automatic/modules/sd_models_compile.py

256 lines
13 KiB
Python

import time
import logging
import torch
from modules import shared, devices
from installer import setup_logging
#Used by OpenVINO, can be used with TensorRT or Olive
class CompiledModelState:
def __init__(self):
self.is_compiled = False
self.model_hash_str = ""
self.first_pass = True
self.first_pass_refiner = True
self.first_pass_vae = True
self.height = 512
self.width = 512
self.batch_size = 1
self.partition_id = 0
self.cn_model = []
self.lora_model = []
self.compiled_cache = {}
self.partitioned_modules = {}
def ipex_optimize(sd_model):
try:
t0 = time.time()
def ipex_optimize_model(model):
import intel_extension_for_pytorch as ipex # pylint: disable=import-error, unused-import
model.eval()
model.training = False
model = ipex.optimize(model, dtype=devices.dtype, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
return model
if "Model" in shared.opts.ipex_optimize:
if hasattr(sd_model, 'unet'):
sd_model.unet = ipex_optimize_model(sd_model.unet)
elif hasattr(sd_model, 'transformer'):
sd_model.transformer = ipex_optimize_model(sd_model.transformer)
else:
shared.log.warning('IPEX Optimize enabled but model has no Unet or Transformer')
if "VAE" in shared.opts.ipex_optimize:
if hasattr(sd_model, 'vae'):
sd_model.vae = ipex_optimize_model(sd_model.vae)
elif hasattr(sd_model, 'movq'):
sd_model.movq = ipex_optimize_model(sd_model.movq)
else:
shared.log.warning('Compress VAE Weights enabled but model has no VAE')
if "Text Encoder" in shared.opts.ipex_optimize:
if hasattr(sd_model, 'text_encoder'):
sd_model.text_encoder = ipex_optimize_model(sd_model.text_encoder)
if hasattr(sd_model, 'text_encoder_2'):
sd_model.text_encoder_2 = ipex_optimize_model(sd_model.text_encoder_2)
else:
shared.log.warning('IPEX Optimize Text Encoder Weights enabled but model has no Text Encoder')
t1 = time.time()
shared.log.info(f"IPEX Optimize: time={t1-t0:.2f}")
return sd_model
except Exception as e:
shared.log.warning(f"IPEX Optimize: error: {e}")
def nncf_compress_weights(sd_model):
try:
t0 = time.time()
import nncf
shared.compiled_model_state = CompiledModelState()
shared.compiled_model_state.is_compiled = True
if "Model" in shared.opts.nncf_compress_weights:
if hasattr(sd_model, 'unet') and hasattr(sd_model.unet, 'config'):
sd_model.unet = nncf.compress_weights(sd_model.unet)
elif hasattr(sd_model, 'transformer') and hasattr(sd_model.transformer, 'config'):
sd_model.transformer = nncf.compress_weights(sd_model.transformer)
else:
shared.log.warning('Compress Weights enabled but model has no Unet or Transformer')
if "VAE" in shared.opts.nncf_compress_weights:
if hasattr(sd_model, 'vae') and hasattr(sd_model.vae, 'decode'):
sd_model.vae = nncf.compress_weights(sd_model.vae)
elif hasattr(sd_model, 'movq') and hasattr(sd_model.movq, 'decode'):
sd_model.movq = nncf.compress_weights(sd_model.movq)
else:
shared.log.warning('Compress VAE Weights enabled but model has no VAE')
if "Text Encoder" in shared.opts.nncf_compress_weights:
if hasattr(sd_model, 'text_encoder') and hasattr(sd_model.text_encoder, 'config'):
sd_model.text_encoder = nncf.compress_weights(sd_model.text_encoder)
if hasattr(sd_model, 'text_encoder_2') and hasattr(sd_model.text_encoder_2, 'config'):
sd_model.text_encoder_2 = nncf.compress_weights(sd_model.text_encoder_2)
else:
shared.log.warning('Compress VAE Text Encoder Weights enabled but model has no Text Encoder')
t1 = time.time()
shared.log.info(f"Compress Weights: time={t1-t0:.2f}")
return sd_model
except Exception as e:
shared.log.warning(f"Compress Weights: error: {e}")
def optimize_openvino(sd_model):
try:
from modules.intel.openvino import openvino_fx # pylint: disable=unused-import
