• 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
mrq pushed to master at mrq/vall-e 2024-10-31 18:20:50 +00:00
76ebef45dc off-by-one...
mrq pushed to master at mrq/vall-e 2024-10-31 03:45:09 +00:00
mrq pushed to master at mrq/vall-e 2024-10-31 01:01:48 +00:00
a22534e8f4 layer skip training implemented (need to gut the inferencing from the repo, and to actually see if the model can benefit from this)
mrq pushed to new at mrq/tortoise-tts 2024-10-27 14:00:50 +00:00
d1e811e6ea sane download bar
mrq pushed to master at mrq/vall-e 2024-10-26 05:09:02 +00:00
4049f51ba9 added option to load lora directly from the model file itself with --lora
mrq pushed to master at mrq/vall-e 2024-10-26 03:12:55 +00:00
023c3af331 updated readme to reflect changes
mrq pushed to master at mrq/vall-e 2024-10-26 03:11:05 +00:00
ccf71dc1b6 added option to load from a model state dict directly instead of a yaml (to-do: do this for LoRAs too), automatically download the default model if none is provided
mrq pushed to master at mrq/vall-e 2024-10-26 01:35:01 +00:00
a96f5aee32 adjusted how i want to pass eval kwargs
mrq pushed to master at mrq/vall-e 2024-10-23 04:59:43 +00:00
92e6bff6dc actually ar temp 0.5 with rep pen 1.125 seems to have the benefits of better outputs without it degrading some of the time but not all the time
mrq pushed to master at mrq/vall-e 2024-10-23 03:02:08 +00:00
8920e5e86b actually have beam_width in the webUI work
mrq pushed to master at mrq/vall-e 2024-10-23 01:10:33 +00:00
910571ad34 too brainlet to diagnose why low temp / greedy sampling is randomly unstable some of the time
mrq pushed to master at mrq/vall-e 2024-10-22 23:08:54 +00:00
8eb9a4056b modified default arguments (ar temp = 0 and rep pen = 1.125 seems to be stable, at least given the few things i tested), do not pass top k/top p/min p to NAR even though technically none of those things should matter when greedy sampling
mrq pushed to master at mrq/vall-e 2024-10-22 00:47:54 +00:00
1a02cd5bce modify demo template to say F5 instead of YourTTS, swap LoRA comparison around to make the lora'd the base file, and the no-lora the suffix'd file
mrq pushed to master at mrq/vall-e 2024-10-18 22:19:13 +00:00
mrq pushed to master at mrq/vall-e 2024-10-18 22:16:57 +00:00
mrq pushed to master at mrq/vall-e 2024-10-18 22:15:44 +00:00
71731ed785 added prefixing with silence (was to test something, currently hidden under cfg.experimental=True)
mrq pushed to master at mrq/vall-e 2024-10-18 21:57:34 +00:00
6b04c13c56 print warning if audio promtpless inferencing with low AR temp (it really doesn't like low temps / greedy sampling)
mrq pushed to master at mrq/vall-e 2024-10-18 21:54:46 +00:00
c8f31db1de default to greedy sample AR (i should probably test this more but it seems to pass my harvard sentences and tongue twisters)
mrq pushed to master at mrq/vall-e 2024-10-18 21:50:51 +00:00
fc8dfd8617 made greedy AR sampling viable (and preferable), with caveats (per comment in vall_e.models.ar_nar)
mrq pushed to master at mrq/vall-e 2024-10-18 18:15:29 +00:00
07f4935a75 more tweaks