this is release candidate 2.3.3-rc1

Incorporates a modified version of the dialog-based invoke.sh script
suggested by JoshuaKimsey:
https://discord.com/channels/1020123559063990373/1089119602425995304

docs/features/TEXTUAL_INVERSION.md

@@ -154,8 +154,11 @@ training sets will converge with 2000-3000 steps.
 
 This adjusts how many training images are processed simultaneously in
 each step. Higher values will cause the training process to run more
-quickly, but use more memory. The default size will run with GPUs with
-as little as 12 GB.
+quickly, but use more memory. The default size is selected based on
+whether you have the `xformers` memory-efficient attention library
+installed. If `xformers` is available, the batch size will be 8,
+otherwise 3.  These values were chosen to allow training to run with
+GPUs with as little as 12 GB VRAM.
 
 ### Learning rate
 
@@ -172,8 +175,10 @@ learning rate to improve performance.
 
 ### Use xformers acceleration
 
-This will activate XFormers memory-efficient attention. You need to
-have XFormers installed for this to have an effect.
+This will activate XFormers memory-efficient attention, which will
+reduce memory requirements by half or more and allow you to select a
+higher batch size. You need to have XFormers installed for this to
+have an effect.
 
 ### Learning rate scheduler
 
@@ -250,6 +255,49 @@ invokeai-ti \
        --only_save_embeds
 ```
 
+## Using Distributed Training
+
+If you have multiple GPUs on one machine, or a cluster of GPU-enabled
+machines, you can activate distributed training. See the [HuggingFace
+Accelerate pages](https://huggingface.co/docs/accelerate/index) for
+full information, but the basic recipe is:
+
+1. Enter the InvokeAI developer's console command line by selecting
+option [8] from the `invoke.sh`/`invoke.bat` script.
+
+2. Configurate Accelerate using `accelerate config`:
+```sh
+accelerate config
+```
+This will guide you through the configuration process, including
+specifying how many machines you will run training on and the number
+of GPUs pe rmachine.
+
+You only need to do this once.
+
+3. Launch training from the command line using `accelerate launch`. Be sure
+that your current working directory is the InvokeAI root directory (usually
+named `invokeai` in your home directory):
+
+```sh
+accelerate launch .venv/bin/invokeai-ti \
+       --model=stable-diffusion-1.5 \
+       --resolution=512 \
+       --learnable_property=object \
+       --initializer_token='*' \
+       --placeholder_token='<shraddha>' \
+       --train_data_dir=/home/lstein/invokeai/text-inversion-training-data/shraddha \
+       --output_dir=/home/lstein/invokeai/text-inversion-training/shraddha \
+       --scale_lr \
+       --train_batch_size=10 \
+       --gradient_accumulation_steps=4 \
+       --max_train_steps=2000 \
+       --learning_rate=0.0005 \
+       --lr_scheduler=constant \
+       --mixed_precision=fp16 \
+       --only_save_embeds
+```
+
 ## Using Embeddings
 
 After training completes, the resultant embeddings will be saved into your `$INVOKEAI_ROOT/embeddings/<trigger word>/learned_embeds.bin`.

installer/templates/invoke.sh.in

@@ -1,5 +1,8 @@
 #!/bin/bash
 
+# coauthored by Lincoln Stein, Eugene Brodsky and JoshuaKimsey
+# Copyright 2023, The InvokeAI Development Team
+
 ####
 # This launch script assumes that:
 # 1. it is located in the runtime directory,
@@ -18,78 +21,135 @@ cd "$scriptdir"
 . .venv/bin/activate
 
 export INVOKEAI_ROOT="$scriptdir"
+PARAMS=$@
 
 # set required env var for torch on mac MPS
 if [ "$(uname -s)" == "Darwin" ]; then
     export PYTORCH_ENABLE_MPS_FALLBACK=1
 fi
 
-while true
-do
-if [ "$0" != "bash" ]; then
-    echo "Do you want to generate images using the"
-    echo "1. command-line interface"
-    echo "2. browser-based UI"
-    echo "3. run textual inversion training"
-    echo "4. merge models (diffusers type only)"
-    echo "5. download and install models"
-    echo "6. change InvokeAI startup options"
-    echo "7. re-run the configure script to fix a broken install"
-    echo "8. open the developer console"
-    echo "9. update InvokeAI"
-    echo "10. command-line help"
-    echo "Q - Quit"
-    echo ""
-    read -p "Please enter 1-10, Q: [2] " yn
-    choice=${yn:='2'}
-    case $choice in
-        1)
-            echo "Starting the InvokeAI command-line..."