torch._dynamo.eval_frame.check_if_dynamo_supported = lambda: True # pylint: disable=protected-access
if shared.compiled_model_state is not None:
shared.compiled_model_state.compiled_cache.clear()
shared.compiled_model_state.partitioned_modules.clear()
shared.compiled_model_state = CompiledModelState()
shared.compiled_model_state.is_compiled = True
shared.compiled_model_state.first_pass = True if not shared.opts.cuda_compile_precompile else False
shared.compiled_model_state.first_pass_vae = True if not shared.opts.cuda_compile_precompile else False
shared.compiled_model_state.first_pass_refiner = True if not shared.opts.cuda_compile_precompile else False
sd_model.has_accelerate = True
except Exception as e:
shared.log.warning(f"Model compile: task=OpenVINO: {e}")
return sd_model
def compile_stablefast(sd_model):
try:
import sfast.compilers.stable_diffusion_pipeline_compiler as sf
except Exception as e:
shared.log.warning(f'Model compile using stable-fast: {e}')
return sd_model
config = sf.CompilationConfig.Default()
try:
import xformers # pylint: disable=unused-import
config.enable_xformers = True
except Exception:
pass
try:
import triton # pylint: disable=unused-import
config.enable_triton = True
except Exception:
pass
import warnings
warnings.filterwarnings("ignore", category=torch.jit.TracerWarning)
config.enable_cuda_graph = shared.opts.cuda_compile_fullgraph
config.enable_jit_freeze = shared.opts.diffusers_eval
config.memory_format = torch.channels_last if shared.opts.opt_channelslast else torch.contiguous_format
# config.trace_scheduler = False
# config.enable_cnn_optimization
# config.prefer_lowp_gemm
try:
t0 = time.time()
sd_model = sf.compile(sd_model, config)
setup_logging() # compile messes with logging so reset is needed
if shared.opts.cuda_compile_precompile:
sd_model("dummy prompt")
t1 = time.time()
shared.log.info(f"Model compile: task=Stable-fast config={config.__dict__} time={t1-t0:.2f}")
except Exception as e:
shared.log.info(f"Model compile: task=Stable-fast error: {e}")
return sd_model
def compile_torch(sd_model):
try:
import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
torch._dynamo.reset() # pylint: disable=protected-access
shared.log.debug(f"Model compile available backends: {torch._dynamo.list_backends()}") # pylint: disable=protected-access
if shared.opts.cuda_compile_backend == "openvino_fx":
sd_model = optimize_openvino(sd_model)
elif shared.opts.cuda_compile_backend == "olive-ai":
if shared.compiled_model_state is None:
shared.compiled_model_state = CompiledModelState()
return sd_model
log_level = logging.WARNING if shared.opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access
if hasattr(torch, '_logging'):
torch._logging.set_logs(dynamo=log_level, aot=log_level, inductor=log_level) # pylint: disable=protected-access
torch._dynamo.config.verbose = shared.opts.cuda_compile_verbose # pylint: disable=protected-access
torch._dynamo.config.suppress_errors = shared.opts.cuda_compile_errors # pylint: disable=protected-access
try:
torch._inductor.config.conv_1x1_as_mm = True # pylint: disable=protected-access
torch._inductor.config.coordinate_descent_tuning = True # pylint: disable=protected-access
torch._inductor.config.epilogue_fusion = False # pylint: disable=protected-access
torch._inductor.config.coordinate_descent_check_all_directions = True # pylint: disable=protected-access
torch._inductor.config.use_mixed_mm = True # pylint: disable=protected-access
# torch._inductor.config.force_fuse_int_mm_with_mul = True # pylint: disable=protected-access
except Exception as e:
shared.log.error(f"Torch inductor config error: {e}")
t0 = time.time()
if "Model" in shared.opts.cuda_compile:
if hasattr(sd_model, 'unet') and hasattr(sd_model.unet, 'config'):
