Commits · 369e8e2e1f43c92c2e4275ce1f940dd3c6e67ba7 · school / Capstone Design / 01 / AudioLM

Browse files

This project is mirrored from https://github.com/lucidrains/audiolm-pytorch.git. Pull mirroring failed May 17, 2023.
Repository mirroring has been paused due to too many failed attempts. It can be resumed by a project maintainer or owner.
Last successful update May 15, 2023.

Mar 03, 2023
- allow for the structured quantized dropout from encodec paper · 369e8e2e
  Phil Wang authored Mar 02, 2023
  
  View commits for tag 0.21.2 0.21.2
  
  369e8e2e
- introduce attn_dynamic_pos_bias for soundstream, which should have the best... · f54e19d5
  Phil Wang authored Mar 02, 2023
  View commits for tag 0.21.0 0.21.0
  
  f54e19d5
- another local attention bump · d4680bed
  Phil Wang authored Mar 02, 2023
  
  View commits for tag 0.20.0 0.20.0
  
  d4680bed
- allow for configuring xpos scale base, as @ilya16 reports that at about 3... · 06602256
  Phil Wang authored Mar 02, 2023
  View commits for tag 0.19.2 0.19.2
  
  06602256
- Merge pull request #117 from zhvng/rm_unused_code · 3e4f582b
  Phil Wang authored Mar 02, 2023
  3e4f582b
Mar 02, 2023
- remove unused MultiHeadedEMABlock · c1de7f22
  zhvng authored Mar 02, 2023
  
  c1de7f22
- add quantize embeddings to facilitate learning and some product management · 7e957b90
  Phil Wang authored Mar 01, 2023
  
  View commits for tag 0.19.1 0.19.1
  
  7e957b90
- remove use of eos in fine transformer · 65495ad5
  Phil Wang authored Mar 01, 2023
  
  View commits for tag 0.19.0 0.19.0
  
  65495ad5
- add some asserts · b6eeb93f
  Phil Wang authored Mar 01, 2023
  
  View commits for tag 0.18.2 0.18.2
  
  b6eeb93f
Mar 01, 2023
- fix length extrapolation for rotary embeddings in local attention package · 1b15751a
  Phil Wang authored Feb 28, 2023
  
  View commits for tag 0.18.1 0.18.1
  
  1b15751a
- remove multihead EMA based on feedback from @ilya16 · a94e32f4
  Phil Wang authored Feb 28, 2023
  
  View commits for tag 0.18.0 0.18.0
  
  a94e32f4
Feb 28, 2023
- expose the soundstream configs as a property · 4daad9e7
  Phil Wang authored Feb 27, 2023
  
  View commits for tag 0.17.1 0.17.1
  
  4daad9e7
Feb 25, 2023
- update readme to use the new soundstream init_and_load_from class method · 8cb662b9
  Phil Wang authored Feb 24, 2023
  
  8cb662b9
- release saving of soundstream configurations within checkpoint, and... · 8d8b255c
  Phil Wang authored Feb 24, 2023
  View commits for tag 0.17.0 0.17.0
  
  8d8b255c
- make sure soundstream load function loads the ema version if available, also... · 5889e518
  Phil Wang authored Feb 24, 2023
  View commits for tag 0.16.2 0.16.2
  
  5889e518
Feb 24, 2023
- get the default parameters for soundstream, as well as AudioLMSoundStream and... · 21a07ab9
  Phil Wang authored Feb 23, 2023
  View commits for tag 0.16.1 0.16.1
  
  21a07ab9
- allow researchers to easily use original audiolm soundstream config · 9b8e702f
  Phil Wang authored Feb 23, 2023
  
  View commits for tag 0.15.9 0.15.9
  
  9b8e702f
- missing import for soundstream file · 043dae08
  Phil Wang authored Feb 23, 2023
  
