update

.gitignore

@@ -1,14 +1,16 @@
 # defaults
 __pycache__
 /params.txt
+/cache.json
 
-# all models
+# all models and temp files
 *.ckpt
 *.safetensors
 *.pth
 *.pt
 *.bin
 *.optim
+*.lock
 
 # all dynamic stuff
 /repositories/**/*

README.md

@@ -20,8 +20,9 @@ Fork adds extra functionality:
 
 Fork does differ in few things:
 
-- Different start script  
+- Additional simplified start script  
   > ./automatic.sh  
+  Existing `webui.sh` still exists for fresh installs to auto-install dependencies, etc.  
 - Drops compatibility with `python` **3.7** and requires **3.10**  
 - Updated **Python** libraries to latest known compatible versions  
   e.g. `accelerate`, `transformers`, `numpy`, etc.  
@@ -35,14 +36,10 @@ Fork does differ in few things:
 - Modified training templates  
 
 Only Python library which is not auto-updated is `PyTorch` itself as that is very system specific  
-I'm currently using **PyTorch 2.0-nightly** compiled with **CUDA 11.8** and with **Triton** optimizations:
+For some Torch optimizations notes, see Wiki
 
-> pip3 install --pre torch torchvision torchaudio torchtriton --extra-index-url https://download.pytorch.org/whl/nightly/cu118 --force 
-> pip show torch  
-> 2.0.0.dev20230113+cu118  
-
-- Fork is compatible with regular **PyTorch 1.13**,  
-  But if you're using **PyTorch 2.0** models will be auto-compiled and optimized on load using `max-tune`  
+Fork is compatible with regular **PyTorch 1.13** as well as pre-release of **PyTorch 2.0**  
+See [Wiki](https://github.com/vladmandic/automatic/wiki) for **Torch** optimization notes
 
 
 <br>

TODO.md

@@ -32,9 +32,8 @@ Things I'm looking into...
 Stuff to be fixed...
 
 - Torch 2.0 model compile
-- Add DeepSpeed support
+- Add DeepSpeed support for training (`accelerate config`)
 - Reconnect WebUI
-- Model reload performance
 
 Tech that can be integrated as part of the core workflow...
 

automatic.sh

@@ -4,5 +4,5 @@ export TF_CPP_MIN_LOG_LEVEL=2
 export FORCE_CUDA="1"
 export ACCELERATE="True"
 export ATTN_PRECISION=fp16export PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
-python launch.py --api --xformers --disable-console-progressbars 
+python launch.py --api --xformers --disable-console-progressbars
 # python launch.py --api  --disable-console-progressbars --opt-channelslast