sd_model.unet = torch.compile(sd_model.unet, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph)
elif hasattr(sd_model, 'transformer') and hasattr(sd_model.transformer, 'config'):
sd_model.transformer = torch.compile(sd_model.transformer, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph)
else:
shared.log.warning('Model compile enabled but model has no Unet or Transformer')
if "VAE" in shared.opts.cuda_compile:
if hasattr(sd_model, 'vae') and hasattr(sd_model.vae, 'decode'):
sd_model.vae.decode = torch.compile(sd_model.vae.decode, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph)
elif hasattr(sd_model, 'movq') and hasattr(sd_model.movq, 'decode'):
sd_model.movq.decode = torch.compile(sd_model.movq.decode, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph)
else:
shared.log.warning('Model compile enabled but model has no VAE')
if "Text Encoder" in shared.opts.cuda_compile:
if hasattr(sd_model, 'text_encoder') and hasattr(sd_model.text_encoder, 'config'):
sd_model.text_encoder = torch.compile(sd_model.text_encoder, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph)
if hasattr(sd_model, 'text_encoder_2') and hasattr(sd_model.text_encoder_2, 'config'):
sd_model.text_encoder_2 = torch.compile(sd_model.text_encoder_2, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph)
else:
shared.log.warning('Text Encoder compile enabled but model has no Text Encoder')
setup_logging() # compile messes with logging so reset is needed
if shared.opts.cuda_compile_precompile:
sd_model("dummy prompt")
t1 = time.time()
shared.log.info(f"Model compile: time={t1-t0:.2f}")
except Exception as e:
shared.log.warning(f"Model compile error: {e}")
return sd_model
def compile_diffusers(sd_model):
if shared.opts.ipex_optimize:
sd_model = ipex_optimize(sd_model)
if shared.opts.nncf_compress_weights and not (shared.opts.cuda_compile and shared.opts.cuda_compile_backend == "openvino_fx"):
sd_model = nncf_compress_weights(sd_model)
if not shared.opts.cuda_compile:
return sd_model
if shared.opts.cuda_compile_backend == 'none':
shared.log.warning('Model compile enabled but no backend specified')
return sd_model
shared.log.info(f"Model compile: pipeline={sd_model.__class__.__name__} mode={shared.opts.cuda_compile_mode} backend={shared.opts.cuda_compile_backend} fullgraph={shared.opts.cuda_compile_fullgraph} compile={shared.opts.cuda_compile}")
if shared.opts.cuda_compile_backend == 'stable-fast':
sd_model = compile_stablefast(sd_model)
else:
sd_model = compile_torch(sd_model)
return sd_model
def dynamic_quantization(sd_model):
try:
from torchao.quantization import quant_api
except Exception as e:
shared.log.error(f"Model dynamic quantization not supported: {e}")
return sd_model
def dynamic_quant_filter_fn(mod, *args): # pylint: disable=unused-argument
return (isinstance(mod, torch.nn.Linear) and mod.in_features > 16 and (mod.in_features, mod.out_features)
not in [(1280, 640), (1920, 1280), (1920, 640), (2048, 1280), (2048, 2560), (2560, 1280), (256, 128), (2816, 1280), (320, 640), (512, 1536), (512, 256), (512, 512), (640, 1280), (640, 1920), (640, 320), (640, 5120), (640, 640), (960, 320), (960, 640)])
def conv_filter_fn(mod, *args): # pylint: disable=unused-argument
return (isinstance(mod, torch.nn.Conv2d) and mod.kernel_size == (1, 1) and 128 in [mod.in_channels, mod.out_channels])
shared.log.info(f"Model dynamic quantization: pipeline={sd_model.__class__.__name__}")
try:
quant_api.swap_conv2d_1x1_to_linear(sd_model.unet, conv_filter_fn)
quant_api.swap_conv2d_1x1_to_linear(sd_model.vae, conv_filter_fn)
quant_api.apply_dynamic_quant(sd_model.unet, dynamic_quant_filter_fn)
quant_api.apply_dynamic_quant(sd_model.vae, dynamic_quant_filter_fn)
except Exception as e:
shared.log.error(f"Model dynamic quantization error: {e}")
return sd_model