About LSTM models

Hi, Does anyone having a model with LSTM Layers in those and also that works with TVM for just for testing.


MxNet test cases cover the LSTM model.

i didn’t get you, could you please tell more about this

The test example provides an example of LSTM model defined in the MxNet that can be converted and compiled by TVM.

1 Like

I have a LSTM model written with pytorch, and first i convert it to onnx model, this model has a dynamic input shape represent as: [batch_size, seq_number], so when i compile this model with: relay.frontend.from_onnx(onnx_model), there will convert the dynamic shape with type Any. so when execute at ./relay/frontend/onnx.py: X_steps = unbind(X, axis=0)./relay/frontend/common.py: _op.split(data, selections, axis), there raise exception: Any type has no len attributes. so, does tvm not support dynamic input shape for lstm?

When i tried like a year ago it was not having dynamic input shape, i am not sure it has been updated. may be you could create dummy model and i could try it out.