-            invokeai $@
+do_choice() {
+    case $1 in
+	1)
+            echo "Generate images with a browser-based interface"
+            clear
+            invokeai --web $PARAMS
             ;;
-        2)
-            echo "Starting the InvokeAI browser-based UI..."
-            invokeai --web $@
+	2)
+            echo "Generate images using a command-line interface"
+            clear
+            invokeai $PARAMS
             ;;
-        3)
-            echo "Starting Textual Inversion:"
-            invokeai-ti --gui $@
+	3)
+            echo "Textual inversion training"
+            clear
+            invokeai-ti --gui $PARAMS
             ;;
-        4)
-            echo "Merging Models:"
-            invokeai-merge --gui $@
+	4)
+            echo "Merge models (diffusers type only)"
+            clear
+            invokeai-merge --gui $PARAMS
             ;;
-        5)
+	5)
+            echo "Download and install models"
+            clear
             invokeai-model-install --root ${INVOKEAI_ROOT}
             ;;
-        6)
+	6)
+            echo "Change InvokeAI startup options"
+            clear
             invokeai-configure --root ${INVOKEAI_ROOT} --skip-sd-weights --skip-support-models
             ;;
-        7)
+	7)
+            echo "Re-run the configure script to fix a broken install"
+            clear
             invokeai-configure --root ${INVOKEAI_ROOT} --yes --default_only
-	    ;;
+            ;;
 	8)
-	    echo "Developer Console:"
+            echo "Open the developer console"
+            clear
             file_name=$(basename "${BASH_SOURCE[0]}")
             bash --init-file "$file_name"
             ;;
-        9)
-	    echo "Update:"
+	9)
+            echo "Update InvokeAI"
+            clear
             invokeai-update
             ;;
-        10)
+	10)
+            echo "Command-line help"
+            clear
             invokeai --help
             ;;
-	[qQ])
-            exit 0
+	*)
+            echo "Exiting..."
+            exit
             ;;
-        *)
-            echo "Invalid selection"
-            exit;;
     esac
+    clear
+}
+    
+do_dialog() {
+    while true
+    do
+	options=(
+	    1 "Generate images with a browser-based interface"
+	    2 "Generate images using a command-line interface"
+	    3 "Textual inversion training"
+	    4 "Merge models (diffusers type only)"
+	    5 "Download and install models"
+	    6 "Change InvokeAI startup options"
+	    7 "Re-run the configure script to fix a broken install"
+	    8 "Open the developer console"
+	    9 "Update InvokeAI"
+	    10 "Command-line help")
+	
+        choice=$(dialog --clear \
+                        --backtitle "InvokeAI" \
+                        --title "What you like to run?" \
+                        --menu "Select an option:" \
+                        0 0 0 \
+                        "${options[@]}" \
+                        2>&1 >/dev/tty) || clear
+	do_choice "$choice"
+    done
+    clear
+}
+
+do_line_input() {
+    echo " ** For a more attractive experience, please install the 'dialog' utility. **"
+    echo ""
+    while true
+    do
+	echo "Do you want to generate images using the"
+	echo "1. browser-based UI"
+	echo "2. command-line interface"
+	echo "3. run textual inversion training"
+	echo "4. merge models (diffusers type only)"
+	echo "5. download and install models"
+	echo "6. change InvokeAI startup options"
+	echo "7. re-run the configure script to fix a broken install"
+	echo "8. open the developer console"
+	echo "9. update InvokeAI"
+	echo "10. command-line help"
+	echo "Q - Quit"
+	echo ""
+	read -p "Please enter 1-10, Q: [1] " yn
+	choice=${yn:='1'}
+	do_choice $choice
+    done
+}
+
+if [ "$0" != "bash" ]; then
+    # Dialog seems to be a standard installtion for most Linux distros, but this checks to ensure it is present regardless
+    if command -v dialog &> /dev/null ; then
+	do_dialog
+    else
+	do_line_input
+    fi
 else # in developer console
     python --version
     echo "Press ^D to exit"
     export PS1="(InvokeAI) \u@\h \w> "
 fi
-done
+

ldm/invoke/_version.py

@@ -1,2 +1,2 @@
 
-__version__='2.3.2'
+__version__='2.3.3-rc1'

ldm/invoke/ckpt_to_diffuser.py

@@ -327,10 +327,10 @@ def convert_ldm_unet_checkpoint(checkpoint, config, path=None, extract_ema=False
     unet_key = "model.diffusion_model."
