Have you solved the problem? No OpKernel was registered to support Op 'SegmentSum'

Question

Have you solved the problem? No OpKernel was registered to support Op 'SegmentSum'

painmoth opened this issue 8 years ago · 11 comments

painmoth commented 8 years ago

tensorflow/tensorflow#13897

hello，I have forked your project and encounter the same problem

Answer 1 · 2018-04-23T09:06:54.000Z

This problem has been fixed a long time ago. Please update your Tensorflow.

Answer 2 · 2018-04-23T09:12:33.000Z

But my tensorflow is up to date(1.7）

Answer 3 · 2018-04-23T09:17:38.000Z

Please put your errors here.

Answer 4 · 2018-04-23T09:33:29.000Z

I have tried 2 way, both have some errors, thanks
tensorflow-predictor-cpp/demo/deep_model]$sh ./predict.sh

#./predict.sh
export TF_CPP_MIN_LOG_LEVEL=3
#../../bin/deep_model.bin "9,6,116,152,179" "./data/dict.data" "./model/graph.pb"
../../bin/deep_model.bin "9,6,116,152,179" "./data/dict.data" "./model/freeze_graph.pb"

ERROR message:
Invalid argument: Multiple OpKernel registrations match NodeDef 'emb_179/embedding_lookup_sparse = SegmentSum[T=DT_FLOAT, Tindices=DT_INT32](emb_179/embedding_lookup_sparse/mul, emb_179/embedding_lookup_sparse/Cast)': 'op: "SegmentSum" device_type: "CPU" constraint { name: "T" allowed_values { list { type: DT_FLOAT } } } constraint { name: "Tindices" allowed_values { list { type: DT_INT32 } } }' and 'op: "SegmentSum" device_type: "CPU" constraint { name: "T" allowed_values { list { type: DT_FLOAT } } } constraint { name: "Tindices" allowed_values { list { type: DT_INT32 } } }'
[[Node: emb_179/embedding_lookup_sparse = SegmentSum[T=DT_FLOAT, Tindices=DT_INT32](emb_179/embedding_lookup_sparse/mul, emb_179/embedding_lookup_sparse/Cast)]]

#./predict.sh
export TF_CPP_MIN_LOG_LEVEL=3
../../bin/deep_model.bin "9,6,116,152,179" "./data/dict.data" "./model/graph.pb"
#../../bin/deep_model.bin "9,6,116,152,179" "./data/dict.data" "./model/freeze_graph.pb"

ERROR message:
Invalid argument: No OpKernel was registered to support Op 'L2Loss' with these attrs. Registered devices: [CPU], Registered kernels:

 [[Node: get_regularization_penalty_1/l2_regularizer_1/L2Loss = L2Loss[T=DT_FLOAT](outlayer/bias_out/read)]]

Answer 5 · 2018-04-23T10:21:41.000Z

Check whether tensorflow/core/kernels/segment_reduction_ops.cc exists in the file tensorflow/contrib/makefile/tf_op_files.txt. Add it if not.

Answer 6 · 2018-04-23T11:11:26.000Z

yes, I have checked...

cat tensorflow/contrib/makefile/tf_op_files.txt|grep -in "tensorflow/core/kernels/segment_reduction_ops.cc"
47:tensorflow/core/kernels/segment_reduction_ops.cc
302:tensorflow/core/kernels/segment_reduction_ops.cc

Answer 7 · 2018-04-23T13:59:23.000Z

May I give you a call or add my wechat 13466781043?

Answer 8 · 2018-04-24T06:49:15.000Z

I tested it again and don't find your problem. My env is:

installed tensorflow version: 1.7.0
tensorflow source code used in `build_all_linux.sh`: 
        https://github.com/formath/tensorflow.git   branch: fix

Answer 9 · 2018-04-25T07:47:13.000Z

Thanks a lot for your help! I have rebuild it， but may be the essential problem is the same.

Session created successfully
Load graph protobuf successfully
Not found: Op type not registered 'HashTableV2' in binary running on MacbookPro. Make sure the Op and Kernel are registered in the binary running in this process.

Answer 10 · 2018-04-25T08:08:49.000Z

I have fixed this problem in the branch fix. Did you use the same env listed above?

Answer 11 · 2018-04-25T11:14:15.000Z

It's ok now，thanks a lot for your help! Must use you fixed version branch，https://github.com/formath/tensorflow.git origin/fix