Commit Graph

2095 Commits

Author SHA1 Message Date
James Betker
79a5b54e57 :( 2022-07-09 08:15:02 -06:00
James Betker
99adc846c7 Restore causal decoding 2022-07-09 08:11:08 -06:00
James Betker
5138d61767 Restore old MDF functionality for cheater gen 2022-07-09 08:01:32 -06:00
James Betker
694400a45b implement causal sampling for standard p_sampling 2022-07-09 08:01:03 -06:00
James Betker
55b9f31825 fix mel outputs 2022-07-08 19:51:12 -06:00
James Betker
b99af89c8f Support causal diffusion in inference 2022-07-08 14:27:19 -06:00
James Betker
ba1699cee2 Improve mdf 2022-07-08 12:30:22 -06:00
James Betker
7b4dcbf136 Support causal diffusion! 2022-07-08 12:30:05 -06:00
James Betker
78bba690de auto grad "lr" scaling 2022-07-08 00:38:25 -06:00
James Betker
e5d97dfd56 misc 2022-07-08 00:37:53 -06:00
James Betker
611316faab
Merge pull request #9 from neonbjb:causal
Add causal timestep adjustments
2022-07-07 15:18:19 -06:00
James Betker
5f575b5d3c
Add causal timestep adjustments 2022-07-07 15:17:47 -06:00
James Betker
28d5b6a80a optionally disable checkpointing in x_transformers (and make it so with the cond_encoder in tfdpc_v5) 2022-07-06 16:55:57 -06:00
James Betker
48270272e7 Use corner alignment for linear interpolation in TFDPC and TFD12
I noticed from experimentation that when this is not enabled, the interpolation edges are "sticky",
which is to say there is more variance in the center of the interpolation than at the edges.
2022-07-06 16:45:03 -06:00
James Betker
5816a4595e ugh 2022-07-05 11:14:09 -06:00
James Betker
7440e43531 Fix some bugs 2022-07-05 10:37:47 -06:00
James Betker
2b128730e7 Improve conditioning separation logic 2022-07-05 10:30:28 -06:00
James Betker
802998674e Fix another edge case 2022-07-04 16:47:57 -06:00
James Betker
808a1a4a31 eval cheater_ar 2022-07-04 08:50:09 -06:00
James Betker
bac9a8b728 Make MDF compatible with ar_prior models 2022-07-04 08:38:47 -06:00
James Betker
455943779b Fix bug in conditioning segment fetching 2022-07-04 08:16:14 -06:00
James Betker
e5859acff7 Rework tfdpc_v5 further.. 2022-07-03 18:19:01 -06:00
James Betker
6cc0405c7c generate longer mels? 2022-07-03 17:54:32 -06:00
James Betker
58f26b1900 mods to support cheater ar prior in tfd12 2022-07-03 17:54:22 -06:00
James Betker
286918c581 conditioning masking is random 2022-07-01 21:43:30 -06:00
James Betker
e06ee1b6f3 music_joiner script 2022-07-01 00:44:48 -06:00
James Betker
1953887122 Add conditoning_masking to tfdpcv5 2022-07-01 00:44:40 -06:00
James Betker
4c3413d008 Support aac datatypes 2022-07-01 00:44:20 -06:00
James Betker
f5c246b879 AR cheater gen & centroid injector 2022-06-28 23:52:54 -06:00
James Betker
43ea259228 Fix MDF for cosine schedule 2022-06-28 17:29:21 -06:00
James Betker
861aa0e139 clustering script and injector 2022-06-28 17:07:56 -06:00
James Betker
ee3b426dae Ffix tfdpc_v5 conditioning 2022-06-27 14:06:09 -06:00
James Betker
0278576f37 Update mdf to be compatible with cheater_gen 2022-06-27 10:11:23 -06:00
James Betker
69b614e08a tfdpc5 2022-06-26 19:46:57 -06:00
James Betker
d0f2560396 fix 2022-06-25 21:22:08 -06:00
James Betker
f12f0200d6 tfdpc_v4
parametric efficiency improvements and lets try feeding the timestep into the conditioning encoder
2022-06-25 21:17:00 -06:00
James Betker
42de09d983 tfdpc_v3 inference 2022-06-25 21:16:13 -06:00
James Betker
7a9c4310e8 support reading cheaters directly 2022-06-23 11:39:10 -06:00
James Betker
b210e5025c Le encoder shalt always be frozen. 2022-06-23 11:34:46 -06:00
James Betker
aeff1a4cc7 divided by zero 2022-06-21 20:26:19 -06:00
James Betker
f0117150d0 produce correct clip_lengths.. 2022-06-21 20:21:12 -06:00
James Betker
4a1f3aba31 come on guys... :(( 2022-06-21 20:12:54 -06:00
James Betker
fcfb3a1525 fix 2022-06-21 20:09:59 -06:00
James Betker
24e60bd510 report quantile losses for diffusion 2022-06-21 20:04:16 -06:00
James Betker
1394213f1e allow variable size crops 2022-06-21 19:48:07 -06:00
James Betker
3330fa2c10 tfdpc_v3 2022-06-20 15:37:48 -06:00
James Betker
2209b4f301 tfdpc2 2022-06-20 09:36:21 -06:00
James Betker