     # at least a 100 parameters have to start with `model_ema` in order for the checkpoint to be EMA
     if sum(k.startswith("model_ema") for k in keys) > 100:
-        print(f"  | Checkpoint {path} has both EMA and non-EMA weights.")
+        print(f"   | Checkpoint {path} has both EMA and non-EMA weights.")
         if extract_ema:
             print(
-                '  | Extracting EMA weights (usually better for inference)'
+                '   | Extracting EMA weights (usually better for inference)'
             )
             for key in keys:
                 if key.startswith("model.diffusion_model"):
@@ -338,7 +338,7 @@ def convert_ldm_unet_checkpoint(checkpoint, config, path=None, extract_ema=False
                     unet_state_dict[key.replace(unet_key, "")] = checkpoint.pop(flat_ema_key)
         else:
             print(
-                '  | Extracting only the non-EMA weights (usually better for fine-tuning)'
+                '   | Extracting only the non-EMA weights (usually better for fine-tuning)'
             )
 
     for key in keys:
@@ -809,6 +809,7 @@ def load_pipeline_from_original_stable_diffusion_ckpt(
         vae:AutoencoderKL=None,
         precision:torch.dtype=torch.float32,
         return_generator_pipeline:bool=False,
+        scan_needed:bool=True,
 )->Union[StableDiffusionPipeline,StableDiffusionGeneratorPipeline]:
     '''
     Load a Stable Diffusion pipeline object from a CompVis-style `.ckpt`/`.safetensors` file and (ideally) a `.yaml`
@@ -843,7 +844,12 @@ def load_pipeline_from_original_stable_diffusion_ckpt(
         verbosity = dlogging.get_verbosity()
         dlogging.set_verbosity_error()
 
-        checkpoint = load_file(checkpoint_path) if Path(checkpoint_path).suffix == '.safetensors' else torch.load(checkpoint_path)
+        if Path(checkpoint_path).suffix == '.ckpt':
+            if scan_needed:
+                ModelManager.scan_model(checkpoint_path,checkpoint_path)
+            checkpoint = torch.load(checkpoint_path)
+        else:
+            checkpoint = load_file(checkpoint_path)
         cache_dir = global_cache_dir('hub')
         pipeline_class = StableDiffusionGeneratorPipeline if return_generator_pipeline else StableDiffusionPipeline
 
@@ -851,7 +857,7 @@ def load_pipeline_from_original_stable_diffusion_ckpt(
         if "global_step" in checkpoint:
             global_step = checkpoint["global_step"]
         else:
-            print("  | global_step key not found in model")
+            print("   | global_step key not found in model")
             global_step = None
 
         # sometimes there is a state_dict key and sometimes not
@@ -953,14 +959,14 @@ def load_pipeline_from_original_stable_diffusion_ckpt(
 
         # Convert the VAE model, or use the one passed
         if not vae:
-            print('  | Using checkpoint model\'s original VAE')
+            print('   | Using checkpoint model\'s original VAE')
             vae_config = create_vae_diffusers_config(original_config, image_size=image_size)
             converted_vae_checkpoint = convert_ldm_vae_checkpoint(checkpoint, vae_config)
 
             vae = AutoencoderKL(**vae_config)
             vae.load_state_dict(converted_vae_checkpoint)
         else:
-            print('  | Using external VAE specified in config')
+            print('   | Using VAE specified in config')
 
