Commit Graph

2096 Commits

Author SHA1 Message Date
James Betker
5c6c8f6904 back to convs? 2022-06-11 15:26:10 -06:00
James Betker
b684622b06 tfd10 2022-06-11 14:06:19 -06:00
James Betker
2a787ec910 more mods 2022-06-11 11:44:33 -06:00
James Betker
999a140f9f whoops 2022-06-11 11:22:34 -06:00
James Betker
00b9f332ee rework arch 2022-06-11 11:17:16 -06:00
James Betker
36b5e89a69 Rework the diet blocks a bit 2022-06-11 08:28:10 -06:00
James Betker
0dd3883662 one last update.. 2022-06-11 08:06:29 -06:00
James Betker
41170f97e9 one more adjustment 2022-06-11 08:01:46 -06:00
James Betker
df0cdf1a4f tfd9 returns with some optimizations 2022-06-11 08:00:09 -06:00
James Betker
acfe9cf880 fp16 2022-06-10 22:39:15 -06:00
James Betker
aca9024d9b qcodes 2022-06-10 16:23:08 -06:00
James Betker
38a00f29c0 now theres deprecation warnings, fml 2022-06-10 15:41:39 -06:00
James Betker
561a6b8ff7 damn this sucks 2022-06-10 15:38:59 -06:00
James Betker
0316063e2d . 2022-06-10 15:37:02 -06:00
James Betker
dca16e6447 . 2022-06-10 15:35:36 -06:00
James Betker
ee2827dee9 Debug warmup state 2022-06-10 15:23:31 -06:00
James Betker
6d85fe05f6 :/ oh well. 2022-06-10 15:17:41 -06:00
James Betker
33178e89c4 harharhack 2022-06-10 15:13:24 -06:00
James Betker
7198bd8bd0 forgot other customizations I want to keep 2022-06-10 15:09:05 -06:00
James Betker
8f40108f5b lets try a different tact 2022-06-10 14:51:59 -06:00
James Betker
2158383fa4 Revert previous changes 2022-06-10 14:34:05 -06:00
James Betker
89bd40d39f eval bug fix 2022-06-10 13:51:06 -06:00
James Betker
84469f3538 get rid of encoder checkpointing 2022-06-10 10:50:34 -06:00
James Betker
97b32dd39d try to make tfd8 be able to be trained e2e in quantizer mode 2022-06-10 10:40:56 -06:00
James Betker
e78c4b422c tfd8 2022-06-10 09:24:41 -06:00
James Betker
d98b895307 loss aware fix and report gumbel temperature 2022-06-09 21:56:47 -06:00
James Betker
6e57eaa186 fix bug 2022-06-09 21:52:57 -06:00
James Betker
07bdd865dc some checks 2022-06-09 21:46:32 -06:00
James Betker
34005367fd setup for partial channel diffusion 2022-06-09 21:41:20 -06:00
James Betker
47b34f5cb9 mup work checkin 2022-06-09 21:15:09 -06:00
James Betker
e67e82be2d misc 2022-06-09 21:14:48 -06:00
James Betker
16936881e5 allow freezing the upper quantizer 2022-06-08 18:30:22 -06:00
James Betker
43f225c35c debug gumbel temperature 2022-06-08 12:12:08 -06:00
James Betker
91be38cba3 . 2022-06-08 11:54:46 -06:00
James Betker
dee2b72786 checkpointing bugs, smh 2022-06-08 11:53:10 -06:00
James Betker
c61cd64bc9 network updates 2022-06-08 09:26:59 -06:00
James Betker
5a54d7db11 unet with ar prior 2022-06-07 17:52:36 -06:00
James Betker
5028703b3d ci not required 2022-06-06 09:26:25 -06:00
James Betker
08597bfaf5 fix 2022-06-06 09:21:58 -06:00
James Betker
49568ee16f some updates 2022-06-06 09:13:47 -06:00
James Betker
602df0abbc revert changes to dietattentionblock 2022-06-05 10:06:17 -06:00
James Betker
51d1908e94 update 2022-06-05 09:35:43 -06:00
