import html import os import gradio as gr import modules.textual_inversion.textual_inversion import modules.textual_inversion.preprocess from modules import sd_hijack, shared, devices from modules.hypernetworks import hypernetwork def create_hypernetwork(name, enable_sizes): fn = os.path.join(shared.cmd_opts.hypernetwork_dir, f"{name}.pt") 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.save(fn) shared.reload_hypernetworks() return gr.Dropdown.update(choices=sorted([x for x in shared.hypernetworks.keys()])), f"Created: {fn}", "" def train_hypernetwork(*args): initial_hypernetwork = shared.loaded_hypernetwork assert not shared.cmd_opts.lowvram, 'Training models with lowvram is not possible' try: sd_hijack.undo_optimizations() hypernetwork, filename = modules.hypernetworks.hypernetwork.train_hypernetwork(*args) res = f""" Training {'interrupted' if shared.state.interrupted else 'finished'} at {hypernetwork.step} steps. Hypernetwork saved to {html.escape(filename)} """ return res, "" except Exception: raise finally: shared.loaded_hypernetwork = initial_hypernetwork shared.sd_model.cond_stage_model.to(devices.device) shared.sd_model.first_stage_model.to(devices.device) sd_hijack.apply_optimizations()