youtubednn部署报错

[kinghuin]

Describe the bug(问题描述) hi，这是一个很棒的项目，我在 youtubednn user tower 部署完成后，报错

(Pdb) stub.Predict(request)
*** grpc._channel._InactiveRpcError: <_InactiveRpcError of RPC that terminated with:
        status = StatusCode.INVALID_ARGUMENT
        details = "ConcatOp : Expected concatenating dimensions in the range [-2, 2), but got 2
         [[{{node model_1/concatenate/concat}}]]"
        debug_error_string = "{"created":"@165693","description":"Error received from peer ipv4:*******:80**","file":"src/core/lib/surface/call.cc","file_line":952,"grpc_message":"ConcatOp : Expected concatenating dimensions in the range [-2, 2), but got 2\n\t [[{{node model_1/concatenate/concat}}]]","grpc_status":3}"

期待项目能够完善一下模型部署部分

To Reproduce(复现步骤) Steps to reproduce the behavior:

1. 参考https://github.com/shenweichen/DeepMatch/issues/46#issuecomment-1038597409 修复deepctr的模型导出bug
2. 参考https://github.com/shenweichen/DeepMatch/blob/v0.2.1/examples/run_youtubednn.py#L81 划出用户塔
3. 如果使用tf.keras.models.save_model(user_embedding_model,"./export_model/youtubeuser/1") 则在导出阶段就报错

TypeError: Could not build a TypeSpec for <KerasTensor: shape=(None, 15) dtype=float32 (created by layer 'tf.compat.v1.squeeze')> with type KerasTensor

如果使用tf.saved_model.save(obj=user_embedding_model, export_dir='./export_model/youtube_dnn_user_tower/1',signatures=None) 则顺利导出，但在部署好以后，报错

(Pdb) stub.Predict(request)
*** grpc._channel._InactiveRpcError: <_InactiveRpcError of RPC that terminated with:
        status = StatusCode.INVALID_ARGUMENT
        details = "ConcatOp : Expected concatenating dimensions in the range [-2, 2), but got 2
         [[{{node model_1/concatenate/concat}}]]"
        debug_error_string = "{"created":"@165693","description":"Error received from peer ipv4:*******:80**","file":"src/core/lib/surface/call.cc","file_line":952,"grpc_message":"ConcatOp : Expected concatenating dimensions in the range [-2, 2), but got 2\n\t [[{{node model_1/concatenate/concat}}]]","grpc_status":3}"

怀疑是建模时使用了一些不适合部署的操作

Operating environment(运行环境):

• python 3.7
• tensorflow 1.15.5
• deepmatch 0.2.0

Additional context Add any other context about the problem here.