initial support for training textual inversion

modules/textual_inversion/ui.py

@@ -0,0 +1,32 @@
+import html
+
+import gradio as gr
+
+import modules.textual_inversion.textual_inversion as ti
+from modules import sd_hijack, shared
+
+
+def create_embedding(name, nvpt):
+    filename = ti.create_embedding(name, nvpt)
+
+    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 train_embedding(*args):
+
+    try:
+        sd_hijack.undo_optimizations()
+
+        embedding, filename = ti.train_embedding(*args)
+
+        res = f"""
+Training {'interrupted' if shared.state.interrupted else 'finished'} after {embedding.step} steps.
+Embedding saved to {html.escape(filename)}
+"""
+        return res, ""
+    except Exception:
+        raise
+    finally:
+        sd_hijack.apply_optimizations()