mirror of
https://github.com/vladmandic/automatic
synced 2026-08-25 22:20:46 +02:00
Vladimir Mandic
824 lines
42 KiB
Python
824 lines
42 KiB
Python
from modules import logger
|
|
import os
|
|
import re
|
|
import sys
|
|
import copy
|
|
import json
|
|
import time
|
|
import diffusers
|
|
import transformers
|
|
from installer import installed, install, setup_logging
|
|
from modules.logger import log
|
|
|
|
|
|
ao = None
|
|
bnb = None
|
|
optimum_quanto = None
|
|
trt = None
|
|
quant_last_model_name = None
|
|
quant_last_model_device = None
|
|
debug = os.environ.get('SD_QUANT_DEBUG', None) is not None
|
|
|
|
|
|
def get_quant_type(args):
|
|
if args is not None and "quantization_config" in args:
|
|
return args['quantization_config'].__class__.__name__
|
|
return None
|
|
|
|
|
|
def get_quant(name):
|
|
if "qint8" in name.lower():
|
|
return 'qint8'
|
|
if "qint4" in name.lower():
|
|
return 'qint4'
|
|
if "fp8" in name.lower():
|
|
return 'fp8'
|
|
if "fp4" in name.lower():
|
|
return 'fp4'
|
|
if "nf4" in name.lower():
|
|
return 'nf4'
|
|
if name.endswith('.gguf'):
|
|
return 'gguf'
|
|
return 'none'
|
|
|
|
|
|
def dont_quant():
|
|
from modules import shared
|
|
models_list = re.split(r'[ ,]+', shared.opts.models_not_to_quant)
|
|
models_list = [m.lower().strip() for m in models_list]
|
|
if shared.sd_model_type.lower() in models_list:
|
|
logger.log.debug(f'Quantization: model={shared.sd_model_type} skip')
|
|
return True
|
|
return False
|
|
|
|
|
|
def create_bnb_config(kwargs = None, allow: bool = True, module: str = 'Model', modules_to_not_convert: list = None):
|
|
from modules import shared, devices
|
|
if allow and (module == 'any' or module in shared.opts.bnb_quantization):
|
|
load_bnb()
|
|
if bnb is None:
|
|
return kwargs
|
|
bnb_config = diffusers.BitsAndBytesConfig(
|
|
load_in_8bit=shared.opts.bnb_quantization_type in ['fp8'],
|
|
load_in_4bit=shared.opts.bnb_quantization_type in ['nf4', 'fp4'],
|
|
bnb_4bit_quant_storage=shared.opts.bnb_quantization_storage,
|
|
bnb_4bit_quant_type=shared.opts.bnb_quantization_type,
|
|
bnb_4bit_compute_dtype=devices.dtype,
|
|
llm_int8_skip_modules=modules_to_not_convert,
|
|
)
|
|
log.debug(f'Quantization: module={module} type=bnb dtype={shared.opts.bnb_quantization_type} storage={shared.opts.bnb_quantization_storage}')
|
|
if kwargs is None:
|
|
return bnb_config
|
|
else:
|
|
kwargs['quantization_config'] = bnb_config
|
|
return kwargs
|
|
return kwargs
|
|
|
|
|
|
def create_ao_config(kwargs = None, allow: bool = True, module: str = 'Model', modules_to_not_convert: list = None):
|
|
from modules import shared
|
|
if allow and (shared.opts.torchao_quantization_mode in {'pre', 'auto'}) and (module == 'any' or module in shared.opts.torchao_quantization):
|
|
torchao = load_torchao()
|
|
if torchao is None:
|
|
return kwargs
|
|
if module in {'TE', 'LLM'}:
|
|
ao_config = transformers.TorchAoConfig(quant_type=shared.opts.torchao_quantization_type, modules_to_not_convert=modules_to_not_convert)
|
|
else:
|
|
ao_config = diffusers.TorchAoConfig(shared.opts.torchao_quantization_type, modules_to_not_convert=modules_to_not_convert)
|
|
log.debug(f'Quantization: module={module} type=torchao dtype={shared.opts.torchao_quantization_type}')
|
|
if kwargs is None:
|
|
return ao_config
|
|
else:
|
|
kwargs['quantization_config'] = ao_config
|
|
return kwargs
|
|
return kwargs
|
|
|
|
|
|
def create_quanto_config(kwargs = None, allow: bool = True, module: str = 'Model', modules_to_not_convert: list = None):
|
|
from modules import shared
|
|
if allow and (module == 'any' or module in shared.opts.quanto_quantization):
|
|
load_quanto(silent=True)
|
|
if optimum_quanto is None:
|
|
return kwargs
|
|
if module in {'TE', 'LLM'}:
|
|
quanto_config = transformers.QuantoConfig(weights=shared.opts.quanto_quantization_type, modules_to_not_convert=modules_to_not_convert)
|
|
quanto_config.weights_dtype = quanto_config.weights
|
|
else:
|
|
quanto_config = diffusers.QuantoConfig(weights_dtype=shared.opts.quanto_quantization_type, modules_to_not_convert=modules_to_not_convert)
|
|
quanto_config.activations = None # patch so it works with transformers
|
|
quanto_config.weights = quanto_config.weights_dtype
|
|
log.debug(f'Quantization: module={module} type=quanto dtype={shared.opts.quanto_quantization_type}')
|
|
if kwargs is None:
|
|
return quanto_config
|
|
else:
|
|
kwargs['quantization_config'] = quanto_config
|
|
return kwargs
|
|
return kwargs
|
|
|
|
|
|
def create_trt_config(kwargs = None, allow: bool = True, module: str = 'Model', modules_to_not_convert: list = None):
|
|
from modules import shared
|
|
if allow and (module == 'any' or module in shared.opts.trt_quantization):
|
|
load_trt()
|
|
if trt is None:
|
|
return kwargs
|
|
trt_config_data = {
|
|
"int8": {"quant_type": "INT8", "quant_method": "modelopt", "modules_to_not_convert": []},
|
|
"int4": {"quant_type": "INT4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "modules_to_not_convert": ["conv", "patch_embed"]},