James Betker
f9ebcf11d8 fix2 2022-06-05 01:31:37 -06:00
James Betker
aac92b01b3 fix 2022-06-05 01:27:28 -06:00
James Betker
38d8b17d18 tfd8 gets real verbose grad norm metrics 2022-06-04 23:09:54 -06:00
James Betker
0a9d4d4afc bunch of new stuff 2022-06-04 22:23:08 -06:00
James Betker
8f8b189025 Support legacy vqvae quantizer in music_quantizer 2022-06-04 10:16:24 -06:00
James Betker
4819f15521 undo quantile sampler in music_diffusion_fid 2022-06-04 10:15:31 -06:00
James Betker
a9387179db add channel loss balancing 2022-06-03 15:19:23 -06:00
James Betker
40ba802104 padding 2022-06-03 12:09:59 -06:00
James Betker
581bc7ac5c udmc update 2022-06-03 12:02:22 -06:00
James Betker
a14274c845 fix memory issue 2022-06-03 11:20:09 -06:00
James Betker
9d8c2bddb1 classical unet for music 2022-06-03 11:03:14 -06:00
James Betker
d3a60633a3 codes generation script 2022-06-03 11:02:28 -06:00
James Betker
2f4d990ad1 tfd7 2022-06-02 09:27:40 -06:00
James Betker
b2a83efe50 a few fixes 2022-06-01 16:35:15 -06:00
James Betker
712e0e82f7 fix bug 2022-06-01 14:21:44 -06:00
James Betker
de54be5570 propagate diversity loss 2022-06-01 14:18:50 -06:00
James Betker
4c6ef42b38 freeze quantizer until step 2022-06-01 08:06:05 -06:00
James Betker
64b6ae2f4a fix 2022-06-01 01:01:32 -06:00
James Betker
1ac02acdc3 tfd7 2022-06-01 00:50:40 -06:00
James Betker
e8cb93a4e9 fix size issues 2022-05-31 21:23:26 -06:00
James Betker
8a1b8e3e62 add checkpointing 2022-05-31 21:09:05 -06:00
James Betker
c0db85bf4f music quantizer 2022-05-31 21:06:54 -06:00
James Betker
96da10415e facepalm 2022-05-30 16:48:18 -06:00
James Betker
479c3195f3 make wandb histograms work again 2022-05-30 16:44:39 -06:00
James Betker
29b55d42a5 one more 2022-05-30 16:33:49 -06:00
James Betker
71cf654957 fix unused parameters 2022-05-30 16:31:40 -06:00
James Betker
f7d237a50a train quantizer with diffusion 2022-05-30 16:25:33 -06:00
James Betker
136021bf8d tfd6 2022-05-30 09:09:42 -06:00
James Betker
eab1162d2b hmm.. 2022-05-29 22:32:25 -06:00
James Betker
2e72fddaeb td_tts_2 2022-05-29 22:22:14 -06:00
James Betker
27f347cdd3 gotta project that shit! 2022-05-28 23:19:36 -06:00
James Betker
536c8558ae fix 2022-05-28 22:32:38 -06:00
James Betker
da367da411 df5 2022-05-28 22:30:23 -06:00
James Betker
6b43915eb8 support projecting to vectors 2022-05-28 22:27:45 -06:00
James Betker
86694aef4e tfd5 2022-05-28 22:27:04 -06:00
James Betker
b6b4f10e1b ... 2022-05-28 10:59:03 -06:00
James Betker
0d3b831cf9 big fatty 2022-05-28 10:55:43 -06:00
James Betker
76aeba7843 fail gracefully from mfd 2022-05-27 20:24:16 -06:00
James Betker
f691f5faa1 f 2022-05-27 13:47:05 -06:00
James Betker
031769150d clip adf test dataset 2022-05-27 13:42:52 -06:00
James Betker
31dec016e0 adf 2022-05-27 12:28:04 -06:00
James Betker
b4269af61b fix circular deps 2022-05-27 11:44:27 -06:00
James Betker
34ee1d0bc3 mdf 2022-05-27 11:40:47 -06:00
James Betker
490d39b967 some stuff 2022-05-27 11:40:31 -06:00
James Betker