         # Convert the text model.
         model_type = pipeline_type

ldm/invoke/config/invokeai_update.py

@@ -72,7 +72,7 @@ def main():
         tag = Prompt.ask('Enter an InvokeAI tag or branch name')
 
     print(f':crossed_fingers: Upgrading to [yellow]{tag}[/yellow]')
-    cmd = f'pip install {INVOKE_AI_SRC}/{tag}.zip --use-pep517'
+    cmd = f'pip install {INVOKE_AI_SRC}/{tag}.zip --use-pep517 --upgrade'
     print('')
     print('')
     if os.system(cmd)==0:

ldm/invoke/dynamic_prompts.py

@@ -157,7 +157,7 @@ def _run_invoke(
 ):
     pid = os.getpid()
     logdir.mkdir(parents=True, exist_ok=True)
-    logfile = Path(logdir, f'{time.strftime("%Y-%m-%d-%H:%M:%S")}-pid={pid}.txt')
+    logfile = Path(logdir, f'{time.strftime("%Y-%m-%d_%H-%M-%S")}-pid={pid}.txt')
     print(
         f">> Process {pid} running on GPU {gpu}; logging to {logfile}", file=sys.stderr
     )

ldm/invoke/model_manager.py

@@ -282,13 +282,13 @@ class ModelManager(object):
             self.stack.remove(model_name)
         if delete_files:
             if weights:
-                print(f"** deleting file {weights}")
+                print(f"** Deleting file {weights}")
                 Path(weights).unlink(missing_ok=True)
             elif path:
-                print(f"** deleting directory {path}")
+                print(f"** Deleting directory {path}")
                 rmtree(path, ignore_errors=True)
             elif repo_id:
-                print(f"** deleting the cached model directory for {repo_id}")
+                print(f"** Deleting the cached model directory for {repo_id}")
                 self._delete_model_from_cache(repo_id)
 
     def add_model(
@@ -420,11 +420,6 @@ class ModelManager(object):
                 "NOHASH",
             )
 
-        # scan model
-        self.scan_model(model_name, weights)
-
-        print(f">> Loading {model_name} from {weights}")
-
         # for usage statistics
         if self._has_cuda():
             torch.cuda.reset_peak_memory_stats()
@@ -438,10 +433,13 @@ class ModelManager(object):
             weight_bytes = f.read()
         model_hash = self._cached_sha256(weights, weight_bytes)
         sd = None
-        if weights.endswith(".safetensors"):
-            sd = safetensors.torch.load(weight_bytes)
-        else:
+        
+        if weights.endswith(".ckpt"):
+            self.scan_model(model_name, weights)
             sd = torch.load(io.BytesIO(weight_bytes), map_location="cpu")
+        else:
+            sd = safetensors.torch.load(weight_bytes)
+            
         del weight_bytes
         # merged models from auto11 merge board are flat for some reason
         if "state_dict" in sd:
@@ -464,18 +462,12 @@ class ModelManager(object):
                 vae = os.path.normpath(os.path.join(Globals.root, vae))
             if os.path.exists(vae):
                 print(f"   | Loading VAE weights from: {vae}")
-                vae_ckpt = None
-                vae_dict = None
-                if vae.endswith(".safetensors"):
-                    vae_ckpt = safetensors.torch.load_file(vae)
-                    vae_dict = {k: v for k, v in vae_ckpt.items() if k[0:4] != "loss"}
-                else:
+                if vae.endswith((".ckpt",".pt")):
+                    self.scan_model(vae,vae)
                     vae_ckpt = torch.load(vae, map_location="cpu")
-                    vae_dict = {
-                        k: v
-                        for k, v in vae_ckpt["state_dict"].items()
-                        if k[0:4] != "loss"
-                    }
+                else:
+                    vae_ckpt = safetensors.torch.load_file(vae)
+                vae_dict = {k: v for k, v in vae_ckpt.items() if k[0:4] != "loss"}
                 model.first_stage_model.load_state_dict(vae_dict, strict=False)
             else:
                 print(f"   | VAE file {vae} not found. Skipping.")
