02ebda42f2
#yolo
2022-07-21 00:43:03 +07:00
b92ff8de78
misc
2022-07-20 23:59:32 +07:00
a1743d26aa
Revert "Try to squeeze a bit more performance out of this arch"
...
This reverts commit 767f963392
2022-07-20 23:57:56 +07:00
767f963392
Try to squeeze a bit more performance out of this arch
2022-07-20 23:51:11 +07:00
b9d0f7e6de
simplify parameterization a bit
2022-07-20 23:41:54 +07:00
ee8ceed6da
rework tfd13 further
...
- use a gated activation layer for both attention & convs
- add a relativistic learned position bias. I believe this is similar to the T5 position encodings but it is simpler and learned
- get rid of prepending to the attention matrix - this doesn't really work that well. the model eventually learns to attend one of its heads to these blocks but why not just concat if it is doing that?
2022-07-20 23:28:29 +07:00
40427de8e3
update tfd13 for inference
2022-07-20 21:51:25 +07:00
dbebe18602
Fix ts=0 with new formulation
2022-07-20 12:12:33 +07:00
82bd62019f
diffuse the cascaded prior for continuous sr model
2022-07-20 11:54:09 +07:00
b0e3be0a17
transition to nearest interpolation mode for downsampling
2022-07-20 10:56:17 +07:00
7b3fc79737
iq checkin
2022-07-20 10:19:32 +07:00
9a37f3ba42
reminder to future self
2022-07-20 10:19:15 +07:00
15decfdb98
misc
2022-07-20 10:19:02 +07:00
2997a640b0
fix mdf
2022-07-19 19:39:29 +07:00
c14bf6dfb2
fix conditioning free
2022-07-19 18:04:49 +07:00
fc0b291b21
do masking up proper
2022-07-19 16:32:17 +07:00
b203a7dc97
And remove unused parameters
2022-07-19 15:05:12 +07:00
17a07b2e33
readd one mdf function
2022-07-19 15:04:36 +07:00
c00398e955
scope attention in tfd13 as well
2022-07-19 14:59:43 +07:00
b157b28c7b
tfd14
...
hopefully this helps address the positional dependencies of tfd12
2022-07-19 13:30:05 +07:00
4597447178
add assertions to mel generator script
2022-07-19 11:23:54 +07:00
1b6fe88bcb
spit out overages in GDI
2022-07-19 11:19:59 +07:00
73d7211a4c
fix script
2022-07-19 11:17:43 +07:00
6b1cfe8e66
ugh
2022-07-19 11:14:20 +07:00
da9e47ca0e
new bounds for MEL normalization and multi-resolution SR in MDF
2022-07-19 11:11:46 +07:00
eecb534e66
a few fixes to multiresolution sr
2022-07-19 11:11:15 +07:00
2fb85526bc
mdf cleanup
2022-07-19 09:57:05 +07:00
4aa840a494
be more stringent on min and max ranges in GDI.. this is gonna break somethings probably
2022-07-19 09:14:08 +07:00
eab7dc339d
iq checkin
...
who knows where I'm going with this.. I don't even know sometimes..
2022-07-19 09:13:27 +07:00
625d7b6f38
music joiner checkin
2022-07-18 18:40:25 +07:00
0824708dc7
iq checkin
2022-07-18 18:40:14 +07:00
df27b98730
ddp doesnt like dropout on checkpointed values
2022-07-18 17:17:04 +07:00
8d7692c1e0
uh
2022-07-18 17:15:27 +07:00
c959e530cb
good ole ddp..
2022-07-18 17:13:45 +07:00
cf57c352c8
Another fix
2022-07-18 17:09:13 +07:00
83a4ef4149
default to use input for conditioning & add preprocessed input to GDI
2022-07-18 17:01:19 +07:00
1b4d9567f3
tfd13 for multi-resolution superscaling
2022-07-18 16:36:22 +07:00
1b648abd7c
iq2
2022-07-18 10:12:23 +07:00
7a10c3fed8
commit my own version of vq, with a fix for cosine similarity and support for masking
2022-07-18 10:12:17 +07:00
4ce5c31705
music_joiner update for new big cosine model
2022-07-17 18:24:56 +07:00
cb24bef406
ddim_with_guidance
2022-07-17 18:24:43 +07:00
20ef9cc6b4
iq checkin
...
yeah - I'm at it again...
2022-07-17 18:24:33 +07:00
a073fbfcb8
force tf32 to be enabled (torch 1.12 disables it)
2022-07-16 13:59:07 +07:00
438dcaccc5
misc
2022-07-16 13:58:52 +07:00
e13b1adfdb
:<
2022-07-14 21:52:23 +07:00
51291ab070
Some additional context regularization in tfd
2022-07-14 21:49:47 +07:00
7170ccdfa9
Merge branch 'baktfd'
2022-07-14 21:38:45 +07:00
3b12d348fc
mdf: provide conditioning margin
2022-07-14 21:38:14 +07:00
fa352e2744
also some good assert text
2022-07-14 21:26:22 +07:00
4d53c66602
simplify span selecting logic in tfdpc
2022-07-14 21:25:03 +07:00
James Betker