English | 简体中文
cube-studio is a one-stop cloud-native machine learning platform open sourced by Tencent Music, Currently mainly includes the following functions
- 1、data management: feature store, online and offline features; dataset management, structure data and media data, data label platform
- 2、develop: notebook(vscode/jupyter); docker image management; image build online
- 3、train: pipeline drag and drop online; open template market; distributed computing/training tasks, example tf/pytorch/mxnet/spark/ray/horovod/kaldi/volcano; batch priority scheduling; resource monitoring/alarm/balancing; cron scheduling
- 4、automl: nni, katib, ray
- 5、inference: model manager; serverless traffic control; tf/pytorch/onnx/tensorrt model deploy, tfserving/torchserver/onnxruntime/triton inference; VGPU; load balancing、high availability、elastic scaling
- 6、infra: multi-user; multi-project; multi-cluster; edge cluster mode; blockchain sharing;
https://github.com/tencentmusic/cube-studio/wiki
learning、deploy、consult、contribution、cooperation, join group, wechart id luanpeng1234 remark<open source>, construction guide
tips:
- 1、You can develop your own template, Easy to develop and more suitable for your own scenarios
| template | type | describe |
|---|---|---|
| linux | base | Custom stand-alone operating environment, free to implement all custom stand-alone functions |
| datax | import export | Import and export of heterogeneous data sources |
| media-download | data processing | Distributed download of media files |
| video-audio | data processing | Distributed extraction of audio from video |
| video-img | data processing | Distributed extraction of pictures from video |
| sparkjob | data processing | spark serverless |
| ray | data processing | python ray multi-machine distributed framework |
| volcano | data processing | volcano multi-machine distributed framework |
| xgb | machine learning | xgb model training and inference |
| ray-sklearn | machine learning | sklearn based on ray framework supports multi-machine distributed parallel computing |
| pytorchjob-train | model train | Multi-machine distributed training of pytorch |
| horovod-train | model train | Multi-machine distributed training of horovod |
| tfjob | model train | Multi-machine distributed training of tensorflow |
| tfjob-train | model train | distributed training of tensorflow: plain and runner |
| tfjob-runner | model train | distributed training of tensorflow: runner method |
| tfjob-plain | model train | distributed training of tensorflow: plain method |
| kaldi-train | model train | Multi-machine distributed training of kaldi |
| tf-model-evaluation | model evaluate | distributed model evaluation of tensorflow2.3 |
| tf-offline-predict | model inference | distributed offline model inference of tensorflow2.3 |
| model-offline-predict | model inference | distributed offline model inference of framework |
| deploy-service | model deploy | deploy inference service |
algorithm:
@hujunaifuture 
platform:
@xiaoyangmai 
