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!