@@ -497,9 +489,9 @@ class ModelManager(object):
 
         print(f">> Loading diffusers model from {name_or_path}")
         if using_fp16:
-            print("  | Using faster float16 precision")
+            print("   | Using faster float16 precision")
         else:
-            print("  | Using more accurate float32 precision")
+            print("   | Using more accurate float32 precision")
 
         # TODO: scan weights maybe?
         pipeline_args: dict[str, Any] = dict(
@@ -551,7 +543,7 @@ class ModelManager(object):
         width = pipeline.unet.config.sample_size * pipeline.vae_scale_factor
         height = width
 
-        print(f"  | Default image dimensions = {width} x {height}")
+        print(f"   | Default image dimensions = {width} x {height}")
 
         return pipeline, width, height, model_hash
 
@@ -591,13 +583,14 @@ class ModelManager(object):
         if self._has_cuda():
             torch.cuda.empty_cache()
 
+    @classmethod
     def scan_model(self, model_name, checkpoint):
         """
         Apply picklescanner to the indicated checkpoint and issue a warning
         and option to exit if an infected file is identified.
         """
         # scan model
-        print(f">> Scanning Model: {model_name}")
+        print(f"   | Scanning Model: {model_name}")
         scan_result = scan_file_path(checkpoint)
         if scan_result.infected_files != 0:
             if scan_result.infected_files == 1:
@@ -620,7 +613,7 @@ class ModelManager(object):
                     print("### Exiting InvokeAI")
                     sys.exit()
         else:
-            print(">> Model scanned ok")
+            print("   | Model scanned ok")
 
     def import_diffuser_model(
         self,
@@ -800,19 +793,20 @@ class ModelManager(object):
         print(f">> Probing {thing} for import")
 
         if thing.startswith(("http:", "https:", "ftp:")):
-            print(f"  | {thing} appears to be a URL")
+            print(f"   | {thing} appears to be a URL")
             model_path = self._resolve_path(
                 thing, "models/ldm/stable-diffusion-v1"
             )  # _resolve_path does a download if needed
             is_temporary = True
+
         elif Path(thing).is_file() and thing.endswith((".ckpt", ".safetensors")):
             if Path(thing).stem in ["model", "diffusion_pytorch_model"]:
                 print(
-                    f"  | {Path(thing).name} appears to be part of a diffusers model. Skipping import"
+                    f"   | {Path(thing).name} appears to be part of a diffusers model. Skipping import"
                 )
                 return
             else:
-                print(f"  | {thing} appears to be a checkpoint file on disk")
+                print(f"   | {thing} appears to be a checkpoint file on disk")
                 model_path = self._resolve_path(thing, "models/ldm/stable-diffusion-v1")
 
         elif Path(thing).is_dir() and Path(thing, "model_index.json").exists():
@@ -869,11 +863,12 @@ class ModelManager(object):
             return model_path.stem
 
         # another round of heuristics to guess the correct config file.
-        checkpoint = (
-            safetensors.torch.load_file(model_path)
-            if model_path.suffix == ".safetensors"
-            else torch.load(model_path)
-        )
+        checkpoint = None
+        if model_path.suffix.endswith((".ckpt",".pt")):
+            self.scan_model(model_path,model_path)
+            checkpoint = torch.load(model_path)
+        else:
+            checkpoint = safetensors.torch.load_file(model_path)
         # additional probing needed if no config file provided
         if model_config_file is None:
             model_type = self.probe_model_type(checkpoint)
@@ -918,7 +913,7 @@ class ModelManager(object):
         if model_config_file.name.startswith('v2'):
             convert = True
             print(
-                "  | This SD-v2 model will be converted to diffusers format for use"
+                "   | This SD-v2 model will be converted to diffusers format for use"
             )
 
         if convert:
@@ -933,6 +928,7 @@ class ModelManager(object):
                 model_description=description,
                 original_config_file=model_config_file,
                 commit_to_conf=commit_to_conf,
+                scan_needed=False,
             )