|
|
"fp8": {"quant_type": "FP8", "quant_method": "modelopt", "modules_to_not_convert": []},
|
|
"nf4": {"quant_type": "NF4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "scale_block_quantize": 8, "scale_channel_quantize": -1, "modules_to_not_convert": ["conv"]},
|
|
"nvfp4": {"quant_type": "NVFP4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "modules_to_not_convert": ["conv"]},
|
|
}
|
|
trt_quant_config = trt_config_data[shared.opts.trt_quantization_type].copy()
|
|
if modules_to_not_convert is not None:
|
|
for m in modules_to_not_convert:
|
|
if m not in trt_quant_config['modules_to_not_convert']:
|
|
trt_quant_config['modules_to_not_convert'].append(m)
|
|
trt_config = diffusers.quantizers.quantization_config.NVIDIAModelOptConfig(**trt_quant_config)
|
|
log.debug(f'Quantization: module={module} type=tensorrt dtype={shared.opts.trt_quantization_type}')
|
|
if kwargs is None:
|
|
return trt_config
|
|
else:
|
|
kwargs['quantization_config'] = trt_config
|
|
return kwargs
|
|
return kwargs
|
|
|
|
|
|
def get_sdnq_devices(mode="pre"):
|
|
from modules import devices, shared
|
|
if shared.opts.device_map == "gpu":
|
|
quantization_device = devices.device
|
|
return_device = devices.device
|
|
elif shared.opts.device_map == "cpu":
|
|
quantization_device = devices.cpu
|
|
return_device = devices.cpu
|
|
elif shared.opts.diffusers_offload_mode in {"none", "model"} or (mode == "post" and shared.opts.sdnq_quantize_shuffle_weights):
|
|
quantization_device = devices.device if shared.opts.sdnq_quantize_with_gpu else devices.cpu
|
|
return_device = devices.device
|
|
elif shared.opts.sdnq_quantize_with_gpu:
|
|
quantization_device = devices.device
|
|
return_device = devices.device if shared.opts.diffusers_to_gpu else devices.cpu
|
|
else:
|
|
quantization_device = None
|
|
return_device = None
|
|
return quantization_device, return_device
|
|
|
|
|
|
def create_sdnq_config(kwargs = None, allow: bool = True, module: str = 'Model', weights_dtype: str = None, quantized_matmul_dtype: str = None, modules_to_not_convert: list = None, modules_dtype_dict: dict = None):
|
|
from modules import shared
|
|
if allow and (shared.opts.sdnq_quantize_mode in {'pre', 'auto'}) and (module == 'any' or module in shared.opts.sdnq_quantize_weights):
|
|
from modules.sdnq import SDNQConfig
|
|
from modules.sdnq.common import use_torch_compile as sdnq_use_torch_compile
|
|
|
|
if shared.opts.sdnq_use_quantized_matmul and not sdnq_use_torch_compile:
|
|
logger.log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.')
|
|
shared.opts.sdnq_use_quantized_matmul = False
|
|
|
|
if weights_dtype is None:
|
|
if module in {"TE", "LLM"} and shared.opts.sdnq_quantize_weights_mode_te not in {"Same as model", "default"}:
|
|
weights_dtype = shared.opts.sdnq_quantize_weights_mode_te
|
|
else:
|
|
weights_dtype = shared.opts.sdnq_quantize_weights_mode
|
|
if weights_dtype is None or weights_dtype == 'none':
|
|
return kwargs
|
|
|
|
if quantized_matmul_dtype is None:
|
|
if module in {"TE", "LLM"} and shared.opts.sdnq_quantize_matmul_mode_te not in {"Same as model", "default"}:
|
|
quantized_matmul_dtype = shared.opts.sdnq_quantize_matmul_mode_te
|
|
else:
|
|
quantized_matmul_dtype = shared.opts.sdnq_quantize_matmul_mode
|
|
if quantized_matmul_dtype == "auto":
|
|
quantized_matmul_dtype = None
|
|
|
|
if modules_to_not_convert is None:
|
|
modules_to_not_convert = []
|
|
if modules_dtype_dict is None:
|
|
modules_dtype_dict = {}
|
|
|
|
sdnq_modules_to_not_convert = [m.strip() for m in re.split(';|,| ', shared.opts.sdnq_modules_to_not_convert) if len(m.strip()) > 1]
|
|
if len(sdnq_modules_to_not_convert) > 0:
|
|
modules_to_not_convert.extend(sdnq_modules_to_not_convert)
|
|
|
|
try:
|
|
if len(shared.opts.sdnq_modules_dtype_dict) > 2:
|
|
sdnq_modules_dtype_dict = shared.opts.sdnq_modules_dtype_dict
|
|
if "{" not in sdnq_modules_dtype_dict:
|
|
sdnq_modules_dtype_dict = "{" + sdnq_modules_dtype_dict + "}"
|
|
sdnq_modules_dtype_dict = json.loads(bytes(sdnq_modules_dtype_dict, 'utf-8'))
|
|
for key, value in sdnq_modules_dtype_dict.items():
|
|
if isinstance(value, str):
|
|
value = [m.strip() for m in re.split(';|,| ', value) if len(m.strip()) > 1]
|
|
if key not in modules_dtype_dict.keys():
|
|
modules_dtype_dict[key] = value
|
|
else:
|
|
modules_dtype_dict[key].extend(value)
|
|
except Exception as e:
|
|
log.warning(f'Quantization: SDNQ failed to parse sdnq_modules_dtype_dict: {e}')
|
|
|
|
quantization_device, return_device = get_sdnq_devices(mode="pre")
|
|
|
|
sdnq_config = SDNQConfig(
|
|
weights_dtype=weights_dtype,
|
|
quantized_matmul_dtype=quantized_matmul_dtype,
|
|
group_size=shared.opts.sdnq_quantize_weights_group_size,
|
|
svd_rank=shared.opts.sdnq_svd_rank,
|
|
svd_steps=shared.opts.sdnq_svd_steps,
|
|
dynamic_loss_threshold=shared.opts.sdnq_dynamic_loss_threshold,
|
|
use_svd=shared.opts.sdnq_use_svd,
|
|