5efeee6b97 fix type bug 2022-05-27 11:19:30 -06:00
James Betker
0659fe3d1e tfd3 mods 2022-05-27 11:16:26 -06:00
James Betker
bed3df4888 propagate type 2022-05-27 11:12:03 -06:00
James Betker
c46da0285c Move stuff around 2022-05-27 11:06:58 -06:00
James Betker
9852599b34 tfd5 - with clvp! 2022-05-27 09:49:10 -06:00
James Betker
3db862dd32 adf update 2022-05-27 09:25:53 -06:00
James Betker
8587a18717 fd fix 2022-05-26 20:19:09 -06:00
James Betker
dd13b883ac td4 2022-05-26 14:56:03 -06:00
James Betker
1dbe0b6b2e a 2022-05-26 10:13:27 -06:00
James Betker
aa653115f1 tfd3 2022-05-26 10:09:11 -06:00
James Betker
36c68692a6 forgot to add rotary embeddings 2022-05-26 09:25:42 -06:00
James Betker
8ce48f04ff transformer diffusion 2 2022-05-26 09:08:35 -06:00
James Betker
56f19a23cd fix nh 2022-05-25 12:31:56 -06:00
James Betker
52a20f3aa3 und10 2022-05-25 12:19:21 -06:00
James Betker
00e133afa9 support not_ew too 2022-05-25 08:58:23 -06:00
James Betker
5188866bd5 facepalm 2022-05-24 14:55:51 -06:00
James Betker
8b4b5ffa72 slight rework 2022-05-24 14:38:37 -06:00
James Betker
48aab2babe ressurect ctc code gen with some cool new ideas 2022-05-24 14:02:33 -06:00
James Betker
65b441d74e transformer diffusion 2022-05-24 14:02:05 -06:00
James Betker
1e1bbe1a27 whoops 2022-05-23 12:28:36 -06:00
James Betker
560b83e770 default to residual encoder 2022-05-23 12:24:00 -06:00
James Betker
f432bdf7ae deeper resblock encoder 2022-05-23 11:46:40 -06:00
James Betker
bb36bb4db5 ugh it never ends 2022-05-23 10:58:28 -06:00
James Betker
5d13d38119 allow opt states to be reset 2022-05-23 10:54:37 -06:00
James Betker
f4a97ca0a7 and this 2022-05-23 10:38:28 -06:00
James Betker
874de1775d Update mdf spectral 2022-05-23 10:37:15 -06:00
James Betker
36dd4eb61f no grads for mel injectors 2022-05-23 10:34:53 -06:00
James Betker
dc471f5c6d residual features 2022-05-23 09:58:30 -06:00
James Betker
1f521d6a1d add reconstruction loss to m2v 2022-05-23 09:28:41 -06:00
James Betker
1d758c3bc8 try and make repeated failures recover better 2022-05-23 08:50:27 -06:00
James Betker
2270c89fdc . 2022-05-23 08:47:15 -06:00
James Betker
40f844657b tolong 2022-05-23 08:27:54 -06:00
James Betker
9f16b25ce5 introduce prepadlength 2022-05-23 08:21:27 -06:00
James Betker
10f4a742bd reintroduce attention masks 2022-05-23 08:16:04 -06:00
James Betker
2da8f8a666 fmp 2022-05-23 07:06:25 -06:00
James Betker
68c0afcbcc m2v frequency masking 2022-05-23 07:04:12 -06:00
James Betker
4093e38717 revert flat diffusion back... 2022-05-22 23:10:58 -06:00
James Betker
8f28404645 another fix 2022-05-22 21:32:43 -06:00
James Betker
41809a6330 Add 8x dim reductor 2022-05-22 20:23:16 -06:00
James Betker
07f7be24ce :/ 2022-05-22 20:00:55 -06:00
James Betker
c0bc466aad mdf for shortened mel test 2022-05-22 19:29:20 -06:00
James Betker
1095248caf Revert "retest"
This reverts commit ed7768c73b.
2022-05-22 19:23:01 -06:00
James Betker
2798bfab8c Revert "update mdf for legacy try"
This reverts commit 693b19ea3b.
