Releases: alibaba/EasyRec
Releases · alibaba/EasyRec
v0.4.2
fix some bugs.
release/0.4.0
Major Features and Improvements
- fix final exporter bug under ps-worker-evaluator mode;
- fix fg export bug for multi-task models;
- add dlrm models;
- add vector retrieve function;
- add interface to export checkpoints which could run on alibaba RTP(RealTime serving Platform);
- add support for hive table as data input;
- add support for export embedding variable(kv embedding) separately from dense variables;
- add support for multi optimizers and gradient freezing ability.
Download
release/0.3.0
Release 0.3.0
Major Features and Improvements
- Add support for sequence feature #62
- Update esmm loss function #76
- Add auc.num_thresholds setting #76
- Add Cold start model #80
- Add Collaborative Metric Learning i2i model #80
- Add TextCNN model #80
- Compat with DataScience enviroment #86
- Support offline inference on DataScience #100
- Add Unit Test #91
Bug Fixes and Other Changes
Documentation update
- Add document for Cold start model, Collaborative Metric Learning i2i model and TextCNN model #80
release/0.2.0
Release 0.2.0
Major Features and Improvements
- Support for early stop #1
- Support for stop by num_epoch #5
- Release MultiTask model PLE #8
- Compatible with feature_configs and feature_config.features #48
- import fg_json to easyrec_config #54
- upgrade params in feature config to higher precision #65
- add write_graph option to train config. #65
- Compatible with FG dtype and EasyRec dtype #66
Bug Fixes and Other Changes
- fix hyperparams_builder bug in tf2 #64
- fix default value bug in inference #66
- fix workqueue dtype bug in OdpsRTPInput #67
Documentation update
release/0.1.4
add more models and data processing functions.
release/0.1.0
initial release.