quant_conv=shared.opts.sdnq_quantize_conv_layers,
|
|
use_quantized_matmul=shared.opts.sdnq_use_quantized_matmul,
|
|
use_quantized_matmul_conv=shared.opts.sdnq_use_quantized_matmul_conv,
|
|
use_dynamic_quantization=shared.opts.sdnq_use_dynamic_quantization,
|
|
dequantize_fp32=shared.opts.sdnq_dequantize_fp32,
|
|
non_blocking=shared.opts.diffusers_offload_nonblocking,
|
|
quantization_device=quantization_device,
|
|
return_device=return_device,
|
|
modules_to_not_convert=modules_to_not_convert,
|
|
modules_dtype_dict=modules_dtype_dict.copy(),
|
|
)
|
|
if quantized_matmul_dtype is None:
|
|
quantized_matmul_dtype = "auto" # set for logging
|
|
svd = f'{shared.opts.sdnq_use_svd} rank={shared.opts.sdnq_svd_rank} steps={shared.opts.sdnq_svd_steps}' if shared.opts.sdnq_use_svd else f'{shared.opts.sdnq_use_svd}'
|
|
log.debug(f'Quantization: module="{module}" type=sdnq mode=pre dtype={weights_dtype} svd={svd} dynamic={shared.opts.sdnq_use_dynamic_quantization} group={shared.opts.sdnq_quantize_weights_group_size} loss={shared.opts.sdnq_dynamic_loss_threshold} matmul_dtype={quantized_matmul_dtype} matmul_quant={shared.opts.sdnq_use_quantized_matmul} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} quant_conv={shared.opts.sdnq_quantize_conv_layers} fp32={shared.opts.sdnq_dequantize_fp32} device={quantization_device} return={return_device} use_gpu={shared.opts.sdnq_quantize_with_gpu} map={shared.opts.device_map} offload={shared.opts.diffusers_offload_mode} non_blocking={shared.opts.diffusers_offload_nonblocking} skip_modules={modules_to_not_convert} dict={modules_dtype_dict}')
|
|
if kwargs is None:
|
|
return sdnq_config
|
|
else:
|
|
kwargs['quantization_config'] = sdnq_config
|
|
return kwargs
|
|
return kwargs
|
|
|
|
|
|
def check_quant(module: str = ''):
|
|
from modules import shared
|
|
if module in shared.opts.sdnq_quantize_weights or module in shared.opts.bnb_quantization or module in shared.opts.torchao_quantization or module in shared.opts.quanto_quantization:
|
|
return True
|
|
return False
|
|
|
|
|
|
def check_nunchaku(module: str = ''):
|
|
from modules import shared
|
|
if 'nunchaku' not in shared.opts.sd_model_checkpoint.lower():
|
|
return False
|
|
base_path = shared.opts.sd_model_checkpoint.split('+')[0]
|
|
for v in shared.reference_models.values():
|
|
if v.get('path', '') != base_path:
|
|
continue
|
|
nunchaku_modules = v.get('nunchaku', None)
|
|
if nunchaku_modules is None:
|
|
continue
|
|
if isinstance(nunchaku_modules, bool) and nunchaku_modules:
|
|
nunchaku_modules = ['Model', 'TE']
|
|
if not isinstance(nunchaku_modules, list):
|
|
continue
|
|
if module in nunchaku_modules:
|
|
from modules import mit_nunchaku
|
|
mit_nunchaku.install_nunchaku()
|
|
return mit_nunchaku.ok
|
|
return False
|
|
|
|
|
|
def create_config(kwargs = None, allow: bool = True, module: str = 'Model', modules_to_not_convert: list = None, modules_dtype_dict: dict = None):
|
|
if kwargs is None:
|
|
kwargs = {}
|
|
if module == 'Model' and dont_quant():
|
|
return kwargs
|
|
kwargs = create_sdnq_config(kwargs, allow=allow, module=module, modules_to_not_convert=modules_to_not_convert, modules_dtype_dict=modules_dtype_dict)
|
|
if kwargs is not None and 'quantization_config' in kwargs:
|
|
if debug:
|
|
log.trace(f'Quantization: type=sdnq config={kwargs.get("quantization_config", None)}')
|
|
return kwargs
|
|
kwargs = create_bnb_config(kwargs, allow=allow, module=module, modules_to_not_convert=modules_to_not_convert)
|
|
if kwargs is not None and 'quantization_config' in kwargs:
|
|
if debug:
|
|
log.trace(f'Quantization: type=bnb config={kwargs.get("quantization_config", None)}')
|
|
return kwargs
|
|
kwargs = create_quanto_config(kwargs, allow=allow, module=module, modules_to_not_convert=modules_to_not_convert)
|
|
if kwargs is not None and 'quantization_config' in kwargs:
|
|
if debug:
|
|
log.trace(f'Quantization: type=quanto config={kwargs.get("quantization_config", None)}')
|
|
return kwargs
|
|
kwargs = create_ao_config(kwargs, allow=allow, module=module, modules_to_not_convert=modules_to_not_convert)
|
|
if kwargs is not None and 'quantization_config' in kwargs:
|
|
if debug:
|
|
log.trace(f'Quantization: type=torchao config={kwargs.get("quantization_config", None)}')
|
|
return kwargs
|
|
kwargs = create_trt_config(kwargs, allow=allow, module=module, modules_to_not_convert=modules_to_not_convert)
|
|
if kwargs is not None and 'quantization_config' in kwargs:
|
|
if debug:
|
|
log.trace(f'Quantization: type=tensorrt config={kwargs.get("quantization_config", None)}')
|
|
return kwargs
|
|
return kwargs
|
|
|
|
|
|
def load_torchao(msg='', silent=False):
|
|
global ao # pylint: disable=global-statement
|
|
if ao is not None:
|
|
return ao
|
|
if not installed('torchao'):
|
|
install('torchao==0.10.0', quiet=True)
|
|
log.warning('Quantization: torchao installed please restart')
|
|
try:
|
|
import torchao
|
|
ao = torchao
|
|
