GraphModel contains a node with dynamic op, so model.predict gives error

[sickopickle]

Please make sure that this is a bug. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:bug_template

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow.js): I have a custom model
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Windows 10
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: Ipad Air 5th Gen (unrelated)
• TensorFlow.js installed from (npm or script link): npm
• TensorFlow.js version (use command below): 3.11.0
• Browser version: N/A
• Tensorflow.js Converter Version: 3.18.0

Describe the current behavior

When I try to do model.predict(), I get the following error:

Error: This execution contains the node 'StatefulPartitionedCall/model/bidirectional_2/forward_lstm_2/PartitionedCall/while/exit/_370', which has the dynamic op 'Exit'. Please use model.executeAsync() instead. Alternatively, to avoid the dynamic ops, specify the inputs [StatefulPartitionedCall/model/bidirectional_2/forward_lstm_2/PartitionedCall/TensorArrayV2Stack/TensorListStack]

I didn't know what specifying the inputs meant, so I asked chatGPT, which told me to do model.predict({'StatefulPartitionedCall/model/bidirectional_2/forward_lstm_2/PartitionedCall/TensorArrayV2Stack/TensorListStack':input);

I tried this, and it gave me the same error except with backward_lstm_2 instead of forward_lstm_2, so I changed it to model.predict({'StatefulPartitionedCall/model/bidirectional_2/forward_lstm_2/PartitionedCall/TensorArrayV2Stack/TensorListStack':input, 'StatefulPartitionedCall/model/bidirectional_2/backward_lstm_2/PartitionedCall/TensorArrayV2Stack/TensorListStack':input};

This then resolved the previous errors, but gave me this new error:

Error in matMul: inner shapes (12) and (256) of Tensors with shapes 3,12 and 256,149 and transposeA=false and transposeB=false must match.

I know that the input shape of the LSTM layer is 256, which is probably the reason for this. I do not understand what the original error message of "specify the inputs" refers to, and when I tried to use executeAsync, the output was something like this: {"_A": null, "_x": 0, "_y": 0, "_z": null}.

Describe the expected behavior

Able to do model.predict() succesfully. Additionally, why is there no predictAsync? It is part of the docs, but when I run it, it says it does not exist / undefined.

Standalone code to reproduce the issue

I cannot upload binary and json files for my model, it seems, but the error comes from the Bidirectional LSTMs, so it should reproduce from any BLSTM model.

Other info / logs Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

[gaikwadrahul8]

Hi, @sickopickle

Aplogize for the delayed response and Could you please help me with command which you're using for tfjs-convertor ? I found similar issue #4579 and if you convert the model with --control_flow_v2=true that error should go away please check this comment so could you please try to convert model with --control_flow_v2=true and let us know whether is it resolving your issue or not ?

If issue still persists please let us know with error log if you get different error log ? Thank you!

[google-ml-butler[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you.

[google-ml-butler[bot]]

Closing as stale. Please @mention us if this needs more attention.

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No

[sickopickle]

I did set the control flow v2 as true. It still did not work.

[shamoqianting]

Hi, @sickopickle

Aplogize for the delayed response and Could you please help me with command which you're using for tfjs-convertor ? I found similar issue #4579 and if you convert the model with --control_flow_v2=true that error should go away please check this comment so could you please try to convert model with --control_flow_v2=true and let us know whether is it resolving your issue or not ?

If issue still persists please let us know with error log if you get different error log ? Thank you!

encounter same issue Error: This execution contains the node 'StatefulPartitionedCall/assert_equal_9/Assert/AssertGuard/branch_executed/_142', which has the dynamic op 'Merge'. Please use model.executeAsync() instead. Alternatively, to avoid the dynamic ops, specify the inputs [StatefulPartitionedCall/NoOp] I tried to use --control_flow_v2=true option when I use tensorflowjs-converter. But it didn't work.