{"full_name":"Tencent/ncnn","owner":"Tencent","name":"ncnn","shard":"1","url":"https://github.com/Tencent/ncnn","description":"ncnn is a high-performance neural network inference framework optimized for the mobile platform","homepage":"","topics":["inference","high-preformance","simd","arm-neon","deep-learning","artificial-intelligence","android","ios","ncnn","vulkan","neural-network","caffe","mxnet","pytorch","onnx"],"tag_pages":["android","artificial-intelligence","caffe","deep-learning","inference","ios","mxnet","ncnn","neural-network","onnx","pytorch","simd","vulkan"],"language":"C++","license":"NOASSERTION","avatar":"https://avatars.githubusercontent.com/u/18461506?v=4","created_at":"2017-06-30T10:55:37Z","pushed_at":"2026-11-13T11:58:26Z","metrics":{"stars":24052,"forks":4551,"watchers":566,"issues":3916,"prs":3195,"releases":50,"issues_open":1121,"issues_closed":2795,"prs_open":202,"prs_closed":650,"prs_merged":2343,"discussions":108},"growth":{"day":{"stars":1,"forks":-1,"base_day":"2026-11-15"},"week":{"stars":29,"forks":12,"base_day":"2026-11-04"},"month":{"stars":110,"forks":29,"base_day":"2026-10-17"}},"history":{"dates":["2026-09-06","2026-09-07","2026-09-08","2026-09-09","2026-09-10","2026-09-11","2026-09-12","2026-09-13","2026-09-14","2026-09-15","2026-09-16","2026-09-17","2026-09-18","2026-09-19","2026-09-20","2026-09-21","2026-09-22","2026-09-23","2026-09-24","2026-09-25","2026-09-26","2026-09-27","2026-09-28","2026-09-29","2026-09-30","2026-10-01","2026-10-02","2026-10-03","2026-10-04","2026-10-05","2026-10-06","2026-10-07","2026-10-08","2026-10-09","2026-10-10","2026-10-11","2026-10-12","2026-10-13","2026-10-14","2026-10-15","2026-10-16","2026-10-17","2026-10-18","2026-10-19","2026-10-20","2026-10-21","2026-10-22","2026-10-23","2026-10-24","2026-10-25","2026-10-26","2026-10-27","2026-10-28","2026-10-29","2026-10-30","2026-10-31","2026-11-01","2026-11-02","2026-11-03","2026-11-04","2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[23738,23742,23752,23755,23757,23762,23763,23776,23779,23784,23784,23786,23788,23791,23796,23800,23802,23807,23813,23823,23826,23830,23840,23844,23852,23861,23865,23869,23871,23873,23881,23895,23907,23905,23905,23908,23908,23916,23923,23928,23933,23942,23952,23956,23960,23965,23969,23977,23983,23986,23990,23994,23994,24002,24004,24008,24007,24011,24023,24023,24033,24038,24044,24045,24050,24051,24052],"forks":[4489,4492,4492,4492,4493,4493,4494,4495,4496,4496,4497,4497,4496,4497,4497,4499,4500,4502,4504,4504,4504,4504,4503,4505,4505,4505,4505,4506,4506,4506,4506,4506,4511,4514,4514,4517,4517,4517,4517,4519,4522,4522,4523,4523,4526,4526,4528,4531,4533,4532,4531,4532,4532,4531,4531,4532,4532,4533,4538,4539,4544,4545,4549,4549,4551,4552,4551],"watchers":[569,569,569,569,569,569,569,570,570,570,570,570,571,571,571,571