 v0xie
|
07805cbeee
fix: AttributeError when attempting to reshape rescale by org_module weight
|
il y a 1 an |
 Kohaku-Blueleaf
|
c4afdb7895
For no constraint
|
il y a 1 an |
|
Kohaku-Blueleaf
|
64179c3221
Update network_oft.py
|
il y a 1 an |
|
Kohaku-Blueleaf
|
591470d86d
linting
|
il y a 1 an |
|
Kohaku-Blueleaf
|
a5436a3ac0
Update network_oft.py
|
il y a 1 an |
|
Kohaku-Blueleaf
|
4eb949625c
prevent undefined variable
|
il y a 1 an |
|
Kohaku-Blueleaf
|
5a8dd0c549
Fix rescale
|
il y a 1 an |
|
Kohaku-Blueleaf
|
90441294db
Add rescale mechanism
|
il y a 1 an |
|
v0xie
|
613b0d9548
doc: add boft comment
|
il y a 1 an |
|
v0xie
|
325eaeb584
fix: get boft params from weight shape
|
il y a 1 an |
|
v0xie
|
2f1073dc6e
style: fix lint
|
il y a 1 an |
|
v0xie
|
a4668a16b6
fix: calculate butterfly factor
|
il y a 1 an |
|
v0xie
|
9588721197
feat: support LyCORIS BOFT
|
il y a 1 an |
|
v0xie
|
fd383140cf
fix: wrong devices for eye and constraint
|
il y a 1 an |
|
Kohaku-Blueleaf
|
f8f38c7c28
Fix dtype casting for OFT module
|
il y a 1 an |
|
Kohaku-Blueleaf
|
3772a82a70
better naming and correct order for device.
|
il y a 1 an |
|
Kohaku-Blueleaf
|
8fc67f3851
remove debug print
|
il y a 1 an |
|
Kohaku-Blueleaf
|
265bc26c21
Use self.scale instead of custom finalize
|
il y a 1 an |
|
Kohaku-Blueleaf
|
735c9e8059
Fix network_oft
|
il y a 1 an |
|
v0xie
|
eb667e715a
feat: LyCORIS/kohya OFT network support
|
il y a 1 an |
|
v0xie
|
d6d0b22e66
fix: ignore calc_scale() for COFT which has very small alpha
|
il y a 1 an |
|
v0xie
|
bbf00a96af
refactor: remove unused function
|
il y a 1 an |
|
v0xie
|
329c8bacce
refactor: use same updown for both kohya OFT and LyCORIS diag-oft
|
il y a 1 an |
|
v0xie
|
f6c8201e56
refactor: move factorization to lyco_helpers, separate calc_updown for kohya and kb
|
il y a 1 an |
|
v0xie
|
fe1967a4c4
skip multihead attn for now
|
il y a 1 an |
|
v0xie
|
d727ddfccd
no idea what i'm doing, trying to support both type of OFT, kblueleaf diag_oft has MultiheadAttn which kohya's doesn't?, attempt create new module based off network_lora.py, errors about tensor dim mismatch
|
il y a 1 an |
|
v0xie
|
a2fad6ee05
test implementation based on kohaku diag-oft implementation
|
il y a 1 an |
|
v0xie
|
6523edb8a4
style: conform style
|
il y a 1 an |
|
v0xie
|
3b8515d2c9
fix: multiplier applied twice in finalize_updown
|
il y a 1 an |
|
v0xie
|
4a50c9638c
refactor: remove used OFT functions
|
il y a 1 an |
