Merge branch 'AUTOMATIC1111:master' into Italian

modules/api/api.py

@@ -40,6 +40,7 @@ class Api:
         self.app.add_api_route("/sdapi/v1/extra-batch-images", self.extras_batch_images_api, methods=["POST"], response_model=ExtrasBatchImagesResponse)
         self.app.add_api_route("/sdapi/v1/png-info", self.pnginfoapi, methods=["POST"], response_model=PNGInfoResponse)
         self.app.add_api_route("/sdapi/v1/progress", self.progressapi, methods=["GET"], response_model=ProgressResponse)
+        self.app.add_api_route("/sdapi/v1/interrupt", self.interruptapi, methods=["POST"])
 
     def text2imgapi(self, txt2imgreq: StableDiffusionTxt2ImgProcessingAPI):
         sampler_index = sampler_to_index(txt2imgreq.sampler_index)
@@ -176,6 +177,11 @@ class Api:
 
         return ProgressResponse(progress=progress, eta_relative=eta_relative, state=shared.state.dict(), current_image=current_image)
 
+    def interruptapi(self):
+        shared.state.interrupt()
+
+        return {}
+
     def launch(self, server_name, port):
         self.app.include_router(self.router)
         uvicorn.run(self.app, host=server_name, port=port)

modules/extras.py

@@ -141,7 +141,7 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
                             upscaling_resize_w, upscaling_resize_h, upscaling_crop)
             cache_key = LruCache.Key(image_hash=hash(np.array(image.getdata()).tobytes()),
                                      info_hash=hash(info),
-                                     args_hash=hash(upscale_args))
+                                     args_hash=hash((upscale_args, upscale_first)))
             cached_entry = cached_images.get(cache_key)
             if cached_entry is None:
                 res = upscale(image, *upscale_args)

modules/images.py

@@ -510,6 +510,7 @@ def save_image(image, path, basename, seed=None, prompt=None, extension='png', i
 
     if extension.lower() == '.png':
         pnginfo_data = PngImagePlugin.PngInfo()
+        if opts.enable_pnginfo:
             for k, v in params.pnginfo.items():
                 pnginfo_data.add_text(k, str(v))
 

modules/interrogate.py

@@ -56,9 +56,9 @@ class InterrogateModels:
         import clip
 
         if self.running_on_cpu:
-            model, preprocess = clip.load(clip_model_name, device="cpu")
+            model, preprocess = clip.load(clip_model_name, device="cpu", download_root=shared.cmd_opts.clip_models_path)
         else:
-            model, preprocess = clip.load(clip_model_name)
+            model, preprocess = clip.load(clip_model_name, download_root=shared.cmd_opts.clip_models_path)
 
         model.eval()
         model = model.to(devices.device_interrogate)

modules/shared.py

@@ -51,6 +51,7 @@ parser.add_argument("--realesrgan-models-path", type=str, help="Path to director
 parser.add_argument("--scunet-models-path", type=str, help="Path to directory with ScuNET model file(s).", default=os.path.join(models_path, 'ScuNET'))
 parser.add_argument("--swinir-models-path", type=str, help="Path to directory with SwinIR model file(s).", default=os.path.join(models_path, 'SwinIR'))
 parser.add_argument("--ldsr-models-path", type=str, help="Path to directory with LDSR model file(s).", default=os.path.join(models_path, 'LDSR'))
+parser.add_argument("--clip-models-path", type=str, help="Path to directory with CLIP model file(s).", default=None)
 parser.add_argument("--xformers", action='store_true', help="enable xformers for cross attention layers")
 parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work")
 parser.add_argument("--deepdanbooru", action='store_true', help="enable deepdanbooru interrogator")
@@ -288,11 +289,12 @@ options_templates.update(options_section(('system', "System"), {
 }))
 
 options_templates.update(options_section(('training', "Training"), {
-    "unload_models_when_training": OptionInfo(False, "Move VAE and CLIP to RAM when training hypernetwork. Saves VRAM."),
+    "unload_models_when_training": OptionInfo(False, "Move VAE and CLIP to RAM when training if possible. Saves VRAM."),
     "dataset_filename_word_regex": OptionInfo("", "Filename word regex"),
     "dataset_filename_join_string": OptionInfo(" ", "Filename join string"),
     "training_image_repeats_per_epoch": OptionInfo(1, "Number of repeats for a single input image per epoch; used only for displaying epoch number", gr.Number, {"precision": 0}),
     "training_write_csv_every": OptionInfo(500, "Save an csv containing the loss to log directory every N steps, 0 to disable"),
+    "training_xattention_optimizations": OptionInfo(False, "Use cross attention optimizations while training"),
 }))
 
 options_templates.update(options_section(('sd', "Stable Diffusion"), {

modules/textual_inversion/textual_inversion.py

@@ -235,6 +235,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
     filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding_name}.pt')
 
     log_directory = os.path.join(log_directory, datetime.datetime.now().strftime("%Y-%m-%d"), embedding_name)
+    unload = shared.opts.unload_models_when_training
 
     if save_embedding_every > 0:
         embedding_dir = os.path.join(log_directory, "embeddings")
@@ -272,6 +273,8 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
     shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
     with torch.autocast("cuda"):
         ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=embedding_name, model=shared.sd_model, device=devices.device, template_file=template_file, batch_size=batch_size)
+    if unload:
+        shared.sd_model.first_stage_model.to(devices.cpu)
 
     embedding.vec.requires_grad = True
     optimizer = torch.optim.AdamW([embedding.vec], lr=scheduler.learn_rate)
@@ -328,6 +331,9 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
         if images_dir is not None and steps_done % create_image_every == 0:
             forced_filename = f'{embedding_name}-{steps_done}'
             last_saved_image = os.path.join(images_dir, forced_filename)
+
+            shared.sd_model.first_stage_model.to(devices.device)
+
             p = processing.StableDiffusionProcessingTxt2Img(
                 sd_model=shared.sd_model,
                 do_not_save_grid=True,
@@ -355,6 +361,9 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
             processed = processing.process_images(p)
             image = processed.images[0]
 
+            if unload:
+                shared.sd_model.first_stage_model.to(devices.cpu)
+
             shared.state.current_image = image
 
             if save_image_with_stored_embedding and os.path.exists(last_saved_file) and embedding_yet_to_be_embedded:
@@ -400,6 +409,7 @@ Last saved image: {html.escape(last_saved_image)}<br/>
 
     filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding_name}.pt')
     save_embedding(embedding, checkpoint, embedding_name, filename, remove_cached_checksum=True)
+    shared.sd_model.first_stage_model.to(devices.device)
 
     return embedding, filename
 

modules/textual_inversion/ui.py

@@ -25,7 +25,9 @@ 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)
@@ -38,5 +40,6 @@ Embedding saved to {html.escape(filename)}
     except Exception:
         raise
     finally:
+        if not apply_optimizations:
             sd_hijack.apply_optimizations()