Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-18 16:54:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #2507 from vladmandic/master

Browse Source 
update dev
This commit is contained in:
Vladimir Mandic
2023-11-13 09:28:42 -05:00
committed by GitHub
parent ebf029ad91 f1862579fd
commit 2c36b0051a
6 changed files with 41 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CHANGELOG.md
+1
View File
@@ -17,6 +17,7 @@
- Updated logic for calculating **steps** when using base/hires/refiner workflows
- Safe model offloading for non-standard models
- Fix **DPM SDE** scheduler
- Better support for SD 1.5 **inpainting** models
- Update to `diffusers==0.23.0`
- **Extra networks**
- Use multi-threading for 5x load speedup
cli/run-benchmark.py
+3 -2
View File
@@ -16,7 +16,7 @@ options = Map({
'restore_faces': False,
'prompt': 'photo of two dice on a table',
'negative_prompt': 'foggy, blurry',
'steps': 20,
'steps': 50,
'batch_size': 1,
'n_iter': 1,
'seed': -1,
@@ -28,7 +28,8 @@ options = Map({
# batch = [1, 1, 2, 4, 8, 12, 16, 24, 32, 48, 64, 96, 128]
batch = [1, 1, 2, 4, 8, 12, 16]
# batch = [1, 1, 2, 4, 8, 12, 16]
batch = [4, 4]
oom = 0
modules/processing_diffusers.py
+6 -3
View File
@@ -223,7 +223,6 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
return task_args
def set_pipeline_args(model, prompts: list, negative_prompts: list, prompts_2: typing.Optional[list]=None, negative_prompts_2: typing.Optional[list]=None, desc:str='', **kwargs):
if hasattr(model, "set_progress_bar_config"):
model.set_progress_bar_config(bar_format='Progress {rate_fmt}{postfix} {bar} {percentage:3.0f}% {n_fmt}/{total_fmt} {elapsed} {remaining} ' + '\x1b[38;5;71m' + desc, ncols=80, colour='#327fba')
args = {}
@@ -408,8 +407,12 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
return max(2, int(steps))
# pipeline type is set earlier in processing, but check for sanity
if sd_models.get_diffusers_task(shared.sd_model) != sd_models.DiffusersTaskType.TEXT_2_IMAGE and len(getattr(p, 'init_images' ,[])) == 0: # reset pipeline
shared.sd_model = sd_models.set_diffuser_pipe(shared.sd_model, sd_models.DiffusersTaskType.TEXT_2_IMAGE)
if sd_models.get_diffusers_task(shared.sd_model) != sd_models.DiffusersTaskType.TEXT_2_IMAGE and len(getattr(p, 'init_images' ,[])) == 0:
shared.sd_model = sd_models.set_diffuser_pipe(shared.sd_model, sd_models.DiffusersTaskType.TEXT_2_IMAGE) # reset pipeline
if hasattr(shared.sd_model, 'unet') and hasattr(shared.sd_model.unet, 'config') and hasattr(shared.sd_model.unet.config, 'in_channels') and shared.sd_model.unet.config.in_channels == 9:
shared.sd_model = sd_models.set_diffuser_pipe(shared.sd_model, sd_models.DiffusersTaskType.INPAINTING) # force pipeline
if len(getattr(p, 'init_images' ,[])) == 0:
p.init_images = [TF.to_pil_image(torch.rand((3, p.height, p.width)))]
base_args = set_pipeline_args(
model=shared.sd_model,
prompts=prompts,
modules/sd_models.py
+27 -12
View File
@@ -690,7 +690,7 @@ def compile_diffusers(sd_model):
shared.log.warning(f"IPEX Optimize not supported: {err}")
try:
if shared.opts.cuda_compile and shared.opts.cuda_compile_backend != 'none':
if (shared.opts.cuda_compile or shared.opts.cuda_compile_vae or shared.opts.cuda_compile_upscaler) and shared.opts.cuda_compile_backend != 'none':
shared.log.info(f"Compiling pipeline={sd_model.__class__.__name__} shape={8 * sd_model.unet.config.sample_size} mode={shared.opts.cuda_compile_backend}")
import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
if shared.opts.cuda_compile_backend == "openvino_fx":
@@ -706,11 +706,15 @@ def compile_diffusers(sd_model):
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
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) # pylint: disable=attribute-defined-outside-init
if hasattr(sd_model, 'vae'):
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) # pylint: disable=attribute-defined-outside-init
if hasattr(sd_model, 'movq'):
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) # pylint: disable=attribute-defined-outside-init
if shared.opts.cuda_compile:
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) # pylint: disable=attribute-defined-outside-init
if shared.opts.cuda_compile_vae:
if hasattr(sd_model, 'vae'):
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) # pylint: disable=attribute-defined-outside-init
if hasattr(sd_model, 'movq'):
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) # pylint: disable=attribute-defined-outside-init
from installer import setup_logging
setup_logging()
if shared.opts.cuda_compile_precompile:
sd_model("dummy prompt")
shared.log.info("Complilation done.")