             # in the event that this file was downloaded automatically prior to conversion
             # we do not keep the original .ckpt/.safetensors around
@@ -957,14 +953,15 @@ class ModelManager(object):
         return model_name
 
     def convert_and_import(
-        self,
-        ckpt_path: Path,
-        diffusers_path: Path,
-        model_name=None,
-        model_description=None,
-        vae=None,
-        original_config_file: Path = None,
-        commit_to_conf: Path = None,
+            self,
+            ckpt_path: Path,
+            diffusers_path: Path,
+            model_name=None,
+            model_description=None,
+            vae=None,
+            original_config_file: Path = None,
+            commit_to_conf: Path = None,
+            scan_needed: bool=True,
     ) -> str:
         """
         Convert a legacy ckpt weights file to diffuser model and import
@@ -999,11 +996,12 @@ class ModelManager(object):
                 extract_ema=True,
                 original_config_file=original_config_file,
                 vae=vae_model,
+                scan_needed=scan_needed,
             )
             print(
-                f"  | Success. Optimized model is now located at {str(diffusers_path)}"
+                f"   | Success. Optimized model is now located at {str(diffusers_path)}"
             )
-            print(f"  | Writing new config file entry for {model_name}")
+            print(f"   | Writing new config file entry for {model_name}")
             new_config = dict(
                 path=str(diffusers_path),
                 description=model_description,
@@ -1293,7 +1291,7 @@ class ModelManager(object):
             with open(hashpath) as f:
                 hash = f.read()
             return hash
-        print("  | Calculating sha256 hash of model files")
+        print("   | Calculating sha256 hash of model files")
         tic = time.time()
         sha = hashlib.sha256()
         count = 0
@@ -1305,7 +1303,7 @@ class ModelManager(object):
                         sha.update(chunk)
         hash = sha.hexdigest()
         toc = time.time()
-        print(f"  | sha256 = {hash} ({count} files hashed in", "%4.2fs)" % (toc - tic))
+        print(f"   | sha256 = {hash} ({count} files hashed in", "%4.2fs)" % (toc - tic))
         with open(hashpath, "w") as f:
             f.write(hash)
         return hash
@@ -1350,12 +1348,12 @@ class ModelManager(object):
             local_files_only=not Globals.internet_available,
         )
 
-        print(f"  | Loading diffusers VAE from {name_or_path}")
+        print(f"   | Loading diffusers VAE from {name_or_path}")
         if using_fp16:
             vae_args.update(torch_dtype=torch.float16)
             fp_args_list = [{"revision": "fp16"}, {}]
         else:
-            print("  | Using more accurate float32 precision")
+            print("   | Using more accurate float32 precision")
             fp_args_list = [{}]
 
         vae = None
@@ -1396,7 +1394,7 @@ class ModelManager(object):
                     hashes_to_delete.add(revision.commit_hash)
         strategy = cache_info.delete_revisions(*hashes_to_delete)
         print(
-            f"** deletion of this model is expected to free {strategy.expected_freed_size_str}"
+            f"** Deletion of this model is expected to free {strategy.expected_freed_size_str}"
         )
         strategy.execute()
 

ldm/invoke/pngwriter.py

@@ -30,14 +30,17 @@ class PngWriter:
                 prefix = self._unused_prefix()
             else:
                 with open(next_prefix_file,'r') as file:
-                    prefix=int(file.readline() or int(self._unused_prefix())-1)
-                    prefix+=1
+                    prefix = 0
+                    try:
+                        prefix=int(file.readline())
+                    except (TypeError, ValueError):
+                        prefix=self._unused_prefix()
             with open(next_prefix_file,'w') as file:
-                file.write(str(prefix))
+                file.write(str(prefix+1))
         return f'{prefix:06}'
 
     # gives the next unique prefix in outdir
-    def _unused_prefix(self):
+    def _unused_prefix(self)->int:
         # sort reverse alphabetically until we find max+1
         dirlist = sorted(os.listdir(self.outdir), reverse=True)
         # find the first filename that matches our pattern or return 000000.0.png