fn = f'{sys._getframe(2).f_code.co_name}:{sys._getframe(1).f_code.co_name}' # pylint: disable=protected-access
|
|
log.debug(f'Quantization: type=torchao version={ao.__version__} fn={fn}') # pylint: disable=protected-access
|
|
from diffusers.utils import import_utils
|
|
import_utils.is_torchao_available = lambda: True
|
|
import_utils._torchao_available = True # pylint: disable=protected-access
|
|
return ao
|
|
except Exception as e:
|
|
if len(msg) > 0:
|
|
log.error(f"{msg} failed to import torchao: {e}")
|
|
ao = None
|
|
if not silent:
|
|
raise
|
|
return None
|
|
|
|
|
|
def load_bnb(msg='', silent=False):
|
|
from modules import devices
|
|
global bnb # pylint: disable=global-statement
|
|
if bnb is not None:
|
|
return bnb
|
|
if not installed('bitsandbytes'):
|
|
if devices.backend == 'cuda':
|
|
# forcing a version will uninstall the multi-backend-refactor branch of bnb
|
|
install('bitsandbytes==0.47.0', quiet=True)
|
|
log.warning('Quantization: bitsandbytes installed please restart')
|
|
try:
|
|
import bitsandbytes
|
|
bnb = bitsandbytes
|
|
from diffusers.utils import import_utils
|
|
import_utils._bitsandbytes_available = True # pylint: disable=protected-access
|
|
import_utils._bitsandbytes_version = '0.43.3' # pylint: disable=protected-access
|
|
fn = f'{sys._getframe(3).f_code.co_name}:{sys._getframe(2).f_code.co_name}:{sys._getframe(1).f_code.co_name}' # pylint: disable=protected-access
|
|
log.debug(f'Quantization: type=bitsandbytes version={bnb.__version__} fn={fn}') # pylint: disable=protected-access
|
|
return bnb
|
|
except Exception as e:
|
|
if len(msg) > 0:
|
|
log.error(f"{msg} failed to import bitsandbytes: {e}")
|
|
bnb = None
|
|
if not silent:
|
|
raise
|
|
return None
|
|
|
|
|
|
def load_quanto(msg='', silent=False):
|
|
global optimum_quanto # pylint: disable=global-statement
|
|
if optimum_quanto is not None:
|
|
return optimum_quanto
|
|
if not installed('optimum-quanto'):
|
|
install('optimum-quanto==0.2.7', quiet=True)
|
|
log.warning('Quantization: optimum-quanto installed please restart')
|
|
try:
|
|
from optimum import quanto # pylint: disable=no-name-in-module
|
|
# disable device specific tensors because the model can't be moved between cpu and gpu with them
|
|
quanto.tensor.weights.qbits.WeightQBitsTensor.create = lambda *args, **kwargs: quanto.tensor.weights.qbits.WeightQBitsTensor(*args, **kwargs)
|
|
optimum_quanto = quanto
|
|
fn = f'{sys._getframe(3).f_code.co_name}:{sys._getframe(2).f_code.co_name}:{sys._getframe(1).f_code.co_name}' # pylint: disable=protected-access
|
|
log.debug(f'Quantization: type=quanto version={quanto.__version__} fn={fn}') # pylint: disable=protected-access
|
|
from diffusers.utils import import_utils
|
|
import_utils.is_optimum_quanto_available = lambda: True
|
|
import_utils._optimum_quanto_available = True # pylint: disable=protected-access
|
|
import_utils._optimum_quanto_version = quanto.__version__ # pylint: disable=protected-access
|
|
import_utils._replace_with_quanto_layers = diffusers.quantizers.quanto.utils._replace_with_quanto_layers # pylint: disable=protected-access
|
|
return optimum_quanto
|
|
except Exception as e:
|
|
if len(msg) > 0:
|
|
log.error(f"{msg} failed to import optimum.quanto: {e}")
|
|
optimum_quanto = None
|
|
if not silent:
|
|
raise
|
|
return None
|
|
|
|
|
|
def load_trt(msg='', silent=False):
|
|
global trt # pylint: disable=global-statement
|
|
if trt is not None:
|
|
return trt
|
|
try:
|
|
install('nvidia-modelopt')
|
|
import pydantic
|
|
if pydantic.__version__.startswith('1'):
|
|
log.error('Quantization: type=tensorrt pydantic==2 required')
|
|
return None
|
|
import modelopt
|
|
trt = modelopt
|
|
fn = f'{sys._getframe(3).f_code.co_name}:{sys._getframe(2).f_code.co_name}:{sys._getframe(1).f_code.co_name}' # pylint: disable=protected-access
|
|
log.debug(f'Quantization: type=tensorrt version={trt.__version__} fn={fn}') # pylint: disable=protected-access
|
|
return trt
|
|
except Exception as e:
|
|
if len(msg) > 0:
|
|
log.error(f"{msg} failed to import tensorrt: {e}")
|
|
trt = None
|
|
if not silent:
|
|
raise
|
|
return None
|
|
|
|
|
|
def upcast_non_layerwise_modules(model, dtype): # pylint: disable=unused-argument
|
|
from diffusers.hooks.layerwise_casting import _GO_LC_SUPPORTED_PYTORCH_LAYERS
|
|
model_children = list(model.children())
|
|
if not model_children:
|
|
if not isinstance(model, _GO_LC_SUPPORTED_PYTORCH_LAYERS):
|
|
model = model.to(dtype)
|
|
return model
|
|
for module in model_children:
|
|
has_children = list(module.children())
|
|
if not has_children:
|
|
if not isinstance(module, _GO_LC_SUPPORTED_PYTORCH_LAYERS):
|
|
module = module.to(dtype)
|
|
else:
|
|
module = upcast_non_layerwise_modules(module, dtype)
|
|
return model
|
|
|
|
|
|
def load_fp8_model_layerwise(checkpoint_info, load_model_func, diffusers_load_config):
|
|
model = None
|
|
if isinstance(checkpoint_info, str):
|
|
repo_path = checkpoint_info
|
|
else:
|
|
repo_path = checkpoint_info.path
|
|
try:
|
|
