Questions about LSTM

1

I’m new with both tvm and RNN, and i wanna understand how the lstm recipe here https://github.com/dmlc/tvm/blob/master/topi/recipe/rnn/lstm.py has been written, like i don’t see how it works really.
what do the parameters represent ?

Is it done for only one layer, or it supports multiple ones ?
If it uses only one layer, how can i extend it to support multiple ones?

thanks

2

Recipe means examples. In the LSTM example, we use tvm.scan to apply an LSTM cell scanning over the input sequence.

I don’t quite understand your questions like layers, etc. You might be interested in reading the tutorial first.

3

62f94ceee433b693ef73be231f51ae4291e53880%20(1)
62f94ceee433b693ef73be231f51ae4291e53880%20(1)301×521 27.4 KB

in the option 2 of this image for example, we have 3 layers. And if i understand well what is done actually in tvm is the first option. If it is the case, how can i use the tvm.scan to do that. should use one or multiples scans ?