  View commits for tag 0.15.8 0.15.8
  
  043dae08
- always save version wherever possible · 1b5fdefc
  Phil Wang authored Feb 23, 2023
  
  View commits for tag 0.15.7 0.15.7
  
  1b5fdefc
- default hubert to 16khz · 7596ae0c
  Phil Wang authored Feb 23, 2023
  
  View commits for tag 0.15.5 0.15.5
  
  7596ae0c
- start saving version numbers along with model, so researchers can trace back... · cd94fae5
  Phil Wang authored Feb 23, 2023
  View commits for tag 0.15.4a 0.15.4a
  
  cd94fae5
Feb 22, 2023
- patch · bee80b5e
  Phil Wang authored Feb 21, 2023
  
  View commits for tag 0.15.3 0.15.3
  
  bee80b5e
- fix embarrassing error, thanks to @alexdemartos · e8ee51f1
  Phil Wang authored Feb 21, 2023
  
  e8ee51f1
- take care of soundstream accepting audio without batch dimension · d244a6c9
  Phil Wang authored Feb 21, 2023
  
  View commits for tag 0.15.2 0.15.2
  
  d244a6c9
- more hacky logic to autoresolve confusion around loading from trainer vs soundstream model directly · 255b578f
  Phil Wang authored Feb 21, 2023
  
  View commits for tag 0.15.1 0.15.1
  
  255b578f
Feb 21, 2023
- fix soundstream stride defaults for 24khz, tailored for music lm · f475b095
  Phil Wang authored Feb 20, 2023
  
  View commits for tag 0.15.0 0.15.0
  
  f475b095
- Merge pull request #100 from smcio/fix-soundstream-default-strides · b179f728
  Phil Wang authored Feb 20, 2023
  b179f728
- Modify default SoundStream strides to 480 to match default 24kHz sampling rate · 381e62dd
  smcio authored Feb 21, 2023
  
  381e62dd
Feb 20, 2023
- add ability to set length of audio being trained on in seconds `data_max_length_seconds` · 9733c558
  Phil Wang authored Feb 20, 2023
  
  View commits for tag 0.14.3 0.14.3
  
  9733c558
- do something hacky to resolve issue from past saved models · c53245de
  Phil Wang authored Feb 19, 2023
  
  View commits for tag 0.14.2 0.14.2
  
  c53245de
Feb 18, 2023
- Merge pull request #94 from LWprogramming/extra_pad_id · c37cd2d9
  Phil Wang authored Feb 17, 2023
  c37cd2d9
- remove unused pad_id · b2ed8b9b
  Leon Wu authored Feb 17, 2023
  
  b2ed8b9b
- use map location to cpu when torch.load · aaf0d2d5
  Phil Wang authored Feb 17, 2023
  
  View commits for tag 0.14.1 0.14.1
  
  aaf0d2d5
- release improvements to both multiscale discriminators as well as soundstream trainer @ilya16 · 0abb04d4
  Phil Wang authored Feb 17, 2023
  
  View commits for tag 0.14.0 0.14.0
  
  0abb04d4
- Merge pull request #93 from ilya16/trainer-tweaks · 5fa0e3c6
  Phil Wang authored Feb 17, 2023
  5fa0e3c6
- Merge pull request #92 from ilya16/msdiscr-downsample · 20ac524a
  Phil Wang authored Feb 17, 2023
  20ac524a
Feb 17, 2023
- control ema model with `use_ema` · 95ebad6c
  Ilya Borovik authored Feb 17, 2023
  95ebad6c
- apply the same gradient penalty schedule for stft and multiscale discriminators · 15f83db0
  Ilya Borovik authored Feb 17, 2023
  
  15f83db0
- log all soundstream losses with specified frequency · 151edc9a
  Ilya Borovik authored Feb 17, 2023
  
  151edc9a
- downsample audios for multiscale discriminators · 29cfceb9
  Ilya Borovik authored Feb 17, 2023
  
  29cfceb9