线，版本迭代从 5 天缩短到 1 天。\n\nvoltaML 让这家初创公司在不增硬件、不加班的前提下，把推理速度提升 10 倍，成本砍半，迭代周期缩短 80%。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FVoltaML_voltaML_d44e6c5a.png","VoltaML","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FVoltaML_956cbcad.png",null,"https:\u002F\u002Fgithub.com\u002FVoltaML",[80,84,88,92],{"name":81,"color":82,"percentage":83},"Python","#3572A5",97.6,{"name":85,"color":86,"percentage":87},"Jupyter Notebook","#DA5B0B",2.1,{"name":89,"color":90,"percentage":91},"Shell","#89e051",0.2,{"name":93,"color":94,"percentage":95},"Dockerfile","#384d54",0.1,1180,39,"2026-03-24T11:33:56","Apache-2.0",4,"Linux","必需 NVIDIA GPU，CUDA 11.x（>11.0），TensorRT 8.4.1.2，NVIDIA Driver ≥510","未说明",{"notes":105,"python":103,"dependencies":106},"官方仅提供 Linux 安装说明；Docker 镜像已集成所有依赖，可直接运行；CPU 编译支持 TVM，但 GPU 编译需 TensorRT；示例默认输入 shape 为 (1,3,224,224)",[107,108],"torch==1.12+cu11.x","tensorrt==8.4.1.2",[13,14],"2026-03-27T02:49:30.150509","2026-04-06T09:43:38.429160",[113,118,123,128,133,138],{"id":114,"question_zh":115,"answer_zh":116,"source_url":117},6059,"为什么 CPU 上加速明显，GPU 上反而变慢？","这是已知现象：VoltaNLPCompile 默认针对 CPU 做了大量优化（量化、ONNX 转换），在 GPU 上反而可能 30 倍慢于原生 ONNX。目前官方没有 GPU 专用优化路径。若必须 GPU 训练、CPU 推理且保持 embedding 一致，可尝试：1) 在 GPU 阶段仍用原始 Hugging Face 模型生成 embedding；2) 仅在 CPU 推理阶段使用 VoltaML 优化后的模型；3) 关注 onnxruntime 的 issue #10267，等待官方修复 GPU 性能回退。","https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fissues\u002F10",{"id":119,"question_zh":120,"answer_zh":121,"source_url":122},6060,"本地开发环境安装总是失败，如何解决？","直接本地 pip 安装会遇到大量依赖冲突。官方推荐：使用官方 Docker 镜像启动容器进行开发，可一次性解决 TensorRT、PyTorch、ONNX 等版本兼容问题。若坚持本地安装，可参考用户整理的依赖列表（见 issue #5 正文），但仍需手动处理 transformer_deploy 等缺失包，成功率低。","https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fissues\u002F5",{"id":124,"question_zh":125,"answer_zh":126,"source_url":127},6061,"项目还在维护吗？","原 voltaML 仓库已停止常规更新，团队目前专注于 Stable Diffusion 分支。如需最新功能，请转到 https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML-fast-stable-diffusion\u002Ftree\u002Fexperimental 获取活跃开发版本。","https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fissues\u002F11",{"id":129,"question_zh":130,"answer_zh":131,"source_url":132},6062,"Windows 系统能运行吗？","可以，但需通过 WSL（Windows Subsystem for Linux）环境运行。原生 Windows 暂不支持。","https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fissues\u002F8",{"id":134,"question_zh":135,"answer_zh":136,"source_url":137},6063,"是否支持 OpenCV 模型？","当前仅支持 PyTorch 模型。OpenCV 模型暂不支持；若需求强烈，官方会考虑添加，也欢迎社区 Fork 后自行实现。","https:\u002F\u002Fgithub.com\u002FVoltaML\u002FvoltaML\u002Fissues\u002F2",{"id":139,"question_zh":140,"answer_zh":141,"source_url":127},6064,"已有本地模型，如何避免重复下载？","在 Docker 启动时挂载本地模型目录作为卷（volume），即可跳过在线下载，直接复用已有权重文件。",[]]