Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Sign in / Register
Toggle navigation
S
Stable Diffusion Webui
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Locked Files
Issues
0
Issues
0
List
Boards
Labels
Service Desk
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Security & Compliance
Security & Compliance
Dependency List
License Compliance
Packages
Packages
List
Container Registry
Analytics
Analytics
CI / CD
Code Review
Insights
Issues
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
novelai-storage
Stable Diffusion Webui
Commits
f510a227
Commit
f510a227
authored
Oct 19, 2022
by
AUTOMATIC1111
Committed by
GitHub
Oct 19, 2022
Browse files
Options
Browse Files
Download
Plain Diff
Merge pull request #3086 from discus0434/master
Add settings for multi-layer structure hypernetworks
parents
f894dd55
42fbda83
Changes
4
Show whitespace changes
Inline
Side-by-side
Showing
4 changed files
with
77 additions
and
21 deletions
+77
-21
modules/hypernetworks/hypernetwork.py
modules/hypernetworks/hypernetwork.py
+63
-16
modules/hypernetworks/ui.py
modules/hypernetworks/ui.py
+7
-2
modules/shared.py
modules/shared.py
+1
-1
modules/ui.py
modules/ui.py
+6
-2
No files found.
modules/hypernetworks/hypernetwork.py
View file @
f510a227
...
@@ -22,45 +22,86 @@ from modules.textual_inversion.learn_schedule import LearnRateScheduler
...
@@ -22,45 +22,86 @@ from modules.textual_inversion.learn_schedule import LearnRateScheduler
class
HypernetworkModule
(
torch
.
nn
.
Module
):
class
HypernetworkModule
(
torch
.
nn
.
Module
):
multiplier
=
1.0
multiplier
=
1.0
def
__init__
(
self
,
dim
,
state_dict
=
None
):
def
__init__
(
self
,
dim
,
state_dict
=
None
,
layer_structure
=
None
,
add_layer_norm
=
False
):
super
()
.
__init__
()
super
()
.
__init__
()
if
layer_structure
is
not
None
:
assert
layer_structure
[
0
]
==
1
,
"Multiplier Sequence should start with size 1!"
assert
layer_structure
[
-
1
]
==
1
,
"Multiplier Sequence should end with size 1!"
else
:
layer_structure
=
parse_layer_structure
(
dim
,
state_dict
)
linears
=
[]
for
i
in
range
(
len
(
layer_structure
)
-
1
):
linears
.
append
(
torch
.
nn
.
Linear
(
int
(
dim
*
layer_structure
[
i
]),
int
(
dim
*
layer_structure
[
i
+
1
])))
if
add_layer_norm
:
linears
.
append
(
torch
.
nn
.
LayerNorm
(
int
(
dim
*
layer_structure
[
i
+
1
])))
self
.
linear1
=
torch
.
nn
.
Linear
(
dim
,
dim
*
2
)
self
.
linear
=
torch
.
nn
.
Sequential
(
*
linears
)
self
.
linear2
=
torch
.
nn
.
Linear
(
dim
*
2
,
dim
)
if
state_dict
is
not
None
:
if
state_dict
is
not
None
:
self
.
load_state_dict
(
state_dict
,
strict
=
True
)
try
:
self
.
load_state_dict
(
state_dict
)
except
RuntimeError
:
self
.
try_load_previous
(
state_dict
)
else
:
else
:
for
layer
in
self
.
linear
:
self
.
linear1
.
weight
.
data
.
normal_
(
mean
=
0.0
,
std
=
0.01
)
layer
.
weight
.
data
.
normal_
(
mean
=
0.0
,
std
=
0.01
)
self
.
linear1
.
bias
.
data
.
zero_
()
layer
.
bias
.
data
.
zero_
()
self
.
linear2
.
weight
.
data
.
normal_
(
mean
=
0.0
,
std
=
0.01
)
self
.
linear2
.
bias
.
data
.
zero_
()
self
.
to
(
devices
.
device
)
self
.
to
(
devices
.
device
)
def
try_load_previous
(
self
,
state_dict
):
states
=
self
.
state_dict
()
states
[
'linear.0.bias'
]
.
copy_
(
state_dict
[
'linear1.bias'
])
states
[
'linear.0.weight'
]
.