,571,571,571,570,570,570,570,570,570,570,570,570,570,570,570,570,570,570,570,570,570,570,570,569,569,569,569,569,569,569,569,569,569,568,568,568,566,566,566,566,566,566,566,566,null,566,566,566,566,566,566],"issues_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,1117,1120,1120,1121,1121,1121],"prs_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,190,197,198,201,203,202]},"boards":{},"same_owner":[{"full_name":"Tencent/AI-Infra-Guard","name":"AI-Infra-Guard","shard":"1","boards":{}},{"full_name":"Tencent/BrowserSkill","name":"BrowserSkill","shard":"1","boards":{}},{"full_name":"Tencent/LoopForge","name":"LoopForge","shard":"1","boards":{"daily":12,"weekly":86,"daily_rate":1,"weekly_rate":1,"monthly_rate":5,"daily_forks":35,"daily_watchers":190}},{"full_name":"Tencent/MMKV","name":"MMKV","shard":"1","boards":{}},{"full_name":"Tencent/QMUI_Android","name":"QMUI_Android","shard":"1","boards":{}},{"full_name":"Tencent/TNN","name":"TNN","shard":"1","boards":{}},{"full_name":"Tencent/WeKnora","name":"WeKnora","shard":"1","boards":{"monthly":169}},{"full_name":"Tencent/cherry-markdown","name":"cherry-markdown","shard":"1","boards":{}},{"full_name":"Tencent/lemon-cleaner","name":"lemon-cleaner","shard":"1","boards":{}},{"full_name":"Tencent/libpag","name":"libpag","shard":"1","boards":{}},{"full_name":"Tencent/mars","name":"mars","shard":"1","boards":{}},{"full_name":"Tencent/matrix","name":"matrix","shard":"1","boards":{}},{"full_name":"Tencent/omi","name":"omi","shard":"1","boards":{}},{"full_name":"Tencent/puerts","name":"puerts","shard":"1","boards":{}},{"full_name":"Tencent/teamai-cli","name":"teamai-cli","shard":"1","boards":{}},{"full_name":"Tencent/tinker","name":"tinker","shard":"1","boards":{}},{"full_name":"Tencent/tmagic-editor","name":"tmagic-editor","shard":"1","boards":{}},{"full_name":"Tencent/vConsole","name":"vConsole","shard":"1","boards":{}},{"full_name":"Tencent/wepy","name":"wepy","shard":"1","boards":{}}],"issues":[{"number":3811,"title":"onnx2ncnn 失败","comments":33,"created_at":"2022-05-20T03:02:37Z"},{"number":5,"title":"后续是否会支持将 tensorflow pb模型转为ncnn的模型呢?","comments":30,"created_at":"2017-07-24T07:26:54Z"},{"number":4811,"title":"Convolution1D and Deconvolution1D layers","comments":20,"created_at":"2023-06-18T11:44:39Z"},{"number":3466,"title":"Fail to convert yolov5 (5.0 or 6.0) to ncnn model","comments":18,"created_at":"2021-12-29T12:23:31Z"},{"number":4679,"title":"行人重识别这个项目模型转为ONNX在转为NCNN后识别错误。","comments":17,"created_at":"2023-04-28T09:12:56Z"},{"number":3866,"title":"语音分帧的流式处理","comments":17,"created_at":"2022-05-31T07:20:10Z"},{"number":1747,"title":"修改backbone后的Yolact模型转ncnn，输出错误","comments":17,"created_at":"2020-05-09T07:11:35Z"},{"number":3846,"title":"in.substract_mean_normalize(mean_vals, norm_vals)报告空指针异常","comments":16,"created_at":"2022-05-26T13:59:07Z"}],"discussions":[{"number":7238,"title":"YOLOVAL: Arena Breakout ONNX-to-ncnn CPU check + 32 examples (paid models)","comments":0,"category":"Show and tell","created_at":"2026-11-04T19:18:54Z"},{"number":7130,"title":"DAT2 transformer (4x UltraSharp V2) running fully on the Android GPU with ncnn + Vulkan","comments":0,"category":"Show and tell","created_at":"2026-10-16T21:24:40Z"},{"number":7004,"title":"SeedVR2-ncnn：SeedVR2 的 ncnn Vulkan C++ 实现与低显存实验","comments":0,"category":"Show and tell","created_at":"2026-09-14T12:58:11Z"},{"number":7002,"title":"【腾讯犀牛鸟2026】ERNIE-Image ncnn Vulkan Port","comments":0,"category":"Show and tell","created_at":"2026-09-14T07:00:25Z"},{"number":7001,"title":"Seedvr2在ncnn-valkan后端上的推理实现","comments":0,"category":"Show and tell","created_at":"2026-09-13T17:58:56Z"},{"number":6999,"title":"[腾讯犀牛鸟2026 #6790] Qwen3-ASR-0.6B ncnn C++ 端到端移植与跨平台验证","comments":0,"category":"Show and tell","created_at":"2026-09-13T15:01:39Z"},{"number":6998,"title":"【腾讯犀牛鸟2026】ERNIE-Image ncnn Vulkan Port","comments":0,"category":"Show and tell","created_at":"2026-09-13T14:31:25Z"},{"number":6997,"title":"[2026 犀牛鸟] ncnn现已支持wan 2.2 TI2V 5B","comments":0,"category":"Show and tell","created_at":"2026-09-12T11:57:52Z"}],"latest_release":{"name":"android harmonyos ios macos linux windows webassembly watchos tvos visionos 预编译库 20261020 7e36d25","tag":"20261020","published_at":"2026-10-20T03:40:10Z","body":"编译版本，默认配置，android-ndk-r30，ohos-sdk-5.0.3，xcode 16.4，ubuntu-22.04，ubuntu-24.04，ubuntu-26.04，vs2015，vs2017，vs2019，vs2022，vs2026，emscripten-6.0.12\r\n| file | content | arch |\r\n|---|---|---|\r\n|ncnn-full-source.zip |包含全部 submodule 代码的完整源码 | | \r\n|ncnn-android.zip | android 静态库/动态库 | armeabi-v7a + arm64-v8a + x86 + x86_64 + riscv64 |\r\n|ncnn-android-vulkan.zip | android 静态库/动态库，支持 GPU | armeabi-v7a + arm64-v8a + x86 + x86_64 + riscv64 |\r\n|ncnn-harmonyos.zip | harmonyos 静态库/动态库 | armeabi-v7a + arm64-v8a + x86_64 |\r\n|ncnn-harmonyos-vulkan.zip | harmonyos 静态库/动态库，支持 GPU | armeabi-v7a + arm64-v8a + x86_64 |\r\n|ncnn-apple.zip | apple xcframework，ios + ios-simulator + macos + mac-catalyst + watchos + watchos-simulator + tvos + tvos-simulator + visionos + visionos-simulator | arm64 + arm64e + x86_64 |\r\n|ncnn-apple-vulkan.zip | apple xcframework，ios + ios-simulator + macos + mac-catalyst + watchos + watchos-simulator + tvos + tvos-simulator + visionos + visionos-simulator，支持 GPU | arm64 + arm64e + x86_64 |\r\n|ncnn-ios.zip | ios 静态库 | arm64 |\r\n|ncnn-ios-vulkan.zip | ios 静态库，支持 GPU | arm64 |\r\n|ncnn-ios-simulator.zip | ios simulator 静态库 | x86_64 + arm64 |\r\n|ncnn-ios-simulator-vulkan.zip | ios simulator 静态库，支持 GPU | x86_64 + arm64 |\r\n|ncnn-macos.zip | macos 静态库 | x86_64 + arm64 |\r\n|ncnn-macos-vulkan.zip | macos 