2022-05-22 19:22:44 -06:00
James Betker
693b19ea3b update mdf for legacy try 2022-05-22 16:37:02 -06:00
James Betker
ed7768c73b retest 2022-05-22 16:30:09 -06:00
James Betker
2dd0b9e6e9 mel_head should be optional 2022-05-22 12:25:45 -06:00
James Betker
37640e9759 Squelch annoying warning 2022-05-22 12:12:40 -06:00
James Betker
0c60f22197 fix unused parameters 2022-05-22 08:16:31 -06:00
James Betker
57d6f6d366 Big rework of flat_diffusion
Back to the drawing board, boys. Time to waste some resources catching bugs....
2022-05-22 08:09:33 -06:00
James Betker
db38672dae precompute diffusion embeddings for from_codes 2022-05-22 06:45:57 -06:00
James Betker
ea21a8b107 Update music_diffusion_fid to support waveform diffusion from codes 2022-05-22 05:23:54 -06:00
James Betker
e0bf3a0ddc Save myself some time in the future 2022-05-20 17:18:35 -06:00
James Betker
be937d202e new attempt 2022-05-20 17:04:22 -06:00
James Betker
968660c248 another update 2022-05-20 11:25:00 -06:00
James Betker
28f950b7d3 fix 2022-05-20 11:18:52 -06:00
James Betker
b317c68ac9 fix 2022-05-20 11:12:53 -06:00
James Betker
3121bc4e43 flat diffusion 2022-05-20 11:01:48 -06:00
James Betker
e9fb2ead9a m2v stuff 2022-05-20 11:01:17 -06:00
James Betker
c9c16e3b01 misc updates 2022-05-19 13:39:32 -06:00
James Betker
10378fc37f make codebooks specifiable 2022-05-18 11:07:12 -06:00
James Betker
efc2657b48 fiddle with init 2022-05-18 10:56:01 -06:00
James Betker
208a703080 use gelu act 2022-05-18 09:34:01 -06:00
James Betker
b2b37453df make the codebook bigger 2022-05-17 20:58:56 -06:00
James Betker
9a9c3cafba Make feature encoder a bit more descriptive 2022-05-17 18:14:52 -06:00
James Betker
ee364f4eeb just take the mean... 2022-05-17 18:09:23 -06:00
James Betker
6130391a85 fix div 2022-05-17 18:04:20 -06:00
James Betker
7213ad2b89 Do grad reduction 2022-05-17 17:59:40 -06:00
James Betker
7c82e18c6c darn mpi 2022-05-17 17:16:09 -06:00
James Betker
88ec0512f7 Scale losses 2022-05-17 17:12:20 -06:00
James Betker
a6397ce84a Fix incorrect projections 2022-05-17 16:53:52 -06:00
James Betker
c37fc3b4ed m2v grad norm groups 2022-05-17 16:29:36 -06:00
James Betker
c1bdb4f9a1 degrade gumbel softmax over time 2022-05-17 16:23:04 -06:00
James Betker
3853f37257 stable layernorm 2022-05-17 16:07:03 -06:00
James Betker
6a2c29f596 Fix inverted logic 2022-05-17 15:39:07 -06:00
James Betker
519151d83f m2v 2022-05-17 15:37:59 -06:00
James Betker
d1de94d75c Stash mel2vec work (gonna throw it all away..) 2022-05-17 12:35:01 -06:00
James Betker
8202b9f39c some stuff 2022-05-15 21:50:54 -06:00
James Betker
ab5acead0e add exp loss for diffusion models 2022-05-15 21:50:38 -06:00
James Betker
ee218ab9b7 uv3 2022-05-13 17:57:47 -06:00
James Betker
eb64d18075 Fix phoneme tokenizer 2022-05-13 17:56:26 -06:00
James Betker
51f8c1bced phonetic dataset 2022-05-12 11:57:28 -06:00
James Betker
3d7e2a2846 fix collection 2022-05-11 21:50:05 -06:00
James Betker
ba2b71796a k 2022-05-11 21:20:06 -06:00
James Betker
efa737b685 re-add distributed collect to clvp 2022-05-11 21:14:18 -06:00
James Betker
545453077e uv3 2022-05-09 15:36:22 -06:00
James Betker
96a5cc66ee uv3 2022-05-09 15:35:51 -06:00
