Commits · 0ec7667bed37ddc7955241fbbdbfdc4644b73f07 · school / Capstone Design / 01 / AudioLM - Pytorch with flask · GitLab

This is an archived project. Repository and other project resources are read-only.

Nov 05, 2022
- handle projection of fine and coarse logits correctly in the final transformer in the hierarchy · 0ec7667b
  Phil Wang authored Nov 04, 2022
  
  0ec7667b
Nov 04, 2022
- todo · a9efd2d9
  Phil Wang authored Nov 03, 2022
  
  a9efd2d9
- always use shazeer glu feedforward · 95280273
  Phil Wang authored Nov 03, 2022
  
  95280273
- get some code for fine transformer (last stage) into place · 9c6785b8
  Phil Wang authored Nov 03, 2022
  
  9c6785b8
- update to latest vq library, make sure residual quantizer can dropout during... · 90904ee5
  Phil Wang authored Nov 03, 2022
  90904ee5
Nov 01, 2022
- will use gradient penalty on discriminator, as had much success with that in stylegan · 3ae3a0dd
  Phil Wang authored Oct 31, 2022
  
  3ae3a0dd
Oct 31, 2022
- Merge pull request #12 from shanguanma/main · 0a714012
  Phil Wang authored Oct 30, 2022
  0a714012
- fixed a errors of STFTDiscriminator · 74a9e42e
  shanguanma authored Oct 31, 2022
  
  74a9e42e
- residual vq ready for distributed training · 2a251506
  Phil Wang authored Oct 30, 2022
  
  View commits for tag 0.0.3 0.0.3
  
  2a251506
Oct 30, 2022
- use l1 loss for "feature" loss · 8910dbfe
  Phil Wang authored Oct 29, 2022
  
  View commits for tag 0.0.2 0.0.2
  
  8910dbfe
- Merge pull request #9 from aabzaliev/main · 532d8e95
  Phil Wang authored Oct 29, 2022
  532d8e95
- include stft discriminator features into feature loss · 42b334cf
  abzaliev authored Oct 29, 2022
  
  42b334cf
Oct 29, 2022
- add the "feature" loss as mentioned in the paper with customizable weight · d0e1f681
  Phil Wang authored Oct 28, 2022
  
  d0e1f681
Oct 28, 2022
- remove padding in causal convtranspose1d thanks to @NPN for catching that · 6810d515
  Phil Wang authored Oct 27, 2022
  
  6810d515
- first pass at stft discriminator · 6bd8b8fa
  Phil Wang authored Oct 27, 2022
  
  6bd8b8fa
- trims output on right for causal convtranspose1d, thanks to @NPN in... · d9c6ba40
  Phil Wang authored Oct 27, 2022
  d9c6ba40
Oct 27, 2022
- complete the multi-scale discriminators for soundstream · 4acee04e
  Phil Wang authored Oct 26, 2022
  
  4acee04e
- prep for adversarial learning in soundstream · ee51e21c
  Phil Wang authored Oct 26, 2022
  
  ee51e21c
- cite soundstream · c5f983b8
  Phil Wang authored Oct 26, 2022
  
  c5f983b8
- add kmeans init and expiry of stale codes as detailed in soundstream paper · 56bcd108
  Phil Wang authored Oct 26, 2022
  
  56bcd108
- will need ema for soundstream · 8b78d101
  Phil Wang authored Oct 26, 2022
  
  8b78d101
Oct 26, 2022
- go with normal convtranspose1d, causal may be not possible? · 9d832142
  Phil Wang authored Oct 25, 2022
  
  9d832142
- give causal convtranspose another shot · c3ed8db5
  Phil Wang authored Oct 25, 2022
  
  c3ed8db5
- get basic encoder, residual quantization, and decoder code for soundstream... · 3bdca366
  Phil Wang authored Oct 25, 2022
  3bdca366
Oct 25, 2022
- causal mask needed in attention · 9cd86d25
  Phil Wang authored Oct 24, 2022
  
  9cd86d25
- they use three causal attention networks for semantic, coarse, fine. prepare... · 277dabf5
  Phil Wang authored Oct 24, 2022
  277dabf5
Oct 20, 2022
- complete basic attention encoder · 9b1fef52
  Phil Wang authored Oct 19, 2022
  
  9b1fef52
Oct 18, 2022
- attention, but what else? · d3bb0612
  Phil Wang authored Oct 17, 2022
  
  d3bb0612
Sep 11, 2022
- diagram · 0e3962db
  Phil Wang authored Sep 10, 2022
  
  0e3962db
- use vq library for residual quantization in soundstream · 17e4b200
  Phil Wang authored Sep 10, 2022
  
  17e4b200
- scaffold · 2cba4a09
  Phil Wang authored Sep 10, 2022
  
  2cba4a09
Sep 10, 2022
- wip · 022ba972
  Phil Wang authored Sep 09, 2022
  
  022ba972
- wip · 70326046
  Phil Wang authored Sep 09, 2022
  
  70326046
- Initial commit · aff5977d
  Phil Wang authored Sep 09, 2022
  
  aff5977d