@@ -45,8 +48,7 @@ class PngWriter:
             (f for f in dirlist if re.match('^(\d+)\..*\.png', f)),
             '0000000.0.png',
         )
-        basecount = int(existing_name.split('.', 1)[0]) + 1
-        return f'{basecount:06}'
+        return int(existing_name.split('.', 1)[0]) + 1
 
     # saves image named _image_ to outdir/name, writing metadata from prompt
     # returns full path of output

ldm/invoke/training/textual_inversion.py

@@ -17,6 +17,7 @@ from pathlib import Path
 from typing import List, Tuple
 
 import npyscreen
+from diffusers.utils.import_utils import is_xformers_available
 from npyscreen import widget
 from omegaconf import OmegaConf
 
@@ -29,7 +30,7 @@ from ldm.invoke.training.textual_inversion_training import (
 TRAINING_DATA = "text-inversion-training-data"
 TRAINING_DIR = "text-inversion-output"
 CONF_FILE = "preferences.conf"
-
+XFORMERS_AVAILABLE = is_xformers_available()
 
 class textualInversionForm(npyscreen.FormMultiPageAction):
     resolutions = [512, 768, 1024]
@@ -178,7 +179,7 @@ class textualInversionForm(npyscreen.FormMultiPageAction):
             out_of=10000,
             step=500,
             lowest=1,
-            value=saved_args.get("max_train_steps", 3000),
+            value=saved_args.get("max_train_steps", 2500),
             scroll_exit=True,
         )
         self.train_batch_size = self.add_widget_intelligent(
@@ -187,7 +188,7 @@ class textualInversionForm(npyscreen.FormMultiPageAction):
             out_of=50,
             step=1,
             lowest=1,
-            value=saved_args.get("train_batch_size", 8),
+            value=saved_args.get("train_batch_size", 8 if XFORMERS_AVAILABLE else 3),
             scroll_exit=True,
         )
         self.gradient_accumulation_steps = self.add_widget_intelligent(
@@ -225,7 +226,7 @@ class textualInversionForm(npyscreen.FormMultiPageAction):
         self.enable_xformers_memory_efficient_attention = self.add_widget_intelligent(
             npyscreen.Checkbox,
             name="Use xformers acceleration",
-            value=saved_args.get("enable_xformers_memory_efficient_attention", False),
+            value=saved_args.get("enable_xformers_memory_efficient_attention", XFORMERS_AVAILABLE),
             scroll_exit=True,
         )
         self.lr_scheduler = self.add_widget_intelligent(
@@ -428,8 +429,7 @@ def do_front_end(args: Namespace):
             print(str(e))
             print("** DETAILS:")
             print(traceback.format_exc())
-
-
+            
 def main():
     args = parse_args()
     global_set_root(args.root_dir or Globals.root)

ldm/invoke/training/textual_inversion_training.py

@@ -67,7 +67,7 @@ else:
         "nearest": PIL.Image.NEAREST,
     }
 # ------------------------------------------------------------------------------
-
+XFORMERS_AVAILABLE = is_xformers_available
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
 check_min_version("0.10.0.dev0")
@@ -227,7 +227,7 @@ def parse_args():
     training_group.add_argument(
         "--train_batch_size",
         type=int,
-        default=16,
+        default=8 if XFORMERS_AVAILABLE else 3,
         help="Batch size (per device) for the training dataloader.",
     )
     training_group.add_argument("--num_train_epochs", type=int, default=100)
@@ -324,6 +324,7 @@ def parse_args():
     parser.add_argument(
         "--enable_xformers_memory_efficient_attention",
         action="store_true",
+        default=XFORMERS_AVAILABLE,
         help="Whether or not to use xformers.",
     )
 
@@ -536,7 +537,7 @@ def do_textual_inversion_training(
     seed: int = None,
     resolution: int = 512,
     center_crop: bool = False,
-    train_batch_size: int = 16,
+    train_batch_size: int = 4,
     num_train_epochs: int = 100,
     max_train_steps: int = 5000,
     gradient_accumulation_steps: int = 1,

pyproject.toml

@@ -70,7 +70,7 @@ dependencies = [
   "taming-transformers-rom1504",
   "test-tube>=0.7.5",
   "torch-fidelity",
-  "torch>=1.13.1",
+  "torch~=1.13.1",
   "torchmetrics",
   "torchvision>=0.14.1",
   "transformers~=4.26",
@@ -147,7 +147,7 @@ version = {attr = "ldm.invoke.__version__"}
 
 [tool.setuptools.package-data]
 "invokeai.assets.web" = ["**.png"]
-"invokeai.configs" = ["**.example", "**.txt", "**.yaml"]
+"invokeai.configs" = ["**.example", "**.txt", "**.yaml", "**/*.yaml"]
 "invokeai.frontend.dist" = ["**"]
 
 [tool.black]