config.json

@@ -1,164 +0,0 @@
-{
-    "samples_save": true,
-    "samples_format": "jpg",
-    "samples_filename_pattern": "",
-    "save_images_add_number": true,
-    "grid_save": true,
-    "grid_format": "jpg",
-    "grid_extended_filename": true,
-    "grid_only_if_multiple": true,
-    "grid_prevent_empty_spots": false,
-    "n_rows": -1,
-    "enable_pnginfo": true,
-    "save_txt": false,
-    "save_images_before_face_restoration": false,
-    "save_images_before_highres_fix": false,
-    "save_images_before_color_correction": false,
-    "jpeg_quality": 80,
-    "export_for_4chan": false,
-    "use_original_name_batch": true,
-    "use_upscaler_name_as_suffix": true,
-    "save_selected_only": true,
-    "do_not_add_watermark": true,
-    "temp_dir": "",
-    "clean_temp_dir_at_start": true,
-    "outdir_samples": "",
-    "outdir_txt2img_samples": "outputs/text",
-    "outdir_img2img_samples": "outputs/image",
-    "outdir_extras_samples": "outputs/extras",
-    "outdir_grids": "",
-    "outdir_txt2img_grids": "outputs/grids",
-    "outdir_img2img_grids": "outputs/grids",
-    "outdir_save": "outputs/save",
-    "save_to_dirs": false,
-    "grid_save_to_dirs": false,
-    "use_save_to_dirs_for_ui": false,
-    "directories_filename_pattern": "",
-    "directories_max_prompt_words": 16,
-    "ESRGAN_tile": 192,
-    "ESRGAN_tile_overlap": 8,
-    "realesrgan_enabled_models": [
-        "R-ESRGAN 4x+"
-    ],
-    "upscaler_for_img2img": "SwinIR_4x",
-    "use_scale_latent_for_hires_fix": false,
-    "face_restoration_model": "CodeFormer",
-    "code_former_weight": 0.15,
-    "face_restoration_unload": false,
-    "memmon_poll_rate": 1,
-    "samples_log_stdout": false,
-    "multiple_tqdm": true,
-    "unload_models_when_training": false,
-    "pin_memory": true,
-    "save_optimizer_state": false,
-    "dataset_filename_word_regex": "",
-    "dataset_filename_join_string": " ",
-    "training_image_repeats_per_epoch": 1,
-    "training_write_csv_every": 10.0,
-    "training_xattention_optimizations": false,
-    "sd_model_checkpoint": "db-analog-diffusion-v10.ckpt",
-    "sd_checkpoint_cache": 0,
-    "sd_vae": "vae-ft-mse-840000-ema-pruned.ckpt",
-    "sd_vae_as_default": false,
-    "sd_hypernetwork": "None",
-    "sd_hypernetwork_strength": 1.0,
-    "inpainting_mask_weight": 1.0,
-    "initial_noise_multiplier": 1.0,
-    "img2img_color_correction": false,
-    "img2img_fix_steps": false,
-    "img2img_background_color": "#ffffff",
-    "enable_quantization": true,
-    "enable_emphasis": true,
-    "use_old_emphasis_implementation": false,
-    "enable_batch_seeds": true,
-    "comma_padding_backtrack": 20,
-    "CLIP_stop_at_last_layers": 1,
-    "random_artist_categories": [],
-    "interrogate_keep_models_in_memory": false,
-    "interrogate_use_builtin_artists": false,
-    "interrogate_return_ranks": true,
-    "interrogate_clip_num_beams": 1,
-    "interrogate_clip_min_length": 32,
-    "interrogate_clip_max_length": 1024,
-    "interrogate_clip_dict_limit": 0.0,
-    "interrogate_deepbooru_score_threshold": 0.7,
-    "deepbooru_sort_alpha": false,
-    "deepbooru_use_spaces": false,
-    "deepbooru_escape": true,
-    "deepbooru_filter_tags": "",
-    "show_progressbar": true,
-    "show_progress_every_n_steps": -1,
-    "show_progress_type": "Full",
-    "show_progress_grid": true,
-    "return_grid": true,
-    "do_not_show_images": false,
-    "add_model_hash_to_info": true,
-    "add_model_name_to_info": true,
-    "disable_weights_auto_swap": false,
-    "send_seed": true,
-    "send_size": true,
-    "font": "",
-    "js_modal_lightbox": true,
-    "js_modal_lightbox_initially_zoomed": true,
-    "show_progress_in_title": true,
-    "quicksettings": "sd_model_checkpoint",
-    "localization": "None",
-    "hide_samplers": [
-        "Euler",
-        "LMS",
-        "Heun",
-        "DPM2",
-        "DPM2 a",
-        "DPM++ 2M",
-        "DPM fast",
-        "DPM adaptive",
-        "DPM++ 2S a Karras",
-        "DDIM",
-        "PLMS",
-        "DPM++ 2S a",
-        "DPM++ SDE Karras",
-        "DPM2 a Karras"
-    ],
-    "eta_ddim": 0.0,
-    "eta_ancestral": 1.0,
-    "ddim_discretize": "uniform",
-    "s_churn": 0.0,
-    "s_tmin": 0.0,
-    "s_noise": 1.0,
-    "eta_noise_seed_delta": 0,
-    "disabled_extensions": [
-        "sdweb-merge-board",
-        "roll-artist"
-    ],
-    "ldsr_steps": 100,
-    "ldsr_cached": false,
-    "SWIN_tile": 192,
-    "SWIN_tile_overlap": 8,
-    "inspiration_max_samples": 20,
-    "inspiration_rows_num": 16,
-    "inspiration_cols_num": 6,
-    "images_history_preload": false,
-    "images_record_paths": true,
-    "images_delete_message": true,
-    "images_history_page_columns": 6.0,
-    "images_history_page_rows": 20.0,
-    "images_history_pages_perload": 20.0,
-    "sd_vae_checkpoint_cache": 0,
-    "use_old_karras_scheduler_sigmas": false,
-    "samplers_in_dropdown": true,
-    "aesthetic_scorer_enabled": false,
-    "aesthetic_scorer_clip_model": "ViT-L/14",
-    "dimensions_and_batch_together": true,
-    "ui_reorder": "sampler, dimensions, cfg, seed, checkboxes, hires_fix, batch, scripts",
-    "always_discard_next_to_last_sigma": false,
-    "clip_models_path": "models/CLIP",
-    "save_training_settings_to_txt": true,
-    "use_old_hires_fix_width_height": false,
-    "training_enable_tensorboard": true,
-    "training_tensorboard_save_images": true,
-    "training_tensorboard_flush_every": 60.0,
-    "sd_checkpoint_hash": "93715dbf32da7b9b97310a805118766eddd89f25464b453f501c06801816447a",
-    "print_hypernet_extra": false,
-    "live_previews_enable": true,
-    "live_preview_content": "Prompt"
-}

modules/sd_hijack.py

@@ -103,14 +103,19 @@ class StableDiffusionModelHijack:
             m.cond_stage_model.model.token_embedding = EmbeddingsWithFixes(m.cond_stage_model.model.token_embedding, self)
             m.cond_stage_model = sd_hijack_open_clip.FrozenOpenCLIPEmbedderWithCustomWords(m.cond_stage_model, self)
 
+        """
         try:
             import time
             t0 = time.time()
-            m = torch.compile(m, mode="max-autotune", fullgraph=True)
+            torch._dynamo.config.verbose = True
+            torch.backends.cudnn.benchmark = True
+            m.model = torch.compile(m.model, mode="max-autotune", fullgraph=False)
+            m = torch.compile(m, mode="max-autotune", fullgraph=False)
             t1 = time.time()
             print(f"Model compiled in {round(t1 - t0, 2)} sec")
         except Exception as err:
             print(f"Model compile not supported: {err}")
+        """
 
         self.optimization_method = apply_optimizations()