-
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
74abd5afad
code cleanup, lots of config args so I don't litter up the place with commented out experiments
D:\SteamLibrary\steamapps\common\xVATrainer\resources\app\datasets\halo_cortana\wavs
Per the documentation:
…
Well, to start with, you're on an outdated version, so I…
779c595b99
tons of fixups (proper input tensor sizes, integrated BnB as much as I can, GC every step because it eat up VRAM), added script to download already prepared dataset
I'm stupid. To spare the gory details:
- BitsAndBytes obviously won't work on everything as inputs are already ints, I don't know what brain worms made me forget this
- which begs the…
what dataset you use currently.
LibriSpeech-Finetuning. I have a rather nasty way of preparing it that I need to streamline better (or just provide the already quantized audio and phonemized…
I only can give rough estimates given:
- cards I've tested (6GiB 2060, 16GiB A4000, 16+16GiB 6800XTs, 80GiB A100)
- semi-sane batch sizes
- using bitsandbytes
- a semi-normally distributed (by…
Use a small subset then.
The other main problem, I imagine, is using too large of a dataset for latents and expecting things to be peachy keen when you're just muddying up shit when it's all…
Must be something with DLAS; I actually used the same latents file with the old and new model when testing.
Strange. I suppose I'll have my 2060 bake up a finetune throughout the day for…
48a4a5d3cc
provided transcription ahead of time of the librispeech_finetune dataset, so you don't need to spend an hour transcribing it with whisperX + VAD
9d743cdade
properly parse spaces, added option to define optimizer in yaml because I actually kept adding it in my testing copy and not the repo copy
444bcdaf62
my sanitizer actually did work, it was just batch sizes leading to problems when transcribing