mirror of https://github.com/vladmandic/automatic
36 lines
1.5 KiB
Python
36 lines
1.5 KiB
Python
import html
|
|
import gradio as gr
|
|
import modules.textual_inversion.textual_inversion
|
|
import modules.textual_inversion.preprocess
|
|
from modules import shared
|
|
|
|
|
|
def create_embedding(name, initialization_text, nvpt, overwrite_old):
|
|
from modules import sd_hijack
|
|
filename = modules.textual_inversion.textual_inversion.create_embedding(name, nvpt, overwrite_old, init_text=initialization_text)
|
|
sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings()
|
|
return gr.Dropdown.update(choices=sorted(sd_hijack.model_hijack.embedding_db.word_embeddings.keys())), f"Created: {filename}", ""
|
|
|
|
|
|
def preprocess(*args):
|
|
modules.textual_inversion.preprocess.preprocess(*args)
|
|
return f"Preprocessing {'interrupted' if shared.state.interrupted else 'finished'}.", ""
|
|
|
|
|
|
def train_embedding(*args):
|
|
from modules import sd_hijack
|
|
assert not shared.cmd_opts.lowvram, 'Training models with lowvram not possible'
|
|
apply_optimizations = False
|
|
try:
|
|
if not apply_optimizations:
|
|
sd_hijack.undo_optimizations()
|
|
embedding, filename = modules.textual_inversion.textual_inversion.train_embedding(*args)
|
|
res = f"Training {'interrupted' if shared.state.interrupted else 'finished'} at {embedding.step} steps. Embedding saved to {html.escape(filename)}"
|
|
return res, ""
|
|
except Exception as e:
|
|
shared.log.error(f"Exception in train_embedding: {e}")
|
|
raise RuntimeError from e
|
|
finally:
|
|
if not apply_optimizations:
|
|
sd_hijack.apply_optimizations()
|