import torch
|
|
from modules import devices, shared
|
|
from diffusers.quantizers import quantization_config
|
|
if not hasattr(quantization_config.QuantizationMethod, 'LAYERWISE'):
|
|
setattr(quantization_config.QuantizationMethod, 'LAYERWISE', 'layerwise') # noqa: B010
|
|
if "e5m2" in repo_path.lower():
|
|
storage_dtype = torch.float8_e5m2
|
|
else:
|
|
storage_dtype= torch.float8_e4m3fn
|
|
load_args = diffusers_load_config.copy()
|
|
load_args["torch_dtype"] = storage_dtype
|
|
model = load_model_func(repo_path, **load_args)
|
|
model = upcast_non_layerwise_modules(model, devices.dtype)
|
|
model._skip_layerwise_casting_patterns = None # pylint: disable=protected-access
|
|
model.enable_layerwise_casting(compute_dtype=devices.dtype, storage_dtype=storage_dtype, non_blocking=shared.opts.diffusers_offload_nonblocking, skip_modules_pattern=[])
|
|
model.layerwise_storage_dtype = storage_dtype
|
|
model.quantization_method = 'LayerWise'
|
|
except Exception as e:
|
|
log.error(f"Load model: Failed to load FP8 model: {e}")
|
|
model = None
|
|
return model
|
|
|
|
|
|
def apply_layerwise(sd_model, quiet:bool=False):
|
|
import torch
|
|
from diffusers.quantizers import quantization_config
|
|
from modules import shared, devices, sd_models
|
|
if shared.opts.layerwise_quantization_storage == 'float8_e4m3fn' and hasattr(torch, 'float8_e4m3fn'):
|
|
storage_dtype = torch.float8_e4m3fn
|
|
elif shared.opts.layerwise_quantization_storage == 'float8_e5m2' and hasattr(torch, 'float8_e5m2'):
|
|
storage_dtype = torch.float8_e5m2
|
|
else:
|
|
storage_dtype = None
|
|
log.warning(f'Quantization: type=layerwise storage={shared.opts.layerwise_quantization_storage} not supported')
|
|
return
|
|
if not hasattr(quantization_config.QuantizationMethod, 'LAYERWISE'):
|
|
setattr(quantization_config.QuantizationMethod, 'LAYERWISE', 'layerwise') # noqa: B010
|
|
for module in sd_models.get_signature(sd_model).keys():
|
|
if not hasattr(sd_model, module):
|
|
continue
|
|
try:
|
|
cls = getattr(sd_model, module).__class__.__name__
|
|
m = getattr(sd_model, module)
|
|
if getattr(m, "quantization_method", None) in {'LayerWise', quantization_config.QuantizationMethod.LAYERWISE}: # pylint: disable=no-member
|
|
storage_dtype = getattr(m, "layerwise_storage_dtype", storage_dtype)
|
|
m.enable_layerwise_casting(compute_dtype=devices.dtype, storage_dtype=storage_dtype, non_blocking=shared.opts.diffusers_offload_nonblocking)
|
|
elif module.startswith('unet') and ('Model' in shared.opts.layerwise_quantization):
|
|
if hasattr(m, 'enable_layerwise_casting'):
|
|
m.enable_layerwise_casting(compute_dtype=devices.dtype, storage_dtype=storage_dtype, non_blocking=shared.opts.diffusers_offload_nonblocking)
|
|
m.layerwise_storage_dtype = storage_dtype
|
|
m.quantization_method = 'LayerWise'
|
|
log.quiet(quiet, f'Quantization: type=layerwise module={module} cls={cls} storage={storage_dtype} compute={devices.dtype} blocking={not shared.opts.diffusers_offload_nonblocking}')
|
|
elif module.startswith('transformer') and ('Model' in shared.opts.layerwise_quantization):
|
|
if hasattr(m, 'enable_layerwise_casting'):
|
|
m.enable_layerwise_casting(compute_dtype=devices.dtype, storage_dtype=storage_dtype, non_blocking=shared.opts.diffusers_offload_nonblocking)
|
|
m.layerwise_storage_dtype = storage_dtype
|
|
m.quantization_method = 'LayerWise'
|
|
log.quiet(quiet, f'Quantization: type=layerwise module={module} cls={cls} storage={storage_dtype} compute={devices.dtype} blocking={not shared.opts.diffusers_offload_nonblocking}')
|
|
elif module.startswith('text_encoder') and ('TE' in shared.opts.layerwise_quantization) and ('clip' not in cls.lower()):
|
|
if hasattr(m, 'enable_layerwise_casting'):
|
|
m.enable_layerwise_casting(compute_dtype=devices.dtype, storage_dtype=storage_dtype, non_blocking=shared.opts.diffusers_offload_nonblocking)
|
|
m.layerwise_storage_dtype = storage_dtype
|
|
m.quantization_method = quantization_config.QuantizationMethod.LAYERWISE # pylint: disable=no-member
|
|
log.quiet(quiet, f'Quantization: type=layerwise module={module} cls={cls} storage={storage_dtype} compute={devices.dtype} blocking={not shared.opts.diffusers_offload_nonblocking}')
|
|
except Exception as e:
|
|
if 'Hook with name' not in str(e):
|
|
log.error(f'Quantization: type=layerwise {e}')
|
|
|
|
|
|
def sdnq_quantize_model(model, op=None, sd_model=None, do_gc: bool = True, weights_dtype: str = None, quantized_matmul_dtype: str = None, modules_to_not_convert: list = None, modules_dtype_dict: dict = None):
|
|
global quant_last_model_name, quant_last_model_device # pylint: disable=global-statement
|
|
from modules import devices, shared, timer
|
|
from modules.sdnq import sdnq_post_load_quant
|
|
from modules.sdnq.common import use_torch_compile as sdnq_use_torch_compile
|
|
|
|
if shared.opts.sdnq_use_quantized_matmul and not sdnq_use_torch_compile:
|
|
logger.log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.')
