| import html | |
| import gradio as gr | |
| import modules.textual_inversion.textual_inversion | |
| import modules.textual_inversion.preprocess | |
| from modules import sd_hijack, shared | |
| def create_embedding(name, initialization_text, nvpt, overwrite_old): | |
| 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): | |
| assert not shared.cmd_opts.lowvram, 'Training models with lowvram not possible' | |
| apply_optimizations = shared.opts.training_xattention_optimizations | |
| 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: | |
| raise | |
| finally: | |
| if not apply_optimizations: | |
| sd_hijack.apply_optimizations() | |