mrq/DL-Art-School
1
3
Fork 5
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Commit Graph

  • 74dd095326 a James Betker 2022-05-08 18:54:09 -0600
  • 1177c35dec music fid updates James Betker 2022-05-08 18:49:39 -0600
  • 7812c23c7a revert fill_gaps back to old masking behavior James Betker 2022-05-08 00:10:19 -0600
  • 58ed27d7a8 new gap_filler James Betker 2022-05-07 12:44:23 -0600
  • 6c8032b4be more work James Betker 2022-05-06 21:56:49 -0600
  • f541610256 contrastive_audio James Betker 2022-05-06 16:37:22 -0600
  • 79543e5488 Simpler form of the wavegen model James Betker 2022-05-06 16:37:04 -0600
  • d8925ccde5 few things with gap filling James Betker 2022-05-06 14:33:44 -0600
  • b83b53cf84 norm mel James Betker 2022-05-06 00:49:54 -0600
  • b13d983c24 and mel_head James Betker 2022-05-06 00:25:27 -0600
  • d5fb79564a remove mel_pred James Betker 2022-05-06 00:24:05 -0600
  • e9bb692490 fixed aligned_latent James Betker 2022-05-06 00:20:21 -0600
  • 1609101a42 musical gap filler James Betker 2022-05-05 16:47:08 -0600
  • d66ab2d28c Remove unused waveform_gens James Betker 2022-05-04 21:06:54 -0600
  • 47662b9ec5 some random crap James Betker 2022-05-04 20:29:23 -0600
  • 6655f7845a add pixel shuffling for 1d cases James Betker 2022-05-04 08:03:09 -0600
  • c42c53e75a Add a trainable network for converting a normal distribution into a latent space James Betker 2022-05-02 09:47:30 -0600
  • e402089556 abstractify James Betker 2022-05-02 00:11:26 -0600
  • ab219fbefb output variance James Betker 2022-05-02 00:10:33 -0600
  • 3b074aac34 add checkpointing James Betker 2022-05-02 00:07:42 -0600
  • ae5f934ea1 diffwave James Betker 2022-05-02 00:05:04 -0600
  • f4254609c1 MDF James Betker 2022-05-01 23:04:56 -0600
  • b712d3b72b break out get_conditioning_latent from unified_voice James Betker 2022-05-01 23:04:44 -0600
  • afa2df57c9 gen3 James Betker 2022-04-30 10:41:38 -0600
  • 64c7582bf5 full pipeline James Betker 2022-04-28 22:47:26 -0600
  • 8aa6651fc7 fix surrogate loss return in waveform_gen2 James Betker 2022-04-28 10:10:11 -0600
  • e208d9fb80 gate augmentations with a flag James Betker 2022-04-28 10:09:22 -0600
  • 3f67cb2023 music diffusion fid adjustments James Betker 2022-04-28 10:08:55 -0600
  • ab8176b217 audio prep misc James Betker 2022-04-28 10:08:38 -0600
  • f02b01bd9d reverse univnet classifier James Betker 2022-04-20 21:37:55 -0600
  • 9df85c902e New gen2 James Betker 2022-04-20 21:37:34 -0600
  • b1c2c48720 music diffusion fid James Betker 2022-04-20 00:28:03 -0600
  • 084b1c1527 file splitter James Betker 2022-04-20 00:27:49 -0600
  • b4549eed9f uv2 fix James Betker 2022-04-20 00:27:38 -0600
  • 24fdafd855 fix2 James Betker 2022-04-20 00:03:29 -0600
  • 0af0051399 fix James Betker 2022-04-20 00:01:57 -0600
  • 419f4d37bd gen2 music James Betker 2022-04-19 23:38:37 -0600
  • c85ab738c5 paired fix James Betker 2022-04-16 23:41:57 -0600
  • 8fe0dff33c support tts typing James Betker 2022-04-16 23:36:57 -0600
  • 48cb6a5abd misc James Betker 2022-04-16 20:28:04 -0600
  • 147478a148 cvvp James Betker 2022-04-16 20:27:46 -0600
  • 546ecd5aeb music! James Betker 2022-04-15 21:21:37 -0600
  • 254357724d gradprop James Betker 2022-04-15 09:37:20 -0600
  • fbf1f4f637 update James Betker 2022-04-15 09:34:44 -0600
  • 82aad335ba add distributued logic for loss James Betker 2022-04-15 09:31:48 -0600
  • efe12cb816 Update clvp to add masking probabilities in conditioning and to support code inputs James Betker 2022-04-15 09:11:23 -0600
  • 3cad1b8114 more fixes James Betker 2022-04-11 15:18:44 -0600
  • 6dea7da7a8 another fix James Betker 2022-04-11 12:29:43 -0600
  • f2c172291f fix audio_diffusion_fid for autoregressive latent inputs James Betker 2022-04-11 12:08:15 -0600
  • 8ea5c307fb Fixes for training the diffusion model on autoregressive inputs James Betker 2022-04-11 11:02:44 -0600
  • a3622462c1 Change latent_conditioner back James Betker 2022-04-11 09:00:13 -0600
  • 03d0b90bda fixes James Betker 2022-04-10 21:02:12 -0600
  • 19ca5b26c1 Remove flat0 and move it into flat James Betker 2022-04-10 21:01:59 -0600