copy_
(
state_dict
[
'linear1.weight'
])
states
[
'linear.1.bias'
]
.
copy_
(
state_dict
[
'linear2.bias'
])
states
[
'linear.1.weight'
]
.
copy_
(
state_dict
[
'linear2.weight'
])
def
forward
(
self
,
x
):
def
forward
(
self
,
x
):
return
x
+
(
self
.
linear2
(
self
.
linear1
(
x
)))
*
self
.
multiplier
return
x
+
self
.
linear
(
x
)
*
self
.
multiplier
def
trainables
(
self
):
layer_structure
=
[]
for
layer
in
self
.
linear
:
layer_structure
+=
[
layer
.
weight
,
layer
.
bias
]
return
layer_structure
def
apply_strength
(
value
=
None
):
def
apply_strength
(
value
=
None
):
HypernetworkModule
.
multiplier
=
value
if
value
is
not
None
else
shared
.
opts
.
sd_hypernetwork_strength
HypernetworkModule
.
multiplier
=
value
if
value
is
not
None
else
shared
.
opts
.
sd_hypernetwork_strength
def
parse_layer_structure
(
dim
,
state_dict
):
i
=
0
layer_structure
=
[
1
]
while
(
key
:
=
"linear.{}.weight"
.
format
(
i
))
in
state_dict
:
weight
=
state_dict
[
key
]
layer_structure
.
append
(
len
(
weight
)
//
dim
)
i
+=
1
return
layer_structure
class
Hypernetwork
:
class
Hypernetwork
:
filename
=
None
filename
=
None
name
=
None
name
=
None
def
__init__
(
self
,
name
=
None
,
enable_sizes
=
None
):
def
__init__
(
self
,
name
=
None
,
enable_sizes
=
None
,
layer_structure
=
None
,
add_layer_norm
=
False
):
self
.
filename
=
None
self
.
filename
=
None
self
.
name
=
name
self
.
name
=
name
self
.
layers
=
{}
self
.
layers
=
{}
self
.
step
=
0
self
.
step
=
0
self
.
sd_checkpoint
=
None
self
.
sd_checkpoint
=
None
self
.
sd_checkpoint_name
=
None
self
.
sd_checkpoint_name
=
None
self
.
layer_structure
=
layer_structure
self
.
add_layer_norm
=
add_layer_norm
for
size
in
enable_sizes
or
[]:
for
size
in
enable_sizes
or
[]:
self
.
layers
[
size
]
=
(
HypernetworkModule
(
size
),
HypernetworkModule
(
size
))
self
.
layers
[
size
]
=
(
HypernetworkModule
(
size
,
None
,
self
.
layer_structure
,
self
.
add_layer_norm
),
HypernetworkModule
(
size
,
None
,
self
.
layer_structure
,
self
.
add_layer_norm
),
)
def
weights
(
self
):
def
weights
(
self
):
res
=
[]
res
=
[]
...
@@ -68,7 +109,7 @@ class Hypernetwork:
...
@@ -68,7 +109,7 @@ class Hypernetwork:
for
k
,
layers
in
self
.
layers
.
items
():
for
k
,
layers
in
self
.
layers
.
items
():
for
layer
in
layers
:
for
layer
in
layers
:
layer
.
train
()
layer
.
train
()
res
+=
[
layer
.
linear1
.
weight
,
layer
.
linear1
.
bias
,
layer
.
linear2
.
weight
,
layer
.
linear2
.
bias
]
res
+=
layer
.
trainables
()
return
res
return
res
...
@@ -80,6 +121,8 @@ class Hypernetwork:
...
@@ -80,6 +121,8 @@ class Hypernetwork:
state_dict
[
'step'
]
=
self
.
step
state_dict
[
'step'
]
=
self
.
step
state_dict
[
'name'
]
=
self
.
name
state_dict
[
'name'
]
=
self
.
name
state_dict
[
'layer_structure'
]
=
self
.
layer_structure
state_dict
[
'is_layer_norm'
]
=
self
.
add_layer_norm
state_dict
[
'sd_checkpoint'
]
=
self
.
sd_checkpoint
state_dict
[
'sd_checkpoint'
]
=
self
.
sd_checkpoint
state_dict
[
'sd_checkpoint_name'
]
=
self
.
sd_checkpoint_name
state_dict
[
'sd_checkpoint_name'
]
=
self
.
sd_checkpoint_name
...
