Convert model LSTM with bidirectional=True get an error in step Compiler

General
1

image
image1779×272 38.6 KB

Here my simple model
image
image553×831 25.8 KB

I think LSTM with bidirectional=True not yet support in Hailo AI Chip hardware

2

Hello @vanvuong0440,

We do support bidirectional LSTM layers, but only for models parsed through ONNX or PyTorch. If your model is built in TensorFlow, we recommend converting it to ONNX for compatibility.

Alternatively, you can simplify the model:

  1. Option 1: Replace the bidirectional LSTM with two single-direction LSTMs (forward and backward) and concatenate their outputs. This is a viable approach if you want to stay within the TensorFlow framework.

  2. Option 2: Reduce the number of units in your bidirectional LSTM from 128 to 64, which could help the model fit the Hailo hardware constraints.

Let us know if you need further guidance on any of these steps!