Right now, TensorFlow 2.x is used for production systems.

We are starting right now to unify the training processed and experiments with a focus on PyTorch for the experimental phase. Models will be exported, if required, to TF via onyx which is possible since we only use standard architectures for ConvNets (Inception or Residual Nets).

The minimal requirement version for CUDA is 10.x.

All components should be located in 'incubator' with a v5- prefix, these are the git repos for existing components:

• v5-backend: Database back-end https://git.picalike.corpex-kunden.de/incubator/v5-backend
• v5-extractor: Feature storage and extraction https://git.picalike.corpex-kunden.de/incubator/v5-extractor Extractor Wiki
• v5-sim-api: Front-end: https://git.picalike.corpex-kunden.de/incubator/v5-sim-api
• billing: https://git.picalike.corpex-kunden.de/picalike/billingv5
• middleware: https://git.picalike.corpex-kunden.de/picalike/billingv5middleware

HINT: Every git now in incubator or that is now prefixed with v5_ should be migrated.

• gpu03: the 2.2.0-gpu-jupyter is ready to use
• grapex: is blocked exclusively for the boss
• grapu: needs an update (nvidia+cuda+docker)