|
|
shared.opts.sdnq_use_quantized_matmul = False
|
|
|
|
if weights_dtype is None:
|
|
if (op is not None) and ("text_encoder" in op or op in {"TE", "LLM"}) and (shared.opts.sdnq_quantize_weights_mode_te not in {"Same as model", "default"}):
|
|
weights_dtype = shared.opts.sdnq_quantize_weights_mode_te
|
|
else:
|
|
weights_dtype = shared.opts.sdnq_quantize_weights_mode
|
|
if weights_dtype is None or weights_dtype == 'none':
|
|
return model
|
|
|
|
if quantized_matmul_dtype is None:
|
|
if (op is not None) and ("text_encoder" in op or op in {"TE", "LLM"}) and (shared.opts.sdnq_quantize_matmul_mode_te not in {"Same as model", "default"}):
|
|
quantized_matmul_dtype = shared.opts.sdnq_quantize_matmul_mode_te
|
|
else:
|
|
quantized_matmul_dtype = shared.opts.sdnq_quantize_matmul_mode
|
|
if quantized_matmul_dtype == "auto":
|
|
quantized_matmul_dtype = None
|
|
|
|
quantization_device, return_device = get_sdnq_devices(mode="post")
|
|
|
|
if modules_to_not_convert is None:
|
|
modules_to_not_convert = []
|
|
if modules_dtype_dict is None:
|
|
modules_dtype_dict = {}
|
|
|
|
sdnq_modules_to_not_convert = [m.strip() for m in re.split(';|,| ', shared.opts.sdnq_modules_to_not_convert) if len(m.strip()) > 1]
|
|
if len(sdnq_modules_to_not_convert) > 0:
|
|
modules_to_not_convert.extend(sdnq_modules_to_not_convert)
|
|
|
|
try:
|
|
if len(shared.opts.sdnq_modules_dtype_dict) > 2:
|
|
sdnq_modules_dtype_dict = shared.opts.sdnq_modules_dtype_dict
|
|
if "{" not in sdnq_modules_dtype_dict:
|
|
sdnq_modules_dtype_dict = "{" + sdnq_modules_dtype_dict + "}"
|
|
sdnq_modules_dtype_dict = json.loads(bytes(sdnq_modules_dtype_dict, 'utf-8'))
|
|
for key, value in sdnq_modules_dtype_dict.items():
|
|
if isinstance(value, str):
|
|
value = [m.strip() for m in re.split(';|,| ', value) if len(m.strip()) > 1]
|
|
if key not in modules_dtype_dict.keys():
|
|
modules_dtype_dict[key] = value
|
|
else:
|
|
modules_dtype_dict[key].extend(value)
|
|
except Exception as e:
|
|
log.warning(f'Quantization: SDNQ failed to parse sdnq_modules_dtype_dict: {e}')
|
|
|
|
t0 = time.time()
|
|
|
|
model = sdnq_post_load_quant(
|
|
model,
|
|
weights_dtype=weights_dtype,
|
|
quantized_matmul_dtype=quantized_matmul_dtype,
|
|
torch_dtype=devices.dtype,
|
|
group_size=shared.opts.sdnq_quantize_weights_group_size,
|
|
svd_rank=shared.opts.sdnq_svd_rank,
|
|
svd_steps=shared.opts.sdnq_svd_steps,
|
|
dynamic_loss_threshold=shared.opts.sdnq_dynamic_loss_threshold,
|
|
use_svd=shared.opts.sdnq_use_svd,
|
|
quant_conv=shared.opts.sdnq_quantize_conv_layers,
|
|
use_quantized_matmul=shared.opts.sdnq_use_quantized_matmul,
|
|
use_quantized_matmul_conv=shared.opts.sdnq_use_quantized_matmul_conv,
|
|
use_dynamic_quantization=shared.opts.sdnq_use_dynamic_quantization,
|
|
dequantize_fp32=shared.opts.sdnq_dequantize_fp32,
|
|
non_blocking=shared.opts.diffusers_offload_nonblocking,
|
|
quantization_device=quantization_device,
|
|
return_device=return_device,
|
|
modules_to_not_convert=modules_to_not_convert,
|
|
modules_dtype_dict=modules_dtype_dict.copy(),
|
|
)
|
|
|
|
t1 = time.time()
|
|
timer.load.add('sdnq', t1 - t0)
|
|
|
|
if op is not None and shared.opts.sdnq_quantize_shuffle_weights:
|
|
if quant_last_model_name is not None:
|
|
if "." in quant_last_model_name:
|
|
last_model_names = quant_last_model_name.split(".")
|
|
getattr(getattr(sd_model, last_model_names[0]), last_model_names[1]).to(quant_last_model_device)
|
|
else:
|
|
getattr(sd_model, quant_last_model_name).to(quant_last_model_device)
|
|
if do_gc:
|
|
devices.torch_gc(force=True, reason='sdnq')
|
|
if shared.cmd_opts.medvram or shared.cmd_opts.lowvram or shared.opts.diffusers_offload_mode != "none":
|
|
quant_last_model_name = op
|
|
quant_last_model_device = model.device
|
|
else:
|
|
quant_last_model_name = None
|
|
quant_last_model_device = None
|
|
model.to(devices.device)
|
|
elif (shared.opts.diffusers_offload_mode != "none") and (not shared.opts.diffusers_to_gpu):
|
|
model = model.to(devices.cpu)
|
|
if do_gc:
|
|
devices.torch_gc(force=True, reason='sdnq')
|
|
|
|
if quantized_matmul_dtype is None:
|
|
quantized_matmul_dtype = "auto" # set for logging
|
|
log.debug(f'Quantization: module="{op if op is not None else model.__class__}" type=sdnq mode=post dtype={weights_dtype} matmul_dtype={quantized_matmul_dtype} matmul={shared.opts.sdnq_use_quantized_matmul} svd={shared.opts.sdnq_use_svd}dynamic={shared.opts.sdnq_use_dynamic_quantization}:group={shared.opts.sdnq_quantize_weights_group_size}:rank={shared.opts.sdnq_svd_rank}:steps={shared.opts.sdnq_svd_steps}:loss={shared.opts.sdnq_dynamic_loss_threshold} quant_conv={shared.opts.sdnq_quantize_conv_layers} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} fp32={shared.opts.sdnq_dequantize_fp32} gpu={shared.opts.sdnq_quantize_with_gpu} device={quantization_device} return={return_device} map={shared.opts.device_map} non_blocking={shared.opts.diffusers_offload_nonblocking} modules_skip={modules_to_not_convert} modules_dtype={modules_dtype_dict}')
|
|
return model
|
|
|
|
|
|
def sdnq_quantize_weights(sd_model):
|
|
try:
|
|
t0 = time.time()
|
|
from modules import shared, devices, sd_models
|
|
log.debug(f"Quantization: type=SDNQ modules={shared.opts.sdnq_quantize_weights} dtype={shared.opts.sdnq_quantize_weights_mode} dtype_te={shared.opts.sdnq_quantize_weights_mode_te} offload={shared.opts.diffusers_offload_mode} pre_forward={shared.opts.diffusers_offload_pre}")
|
|
global quant_last_model_name, quant_last_model_device # pylint: disable=global-statement
|
|
|
|
sd_model = sd_models.apply_function_to_model(sd_model, sdnq_quantize_model, shared.opts.sdnq_quantize_weights, op="sdnq")
|
|
if quant_last_model_name is not None:
|
|
if "." in quant_last_model_name:
|
|
last_model_names = quant_last_model_name.split(".")