静态库，支持 GPU | x86_64 + arm64 |\r\n|ncnn-mac-catalyst.zip | mac catalyst 静态库 | x86_64 + arm64 |\r\n|ncnn-mac-catalyst-vulkan.zip | mac catalyst 静态库，支持 GPU | x86_64 + arm64 |\r\n|ncnn-watchos.zip | watchos 静态库 | armv7k + arm64_32 |\r\n|ncnn-watchos-simulator.zip | watchos simulator 静态库 | x86_64 + arm64 |\r\n|ncnn-tvos.zip | tvos 静态库 | x86_64 + arm64 |\r\n|ncnn-tvos-vulkan.zip | tvos 静态库，支持 GPU | x86_64 + arm64 |\r\n|ncnn-tvos-simulator.zip | tvos simulator 静态库 | x86_64 + arm64 |\r\n|ncnn-tvos-simulator-vulkan.zip | tvos simulator 静态库，支持 GPU | x86_64 + arm64 |\r\n|ncnn-visionos.zip | visionos 静态库 | arm64 |\r\n|ncnn-visionos-vulkan.zip | visionos 静态库，支持 GPU | arm64 |\r\n|ncnn-visionos-simulator.zip | visionos simulator 静态库 | x86_64 + arm64 |\r\n|ncnn-visionos-simulator-vulkan.zip | visionos simulator 静态库，支持 GPU | x86_64 + arm64 |\r\n|ncnn-ubuntu.zip | ubuntu linux 静态库/动态库，支持 GPU，模型转换工具 | x86_64 |\r\n|ncnn-windows.zip | windows 静态库/动态库，支持 GPU，模型转换工具 | x86 + x64 + arm + arm64 |\r\n|ncnn-webassembly.zip | webassembly 静态库 | wasm32 + simd + threads + simd-threads |\r\n|ncnn-webassembly64.zip | webassembly64 静态库 | wasm64 + simd + threads + simd-threads |\r\n\r\n## 重点概览\r\n\r\n- 新增原生 batch 推理基础设施，Mat/VkMat、CPU/Vulkan 调度、C/Python 接口和 pnnx 转换同步扩展；Reshape 支持 batch 与普通维度之间的重组，进一步完善动态 reshape、flatten、unflatten 的 batch 语义。（@futz12）\r\n- Vulkan 后端新增 Convolution、ConvolutionDepthWise、InnerProduct、GEMM 的 int8 推理路径，配套 integer dot product、cooperative matrix、Winograd 和 int8 packing，量化模型不再因存在 int8 层而整体关闭 GPU。\r\n- 新增面向 LLM 的 int4/int6/int8 块量化、`ncnnllm2table` / `ncnnllm2int` 工具和五大 CPU 后端的 int8 块量化优化，并引入会话级 KV cache allocator、容量复用及 bf16 模型权重存储。\r\n- x86 后端新增 fp32/bf16 SDPA FlashAttention prefill/decode，扩展 AVX2、AVX-NE-CONVERT、FMA/FMA4 dispatch，并补齐 CumulativeSum、Reduction、Normalize、Permute 的 SIMD 实现。（@crafcat7）\r\n- RISC-V 补齐 packed int8 Convolution、im2col GEMM、Winograd、SDPA/MultiHeadAttention，继续优化 fp16 GEMM 和常见激活，并新增 BF16 能力检测及 Cast/Clip 转换路径。（@Deepdive543443, @futz12, @darkavatar23）\r\n- Vulkan LayerNorm、RMSNorm、GroupNorm、InstanceNorm 新增 subgroup 归约，cooperative matrix 继续优化访存、对齐和 Adreno 兼容；同步修复 bf16 舍入、归一化系数精度及 Tanh/GELU 大输入溢出问题。（@futz12）\r\n- 模型、参数与图加载增加系统性校验，新增默认开启的 `NCNN_VALIDATION`，修复负数索引、长度/数值溢出、非法层参数、循环图、截断 mmap 权重及失败加载后的资源清理问题。（@aleff-github, @professor-moody, @beilzx, @kaltsit-10, @ktz03, @Dr3ladwyn, @FluxMind-O, @lxcxjxhx）\r\n- pnnx 完善 MelSpectrogram、多前导维度 Linear、全局池化布局、ONNX fp16/bool 常量和 Transformer attention 融合，修复插值与动态 batch 转换，CI 测试矩阵扩展至 PyTorch 2.14。（@futz12, @Lfan-ke, @ktz03, @giuliocorradi, @Acture, @xy200303）\r\n- 发布流程新增 Ubuntu 26.04、VS2026、wasm64 和 Android/iOS Python wheel，WebAssembly 更新至 Emscripten 6.0.12 并使用原生 LLVM OpenMP；Python 的 OpenCV 依赖改为可选项。\r\n\r\n## Vulkan / GPU\r\n\r\n- 新增 Vulkan Convolution、ConvolutionDep","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}