MindSpore Recommender 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 recommendation models online learning and realize minute-level end-to-end incremental training + model update.
• Support HBM-DRAM-SSD multilevel distributed feature cache, realize TB level recommendation models training.

API Change

• 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:

lizhenyu, hewei, wangrui, hangangqiang, xupan, chengang, zhoupeichen, gaoyong, zhangxiaohan.

Contributions of any kind are welcome!