refactor InvokeAIAppConfig

docs/features/CONFIGURATION.md

@@ -175,22 +175,27 @@ These configuration settings allow you to enable and disable various InvokeAI fe
 | `internet_available` | `true`     | When a resource is not available locally, try to fetch it via the internet |
 | `log_tokenization` | `false`      | Before each text2image generation, print a color-coded representation of the prompt to the console; this can help understand why a prompt is not working as expected |
 | `patchmatch` | `true`     | Activate the "patchmatch" algorithm for improved inpainting |
-| `restore`    | `true`     | Activate the facial restoration features (DEPRECATED; restoration features will be removed in 3.0.0) |
 
-### Memory/Performance
+### Generation
 
 These options tune InvokeAI's memory and performance characteristics.
 
-| Setting  | Default Value  |  Description |
+| Setting               | Default Value | Description                                                                                                                                                                                                                                                      |
-|----------|----------------|--------------|
+|-----------------------|---------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-| `always_use_cpu`     | `false`      | Use the CPU to generate images, even if a GPU is available |
+| `sequential_guidance` | `false`       | Calculate guidance in serial rather than in parallel, lowering memory requirements at the cost of some performance loss                                                                                                                                          |
-| `free_gpu_mem`       | `false`      | Aggressively free up GPU memory after each operation; this will allow you to run in low-VRAM environments with some performance penalties |
+| `attention_type`      | `auto`        | Select the type of attention to use. One of `auto`,`normal`,`xformers`,`sliced`, or `torch-sdp`                                                                                                                                                                  |
-| `max_cache_size`       | `6`      | Amount of CPU RAM (in GB) to reserve for caching models in memory; more cache allows you to keep models in memory and switch among them quickly |
+| `attention_slice_size` | `auto`       | When "sliced" attention is selected, set the slice size. One of `auto`, `max` or the integers 1-8|
-| `max_vram_cache_size`  | `2.75`   | Amount of GPU VRAM (in GB) to reserve for caching models in VRAM; more cache speeds up generation but reduces the size of the images that can be generated. This can be set to zero to maximize the amount of memory available for generation. |
+| `force_tiled_decode`  | `false`       | Force the VAE step to decode in tiles, reducing memory consumption at the cost of performance |
-| `precision`       | `auto`      | Floating point precision. One of `auto`, `float16` or `float32`. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system |
+
-| `sequential_guidance`     | `false`      | Calculate guidance in serial rather than in parallel, lowering memory requirements at the cost of some performance loss |
+### Device
-| `xformers_enabled`        | `true`      | If the x-formers memory-efficient attention module is installed, activate it for better memory usage and generation speed|
+
-| `tiled_decode`            | `false`     | If true, then during the VAE decoding phase the image will be decoded a section at a time, reducing memory consumption at the cost of a performance hit |
+These options configure the generation execution device.
+
+| Setting               | Default Value | Description                                                                                                                                                                                                                                                      |
+|-----------------------|---------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| `device`           | `auto`        | Preferred execution device. One of `auto`, `cpu`, `cuda`, `cuda:1`, `mps`. `auto` will choose the device depending on the hardware platform and the installed torch capabilities. |
+| `precision`           | `auto`        | Floating point precision. One of `auto`, `float16` or `float32`. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system |
+
 
 ### Paths
 

invokeai/app/services/config/__init__.py

@@ -0,0 +1,8 @@
+"""
+Init file for InvokeAI configure package
+"""
+
+from .invokeai_config import (
+    InvokeAIAppConfig,
+    get_invokeai_config,
+)