|
|
getattr(getattr(sd_model, last_model_names[0]), last_model_names[1]).to(quant_last_model_device)
|
|
else:
|
|
getattr(sd_model, quant_last_model_name).to(quant_last_model_device)
|
|
devices.torch_gc(force=True, reason='sdnq')
|
|
quant_last_model_name = None
|
|
quant_last_model_device = None
|
|
|
|
t1 = time.time()
|
|
log.info(f"Quantization: type=SDNQ time={t1-t0:.2f}")
|
|
except Exception as e:
|
|
log.warning(f"Quantization: type=SDNQ {e}")
|
|
from modules import errors
|
|
errors.display(e, 'Quantization')
|
|
return sd_model
|
|
|
|
|
|
def optimum_quanto_model(model, op=None, sd_model=None, weights=None, activations=None):
|
|
from modules import devices, shared
|
|
quanto = load_quanto('Quantize model: type=Optimum Quanto')
|
|
global quant_last_model_name, quant_last_model_device # pylint: disable=global-statement
|
|
if model.__class__.__name__ in {"FluxTransformer2DModel", "ChromaTransformer2DModel"}: # LayerNorm is not supported
|
|
exclude_list = ["transformer_blocks.*.norm1.norm", "transformer_blocks.*.norm2", "transformer_blocks.*.norm1_context.norm", "transformer_blocks.*.norm2_context", "single_transformer_blocks.*.norm.norm", "norm_out.norm"]
|
|
if model.__class__.__name__ == "ChromaTransformer2DModel":
|
|
# we ignore the distilled guidance layer because it degrades quality too much
|
|
# see: https://github.com/huggingface/diffusers/pull/11698#issuecomment-2969717180 for more details
|
|
exclude_list.append("distilled_guidance_layer.*")
|
|
elif model.__class__.__name__ == "QwenImageTransformer2DModel":
|
|
exclude_list = ["transformer_blocks.0.img_mod.1.weight", "time_text_embed", "img_in", "txt_in", "proj_out", "norm_out", "pos_embed"]
|
|
else:
|
|
exclude_list = None
|
|
weights = getattr(quanto, weights) if weights is not None else getattr(quanto, shared.opts.optimum_quanto_weights_type)
|
|
if activations is not None:
|
|
activations = getattr(quanto, activations) if activations != 'none' else None
|
|
elif shared.opts.optimum_quanto_activations_type != 'none':
|
|
activations = getattr(quanto, shared.opts.optimum_quanto_activations_type)
|
|
else:
|
|
activations = None
|
|
model.eval()
|
|
backup_embeddings = None
|
|
if hasattr(model, "get_input_embeddings"):
|
|
backup_embeddings = copy.deepcopy(model.get_input_embeddings())
|
|
quanto.quantize(model, weights=weights, activations=activations, exclude=exclude_list)
|
|
quanto.freeze(model)
|
|
if hasattr(model, "set_input_embeddings") and backup_embeddings is not None:
|
|
model.set_input_embeddings(backup_embeddings)
|
|
if op is not None and shared.opts.optimum_quanto_shuffle_weights:
|
|
if quant_last_model_name is not None:
|
|
if "." in quant_last_model_name:
|
|
last_model_names = quant_last_model_name.split(".")
|
|
getattr(getattr(sd_model, last_model_names[0]), last_model_names[1]).to(quant_last_model_device)
|
|
else:
|
|
getattr(sd_model, quant_last_model_name).to(quant_last_model_device)
|
|
devices.torch_gc(force=True, reason='quanto')
|
|
if shared.cmd_opts.medvram or shared.cmd_opts.lowvram or shared.opts.diffusers_offload_mode != "none":
|
|
quant_last_model_name = op
|
|
quant_last_model_device = model.device
|
|
else:
|
|
quant_last_model_name = None
|
|
quant_last_model_device = None
|
|
model.to(devices.device)
|
|
devices.torch_gc(force=True, reason='quanto')
|
|
return model
|
|
|
|
|
|
def optimum_quanto_weights(sd_model):
|
|
try:
|
|
t0 = time.time()
|
|
from modules import shared, devices, sd_models
|
|
if shared.opts.diffusers_offload_mode in {"balanced", "sequential"}:
|
|
log.warning(f"Quantization: type=Optimum.quanto offload={shared.opts.diffusers_offload_mode} not compatible")
|
|
return sd_model
|
|
log.info(f"Quantization: type=Optimum.quanto: modules={shared.opts.optimum_quanto_weights}")
|
|
global quant_last_model_name, quant_last_model_device # pylint: disable=global-statement
|
|
quanto = load_quanto()
|
|
|
|
sd_model = sd_models.apply_function_to_model(sd_model, optimum_quanto_model, shared.opts.optimum_quanto_weights, op="optimum-quanto")
|
|
if quant_last_model_name is not None:
|
|
if "." in quant_last_model_name:
|
|
last_model_names = quant_last_model_name.split(".")
