Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Contribute to GitLab
Sign in / Register
Toggle navigation
S
stable-diffusion-webui
Project
Project
Details
Activity
Cycle Analytics
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Charts
Issues
0
Issues
0
List
Board
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Charts
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Charts
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
Administrator
stable-diffusion-webui
Commits
cffc240a
Commit
cffc240a
authored
Oct 23, 2022
by
Nerogar
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
fixed textual inversion training with inpainting models
parent
198a1ffc
Show whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
26 additions
and
1 deletion
+26
-1
textual_inversion.py
modules/textual_inversion/textual_inversion.py
+26
-1
No files found.
modules/textual_inversion/textual_inversion.py
View file @
cffc240a
...
@@ -224,6 +224,26 @@ def validate_train_inputs(model_name, learn_rate, batch_size, data_root, templat
...
@@ -224,6 +224,26 @@ def validate_train_inputs(model_name, learn_rate, batch_size, data_root, templat
if
save_model_every
or
create_image_every
:
if
save_model_every
or
create_image_every
:
assert
log_directory
,
"Log directory is empty"
assert
log_directory
,
"Log directory is empty"
def
create_dummy_mask
(
x
,
width
=
None
,
height
=
None
):
if
shared
.
sd_model
.
model
.
conditioning_key
in
{
'hybrid'
,
'concat'
}:
# The "masked-image" in this case will just be all zeros since the entire image is masked.
image_conditioning
=
torch
.
zeros
(
x
.
shape
[
0
],
3
,
height
,
width
,
device
=
x
.
device
)
image_conditioning
=
shared
.
sd_model
.
get_first_stage_encoding
(
shared
.
sd_model
.
encode_first_stage
(
image_conditioning
))
# Add the fake full 1s mask to the first dimension.
image_conditioning
=
torch
.
nn
.
functional
.
pad
(
image_conditioning
,
(
0
,
0
,
0
,
0
,
1
,
0
),
value
=
1.0
)
image_conditioning
=
image_conditioning
.
to
(
x
.
dtype
)
else
:
# Dummy zero conditioning if we're not using inpainting model.
# Still takes up a bit of memory, but no encoder call.
# Pretty sure we can just make this a 1x1 image since its not going to be used besides its batch size.
image_conditioning
=
torch
.
zeros
(
x
.
shape
[
0
],
5
,
1
,
1
,
dtype
=
x
.
dtype
,
device
=
x
.
device
)
return
image_conditioning
def
train_embedding
(
embedding_name
,
learn_rate
,
batch_size
,
data_root
,
log_directory
,
training_width
,
training_height
,
steps
,
create_image_every
,
save_embedding_every
,
template_file
,
save_image_with_stored_embedding
,
preview_from_txt2img
,
preview_prompt
,
preview_negative_prompt
,
preview_steps
,
preview_sampler_index
,
preview_cfg_scale
,
preview_seed
,
preview_width
,
preview_height
):
def
train_embedding
(
embedding_name
,
learn_rate
,
batch_size
,
data_root
,
log_directory
,
training_width
,
training_height
,
steps
,
create_image_every
,
save_embedding_every
,
template_file
,
save_image_with_stored_embedding
,
preview_from_txt2img
,
preview_prompt
,
preview_negative_prompt
,
preview_steps
,
preview_sampler_index
,
preview_cfg_scale
,
preview_seed
,
preview_width
,
preview_height
):
save_embedding_every
=
save_embedding_every
or
0
save_embedding_every
=
save_embedding_every
or
0
create_image_every
=
create_image_every
or
0
create_image_every
=
create_image_every
or
0
...
@@ -286,6 +306,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
...
@@ -286,6 +306,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
forced_filename
=
"<none>"
forced_filename
=
"<none>"
embedding_yet_to_be_embedded
=
False
embedding_yet_to_be_embedded
=
False
img_c
=
None
pbar
=
tqdm
.
tqdm
(
enumerate
(
ds
),
total
=
steps
-
ititial_step
)
pbar
=
tqdm
.
tqdm
(
enumerate
(
ds
),
total
=
steps
-
ititial_step
)
for
i
,
entries
in
pbar
:
for
i
,
entries
in
pbar
:
embedding
.
step
=
i
+
ititial_step
embedding
.
step
=
i
+
ititial_step
...
@@ -299,8 +320,12 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
...
@@ -299,8 +320,12 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
with
torch
.
autocast
(
"cuda"
):
with
torch
.
autocast
(
"cuda"
):
c
=
cond_model
([
entry
.
cond_text
for
entry
in
entries
])
c
=
cond_model
([
entry
.
cond_text
for
entry
in
entries
])
if
img_c
is
None
:
img_c
=
create_dummy_mask
(
c
,
training_width
,
training_height
)
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
]
cond
=
{
"c_concat"
:
[
img_c
],
"c_crossattn"
:
[
c
]}
loss
=
shared
.
sd_model
(
x
,
cond
)[
0
]
del
x
del
x
losses
[
embedding
.
step
%
losses
.
shape
[
0
]]
=
loss
.
item
()
losses
[
embedding
.
step
%
losses
.
shape
[
0
]]
=
loss
.
item
()
...
...
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