  • 81c952a00a undo relative James Betker 2022-04-08 16:32:52 -0600
  • 944b4c3335 more undos James Betker 2022-04-08 16:31:08 -0600
  • 032983e2ed fix bug and allow position encodings to be trained separately from the rest of the model James Betker 2022-04-08 16:26:01 -0600
  • 09ab1aa9bc revert rotary embeddings work James Betker 2022-04-08 16:18:35 -0600
  • 2fb9ffb0aa Align autoregressive text using start and stop tokens James Betker 2022-04-08 09:41:59 -0600
  • 628569af7b Another fix James Betker 2022-04-08 09:41:18 -0600
  • 423293e518 fix xtransformers bug James Betker 2022-04-08 09:12:46 -0600
  • 048f6f729a remove lightweight_gan James Betker 2022-04-07 23:12:08 -0700
  • e634996a9c autoregressive_codegen: support key_value caching for faster inference James Betker 2022-04-07 23:08:46 -0700
  • d05e162f95 reformat x_transformers James Betker 2022-04-07 23:08:03 -0700
  • 7c578eb59b Fix inference in new autoregressive_codegen James Betker 2022-04-07 21:22:46 -0600
  • 3f8d7955ef unified_voice with rotary embeddings James Betker 2022-04-07 20:11:14 -0600
  • 573e5552b9 CLVP v1 James Betker 2022-04-07 20:10:57 -0600
  • 71b73db044 clean up James Betker 2022-04-07 11:34:10 -0600
  • 6fc4f49e86 some dumb stuff James Betker 2022-04-07 11:32:34 -0600
  • e6387c7613 Fix eval logic to not run immediately James Betker 2022-04-07 11:29:57 -0600
  • 305dc95e4b cg2 James Betker 2022-04-06 21:24:36 -0600
  • e011166dd6 autoregressive_codegen r3 James Betker 2022-04-06 21:04:23 -0600
  • 33ef17e9e5 fix context James Betker 2022-04-06 00:45:42 -0600
  • 37bdfe82b2 Modify x_transformers to do checkpointing and use relative positional biases James Betker 2022-04-06 00:35:29 -0600
  • 09879b434d bring in x_transformers James Betker 2022-04-06 00:21:58 -0600
  • 3d916e7687 Fix evaluation when using multiple batch sizes James Betker 2022-04-05 07:51:09 -0600
  • 572d137589 track iteration rate James Betker 2022-04-04 12:33:25 -0600
  • 4cdb0169d0 update training data encountered when using force_start_step James Betker 2022-04-04 12:25:00 -0600
  • cdd12ff46c Add code validation to autoregressive_codegen James Betker 2022-04-04 09:51:41 -0600
  • 99de63a922 man I'm really on it tonight.... James Betker 2022-04-02 22:01:33 -0600
  • a4bdc80933 moikmadsf James Betker 2022-04-02 21:59:50 -0600
  • 1cf20b7337 sdfds James Betker 2022-04-02 21:58:09 -0600
  • b6afc4d542 dsfa James Betker 2022-04-02 21:57:00 -0600
  • 4c6bdfc9e2 get rid of relative position embeddings, which do not work with DDP & checkpointing James Betker 2022-04-02 21:55:32 -0600
  • b6d62aca5d add inference model on top of codegen James Betker 2022-04-02 21:25:10 -0600
  • 2b6ff09225 autoregressive_codegen v1 James Betker 2022-04-02 15:07:39 -0600
  • 00767219fc undo latent converter change James Betker 2022-04-01 20:46:27 -0600
  • 55c86e02c7 Flat fix James Betker 2022-04-01 19:13:33 -0600
  • 8623c51902 fix bug James Betker 2022-04-01 16:11:34 -0600
  • 035bcd9f6c fwd fix James Betker 2022-04-01 16:03:07 -0600
  • f6a8b0a5ca prep flat0 for feeding from autoregressive_latent_converter James Betker 2022-04-01 15:53:45 -0600
  • 3e97abc8a9 update flat0 to break out timestep-independent inference steps James Betker 2022-04-01 14:38:53 -0600
  • a6181a489b Fix loss gapping caused by poor gradients into mel_pred James Betker 2022-03-26 22:49:14 -0600
  • 0070867d0f inference script for diffusion image models James Betker 2022-03-26 22:48:24 -0600
  • 1feade23ff support x-transformers in text_voice_clip and support relative positional embeddings James Betker 2022-03-26 22:48:10 -0600
  • 9b90472e15 feed direct inputs into gd James Betker 2022-03-26 08:36:19 -0600
  • 6909f196b4 make code pred returns optional James Betker 2022-03-26 08:33:30 -0600
  • 2a29a71c37 attempt to force meaningful codes by adding a surrogate loss James Betker 2022-03-26 08:31:40 -0600
  • 45804177b8 more stuff James Betker 2022-03-25 00:03:18 -0600
  • d4218d8443 mods James Betker 2022-03-24 23:31:20 -0600
  • 9c79fec734 update adf James Betker 2022-03-24 21:20:29 -0600