@@ -899,15 +903,26 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
if model_type.startswith('Stable Diffusion'):
diffusers_load_config['force_zeros_for_empty_prompt '] = shared.opts.diffusers_force_zeros
diffusers_load_config['requires_aesthetics_score'] = shared.opts.diffusers_aesthetics_score
diffusers_load_config['config_files'] = {
'v1': 'configs/v1-inference.yaml',
'v2': 'configs/v2-inference-768-v.yaml',
'xl': 'configs/sd_xl_base.yaml',
'xl_refiner': 'configs/sd_xl_refiner.yaml',
}
if 'inpainting' in checkpoint_info.path.lower():
diffusers_load_config['config_files'] = {
'v1': 'configs/v1-inpainting-inference.yaml',
'v2': 'configs/v2-inference-768-v.yaml',
'xl': 'configs/sd_xl_base.yaml',
'xl_refiner': 'configs/sd_xl_refiner.yaml',
}
else:
diffusers_load_config['config_files'] = {
'v1': 'configs/v1-inference.yaml',
'v2': 'configs/v2-inference-768-v.yaml',
'xl': 'configs/sd_xl_base.yaml',
'xl_refiner': 'configs/sd_xl_refiner.yaml',
}
if hasattr(pipeline, 'from_single_file'):
diffusers_load_config['use_safetensors'] = True
sd_model = pipeline.from_single_file(checkpoint_info.path, **diffusers_load_config)
if sd_model is not None and hasattr(sd_model, 'unet') and hasattr(sd_model.unet, 'config') and 'inpainting' in checkpoint_info.path.lower():
shared.log.debug('Model patch: type=inpaint')
sd_model.unet.config.in_channels = 9
elif hasattr(pipeline, 'from_ckpt'):
sd_model = pipeline.from_ckpt(checkpoint_info.path, **diffusers_load_config)
else:
modules/shared.py
+3 -2
View File
@@ -285,8 +285,9 @@ options_templates.update(options_section(('cuda', "Compute Settings"), {
"torch_gc_threshold": OptionInfo(90, "VRAM usage threshold before running Torch GC to clear up VRAM", gr.Slider, {"minimum": 0, "maximum": 100, "step": 1}),
"cuda_compile_sep": OptionInfo("<h2>Model Compile</h2>", "", gr.HTML),
"cuda_compile": OptionInfo(True if cmd_opts.use_openvino else False, "Enable model compile"),
"cuda_compile_upscaler": OptionInfo(True if cmd_opts.use_openvino else False, "Enable upscaler compile"),
"cuda_compile": OptionInfo(True if cmd_opts.use_openvino else False, "Compile UNet"),
"cuda_compile_vae": OptionInfo(True if cmd_opts.use_openvino else False, "Compile VAE"),
"cuda_compile_upscaler": OptionInfo(True if cmd_opts.use_openvino else False, "Compile upscaler"),
"cuda_compile_backend": OptionInfo("openvino_fx" if cmd_opts.use_openvino else "none", "Model compile backend", gr.Radio, {"choices": ['none', 'inductor', 'cudagraphs', 'aot_ts_nvfuser', 'hidet', 'ipex', 'openvino_fx']}),
"cuda_compile_mode": OptionInfo("default", "Model compile mode", gr.Radio, {"choices": ['default', 'reduce-overhead', 'max-autotune']}),
"cuda_compile_fullgraph": OptionInfo(False, "Model compile fullgraph"),
scripts/xyz_grid.py
+1 -1
View File
@@ -254,7 +254,7 @@ axis_options = [
AxisOption("[Second pass] hires steps", int, apply_field("hr_second_pass_steps")),
AxisOption("[Second pass] CFG scale", float, apply_field("image_cfg_scale")),
AxisOption("[Second pass] guidance rescale", float, apply_field("diffusers_guidance_rescale")),
AxisOption("[Refiner] model", str, apply_refiner, fmt=format_value, cost=1.0, choices=lambda: sorted(sd_models.checkpoints_list)),
AxisOption("[Refiner] model", str, apply_refiner, fmt=format_value, cost=1.0, choices=lambda: ['None'] + sorted(sd_models.checkpoints_list)),
AxisOption("[Refiner] refiner start", float, apply_field("refiner_start")),
AxisOption("[Refiner] refiner steps", float, apply_field("refiner_steps")),
AxisOption("[TOME] Token merging ratio (txt2img)", float, apply_override('token_merging_ratio')),