James Betker
b42b4e18de clean up unified voice
- remove unused code
- fix inference model to use the terms "prior" and "posterior" to properly define the modeling order (they were inverted before)
- default some settings I never intend to change in the future
2022-05-09 14:45:49 -06:00
James Betker
9118f58849 uncomment music projector.. 2022-05-09 09:19:26 -06:00
James Betker
74dd095326 a 2022-05-08 18:54:09 -06:00
James Betker
1177c35dec music fid updates 2022-05-08 18:49:39 -06:00
James Betker
7812c23c7a revert fill_gaps back to old masking behavior 2022-05-08 00:10:19 -06:00
James Betker
58ed27d7a8 new gap_filler 2022-05-07 12:44:23 -06:00
James Betker
6c8032b4be more work 2022-05-06 21:56:49 -06:00
James Betker
f541610256 contrastive_audio 2022-05-06 16:37:22 -06:00
James Betker
79543e5488 Simpler form of the wavegen model 2022-05-06 16:37:04 -06:00
James Betker
d8925ccde5 few things with gap filling 2022-05-06 14:33:44 -06:00
James Betker
b83b53cf84 norm mel 2022-05-06 00:49:54 -06:00
James Betker
b13d983c24 and mel_head 2022-05-06 00:25:27 -06:00
James Betker
d5fb79564a remove mel_pred 2022-05-06 00:24:05 -06:00
James Betker
e9bb692490 fixed aligned_latent 2022-05-06 00:20:21 -06:00
James Betker
1609101a42 musical gap filler 2022-05-05 16:47:08 -06:00
James Betker
d66ab2d28c Remove unused waveform_gens 2022-05-04 21:06:54 -06:00
James Betker
47662b9ec5 some random crap 2022-05-04 20:29:23 -06:00
James Betker
6655f7845a add pixel shuffling for 1d cases 2022-05-04 08:03:09 -06:00
James Betker
c42c53e75a Add a trainable network for converting a normal distribution into a latent space 2022-05-02 09:47:30 -06:00
James Betker
e402089556 abstractify 2022-05-02 00:11:26 -06:00
James Betker
ab219fbefb output variance 2022-05-02 00:10:33 -06:00
James Betker
3b074aac34 add checkpointing 2022-05-02 00:07:42 -06:00
James Betker
ae5f934ea1 diffwave 2022-05-02 00:05:04 -06:00
James Betker
f4254609c1 MDF
around and around in circles........
2022-05-01 23:04:56 -06:00
James Betker
b712d3b72b break out get_conditioning_latent from unified_voice 2022-05-01 23:04:44 -06:00
James Betker
afa2df57c9 gen3 2022-04-30 10:41:38 -06:00
James Betker
64c7582bf5 full pipeline 2022-04-28 22:47:26 -06:00
James Betker
8aa6651fc7 fix surrogate loss return in waveform_gen2 2022-04-28 10:10:11 -06:00
James Betker
e208d9fb80 gate augmentations with a flag 2022-04-28 10:09:22 -06:00
James Betker
3f67cb2023 music diffusion fid adjustments 2022-04-28 10:08:55 -06:00
James Betker
ab8176b217 audio prep misc 2022-04-28 10:08:38 -06:00
James Betker
f02b01bd9d reverse univnet classifier 2022-04-20 21:37:55 -06:00
James Betker
9df85c902e New gen2
Which is basically a autoencoder with a giant diffusion appendage attached
2022-04-20 21:37:34 -06:00
James Betker
b1c2c48720 music diffusion fid 2022-04-20 00:28:03 -06:00
James Betker
084b1c1527 file splitter 2022-04-20 00:27:49 -06:00
James Betker
b4549eed9f uv2 fix 2022-04-20 00:27:38 -06:00
James Betker
24fdafd855 fix2 2022-04-20 00:03:29 -06:00
James Betker
0af0051399 fix 2022-04-20 00:01:57 -06:00
James Betker
419f4d37bd gen2 music 2022-04-19 23:38:37 -06:00
James Betker
