 Lengyue
|
4304fab202
Optimize segment guessing
|
2 anni fa |
|
Lengyue
|
1bfec1240e
Add tts text cleaner
|
2 anni fa |
|
Lengyue
|
66219b20cf
Updata llama pretain config (max 100 val epoch)
|
2 anni fa |
|
Lengyue
|
e4839d6e84
Use 4 workers
|
2 anni fa |
|
Lengyue
|
f93e89ab06
Update dataset
|
2 anni fa |
|
Lengyue
|
d16c9cb695
Update huggingface dataset builder
|
2 anni fa |
|
Lengyue
|
a03b1b2767
Optimize logger & add multilingual data
|
2 anni fa |
|
Lengyue
|
f7f2c03282
Support pytorch lightning
|
2 anni fa |
|
Lengyue
|
6d0e669f8c
Rename project to fish-speech
|
2 anni fa |
|
Lengyue
|
423e1c0cfa
Add llama finetune command
|
2 anni fa |
|
Lengyue
|
f4e8633363
Move flash whisper
|
2 anni fa |
|
Lengyue
|
c06704fb03
Optimize script & logger
|
2 anni fa |
|
Lengyue
|
491b1d7125
rename folder
|
2 anni fa |
|
Lengyue
|
b530faccc9
Lint & fix attention mask & support whisper input features
|
2 anni fa |
|
Lengyue
|
254e93e632
Fix inference without attention mask
|
2 anni fa |
|
Lengyue
|
0c084ec779
Apply gradient accumulate to save memory
|
2 anni fa |
|
Lengyue
|
4cfb62a08a
Fix gradient accumulate logging
|
2 anni fa |
|
Lengyue
|
086c82d789
Fix vq whisper training & remove ema in tb logger
|
2 anni fa |
|
Lengyue
|
23109ecb94
Optimize training loop & upgrade transformers
|
2 anni fa |
|
Lengyue
|
7704bb7d09
Fix filename
|
2 anni fa |
|
Lengyue
|
938eca21a6
Rename config & smaller schedule
|
2 anni fa |
|
Lengyue
|
a003e5a390
Add utilities & whisper vq model
|
2 anni fa |
|
Lengyue
|
a114dfab4f
Fix multi lingual whisper
|
2 anni fa |
|
Lengyue
|
9153fc8278
Add whisper loss example
|
2 anni fa |
|
Lengyue
|
1854897e3c
Fix utf-8 encoding
|
2 anni fa |
|
Lengyue
|
0687dd1f1a
Greatly reduced cpu usage
|
2 anni fa |
|
Lengyue
|
18af8a9406
Add whisper with flash attn & use whisper to batch transcribe
|
2 anni fa |
|
Lengyue
|
9920873cf2
Optimize schedule
|
2 anni fa |
|
Lengyue
|
7031d18b3d
Let model fail on non-inf loss
|
2 anni fa |
|
Lengyue
|
5f01798ab4
Fix non-inf gradient
|
2 anni fa |