@@ -94,10 +137,15 @@ class Hypernetwork:
...
@@ -94,10 +137,15 @@ class Hypernetwork:
for
size
,
sd
in
state_dict
.
items
():
for
size
,
sd
in
state_dict
.
items
():
if
type
(
size
)
==
int
:
if
type
(
size
)
==
int
:
self
.
layers
[
size
]
=
(
HypernetworkModule
(
size
,
sd
[
0
]),
HypernetworkModule
(
size
,
sd
[
1
]))
self
.
layers
[
size
]
=
(
HypernetworkModule
(
size
,
sd
[
0
],
state_dict
[
"layer_structure"
],
state_dict
[
"is_layer_norm"
]),
HypernetworkModule
(
size
,
sd
[
1
],
state_dict
[
"layer_structure"
],
state_dict
[
"is_layer_norm"
]),
)
self
.
name
=
state_dict
.
get
(
'name'
,
self
.
name
)
self
.
name
=
state_dict
.
get
(
'name'
,
self
.
name
)
self
.
step
=
state_dict
.
get
(
'step'
,
0
)
self
.
step
=
state_dict
.
get
(
'step'
,
0
)
self
.
layer_structure
=
state_dict
.
get
(
'layer_structure'
,
None
)
self
.
add_layer_norm
=
state_dict
.
get
(
'is_layer_norm'
,
False
)
self
.
sd_checkpoint
=
state_dict
.
get
(
'sd_checkpoint'
,
None
)
self
.
sd_checkpoint
=
state_dict
.
get
(
'sd_checkpoint'
,
None
)
self
.
sd_checkpoint_name
=
state_dict
.
get
(
'sd_checkpoint_name'
,
None
)
self
.
sd_checkpoint_name
=
state_dict
.
get
(
'sd_checkpoint_name'
,
None
)
...
@@ -226,7 +274,6 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
...
@@ -226,7 +274,6 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
shared
.
state
.
textinfo
=
f
"Preparing dataset from {html.escape(data_root)}..."
shared
.
state
.
textinfo
=
f
"Preparing dataset from {html.escape(data_root)}..."
with
torch
.
autocast
(
"cuda"
):
with
torch
.
autocast
(
"cuda"
):
ds
=
modules
.
textual_inversion
.
dataset
.
PersonalizedBase
(
data_root
=
data_root
,
width
=
training_width
,
height
=
training_height
,
repeats
=
shared
.
opts
.
training_image_repeats_per_epoch
,
placeholder_token
=
hypernetwork_name
,
model
=
shared
.
sd_model
,
device
=
devices
.
device
,
template_file
=
template_file
,
include_cond
=
True
,
batch_size
=
batch_size
)
ds
=
modules
.
textual_inversion
.
dataset
.
PersonalizedBase
(
data_root
=
data_root
,
width
=
training_width
,
height
=
training_height
,
repeats
=
shared
.
opts
.
training_image_repeats_per_epoch
,
placeholder_token
=
hypernetwork_name
,
model
=
shared
.
sd_model
,
device
=
devices
.
device
,
template_file
=
template_file
,
include_cond
=
True
,
batch_size
=
batch_size
)
if
unload
:
if
unload
:
shared
.
sd_model
.
cond_stage_model
.
to
(
devices
.
cpu
)
shared
.
sd_model
.
cond_stage_model
.
to
(
devices
.
cpu
)
shared
.
sd_model
.
first_stage_model
.
to
(
devices
.
cpu
)
shared
.
sd_model
.
first_stage_model
.
to
(
devices
.
cpu
)
...
@@ -261,7 +308,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
...
@@ -261,7 +308,7 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, data_root, log
with
torch
.
autocast
(
"cuda"
):
with
torch
.
autocast
(
"cuda"
):
c
=
stack_conds
([
entry
.
cond
for
entry
in
entries
])
.
to
(
devices
.
device
)
c
=
stack_conds
([
entry
.
cond
for
entry
in
entries
])
.
to
(
devices
.
device
)
#
c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
#
c = torch.vstack([entry.cond for entry in entries]).to(devices.device)
x
=
torch
.
stack
([
entry
.
latent
for
entry
in
entries
])
.
to
(
devices
.
device
)
x
=
torch
.
stack
([
entry
.
latent
for
entry
in
entries
])
.
to
(
devices
.
device
)
loss
=
shared
.
sd_model
(
x
,
c
)[
0
]
loss
=
shared
.
sd_model
(
x
,
c
)[
0
]
del
x
del
x
...
