Release Notes

MindSpore Recommender 0.2.0 Release Notes

Major Features and Improvements

  • Support dynamic Embedding and real-time feature addition and deletion to realize dynamic feature admission and eviction as well as incremental recommendation models import and export.

  • Support recommended model online learning and realize minute-level end-to-end incremental training + model update.

  • Support HBM-DRAM-SSD multilevel distributed feature cache and realize TB-level recommendation models training.

API Changed

  • Add API mindspore_rec.RecModel

  • Add API mindspore_rec.HashEmbeddingLookup

Bug Fixes

  • Fix the put operation stuck after multiple put and erase operations on MapParameter.

Contributors

Thanks goes to these wonderful people: hewei, chengang, lizhenyu, wangrui, zhoupeichen, gaoyong, hangangqiang, xupan, zhangxiaohan. Contributions of any kind are welcome!