模型太大如何用easyrec保存不带embedding并在serving的时候直接输入embedding的模型？

模型训练时输入是userId/ItemId这些特征，训练好后模型很大，想把训练好的模型里把Embedding层单独拿出来存Redis，然后推理的时候从redis查询embedding后直接给模型serving推理，这个用easyrec如何实现导出这样的模型？
参考：[embedding层单独拿出来存Redis，Redis放不下就可以放Hbase](https://www.zhihu.com/question/354086469/answer/894235805)

The issue names no files, tests, or EasyRec entry points. Start by locating the model export and serving documentation, then determine whether external embeddings from Redis are supported. Done should be a confirmed, documented workflow for exporting a model without its embedding layer and supplying embeddings during serving.

Written by the indexing model from the issue text.