 v0xie
|
fd383140cf
fix: wrong devices for eye and constraint
|
преди 1 година |
 Kohaku-Blueleaf
|
f8f38c7c28
Fix dtype casting for OFT module
|
преди 1 година |
|
Kohaku-Blueleaf
|
3772a82a70
better naming and correct order for device.
|
преди 1 година |
|
Kohaku-Blueleaf
|
8fc67f3851
remove debug print
|
преди 1 година |
|
Kohaku-Blueleaf
|
265bc26c21
Use self.scale instead of custom finalize
|
преди 1 година |
|
Kohaku-Blueleaf
|
735c9e8059
Fix network_oft
|
преди 1 година |
|
v0xie
|
eb667e715a
feat: LyCORIS/kohya OFT network support
|
преди 1 година |
|
v0xie
|
d6d0b22e66
fix: ignore calc_scale() for COFT which has very small alpha
|
преди 1 година |
|
v0xie
|
bbf00a96af
refactor: remove unused function
|
преди 1 година |
|
v0xie
|
329c8bacce
refactor: use same updown for both kohya OFT and LyCORIS diag-oft
|
преди 1 година |
|
v0xie
|
f6c8201e56
refactor: move factorization to lyco_helpers, separate calc_updown for kohya and kb
|
преди 1 година |
|
v0xie
|
fe1967a4c4
skip multihead attn for now
|
преди 1 година |
|
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
|
преди 1 година |
|
v0xie
|
a2fad6ee05
test implementation based on kohaku diag-oft implementation
|
преди 1 година |
|
v0xie
|
6523edb8a4
style: conform style
|
преди 1 година |
|
v0xie
|
3b8515d2c9
fix: multiplier applied twice in finalize_updown
|
преди 1 година |
|
v0xie
|
4a50c9638c
refactor: remove used OFT functions
|
преди 1 година |
|
v0xie
|
de8ee92ed8
fix: use merge_weight to cache value
|
преди 1 година |
|
v0xie
|
76f5abdbdb
style: cleanup oft
|
преди 1 година |
|
v0xie
|
fce86ab7d7
fix: support multiplier, no forward pass hook
|
преди 1 година |
|
v0xie
|
7683547728
fix: return orig weights during updown, merge weights before forward
|
преди 1 година |
|
v0xie
|
2d8c894b27
refactor: use forward hook instead of custom forward
|
преди 1 година |
|
v0xie
|
0550659ce6
style: fix ambiguous variable name
|
преди 1 година |
|
v0xie
|
d10c4db57e
style: formatting
|
преди 1 година |
|
v0xie
|
321680ccd0
refactor: fix constraint, re-use get_weight
|
преди 1 година |
|
v0xie
|
eb01d7f0e0
faster by calculating R in updown and using cached R in forward
|
преди 1 година |
|
v0xie
|
853e21d98e
faster by using cached R in forward
|
преди 1 година |
|
v0xie
|
1c6efdbba7
inference working but SLOW
|
преди 1 година |
|
v0xie
|
ec718f76b5
wip incorrect OFT implementation
|
преди 1 година |