c85ab738c5 paired fix 2022-04-16 23:41:57 -06:00
James Betker
8fe0dff33c support tts typing 2022-04-16 23:36:57 -06:00
James Betker
48cb6a5abd misc 2022-04-16 20:28:04 -06:00
James Betker
147478a148 cvvp 2022-04-16 20:27:46 -06:00
James Betker
546ecd5aeb music! 2022-04-15 21:21:37 -06:00
James Betker
254357724d gradprop 2022-04-15 09:37:20 -06:00
James Betker
fbf1f4f637 update 2022-04-15 09:34:44 -06:00
James Betker
82aad335ba add distributued logic for loss 2022-04-15 09:31:48 -06:00
James Betker
efe12cb816 Update clvp to add masking probabilities in conditioning and to support code inputs 2022-04-15 09:11:23 -06:00
James Betker
3cad1b8114 more fixes 2022-04-11 15:18:44 -06:00
James Betker
6dea7da7a8 another fix 2022-04-11 12:29:43 -06:00
James Betker
f2c172291f fix audio_diffusion_fid for autoregressive latent inputs 2022-04-11 12:08:15 -06:00
James Betker
8ea5c307fb Fixes for training the diffusion model on autoregressive inputs 2022-04-11 11:02:44 -06:00
James Betker
a3622462c1 Change latent_conditioner back 2022-04-11 09:00:13 -06:00
James Betker
03d0b90bda fixes 2022-04-10 21:02:12 -06:00
James Betker
19ca5b26c1 Remove flat0 and move it into flat 2022-04-10 21:01:59 -06:00
James Betker
81c952a00a undo relative 2022-04-08 16:32:52 -06:00
James Betker
944b4c3335 more undos 2022-04-08 16:31:08 -06:00
James Betker
032983e2ed fix bug and allow position encodings to be trained separately from the rest of the model 2022-04-08 16:26:01 -06:00
James Betker
09ab1aa9bc revert rotary embeddings work
I'm not really sure that this is going to work. I'd rather explore re-using what I've already trained
2022-04-08 16:18:35 -06:00
James Betker
2fb9ffb0aa Align autoregressive text using start and stop tokens 2022-04-08 09:41:59 -06:00
James Betker
628569af7b Another fix 2022-04-08 09:41:18 -06:00
James Betker
423293e518 fix xtransformers bug 2022-04-08 09:12:46 -06:00
James Betker
048f6f729a remove lightweight_gan 2022-04-07 23:12:08 -07:00
James Betker
e634996a9c autoregressive_codegen: support key_value caching for faster inference 2022-04-07 23:08:46 -07:00
James Betker
d05e162f95 reformat x_transformers 2022-04-07 23:08:03 -07:00
James Betker
7c578eb59b Fix inference in new autoregressive_codegen 2022-04-07 21:22:46 -06:00
James Betker
3f8d7955ef unified_voice with rotary embeddings 2022-04-07 20:11:14 -06:00
James Betker
573e5552b9 CLVP v1 2022-04-07 20:10:57 -06:00
James Betker
71b73db044 clean up 2022-04-07 11:34:10 -06:00
James Betker
6fc4f49e86 some dumb stuff 2022-04-07 11:32:34 -06:00
James Betker
e6387c7613 Fix eval logic to not run immediately 2022-04-07 11:29:57 -06:00
James Betker
305dc95e4b cg2 2022-04-06 21:24:36 -06:00
James Betker
e011166dd6 autoregressive_codegen r3 2022-04-06 21:04:23 -06:00
James Betker
33ef17e9e5 fix context 2022-04-06 00:45:42 -06:00
James Betker
37bdfe82b2 Modify x_transformers to do checkpointing and use relative positional biases 2022-04-06 00:35:29 -06:00
James Betker
09879b434d bring in x_transformers 2022-04-06 00:21:58 -06:00
James Betker
3d916e7687 Fix evaluation when using multiple batch sizes 2022-04-05 07:51:09 -06:00
James Betker
572d137589 track iteration rate 2022-04-04 12:33:25 -06:00