Liu Yue

you can use these code to use pytorch 1.8.0 to simulate pytorch1.5.0, when signal_ndim = 1 and onesided=False. for this file, these two functions are totally enough. def _rfft(x, signal_ndim=1,...

> DID you have the code of selective_scan_fwd and selective_scan_bwd in python you may want to refer to this repo: `https://github.com/alxndrTL/mamba.py` 😀

We calc FLOPs based on the ref code, though it is very different from the real speed in practise. ```python def flops_selective_scan_ref(B=1, L=256, D=768, N=16, with_D=True, with_Z=False, with_Group=True, with_complex=False): """...

> * 2L associative operations Thank you for your quick reply. Can you explain that why is there 2*L associative operations, but not L?

Many thanks. I think I've got the answer.

@llmexperiment As addressed by @albertfgu , you can just return `9BLDN` if you only use the core function of selective_scan. For full script: ```python def flops_selective_scan_fn(B=1, L=256, D=768, N=16, with_D=True,...

Thank you very much, @apoorv2904. You are right, and I nearly failed to reproduce the results I have observed before. These days, I have been working on it. (the environment...

I am so glad that you can pay attention to this tiny project! 1. the hidden state is hprefix.view(B, -1, N); in the new code, it has also been returned....