|
|
getattr(getattr(sd_model, last_model_names[0]), last_model_names[1]).to(quant_last_model_device)
|
|
else:
|
|
getattr(sd_model, quant_last_model_name).to(quant_last_model_device)
|
|
devices.torch_gc(force=True, reason='quanto')
|
|
quant_last_model_name = None
|
|
quant_last_model_device = None
|
|
|
|
if shared.opts.optimum_quanto_activations_type != 'none':
|
|
activations = getattr(quanto, shared.opts.optimum_quanto_activations_type)
|
|
else:
|
|
activations = None
|
|
|
|
if activations is not None:
|
|
def optimum_quanto_freeze(model, op=None, sd_model=None): # pylint: disable=unused-argument
|
|
quanto.freeze(model)
|
|
return model
|
|
if shared.opts.diffusers_offload_mode == "model":
|
|
sd_model.enable_model_cpu_offload(device=devices.device)
|
|
if hasattr(sd_model, "encode_prompt"):
|
|
original_encode_prompt = sd_model.encode_prompt
|
|
def encode_prompt(*args, **kwargs):
|
|
embeds = original_encode_prompt(*args, **kwargs)
|
|
sd_model.maybe_free_model_hooks() # Diffusers keeps the TE on VRAM
|
|
return embeds
|
|
sd_model.encode_prompt = encode_prompt
|
|
else:
|
|
sd_models.move_model(sd_model, devices.device)
|
|
with quanto.Calibration(momentum=0.9):
|
|
sd_model(prompt="dummy prompt", num_inference_steps=10)
|
|
sd_model = sd_models.apply_function_to_model(sd_model, optimum_quanto_freeze, shared.opts.optimum_quanto_weights, op="optimum-quanto-freeze")
|
|
if shared.opts.diffusers_offload_mode == "model":
|
|
sd_models.disable_offload(sd_model)
|
|
sd_models.move_model(sd_model, devices.cpu)
|
|
if hasattr(sd_model, "encode_prompt"):
|
|
sd_model.encode_prompt = original_encode_prompt
|
|
devices.torch_gc(force=True, reason='quanto')
|
|
|
|
t1 = time.time()
|
|
log.info(f"Quantization: type=Optimum.quanto time={t1-t0:.2f}")
|
|
except Exception as e:
|
|
log.warning(f"Quantization: type=Optimum.quanto {e}")
|
|
return sd_model
|
|
|
|
|
|
def torchao_quantization(sd_model):
|
|
from modules import shared, devices, sd_models
|
|
torchao = load_torchao()
|
|
q = torchao.quantization
|
|
|
|
fn = getattr(q, shared.opts.torchao_quantization_type, None)
|
|
if fn is None:
|
|
log.error(f"Quantization: type=TorchAO type={shared.opts.torchao_quantization_type} not supported")
|
|
return sd_model
|
|
def torchao_model(model, op=None, sd_model=None): # pylint: disable=unused-argument
|
|
q.quantize_(model, fn(), device=devices.device)
|
|
return model
|
|
|
|
log.info(f"Quantization: type=TorchAO pipe={sd_model.__class__.__name__} quant={shared.opts.torchao_quantization_type} fn={fn} targets={shared.opts.torchao_quantization}")
|
|
try:
|
|
t0 = time.time()
|
|
sd_models.apply_function_to_model(sd_model, torchao_model, shared.opts.torchao_quantization, op="torchao")
|
|
t1 = time.time()
|
|
log.info(f"Quantization: type=TorchAO time={t1-t0:.2f}")
|
|
except Exception as e:
|
|
log.error(f"Quantization: type=TorchAO {e}")
|
|
setup_logging() # torchao uses dynamo which messes with logging so reset is needed
|
|
return sd_model
|
|
|
|
|
|
def get_dit_args(load_config:dict=None, module:str=None, device_map:bool=False, allow_quant:bool=True, modules_to_not_convert: list = None, modules_dtype_dict: dict = None):
|
|
from modules import shared, devices
|
|
config = {} if load_config is None else load_config.copy()
|
|
if 'torch_dtype' not in config:
|
|
config['torch_dtype'] = devices.dtype
|
|
if 'low_cpu_mem_usage' in config:
|
|
del config['low_cpu_mem_usage']
|
|
if 'load_connected_pipeline' in config:
|
|
del config['load_connected_pipeline']
|
|
if 'safety_checker' in config:
|
|
del config['safety_checker']
|
|
if 'requires_safety_checker' in config:
|
|
del config['requires_safety_checker']
|
|
# if 'variant' in config:
|
|
# del config['variant']
|
|
if device_map:
|
|
if devices.backend == 'ipex' and os.environ.get('UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS', '0') != '1' and module in {'TE', 'LLM'}:
|
|
config['device_map'] = 'cpu' # alchemist gpus hits the 4GB allocation limit with transformers, UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS emulates above 4GB allocations
|
|
elif shared.opts.device_map == 'cpu':
|
|
config['device_map'] = 'cpu'
|
|
elif shared.opts.device_map == 'gpu':
|
|
config['device_map'] = devices.device
|
|
if allow_quant:
|
|
quant_args = create_config(module=module, modules_to_not_convert=modules_to_not_convert, modules_dtype_dict=modules_dtype_dict)
|
|
else:
|
|
quant_args = {}
|
|
return config, quant_args
|
|
|
|
|
|
def do_post_load_quant(sd_model, allow=True):
|
|
from modules import shared
|
|
if dont_quant():
|
|
return sd_model
|
|
if shared.opts.sdnq_quantize_weights and (shared.opts.sdnq_quantize_mode == 'post' or (allow and shared.opts.sdnq_quantize_mode == 'auto')):
|
|
logger.log.debug('Load model: post_quant=sdnq')
|
|
sd_model = sdnq_quantize_weights(sd_model)
|
|
if len(shared.opts.optimum_quanto_weights) > 0:
|
|
logger.log.debug('Load model: post_quant=quanto')
|
|
sd_model = optimum_quanto_weights(sd_model)
|
|
if shared.opts.torchao_quantization and (shared.opts.torchao_quantization_mode == 'post' or (allow and shared.opts.torchao_quantization_mode == 'auto')):
|
|
logger.log.debug('Load model: post_quant=torchao')
|
|
sd_model = torchao_quantization(sd_model)
|
|
if shared.opts.layerwise_quantization:
|
|
logger.log.debug('Load model: post_quant=layerwise')
|
|
apply_layerwise(sd_model)
|
|
return sd_model
|