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 2024-11-10 18:44:22 +00:00
9def34cd66 lol
mrq pushed to master at mrq/vall-e 2024-11-10 18:39:20 +00:00
21cd2714a3 lol
mrq pushed to master at mrq/vall-e 2024-11-10 18:34:56 +00:00
7f6e8b1e27 lol
mrq pushed to master at mrq/vall-e 2024-11-10 18:25:58 +00:00
965d7b473d lol
mrq pushed to master at mrq/vall-e 2024-11-10 18:15:36 +00:00
9cb0b6901b unified nar.py into ar_nar.py
mrq pushed to master at mrq/vall-e 2024-11-10 04:53:24 +00:00
a9d2faf2d7 all I can do now until I wait for the model to (re)train for pure NAR
mrq pushed to master at mrq/vall-e 2024-11-10 01:35:51 +00:00
ad7e290a5e ugh (ROCm seems to silently clamp any token value >= logits.shape[-1] for loss calculation, while cuda will throw an assert, making it hard to find this dumb fuckup)
mrq pushed to master at mrq/vall-e 2024-11-10 00:59:54 +00:00
943fe70c10 I don't know why this fixes an assert thrown but it does
mrq pushed to master at mrq/vall-e 2024-11-10 00:08:40 +00:00
f50d92ba6c Almost made a mistake
mrq pushed to master at mrq/vall-e 2024-11-10 00:00:55 +00:00
c6a38693a2 This better work
mrq pushed to master at mrq/vall-e 2024-11-09 21:04:58 +00:00
8b3d1cf70a Something's Wrong
mrq pushed to master at mrq/vall-e 2024-11-09 18:09:31 +00:00
dcd5fecff3 some cleanup while I wait for the NAR-len to train to an acceptable state (currently it performs okay, but only on audo after 3 seconds or so)
mrq pushed to master at mrq/vall-e 2024-11-09 06:08:58 +00:00
69b0b3b854 set timestep tensor to whatever the time embedding's dtype is because it'll gripe under amp
mrq pushed to master at mrq/vall-e 2024-11-09 04:42:08 +00:00
5a09a5f6e9 I forgot about the time embedding...
mrq pushed to master at mrq/vall-e 2024-11-09 04:01:29 +00:00
811b15d280 I suppose I just have a shit training method since the sampler is as solid as I can get it...............
mrq pushed to master at mrq/vall-e 2024-11-08 19:30:32 +00:00
13b54953bd agony
mrq pushed to master at mrq/vall-e 2024-11-08 03:15:03 +00:00
c127c4e488 'borrowed' a sampling scheduler for NAR-len's RVQ level 0 (better than before, but still not good enough)
mrq pushed to master at mrq/vall-e 2024-11-08 01:54:37 +00:00
bfbd19c14f ugh
mrq pushed to master at mrq/vall-e 2024-11-08 00:58:43 +00:00
d17f0ebc7c 'borrowed' a sampling scheduler for NAR-len's RVQ level 0 (better than before, but still not good enough)
mrq pushed to master at mrq/vall-e 2024-11-07 17:27:54 +00:00
e108c54daf new NAR-len training paradigm......