0e5a3f4712 We don't need that encoder either.. 2022-06-19 23:24:42 -06:00
James Betker
56c4a00e71 whoops 2022-06-19 23:22:30 -06:00
James Betker
a659cd865c All the stuff needed for cheater latent generation 2022-06-19 23:12:52 -06:00
James Betker
c5ea2bee52 More mods 2022-06-19 22:30:46 -06:00
James Betker
691ed196da fix codes 2022-06-19 21:29:40 -06:00
James Betker
b9f53a3ff9 don't populate gn5 2022-06-19 21:07:13 -06:00
James Betker
a5d2123daa more cleanup 2022-06-19 21:04:51 -06:00
James Betker
fef1066687 couple more alterations 2022-06-19 20:58:24 -06:00
James Betker
02ead8c05c update params 2022-06-19 20:47:06 -06:00
James Betker
ff8b0533ac gen3 waveform 2022-06-19 19:23:48 -06:00
James Betker
b19b0a74da Merge branch 'master' of https://github.com/neonbjb/DL-Art-School 2022-06-19 18:56:27 -06:00
James Betker
0b9588708c rrdb-inspired waveform gen 2022-06-19 18:56:17 -06:00
James Betker
f425afc965 permute codes 2022-06-19 18:00:30 -06:00
James Betker
90b232f965 gen_long_mels 2022-06-19 17:54:37 -06:00
James Betker
8c8efbe131 fix code_emb 2022-06-19 17:54:08 -06:00
James Betker
368dca18b1 mdf fixes + support for tfd-based waveform gen 2022-06-19 15:07:24 -06:00
James Betker
cb7569ee5e resample real inputs for music_diffusion_fid 2022-06-18 10:40:48 -06:00
James Betker
c000e489fa . 2022-06-17 09:40:11 -06:00
James Betker
7ca532c7cc handle unused encoder parameters 2022-06-17 09:37:07 -06:00
James Betker
e025183bfb and the other ones..
really need to unify this file better.
2022-06-17 09:30:25 -06:00
James Betker
3081c893d4 Don't augment grad scale when the grad don't exist! 2022-06-17 09:27:04 -06:00
James Betker
3efd64ed7a expand codes before the code converters for cheater latents 2022-06-17 09:20:29 -06:00
James Betker
f70b16214d allow more interesting code interpolation 2022-06-17 09:12:44 -06:00
James Betker
87a86ae6a8 integrate tfd12 with cheater network 2022-06-17 09:08:56 -06:00
James Betker
9d7ce42630 add tanh to the end of the latent thingy 2022-06-16 20:31:23 -06:00
James Betker
b7758f25a9 and fix 2022-06-16 20:06:27 -06:00
James Betker
e3287c9d95 Reduce severity of gpt_music reduction 2022-06-16 20:05:05 -06:00
James Betker
41abf776d9 better downsampler 2022-06-16 15:19:08 -06:00
James Betker
28d95e3141 gptmusic work 2022-06-16 15:09:47 -06:00
James Betker
781c43c1fc Clean up old TFD models 2022-06-15 16:49:06 -06:00
James Betker
34d9d5f202 adf for ar-latent tfd 2022-06-15 16:41:08 -06:00
James Betker
157d5d56c3 another debugging fix 2022-06-15 09:19:34 -06:00
James Betker
6fc86bbbe7 get rid of unused param 2022-06-15 09:14:06 -06:00
James Betker
3757ff9526 uv back to tortoise days 2022-06-15 09:04:41 -06:00
James Betker
b51ff8a176 whoops! 2022-06-15 09:01:20 -06:00
James Betker
ff5c03b460 tfd12 with ar prior 2022-06-15 08:58:02 -06:00
James Betker
3f10ce275b some tfd12 fixes to support multivae 2022-06-14 23:53:50 -06:00
James Betker
fae05229ec asdf 2022-06-14 21:52:22 -06:00
James Betker
804b365d5f make adf compatible with 7 gpus 2022-06-14 21:49:26 -06:00
James Betker
6bc19d1328 multivqvae tfd12 2022-06-14 15:58:18 -06:00
James Betker
d29ea0df5e Update ADF to be compatible with classical mel spectrograms 2022-06-14 15:19:52 -06:00
James Betker
c68669e1e1 uv2 add alignment head 2022-06-14 15:18:58 -06:00
James Betker
7ff1fbe2be channel clipper 2022-06-13 20:37:35 -06:00
James Betker
47330d603b Pretrained vqvae option for tfd12.. 2022-06-13 11:19:33 -06:00
James Betker
1fde3e5a08 more reworks 2022-06-13 08:40:23 -06:00
James Betker
7a36668870 whoops! 2022-06-12 21:11:34 -06:00
James Betker
efabcf5008 When ema is on CPU, only update every 10 steps. 2022-06-12 18:34:58 -06:00
James Betker
fc3a7ed5e3 tfd12 2022-06-12 18:09:59 -06:00
James Betker
798166015a provide conditioning ijnput as mel_norm 2022-06-12 14:51:56 -06:00
James Betker
0c95be1624 Fix MDF evaluator for current generation of 2022-06-12 14:41:06 -06:00
James Betker
a3da7f186e add tfd audio diffusion 2022-06-12 13:59:22 -06:00
James Betker
11e70dde14 update tfd11 2022-06-11 17:53:27 -06:00
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