invokeai/app/services/config/base.py

@@ -0,0 +1,239 @@
+# Copyright (c) 2023 Lincoln Stein (https://github.com/lstein) and the InvokeAI Development Team
+
+"""
+Base class for the InvokeAI configuration system.
+It defines a type of pydantic BaseSettings object that
+is able to read and write from an omegaconf-based config file,
+with overriding of settings from environment variables and/or
+the command line.
+"""
+
+from __future__ import annotations
+import argparse
+import os
+import pydoc
+import sys
+from argparse import ArgumentParser
+from omegaconf import OmegaConf, DictConfig, ListConfig
+from pathlib import Path
+from pydantic import BaseSettings
+from typing import ClassVar, Dict, List, Literal, Union, get_origin, get_type_hints, get_args
+
+
+class PagingArgumentParser(argparse.ArgumentParser):
+    """
+    A custom ArgumentParser that uses pydoc to page its output.
+    It also supports reading defaults from an init file.
+    """
+
+    def print_help(self, file=None):
+        text = self.format_help()
+        pydoc.pager(text)
+
+
+class InvokeAISettings(BaseSettings):
+    """
+    Runtime configuration settings in which default values are
+    read from an omegaconf .yaml file.
+    """
+
+    initconf: ClassVar[DictConfig] = None
+    argparse_groups: ClassVar[Dict] = {}
+
+    def parse_args(self, argv: list = sys.argv[1:]):
+        parser = self.get_parser()
+        opt = parser.parse_args(argv)
+        for name in self.__fields__:
+            if name not in self._excluded():
+                value = getattr(opt, name)
+                if isinstance(value, ListConfig):
+                    value = list(value)
+                elif isinstance(value, DictConfig):
+                    value = dict(value)
+                setattr(self, name, value)
+
+    def to_yaml(self) -> str:
+        """
+        Return a YAML string representing our settings. This can be used
+        as the contents of `invokeai.yaml` to restore settings later.
+        """
+        cls = self.__class__
+        type = get_args(get_type_hints(cls)["type"])[0]
+        field_dict = dict({type: dict()})
+        for name, field in self.__fields__.items():
+            if name in cls._excluded_from_yaml():
+                continue
+            category = field.field_info.extra.get("category") or "Uncategorized"
+            value = getattr(self, name)
+            if category not in field_dict[type]:
+                field_dict[type][category] = dict()
+            # keep paths as strings to make it easier to read
+            field_dict[type][category][name] = str(value) if isinstance(value, Path) else value
+        conf = OmegaConf.create(field_dict)
+        return OmegaConf.to_yaml(conf)
+
+    @classmethod
+    def add_parser_arguments(cls, parser):
+        if "type" in get_type_hints(cls):
+            settings_stanza = get_args(get_type_hints(cls)["type"])[0]
+        else:
+            settings_stanza = "Uncategorized"
+
+        env_prefix = cls.Config.env_prefix if hasattr(cls.Config, "env_prefix") else settings_stanza.upper()
+
+        initconf = (
+            cls.initconf.get(settings_stanza)
+            if cls.initconf and settings_stanza in cls.initconf
+            else OmegaConf.create()
+        )
+
+        # create an upcase version of the environment in
+        # order to achieve case-insensitive environment
+        # variables (the way Windows does)
+        upcase_environ = dict()
+        for key, value in os.environ.items():
+            upcase_environ[key.upper()] = value
+
+        fields = cls.__fields__
+        cls.argparse_groups = {}
+
+        for name, field in fields.items():
+            if name not in cls._excluded():
+                current_default = field.default
+
+                category = field.field_info.extra.get("category", "Uncategorized")
+                env_name = env_prefix + "_" + name
+                if category in initconf and name in initconf.get(category):
+                    field.default = initconf.get(category).get(name)
+                if env_name.upper() in upcase_environ:
+                    field.default = upcase_environ[env_name.upper()]
+                cls.add_field_argument(parser, name, field)
+
+                field.default = current_default
+
+    @classmethod
+    def cmd_name(self, command_field: str = "type") -> str:
+        hints = get_type_hints(self)
+        if command_field in hints:
+            return get_args(hints[command_field])[0]
+        else:
+            return "Uncategorized"
+
+    @classmethod
+    def get_parser(cls) -> ArgumentParser:
+        parser = PagingArgumentParser(
+            prog=cls.cmd_name(),
+            description=cls.__doc__,
+        )
+        cls.add_parser_arguments(parser)
+        return parser
+
+    @classmethod
+    def add_subparser(cls, parser: argparse.ArgumentParser):
+        parser.add_parser(cls.cmd_name(), help=cls.__doc__)
+
+    @classmethod
+    def _excluded(self) -> List[str]:
+        # internal fields that shouldn't be exposed as command line options
+        return ["type", "initconf"]
+
+    @classmethod
+    def _excluded_from_yaml(self) -> List[str]:
+        # combination of deprecated parameters and internal ones that shouldn't be exposed as invokeai.yaml options
+        return [
+            "type",
+            "initconf",
+            "version",
+            "from_file",
+            "model",
+            "root",
+            "max_cache_size",
+            "max_vram_cache_size",
+            "always_use_cpu",
+            "free_gpu_mem",
+            "xformers_enabled",
+            "tiled_decode",
+        ]
+
+    class Config:
+        env_file_encoding = "utf-8"
+        arbitrary_types_allowed = True
+        case_sensitive = True
+
+    @classmethod
+    def add_field_argument(cls, command_parser, name: str, field, default_override=None):
+        field_type = get_type_hints(cls).get(name)
+        default = (
+            default_override
+            if default_override is not None
+            else field.default
+            if field.default_factory is None
+            else field.default_factory()
+        )
+        if category := field.field_info.extra.get("category"):
+            if category not in cls.argparse_groups:
+                cls.argparse_groups[category] = command_parser.add_argument_group(category)
+            argparse_group = cls.argparse_groups[category]
+        else:
+            argparse_group = command_parser
+
+        if get_origin(field_type) == Literal:
+            allowed_values = get_args(field.type_)
+            allowed_types = set()
+            for val in allowed_values:
+                allowed_types.add(type(val))
+            allowed_types_list = list(allowed_types)
+            field_type = allowed_types_list[0] if len(allowed_types) == 1 else int_or_float_or_str
+
+            argparse_group.add_argument(
+                f"--{name}",
+                dest=name,
+                type=field_type,
+                default=default,
+                choices=allowed_values,
+                help=field.field_info.description,
+            )
+
+        elif get_origin(field_type) == Union:
+            argparse_group.add_argument(
+                f"--{name}",
+                dest=name,
+                type=int_or_float_or_str,
+                default=default,
+                help=field.field_info.description,
+            )
+
+        elif get_origin(field_type) == list:
+            argparse_group.add_argument(
+                f"--{name}",
+                dest=name,
+                nargs="*",
+                type=field.type_,
+                default=default,
+                action=argparse.BooleanOptionalAction if field.type_ == bool else "store",
+                help=field.field_info.description,
+            )
+        else:
+            argparse_group.add_argument(
+                f"--{name}",
+                dest=name,
+                type=field.type_,
+                default=default,
+                action=argparse.BooleanOptionalAction if field.type_ == bool else "store",
+                help=field.field_info.description,
+            )
+
+
+def int_or_float_or_str(value: str) -> Union[int, float, str]:
+    """
+    Workaround for argparse type checking.
+    """
+    try:
+        return int(value)
+    except:
+        pass
+    try:
+        return float(value)
+    except:
+        pass
+    return str(value)

invokeai/app/services/config/invokeai_config.py

@@ -159,15 +159,13 @@ two configs are kept in separate sections of the config file:
 
 """
 from __future__ import annotations
-import argparse
-import pydoc
 import os
-import sys
+from omegaconf import OmegaConf, DictConfig
-from argparse import ArgumentParser
-from omegaconf import OmegaConf, DictConfig, ListConfig
 from pathlib import Path
-from pydantic import BaseSettings, Field, parse_obj_as
+from pydantic import Field, parse_obj_as
-from typing import Any, ClassVar, Dict, List, Set, Literal, Union, get_origin, get_type_hints, get_args
+from typing import ClassVar, Dict, List, Literal, Union, Optional, get_type_hints
+
+from .base import InvokeAISettings
 
 INIT_FILE = Path("invokeai.yaml")
 DB_FILE = Path("invokeai.db")
@@ -175,204 +173,6 @@ LEGACY_INIT_FILE = Path("invokeai.init")
 DEFAULT_MAX_VRAM = 0.5
 
 
-class InvokeAISettings(BaseSettings):
-    """
-    Runtime configuration settings in which default values are
-    read from an omegaconf .yaml file.
-    """
-
-    initconf: ClassVar[DictConfig] = None
-    argparse_groups: ClassVar[Dict] = {}
-
-    def parse_args(self, argv: list = sys.argv[1:]):
-        parser = self.get_parser()
-        opt = parser.parse_args(argv)
-        for name in self.__fields__:
-            if name not in self._excluded():
-                value = getattr(opt, name)
-                if isinstance(value, ListConfig):
-                    value = list(value)
-                elif isinstance(value, DictConfig):
-                    value = dict(value)
-                setattr(self, name, value)
-
-    def to_yaml(self) -> str:
-        """
-        Return a YAML string representing our settings. This can be used
-        as the contents of `invokeai.yaml` to restore settings later.
-        """
-        cls = self.__class__
-        type = get_args(get_type_hints(cls)["type"])[0]
-        field_dict = dict({type: dict()})
-        for name, field in self.__fields__.items():
-            if name in cls._excluded_from_yaml():
-                continue
-            category = field.field_info.extra.get("category") or "Uncategorized"
-            value = getattr(self, name)
-            if category not in field_dict[type]:
-                field_dict[type][category] = dict()
-            # keep paths as strings to make it easier to read
-            field_dict[type][category][name] = str(value) if isinstance(value, Path) else value
-        conf = OmegaConf.create(field_dict)
-        return OmegaConf.to_yaml(conf)
-
-    @classmethod
-    def add_parser_arguments(cls, parser):
-        if "type" in get_type_hints(cls):
-            settings_stanza = get_args(get_type_hints(cls)["type"])[0]
-        else:
-            settings_stanza = "Uncategorized"
-
-        env_prefix = cls.Config.env_prefix if hasattr(cls.Config, "env_prefix") else settings_stanza.upper()
-
-        initconf = (
-            cls.initconf.get(settings_stanza)
-            if cls.initconf and settings_stanza in cls.initconf
-            else OmegaConf.create()
-        )
-
-        # create an upcase version of the environment in
-        # order to achieve case-insensitive environment
-        # variables (the way Windows does)
-        upcase_environ = dict()
-        for key, value in os.environ.items():
-            upcase_environ[key.upper()] = value
-
-        fields = cls.__fields__
-        cls.argparse_groups = {}
-
-        for name, field in fields.items():
-            if name not in cls._excluded():
-                current_default = field.default
-
-                category = field.field_info.extra.get("category", "Uncategorized")
-                env_name = env_prefix + "_" + name
-                if category in initconf and name in initconf.get(category):
-                    field.default = initconf.get(category).get(name)
-                if env_name.upper() in upcase_environ:
-                    field.default = upcase_environ[env_name.upper()]
-                cls.add_field_argument(parser, name, field)
-
-                field.default = current_default
-
-    @classmethod
-    def cmd_name(self, command_field: str = "type") -> str:
-        hints = get_type_hints(self)
-        if command_field in hints:
-            return get_args(hints[command_field])[0]
-        else:
-            return "Uncategorized"
-
-    @classmethod
-    def get_parser(cls) -> ArgumentParser:
-        parser = PagingArgumentParser(
-            prog=cls.cmd_name(),
-            description=cls.__doc__,
-        )
-        cls.add_parser_arguments(parser)
-        return parser
-
-    @classmethod
-    def add_subparser(cls, parser: argparse.ArgumentParser):
-        parser.add_parser(cls.cmd_name(), help=cls.__doc__)
-
-    @classmethod
-    def _excluded(self) -> List[str]:
-        # internal fields that shouldn't be exposed as command line options
-        return ["type", "initconf"]
-
-    @classmethod
-    def _excluded_from_yaml(self) -> List[str]:
-        # combination of deprecated parameters and internal ones that shouldn't be exposed as invokeai.yaml options
-        return [
-            "type",
-            "initconf",
-            "version",
-            "from_file",
-            "model",
-            "root",
-        ]
-
-    class Config:
-        env_file_encoding = "utf-8"
-        arbitrary_types_allowed = True
-        case_sensitive = True
-
-    @classmethod
-    def add_field_argument(cls, command_parser, name: str, field, default_override=None):
-        field_type = get_type_hints(cls).get(name)
-        default = (
-            default_override
-            if default_override is not None
-            else field.default
-            if field.default_factory is None
-            else field.default_factory()
-        )
-        if category := field.field_info.extra.get("category"):
-            if category not in cls.argparse_groups:
-                cls.argparse_groups[category] = command_parser.add_argument_group(category)
-            argparse_group = cls.argparse_groups[category]
-        else:
-            argparse_group = command_parser
-
-        if get_origin(field_type) == Literal:
-            allowed_values = get_args(field.type_)
-            allowed_types = set()
-            for val in allowed_values:
-                allowed_types.add(type(val))
-            allowed_types_list = list(allowed_types)
-            field_type = allowed_types_list[0] if len(allowed_types) == 1 else int_or_float_or_str
-
-            argparse_group.add_argument(
-                f"--{name}",
-                dest=name,
-                type=field_type,
-                default=default,
-                choices=allowed_values,
-                help=field.field_info.description,
-            )
-            
-        elif get_origin(field_type) == Union:
-            argparse_group.add_argument(
-                f"--{name}",
-                dest=name,
-                type=int_or_float_or_str,
-                default=default,
-                help=field.field_info.description,
-            )
-
-        elif get_origin(field_type) == list:
-            argparse_group.add_argument(
-                f"--{name}",
-                dest=name,
-                nargs="*",
-                type=field.type_,
-                default=default,
-                action=argparse.BooleanOptionalAction if field.type_ == bool else "store",
-                help=field.field_info.description,
-            )
-        else:
-            argparse_group.add_argument(
-                f"--{name}",
-                dest=name,
-                type=field.type_,
-                default=default,
-                action=argparse.BooleanOptionalAction if field.type_ == bool else "store",
-                help=field.field_info.description,
-            )
-
-
-def _find_root() -> Path:
-    venv = Path(os.environ.get("VIRTUAL_ENV") or ".")
-    if os.environ.get("INVOKEAI_ROOT"):
-        root = Path(os.environ["INVOKEAI_ROOT"])
-    elif any([(venv.parent / x).exists() for x in [INIT_FILE, LEGACY_INIT_FILE]]):
-        root = (venv.parent).resolve()
-    else:
-        root = Path("~/invokeai").expanduser().resolve()
-    return root
-
-
 class InvokeAIAppConfig(InvokeAISettings):
     """
     Generate images using Stable Diffusion. Use "invokeai" to launch
@@ -387,6 +187,8 @@ class InvokeAIAppConfig(InvokeAISettings):
 
     # fmt: off
     type: Literal["InvokeAI"] = "InvokeAI"
+
+    # WEB
     host                : str = Field(default="127.0.0.1", description="IP address to bind to", category='Web Server')
     port                : int = Field(default=9090, description="Port to bind to", category='Web Server')
     allow_origins       : List[str] = Field(default=[], description="Allowed CORS origins", category='Web Server')
@@ -394,21 +196,15 @@ class InvokeAIAppConfig(InvokeAISettings):
     allow_methods       : List[str] = Field(default=["*"], description="Methods allowed for CORS", category='Web Server')
     allow_headers       : List[str] = Field(default=["*"], description="Headers allowed for CORS", category='Web Server')
 
+    # FEATURES
     esrgan              : bool = Field(default=True, description="Enable/disable upscaling code", category='Features')
     internet_available  : bool = Field(default=True, description="If true, attempt to download models on the fly; otherwise only use local models", category='Features')
     log_tokenization    : bool = Field(default=False, description="Enable logging of parsed prompt tokens.", category='Features')
     patchmatch          : bool = Field(default=True, description="Enable/disable patchmatch inpaint code", category='Features')
+    ignore_missing_core_models : bool = Field(default=False, description='Ignore missing models in models/core/convert', category='Features')
 
-    ram                 : Union[float,Literal['auto']] = Field(default=6.0, gt=0, description="Maximum memory amount used by model cache for rapid switching (floating point number or 'auto')", category='Cache')
-    vram                : Union[float,Literal['auto']] = Field(default=2.75, ge=0, description="Amount of VRAM reserved for model storage (floating point number or 'auto')", category='Cache')
-    lazy_offload        : bool = Field(default=True, description='Keep models in VRAM until their space is needed', category='Cache')
-    precision           : Literal[tuple(['auto','float16','float32','autocast'])] = Field(default='auto',description='Floating point precision', category='Device')
-    device              : Literal[tuple(['cpu','cuda','mps','cuda','cuda:1','auto'])] = Field(default='auto',description='Generation device', category='Device')
-    sequential_guidance : bool = Field(default=False, description="Whether to calculate guidance in serial instead of in parallel, lowering memory requirements", category='Generation')
-    attention_type      : Literal[tuple(['auto','normal','xformers','sliced','torch-sdp'])] = Field(default='auto', description='Attention type', category='Generation')
-    attention_slice_size: Literal[tuple(['auto','max',1,2,3,4,5,6,7,8])] = Field(default='auto', description='Slice size, valid when attention_type=="sliced"', category='Generation')
-    tiled_decode        : bool = Field(default=False, description="Whether to enable tiled VAE decode (reduces memory consumption with some performance penalty)", category='Generation')
 
+    # PATHS
     root                : Path = Field(default=None, description='InvokeAI runtime root directory', category='Paths')
     autoimport_dir      : Path = Field(default='autoimport', description='Path to a directory of models files to be imported on startup.', category='Paths')
     lora_dir            : Path = Field(default=None, description='Path to a directory of LoRA/LyCORIS models to be imported on startup.', category='Paths')
@@ -419,16 +215,43 @@ class InvokeAIAppConfig(InvokeAISettings):
     legacy_conf_dir     : Path = Field(default='configs/stable-diffusion', description='Path to directory of legacy checkpoint config files', category='Paths')
     db_dir              : Path = Field(default='databases', description='Path to InvokeAI databases directory', category='Paths')
     outdir              : Path = Field(default='outputs', description='Default folder for output images', category='Paths')
-    from_file           : Path = Field(default=None, description='Take command input from the indicated file (command-line client only)', category='Paths')
     use_memory_db       : bool = Field(default=False, description='Use in-memory database for storing image metadata', category='Paths')
-    ignore_missing_core_models : bool = Field(default=False, description='Ignore missing models in models/core/convert', category='Features')
+    from_file           : Path = Field(default=None, description='Take command input from the indicated file (command-line client only)', category='Paths')
 
+    # LOGGING
     log_handlers        : List[str] = Field(default=["console"], description='Log handler. Valid options are "console", "file=<path>", "syslog=path|address:host:port", "http=<url>"', category="Logging")
     # note - would be better to read the log_format values from logging.py, but this creates circular dependencies issues
     log_format          : Literal[tuple(['plain','color','syslog','legacy'])] = Field(default="color", description='Log format. Use "plain" for text-only, "color" for colorized output, "legacy" for 2.3-style logging and "syslog" for syslog-style', category="Logging")
     log_level           : Literal[tuple(["debug","info","warning","error","critical"])] = Field(default="info", description="Emit logging messages at this level or  higher", category="Logging")
 
     version             : bool = Field(default=False, description="Show InvokeAI version and exit", category="Other")
+
+    # CACHE
+    ram                 : Union[float, Literal["auto"]] = Field(default=6.0, gt=0, description="Maximum memory amount used by model cache for rapid switching (floating point number or 'auto')", category="Cache", )
+    vram                : Union[float, Literal["auto"]] = Field(default=0.25, ge=0, description="Amount of VRAM reserved for model storage (floating point number or 'auto')", category="Cache", )
+    lazy_offload        : bool = Field(default=True, description="Keep models in VRAM until their space is needed", category="Cache", )
+
+    # DEVICE
+    device              : Literal[tuple(["auto", "cpu", "cuda", "cuda:1", "mps"])] = Field(default="auto", description="Generation device", category="Device", )
+    precision: Literal[tuple(["auto", "float16", "float32", "autocast"])] = Field(default="auto", description="Floating point precision", category="Device", )
+
+    # GENERATION
+    sequential_guidance : bool = Field(default=False, description="Whether to calculate guidance in serial instead of in parallel, lowering memory requirements", category="Generation", )
+    attention_type      : Literal[tuple(["auto", "normal", "xformers", "sliced", "torch-sdp"])] = Field(default="auto", description="Attention type", category="Generation", )
+    attention_slice_size: Literal[tuple(["auto", "max", 1, 2, 3, 4, 5, 6, 7, 8])] = Field(default="auto", description='Slice size, valid when attention_type=="sliced"', category="Generation", )
+    force_tiled_decode: bool = Field(default=False, description="Whether to enable tiled VAE decode (reduces memory consumption with some performance penalty)", category="Generation",)
+    
+    # DEPRECATED FIELDS - STILL HERE IN ORDER TO OBTAN VALUES FROM PRE-3.1 CONFIG FILES
+    always_use_cpu      : bool = Field(default=False, description="If true, use the CPU for rendering even if a GPU is available.", category='Memory/Performance')
+    free_gpu_mem        : Optional[bool] = Field(default=None, description="If true, purge model from GPU after each generation.", category='Memory/Performance')
+    max_cache_size      : Optional[float] = Field(default=None, gt=0, description="Maximum memory amount used by model cache for rapid switching", category='Memory/Performance')
+    max_vram_cache_size : Optional[float] = Field(default=None, ge=0, description="Amount of VRAM reserved for model storage", category='Memory/Performance')
+    precision           : Literal[tuple(['auto','float16','float32','autocast'])] = Field(default='auto',description='Floating point precision', category='Memory/Performance')
+    sequential_guidance : bool = Field(default=False, description="Whether to calculate guidance in serial instead of in parallel, lowering memory requirements", category='Memory/Performance')
+    xformers_enabled    : bool = Field(default=True, description="Enable/disable memory-efficient attention", category='Memory/Performance')
+    tiled_decode        : bool = Field(default=False, description="Whether to enable tiled VAE decode (reduces memory consumption with some performance penalty)", category='Memory/Performance')
+
+    # See InvokeAIAppConfig subclass below for CACHE and DEVICE categories
     # fmt: on
 
     class Config:
@@ -551,16 +374,6 @@ class InvokeAIAppConfig(InvokeAISettings):
         """Return true if precision set to float32"""
         return self.precision == "float32"
 
-    @property
-    def xformers_enabled(self) -> bool:
-        """Return true if attention_type=='xformers'."""
-        return self.attention_type=='xformers'
-
-    @property
-    def disable_xformers(self) -> bool:
-        """Return true if xformers_enabled is false"""
-        return not self.xformers_enabled
-
     @property
     def try_patchmatch(self) -> bool:
         """Return true if patchmatch true"""
@@ -577,14 +390,25 @@ class InvokeAIAppConfig(InvokeAISettings):
         return True
 
     @property
-    def max_cache_size(self) -> Union[str, float]:
+    def ram_cache_size(self) -> float:
-        """return value of ram attribute."""
+        return self.max_cache_size or self.ram
-        return self.ram
 
     @property
-    def max_vram_cache_size(self) -> Union[str, float]:
+    def vram_cache_size(self) -> float:
-        """return value of vram attribute."""
+        return self.max_vram_cache_size or self.vram
-        return self.vram
+
+    @property
+    def use_cpu(self) -> bool:
+        return self.always_use_cpu or self.device == "cpu"
+
+    @property
+    def disable_xformers(self) -> bool:
+        """
+        Return true if enable_xformers is false (reversed logic)
+        and attention type is not set to xformers.
+        """
+        disabled_in_config = not self.xformers_enabled
+        return disabled_in_config and self.attention_type != "xformers"
 
     @staticmethod
     def find_root() -> Path:
@@ -594,29 +418,6 @@ class InvokeAIAppConfig(InvokeAISettings):
         """
         return _find_root()
 
-    # @property
-    # def attention_slice_size(self) -> Union[str, int]:
-    #     """
-    #     Return one of "auto", "max", or 1-8.
-    #     """
-    #     size = self.attention_slice
-    #     try:
-    #         size= int(size)
-    #         assert size > 0
-    #     except ValueError:
-    #         pass
-    #     return size
-
-class PagingArgumentParser(argparse.ArgumentParser):
-    """
-    A custom ArgumentParser that uses pydoc to page its output.
-    It also supports reading defaults from an init file.
-    """
-
-    def print_help(self, file=None):
-        text = self.format_help()
-        pydoc.pager(text)
-
 
 def get_invokeai_config(**kwargs) -> InvokeAIAppConfig:
     """
@@ -624,17 +425,13 @@ def get_invokeai_config(**kwargs) -> InvokeAIAppConfig:
     """
     return InvokeAIAppConfig.get_config(**kwargs)
 
-def int_or_float_or_str(value:Any) -> Union[int, float, str]:
-    """
-    Workaround for argparse type checking.
-    """
-    try:
-        return int(value)
-    except:
-        pass
-    try:
-        return float(value)
-    except:
-        pass
-    return str(value)
 
+def _find_root() -> Path:
+    venv = Path(os.environ.get("VIRTUAL_ENV") or ".")
+    if os.environ.get("INVOKEAI_ROOT"):
+        root = Path(os.environ["INVOKEAI_ROOT"])
+    elif any([(venv.parent / x).exists() for x in [INIT_FILE, LEGACY_INIT_FILE]]):
+        root = (venv.parent).resolve()
+    else:
+        root = Path("~/invokeai").expanduser().resolve()
+    return root

invokeai/app/services/model_manager_service.py

@@ -322,8 +322,8 @@ class ModelManagerService(ModelManagerServiceBase):
         # configuration value. If present, then the
         # cache size is set to 2.5 GB times
         # the number of max_loaded_models. Otherwise
-        # use new `max_cache_size` config setting
+        # use new `ram_cache_size` config setting
-        max_cache_size = config.max_cache_size if hasattr(config, "max_cache_size") else config.max_loaded_models * 2.5
+        max_cache_size = config.ram_cache_size
 
         logger.debug(f"Maximum RAM cache size: {max_cache_size} GiB")
 

invokeai/backend/install/invokeai_configure.py

@@ -21,6 +21,7 @@ from argparse import Namespace
 from enum import Enum
 from pathlib import Path
 from shutil import get_terminal_size
+from typing import get_type_hints, get_args, Any
 from urllib import request
 
 import npyscreen
@@ -50,6 +51,7 @@ from invokeai.frontend.install.model_install import addModelsForm, process_and_e
 # TO DO - Move all the frontend code into invokeai.frontend.install
 from invokeai.frontend.install.widgets import (
     SingleSelectColumns,
+    MultiSelectColumns,
     CenteredButtonPress,
     FileBox,
     IntTitleSlider,
@@ -72,6 +74,10 @@ warnings.filterwarnings("ignore")
 transformers.logging.set_verbosity_error()
 
 
+def get_literal_fields(field) -> list[Any]:
+    return get_args(get_type_hints(InvokeAIAppConfig).get(field))
+
+
 # --------------------------globals-----------------------
 
 config = InvokeAIAppConfig.get_config()
@@ -81,7 +87,11 @@ Model_dir = "models"
 Default_config_file = config.model_conf_path
 SD_Configs = config.legacy_conf_path
 
-PRECISION_CHOICES = ["auto", "float16", "float32"]
+PRECISION_CHOICES = get_literal_fields("precision")
+DEVICE_CHOICES = get_literal_fields("device")
+ATTENTION_CHOICES = get_literal_fields("attention_type")
+ATTENTION_SLICE_CHOICES = get_literal_fields("attention_slice_size")
+GENERATION_OPT_CHOICES = ["sequential_guidance", "force_tiled_decode", "lazy_offload"]
 GB = 1073741824  # GB in bytes
 HAS_CUDA = torch.cuda.is_available()
 _, MAX_VRAM = torch.cuda.mem_get_info() if HAS_CUDA else (0, 0)
@@ -312,6 +322,7 @@ class editOptsForm(CyclingForm, npyscreen.FormMultiPage):
 Use ctrl-N and ctrl-P to move to the <N>ext and <P>revious fields.
 Use cursor arrows to make a checkbox selection, and space to toggle.
 """
+        self.nextrely -= 1
         for i in textwrap.wrap(label, width=window_width - 6):
             self.add_widget_intelligent(
                 npyscreen.FixedText,
@@ -338,76 +349,130 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
             use_two_lines=False,
             scroll_exit=True,
         )
-        self.nextrely += 1
+
-        self.add_widget_intelligent(
+        # old settings for defaults
-            npyscreen.TitleFixedText,
-            name="GPU Management",
-            begin_entry_at=0,
-            editable=False,
-            color="CONTROL",
-            scroll_exit=True,
-        )
-        self.nextrely -= 1
-        self.free_gpu_mem = self.add_widget_intelligent(
-            npyscreen.Checkbox,
-            name="Free GPU memory after each generation",
-            value=old_opts.free_gpu_mem,
-            max_width=45,
-            relx=5,
-            scroll_exit=True,
-        )
-        self.nextrely -= 1
-        self.xformers_enabled = self.add_widget_intelligent(
-            npyscreen.Checkbox,
-            name="Enable xformers support",
-            value=old_opts.xformers_enabled,
-            max_width=30,
-            relx=50,
-            scroll_exit=True,
-        )
-        self.nextrely -= 1
-        self.always_use_cpu = self.add_widget_intelligent(
-            npyscreen.Checkbox,
-            name="Force CPU to be used on GPU systems",
-            value=old_opts.always_use_cpu,
-            relx=80,
-            scroll_exit=True,
-        )
         precision = old_opts.precision or ("float32" if program_opts.full_precision else "auto")
+        device = old_opts.device
+        attention_type = "xformers" if old_opts.xformers_enabled else old_opts.attention_type
+        attention_slice_size = old_opts.attention_slice_size
+
         self.nextrely += 1
         self.add_widget_intelligent(
             npyscreen.TitleFixedText,
-            name="Floating Point Precision",
+            name="Image Generation Options:",
+            editable=False,
+            color="CONTROL",
+            scroll_exit=True,
+        )
+        self.nextrely -= 2
+        self.generation_options = self.add_widget_intelligent(
+            MultiSelectColumns,
+            columns=3,
+            values=GENERATION_OPT_CHOICES,
+            value=[GENERATION_OPT_CHOICES.index(x) for x in GENERATION_OPT_CHOICES if getattr(old_opts, x)],
+            relx=30,
+            max_height=2,
+            max_width=80,
+            scroll_exit=True,
+        )
+
+        self.add_widget_intelligent(
+            npyscreen.TitleFixedText,
+            name="Floating Point Precision:",
             begin_entry_at=0,
             editable=False,
             color="CONTROL",
             scroll_exit=True,
         )
-        self.nextrely -= 1
+        self.nextrely -= 2
         self.precision = self.add_widget_intelligent(
             SingleSelectColumns,
-            columns=3,
+            columns=len(PRECISION_CHOICES),
             name="Precision",
             values=PRECISION_CHOICES,
             value=PRECISION_CHOICES.index(precision),
             begin_entry_at=3,
             max_height=2,
+            relx=30,
+            max_width=56,
+            scroll_exit=True,
+        )
+        self.add_widget_intelligent(
+            npyscreen.TitleFixedText,
+            name="Generation Device:",
+            begin_entry_at=0,
+            editable=False,
+            color="CONTROL",
+            scroll_exit=True,
+        )
+        self.nextrely -= 2
+        self.device = self.add_widget_intelligent(
+            SingleSelectColumns,
+            columns=len(DEVICE_CHOICES),
+            values=DEVICE_CHOICES,
+            value=DEVICE_CHOICES.index(device),
+            begin_entry_at=3,
+            relx=30,
+            max_height=2,
+            max_width=60,
+            scroll_exit=True,
+        )
+        self.add_widget_intelligent(
+            npyscreen.TitleFixedText,
+            name="Attention Type:",
+            begin_entry_at=0,
+            editable=False,
+            color="CONTROL",
+            scroll_exit=True,
+        )
+        self.nextrely -= 2
+        self.attention_type = self.add_widget_intelligent(
+            SingleSelectColumns,
+            columns=len(ATTENTION_CHOICES),
+            values=ATTENTION_CHOICES,
+            value=ATTENTION_CHOICES.index(attention_type),
+            begin_entry_at=3,
+            max_height=2,
+            relx=30,
             max_width=80,
             scroll_exit=True,
         )
-        self.nextrely += 1
+        self.attention_type.on_changed = self.show_hide_slice_sizes
+        self.attention_slice_label = self.add_widget_intelligent(
+            npyscreen.TitleFixedText,
+            name="Attention Slice Size:",
+            relx=5,
+            editable=False,
+            hidden=True,
+            color="CONTROL",
+            scroll_exit=True,
+        )
+        self.nextrely -= 2
+        self.attention_slice_size = self.add_widget_intelligent(
+            SingleSelectColumns,
+            columns=len(ATTENTION_SLICE_CHOICES),
+            values=ATTENTION_SLICE_CHOICES,
+            value=ATTENTION_SLICE_CHOICES.index(attention_slice_size),
+            begin_entry_at=2,
+            relx=30,
+            hidden=True,
+            max_height=2,
+            max_width=100,
+            scroll_exit=True,
+        )
+
         self.add_widget_intelligent(
             npyscreen.TitleFixedText,
-            name="RAM cache size (GB). Make this at least large enough to hold a single full model.",
+            name="Model RAM cache size (GB). Make this at least large enough to hold a single full model.",
             begin_entry_at=0,
             editable=False,
             color="CONTROL",
             scroll_exit=True,
         )
         self.nextrely -= 1
-        self.max_cache_size = self.add_widget_intelligent(
+        self.ram = self.add_widget_intelligent(
             npyscreen.Slider,
-            value=clip(old_opts.max_cache_size, range=(3.0, MAX_RAM), step=0.5),
+            value=clip(old_opts.ram_cache_size, range=(3.0, MAX_RAM), step=0.5),
             out_of=round(MAX_RAM),
             lowest=0.0,
             step=0.5,
@@ -418,16 +483,16 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
             self.nextrely += 1
             self.add_widget_intelligent(
                 npyscreen.TitleFixedText,
-                name="VRAM cache size (GB). Reserving a small amount of VRAM will modestly speed up the start of image generation.",
+                name="Model VRAM cache size (GB). Reserving a small amount of VRAM will modestly speed up the start of image generation.",
                 begin_entry_at=0,
                 editable=False,
                 color="CONTROL",
                 scroll_exit=True,
             )
             self.nextrely -= 1
-            self.max_vram_cache_size = self.add_widget_intelligent(
+            self.vram = self.add_widget_intelligent(
                 npyscreen.Slider,
-                value=clip(old_opts.max_vram_cache_size, range=(0, MAX_VRAM), step=0.25),
+                value=clip(old_opts.vram_cache_size, range=(0, MAX_VRAM), step=0.25),
                 out_of=round(MAX_VRAM * 2) / 2,
                 lowest=0.0,
                 relx=8,
@@ -435,7 +500,7 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
                 scroll_exit=True,
             )
         else:
-            self.max_vram_cache_size = DummyWidgetValue.zero
+            self.vram_cache_size = DummyWidgetValue.zero
         self.nextrely += 1
         self.outdir = self.add_widget_intelligent(
             FileBox,
@@ -491,6 +556,11 @@ https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENS
             when_pressed_function=self.on_ok,
         )
 
+    def show_hide_slice_sizes(self, value):
+        show = ATTENTION_CHOICES[value[0]] == "sliced"
+        self.attention_slice_label.hidden = not show
+        self.attention_slice_size.hidden = not show
+
     def on_ok(self):
         options = self.marshall_arguments()
         if self.validate_field_values(options):
@@ -524,12 +594,9 @@ https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENS
         new_opts = Namespace()
 
         for attr in [
+            "ram",
+            "vram",
             "outdir",
-            "free_gpu_mem",
-            "max_cache_size",
-            "max_vram_cache_size",
-            "xformers_enabled",
-            "always_use_cpu",
         ]:
             setattr(new_opts, attr, getattr(self, attr).value)
 
@@ -542,6 +609,12 @@ https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/LICENS
         new_opts.hf_token = self.hf_token.value
         new_opts.license_acceptance = self.license_acceptance.value
         new_opts.precision = PRECISION_CHOICES[self.precision.value[0]]
+        new_opts.device = DEVICE_CHOICES[self.device.value[0]]
+        new_opts.attention_type = ATTENTION_CHOICES[self.attention_type.value[0]]
+        new_opts.attention_slice_size = ATTENTION_SLICE_CHOICES[self.attention_slice_size.value]
+        generation_options = [GENERATION_OPT_CHOICES[x] for x in self.generation_options.value]
+        for v in GENERATION_OPT_CHOICES:
+            setattr(new_opts, v, v in generation_options)
 
         return new_opts
 

invokeai/backend/model_management/model_manager.py

@@ -341,7 +341,7 @@ class ModelManager(object):
         self.logger = logger
         self.cache = ModelCache(
             max_cache_size=max_cache_size,
-            max_vram_cache_size=self.app_config.max_vram_cache_size,
+            max_vram_cache_size=self.app_config.vram_cache_size,
             execution_device=device_type,
             precision=precision,
             sequential_offload=sequential_offload,

invokeai/backend/util/devices.py

@@ -17,13 +17,17 @@ config = InvokeAIAppConfig.get_config()
 
 def choose_torch_device() -> torch.device:
     """Convenience routine for guessing which GPU device to run model on"""
-    if config.always_use_cpu:
+    if config.use_cpu:  # legacy setting - force CPU
         return CPU_DEVICE
-    if torch.cuda.is_available():
+    elif config.device == "auto":
-        return torch.device("cuda")
+        if torch.cuda.is_available():
-    if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+            return torch.device("cuda")
-        return torch.device("mps")
+        if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
-    return CPU_DEVICE
+            return torch.device("mps")
+        else:
+            return CPU_DEVICE
+    else:
+        return torch.device(config.device)
 
 
 def choose_precision(device: torch.device) -> str:

invokeai/frontend/install/widgets.py

@@ -18,7 +18,7 @@ from curses import BUTTON2_CLICKED, BUTTON3_CLICKED
 
 # minimum size for UIs
 MIN_COLS = 130
-MIN_LINES = 38
+MIN_LINES = 40
 
 
 class WindowTooSmallException(Exception):
@@ -277,6 +277,9 @@ class SingleSelectColumns(SelectColumnBase, SingleSelectWithChanged):
     def h_cursor_line_right(self, ch):
         self.h_exit_down("bye bye")
 
+    def h_cursor_line_left(self, ch):
+        self.h_exit_up("bye bye")
+
 
 class TextBoxInner(npyscreen.MultiLineEdit):
     def __init__(self, *args, **kwargs):
@@ -324,55 +327,6 @@ class TextBoxInner(npyscreen.MultiLineEdit):
         if bstate & (BUTTON2_CLICKED | BUTTON3_CLICKED):
             self.h_paste()
 
-    # def update(self, clear=True):
-    #     if clear:
-    #         self.clear()
-
-    #     HEIGHT = self.height
-    #     WIDTH = self.width
-    #     # draw box.
-    #     self.parent.curses_pad.hline(self.rely, self.relx, curses.ACS_HLINE, WIDTH)
-    #     self.parent.curses_pad.hline(
-    #         self.rely + HEIGHT, self.relx, curses.ACS_HLINE, WIDTH
-    #     )
-    #     self.parent.curses_pad.vline(
-    #         self.rely, self.relx, curses.ACS_VLINE, self.height
-    #     )
-    #     self.parent.curses_pad.vline(
-    #         self.rely, self.relx + WIDTH, curses.ACS_VLINE, HEIGHT
-    #     )
-
-    # # draw corners
-    # self.parent.curses_pad.addch(
-    #     self.rely,
-    #     self.relx,
-    #     curses.ACS_ULCORNER,
-    # )
-    # self.parent.curses_pad.addch(
-    #     self.rely,
-    #     self.relx + WIDTH,
-    #     curses.ACS_URCORNER,
-    # )
-    # self.parent.curses_pad.addch(
-    #     self.rely + HEIGHT,
-    #     self.relx,
-    #     curses.ACS_LLCORNER,
-    # )
-    # self.parent.curses_pad.addch(
-    #     self.rely + HEIGHT,
-    #     self.relx + WIDTH,
-    #     curses.ACS_LRCORNER,
-    # )
-
-    # # fool our superclass into thinking drawing area is smaller - this is really hacky but it seems to work
-    # (relx, rely, height, width) = (self.relx, self.rely, self.height, self.width)
-    # self.relx += 1
-    # self.rely += 1
-    # self.height -= 1
-    # self.width -= 1
-    # super().update(clear=False)
-    # (self.relx, self.rely, self.height, self.width) = (relx, rely, height, width)
-
 
 class TextBox(npyscreen.BoxTitle):
     _contained_widget = TextBoxInner

tests/test_config.py

@@ -31,6 +31,21 @@ InvokeAI:
 """
 )
 
+init3 = OmegaConf.create(
+    """
+InvokeAI:
+  Generation:
+    sequential_guidance: true
+    attention_type: xformers
+    attention_slice_size: 7
+    forced_tiled_decode: True
+  Device:
+    device: cpu
+  Cache:
+    ram: 1.25
+"""
+)
+
 
 def test_use_init():
     # note that we explicitly set omegaconf dict and argv here
@@ -51,6 +66,16 @@ def test_use_init():
     assert not hasattr(conf2, "invalid_attribute")
 
 
+def test_legacy():
+    conf = InvokeAIAppConfig.get_config()
+    assert conf
+    conf.parse_args(conf=init3, argv=[])
+    assert conf.xformers_enabled
+    assert conf.device == "cpu"
+    assert conf.use_cpu
+    assert conf.ram_cache_size == 1.25
+
+
 def test_argv_override():
     conf = InvokeAIAppConfig.get_config()
     conf.parse_args(conf=init1, argv=["--always_use_cpu", "--max_cache=10"])