From d2f701f909af2f30c40290c540a0ae974776c8d0 Mon Sep 17 00:00:00 2001 From: Phil Wang Date: Sun, 25 Jun 2023 12:39:51 -0700 Subject: [PATCH] product management --- README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/README.md b/README.md index 130781e..c25b423 100644 --- a/README.md +++ b/README.md @@ -109,6 +109,7 @@ generated_audio = model.generate(seconds = 30, batch_size = 2) # generate 30 se - [x] default flash attention to true - [x] remove batchnorm, and just use layernorm, but after the swish (as in normformer paper) +- [ ] add an option to mask out attention to padding (variable lengthed semantic) - should work as is though - [ ] option to return list of audio files when generating - [ ] turn it into a command line tool - [ ] add cross attention and adaptive layernorm conditioning