mrq
18 Followers · 0 Following

  • https://git.ecker.tech/ aims to provide a place to share my efforts while maintaining true ownership of my code, as I do not trust GitHub.

    XMR: 4B9TQdkAkBFYrbj5ztvTx89e5LpucPeTSPzemCihdDi9EBnx7btn8RDNZTBz2zihWsjMnDkzn5As1LU6gLv3KQy8BLsZ8SG

  • Joined on 2022-10-10
Repositories
15
 Projects Packages Public Activity Starred Repositories
mrq pushed to master at mrq/vall-e 2025-03-13 04:12:20 +00:00
6ee505cffd fixed dac
mrq pushed to master at mrq/vall-e 2025-03-13 03:42:12 +00:00
ba5f3d19b4 use the FSQ-targeted encoder/decodede whole-ly as it works for EnCodec too, as the RVQ-targeted encoder/decoder doesnt (and some notes)
2ccf1b5740 actually do duration prediction
Compare 2 commits »
mrq pushed to master at mrq/vall-e 2025-03-12 01:28:22 +00:00
5c512717a6 len prediction for new model (and remove logit normalization since it kills inferencing)
mrq pushed to master at mrq/vall-e 2025-03-11 02:15:00 +00:00
5f98543d4d ughh
mrq pushed to master at mrq/vall-e 2025-03-11 02:12:25 +00:00
8ac03aac8a ugh
mrq pushed to master at mrq/vall-e 2025-03-11 01:55:17 +00:00
5670fcb23f hopefully the final tweaks needed for this bastard of a model
mrq pushed to master at mrq/vall-e 2025-03-08 23:05:50 +00:00
00d1fed217 another optimization (within the dataloader because the similar utterance sampler was mondo slow)
mrq pushed to master at mrq/vall-e 2025-03-08 01:27:48 +00:00
5e9d1a5302 one more time one more time (this normalization isn't a spook)
mrq pushed to master at mrq/vall-e 2025-03-08 01:10:03 +00:00
93044829af one more time (could have sworn i tested it with batch size > 1)
mrq pushed to master at mrq/vall-e 2025-03-08 00:52:19 +00:00
6cea840710 oops
mrq pushed to master at mrq/vall-e 2025-03-08 00:39:32 +00:00
dbd34b6430 add specialized calc_loss because schizo
mrq pushed to master at mrq/vall-e 2025-03-07 20:07:30 +00:00
8d848ed549 handle case of dropping cond for segment mask
mrq pushed to master at mrq/vall-e 2025-03-07 19:50:53 +00:00
89e52b9877 ugh
mrq pushed to master at mrq/vall-e 2025-03-07 19:47:20 +00:00
6afc2b7526 gut feeling to change the attention mask
mrq pushed to master at mrq/vall-e 2025-03-06 23:14:20 +00:00
91ede71cf0 ugh
mrq pushed to master at mrq/vall-e 2025-03-06 23:02:27 +00:00
2dd80a03ff stuff for interfacing with the loss scaler value (because I want to cap it)
mrq pushed to master at mrq/vall-e 2025-03-06 21:39:38 +00:00
a30dffcca7 wandb additions (to-do eventually, upload samples as artifacts)
mrq pushed to master at mrq/vall-e 2025-03-06 21:26:20 +00:00
ec87308d75 final tweaks before training this meme 44khz model for the 3rd time
mrq pushed to master at mrq/vall-e 2025-03-06 20:43:16 +00:00
5cd71ef238 QoL so I can stop having to manually inject different configs
mrq pushed to master at mrq/vall-e 2025-03-05 22:32:02 +00:00
0d809561c6 accuracy k=1 and k=80 because im probably dumb for k=10 as the default since it does not represent any usecase
2fb2b732fc wow that was fast
Compare 2 commits »