...
modules/hypernetworks/ui.py
View file @
f510a227
...
@@ -9,11 +9,16 @@ from modules import sd_hijack, shared, devices
...
@@ -9,11 +9,16 @@ from modules import sd_hijack, shared, devices
from
modules.hypernetworks
import
hypernetwork
from
modules.hypernetworks
import
hypernetwork
def
create_hypernetwork
(
name
,
enable_sizes
):
def
create_hypernetwork
(
name
,
enable_sizes
,
layer_structure
=
None
,
add_layer_norm
=
False
):
fn
=
os
.
path
.
join
(
shared
.
cmd_opts
.
hypernetwork_dir
,
f
"{name}.pt"
)
fn
=
os
.
path
.
join
(
shared
.
cmd_opts
.
hypernetwork_dir
,
f
"{name}.pt"
)
assert
not
os
.
path
.
exists
(
fn
),
f
"file {fn} already exists"
assert
not
os
.
path
.
exists
(
fn
),
f
"file {fn} already exists"
hypernet
=
modules
.
hypernetworks
.
hypernetwork
.
Hypernetwork
(
name
=
name
,
enable_sizes
=
[
int
(
x
)
for
x
in
enable_sizes
])
hypernet
=
modules
.
hypernetworks
.
hypernetwork
.
Hypernetwork
(
name
=
name
,
enable_sizes
=
[
int
(
x
)
for
x
in
enable_sizes
],
layer_structure
=
layer_structure
,
add_layer_norm
=
add_layer_norm
,
)
hypernet
.
save
(
fn
)
hypernet
.
save
(
fn
)
shared
.
reload_hypernetworks
()
shared
.
reload_hypernetworks
()
...
...
modules/shared.py
View file @
f510a227
modules/ui.py
View file @
f510a227
...
@@ -1217,6 +1217,8 @@ def create_ui(wrap_gradio_gpu_call):
...
@@ -1217,6 +1217,8 @@ def create_ui(wrap_gradio_gpu_call):
with
gr
.
Tab
(
label
=
"Create hypernetwork"
):
with
gr
.
Tab
(
label
=
"Create hypernetwork"
):
new_hypernetwork_name
=
gr
.
Textbox
(
label
=
"Name"
)
new_hypernetwork_name
=
gr
.
Textbox
(
label
=
"Name"
)
new_hypernetwork_sizes
=
gr
.
CheckboxGroup
(
label
=
"Modules"
,
value
=
[
"768"
,
"320"
,
"640"
,
"1280"
],
choices
=
[
"768"
,
"320"
,
"640"
,
"1280"
])
new_hypernetwork_sizes
=
gr
.
CheckboxGroup
(
label
=
"Modules"
,
value
=
[
"768"
,
"320"
,
"640"
,
"1280"
],
choices
=
[
"768"
,
"320"
,
"640"
,
"1280"
])
new_hypernetwork_layer_structure
=
gr
.
Dropdown
(
label
=
"Hypernetwork layer structure"
,
choices
=
[(
1
,
2
,
1
),
(
1
,
2
,
2
,
1
),
(
1
,
2
,
4
,
2
,
1
)])
new_hypernetwork_add_layer_norm
=
gr
.
Checkbox
(
label
=
"Add layer normalization"
)
with
gr
.
Row
():
with
gr
.
Row
():
with
gr
.
Column
(
scale
=
3
):
with
gr
.
Column
(
scale
=
3
):
...
@@ -1299,6 +1301,8 @@ def create_ui(wrap_gradio_gpu_call):
...
@@ -1299,6 +1301,8 @@ def create_ui(wrap_gradio_gpu_call):
inputs
=
[
inputs
=
[
new_hypernetwork_name
,
new_hypernetwork_name
,
new_hypernetwork_sizes
,
new_hypernetwork_sizes
,
new_hypernetwork_layer_structure
,
new_hypernetwork_add_layer_norm
,
],
],
outputs
=
[
outputs
=
[
train_hypernetwork_name
,
train_hypernetwork_name
,
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment