mirror of
https://github.com/lllyasviel/stable-diffusion-webui-forge.git
synced 2026-07-21 21:01:24 +08:00
188 lines
9.4 KiB
Python
188 lines
9.4 KiB
Python
# https://github.com/comfyanonymous/ComfyUI/blob/v0.7.0/comfy/cli_args.py
|
|
|
|
import argparse
|
|
import enum
|
|
|
|
|
|
class EnumAction(argparse.Action):
|
|
"""Argparse `action` for handling Enum"""
|
|
|
|
def __init__(self, **kwargs):
|
|
enum_type = kwargs.pop("type", None)
|
|
assert issubclass(enum_type, enum.Enum)
|
|
|
|
choices = tuple(e.value for e in enum_type)
|
|
kwargs.setdefault("choices", choices)
|
|
kwargs.setdefault("metavar", f"[{','.join(choices)}]")
|
|
|
|
super(EnumAction, self).__init__(**kwargs)
|
|
self._enum = enum_type
|
|
|
|
def __call__(self, parser, namespace, values, option_string=None):
|
|
value = self._enum(values)
|
|
setattr(namespace, self.dest, value)
|
|
|
|
|
|
parser = argparse.ArgumentParser(add_help=False)
|
|
|
|
parser.add_argument("--loglevel", type=str, default=None, choices=["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"], help="Set the logging level")
|
|
parser.add_argument("--gpu-device-id", type=int, default=None, metavar="DEVICE_ID", help="Set the id of device to use (all other devices will not be visible)")
|
|
parser.add_argument("--disable-gpu-warning", action="store_true", help="Disable the low VRAM warnings")
|
|
|
|
parser.add_argument("--text-enc-device", type=str, default=None, metavar="DEVICE", help='Set the device to load text encoder (e.g. "cuda:1")')
|
|
parser.add_argument("--vae-device", type=str, default=None, metavar="DEVICE", help='Set the device to load VAE (e.g. "cuda:1")')
|
|
|
|
fp_group = parser.add_mutually_exclusive_group()
|
|
fp_group.add_argument("--force-fp32", action="store_true", help="Force fp32")
|
|
fp_group.add_argument("--force-fp16", action="store_true", help="Force fp16")
|
|
|
|
fpunet_group = parser.add_mutually_exclusive_group()
|
|
fpunet_group.add_argument("--fp32-unet", action="store_true", help="Store the diffusion model in fp32")
|
|
fpunet_group.add_argument("--bf16-unet", action="store_true", help="Store the diffusion model in bf16")
|
|
fpunet_group.add_argument("--fp16-unet", action="store_true", help="Store the diffusion model in fp16")
|
|
fpunet_group.add_argument("--fp8_e4m3fn-unet", action="store_true", help="Store the diffusion model in fp8_e4m3fn")
|
|
fpunet_group.add_argument("--fp8_e5m2-unet", action="store_true", help="Store the diffusion model in fp8_e5m2")
|
|
fpunet_group.add_argument("--fp8_e8m0fnu-unet", action="store_true", help="Store the diffusion model in fp8_e8m0fnu")
|
|
|
|
fpvae_group = parser.add_mutually_exclusive_group()
|
|
fpvae_group.add_argument("--fp32-vae", action="store_true", help="Run the VAE in full precision fp32")
|
|
fpvae_group.add_argument("--bf16-vae", action="store_true", help="Run the VAE in bf16")
|
|
fpvae_group.add_argument("--fp16-vae", action="store_true", help="Run the VAE in fp16 (might cause black images)")
|
|
|
|
parser.add_argument("--cpu-vae", action="store_true", help="Run the VAE on the CPU")
|
|
|
|
fpte_group = parser.add_mutually_exclusive_group()
|
|
fpte_group.add_argument("--fp32-text-enc", action="store_true", help="Store the text encoder in fp32")
|
|
fpte_group.add_argument("--bf16-text-enc", action="store_true", help="Store the text encoder in bf16")
|
|
fpte_group.add_argument("--fp16-text-enc", action="store_true", help="Store the text encoder in fp16")
|
|
fpte_group.add_argument("--fp8_e4m3fn-text-enc", action="store_true", help="Store the text encoder in fp8_e4m3fn")
|
|
fpte_group.add_argument("--fp8_e5m2-text-enc", action="store_true", help="Store the text encoder in fp8_e5m2")
|
|
|
|
parser.add_argument("--cpu-text-enc", action="store_true", help="Run the text encoder on the CPU")
|
|
|
|
parser.add_argument("--use-pytorch-cross-attention", action="store_true", help="Use the PyTorch cross attention (override sageattention/flash_attn/xformers)")
|
|
parser.add_argument("--force-xformers-vae", action="store_true", help="Force VAE to use xformers attention (meant to use with PyTorch cross attention)")
|
|
parser.add_argument("--force-upcast-attention", action="store_true", help="Always upcast to fp32 during attention")
|
|
|
|
parser.add_argument("--sage", action="store_true", help="install sageattention")
|
|
parser.add_argument("--flash", action="store_true", help="install flash_attn")
|
|
parser.add_argument("--xformers", action="store_true", help="install xformers")
|
|
parser.add_argument("--nunchaku", action="store_true", help="install nunchaku for SVDQ inference")
|
|
parser.add_argument("--bnb", action="store_true", help="install bitsandbytes for 4-bit inference")
|
|
parser.add_argument("--onnxruntime-gpu", action="store_true", help="install nightly onnxruntime-gpu with cu130 support")
|
|
|
|
parser.add_argument("--disable-sage", action="store_true", help="disable sageattention")
|
|
parser.add_argument("--disable-flash", action="store_true", help="disable flash_attn")
|
|
parser.add_argument("--disable-xformers", action="store_true", help="disable xformers")
|
|
|
|
parser.add_argument("--directml", type=int, nargs="?", metavar="DIRECTML_DEVICE", const=-1, help="Use torch-directml")
|
|
parser.add_argument("--disable-ipex-optimize", action="store_true", help="Disable ipex.optimize default when loading models with Intel's Extension for PyTorch")
|
|
parser.add_argument("--deterministic", action="store_true", help="Use slower deterministic algorithms when possible")
|
|
|
|
vram_group = parser.add_mutually_exclusive_group()
|
|
vram_group.add_argument("--gpu-only", action="store_true", help="Store and run everything on the GPU")
|
|
vram_group.add_argument("--highvram", action="store_true", help="Keeps models in VRAM after usage")
|
|
vram_group.add_argument("--normalvram", action="store_true", help="Force NORMAL_VRAM in case LOW_VRAM gets automatically enabled")
|
|
vram_group.add_argument("--lowvram", action="store_true", help="Split the diffusion model in parts to use less VRAM")
|
|
vram_group.add_argument("--novram", action="store_true", help="When even LOW_VRAM is still not enough")
|
|
vram_group.add_argument("--cpu", action="store_true", help="Use the CPU for everything (slow)")
|
|
|
|
parser.add_argument("--reserve-vram", type=float, default=None, metavar="GB", help="Set the amount of VRAM you want to reserve for other software (by default some amount is reserved)")
|
|
parser.add_argument("--disable-smart-memory", action="store_true", help="Aggressively offload to RAM instead of keeping models in VRAM when possible")
|
|
parser.add_argument("--force-non-blocking", action="store_true", help="Use non-blocking operations for all applicable tensors")
|
|
|
|
parser.add_argument("--cuda-malloc", action="store_true", help="improve memory allocation")
|
|
parser.add_argument("--cuda-stream", type=int, nargs="?", metavar="NUM_STREAMS", const=2, help="improve offloading")
|
|
parser.add_argument("--pin-shared-memory", action="store_true", help="improve RAM utilization")
|
|
|
|
parser.add_argument("--fast-fp8", action="store_true", help="torch._scaled_mm")
|
|
parser.add_argument("--fast-fp16", action="store_true", help="torch.backends.cuda.matmul.allow_fp16_accumulation")
|
|
parser.add_argument("--autotune", action="store_true", help="torch.backends.cudnn.benchmark")
|
|
|
|
parser.add_argument("--mmap-torch-files", action="store_true", help="Use mmap when loading ckpt/pt files")
|
|
parser.add_argument("--disable-mmap", action="store_true", help="Don't use mmap when loading safetensors")
|
|
|
|
|
|
class SageAttentionFuncs(enum.Enum):
|
|
auto = "auto"
|
|
fp16_triton = "fp16_triton"
|
|
fp16_cuda = "fp16_cuda"
|
|
fp8_cuda = "fp8_cuda"
|
|
|
|
|
|
class Sage_quantization_backend(enum.Enum):
|
|
cuda = "cuda"
|
|
triton = "triton"
|
|
|
|
|
|
class Sage_qk_quant_gran(enum.Enum):
|
|
per_warp = "per_warp"
|
|
per_thread = "per_thread"
|
|
|
|
|
|
class Sage_pv_accum_dtype(enum.Enum):
|
|
fp16 = "fp16"
|
|
fp32 = "fp32"
|
|
fp16fp32 = "fp16+fp32"
|
|
fp32fp32 = "fp32+fp32"
|
|
|
|
|
|
sage2 = parser.add_argument_group(description="SageAttention 2")
|
|
sage2.add_argument("--sage2-function", type=SageAttentionFuncs, default=SageAttentionFuncs.auto, action=EnumAction)
|
|
sage2.add_argument("--sage-quantization-backend", type=Sage_quantization_backend, default=Sage_quantization_backend.triton, action=EnumAction)
|
|
sage2.add_argument("--sage-quant-gran", type=Sage_qk_quant_gran, default=Sage_qk_quant_gran.per_thread, action=EnumAction)
|
|
sage2.add_argument("--sage-accum-dtype", type=Sage_pv_accum_dtype, default=Sage_pv_accum_dtype.fp32, action=EnumAction)
|
|
|
|
|
|
args, _ = parser.parse_known_args()
|
|
|
|
from typing import TYPE_CHECKING
|
|
|
|
if TYPE_CHECKING:
|
|
import os
|
|
|
|
import torch
|
|
|
|
|
|
class _DynamicArgsMeta(type):
|
|
def get(cls, key, default=None):
|
|
return getattr(cls, key, default)
|
|
|
|
def __getitem__(cls, key, default=None):
|
|
return getattr(cls, key, default)
|
|
|
|
def __setitem__(cls, key, value):
|
|
setattr(cls, key, value)
|
|
|
|
def __contains__(cls, key):
|
|
return hasattr(cls, key)
|
|
|
|
|
|
class dynamic_args(metaclass=_DynamicArgsMeta):
|
|
"""Some parameters that are used throughout the Webui"""
|
|
|
|
embedding_dir: "os.PathLike" = None
|
|
"""set in modules/sd_models/forge_model_reload"""
|
|
forge_unet_storage_dtype: "torch.dtype" = None
|
|
"""set in modules/sd_models/forge_model_reload"""
|
|
online_lora: bool = False
|
|
"""patch LoRAs on-the-fly"""
|
|
kontext: bool = False
|
|
"""Flux Kontext"""
|
|
edit: bool = False
|
|
"""Qwen-Image-Edit"""
|
|
nunchaku: bool = False
|
|
"""Nunchaku (SVDQ) Models"""
|
|
klein: bool = False
|
|
"""Flux.2 Klein"""
|
|
wan: bool = False
|
|
"""Wan 2.2"""
|
|
ref_latents: list["torch.Tensor"] = []
|
|
"""Reference Latent(s) for Flux Kontext / Qwen-Image-Edit / Flux.2 Klein"""
|
|
concat_latent: "torch.Tensor" = None
|
|
"""Input Latent for Wan 2.2 I2V"""
|
|
is_referencing: bool = False
|
|
"""Appending Reference Latent(s) (by. ImageStitch)"""
|
|
ops: str = None
|
|
"""Operations for the Diffusion Model"""
|