diff options
Diffstat (limited to 'modules')
| -rw-r--r-- | modules/cache.py | 5 | ||||
| -rw-r--r-- | modules/cmd_args.py | 1 | ||||
| -rw-r--r-- | modules/errors.py | 2 | ||||
| -rw-r--r-- | modules/interrogate.py | 5 | ||||
| -rw-r--r-- | modules/lowvram.py | 18 | ||||
| -rw-r--r--[-rwxr-xr-x] | modules/processing.py | 22 | ||||
| -rw-r--r-- | modules/prompt_parser.py | 41 | ||||
| -rw-r--r-- | modules/realesrgan_model.py | 1 | ||||
| -rw-r--r-- | modules/sd_models.py | 19 | ||||
| -rw-r--r-- | modules/sd_models_types.py | 31 | ||||
| -rw-r--r-- | modules/sd_unet.py | 2 | ||||
| -rw-r--r-- | modules/sd_vae.py | 4 | ||||
| -rw-r--r-- | modules/shared.py | 7 | ||||
| -rw-r--r-- | modules/shared_options.py | 1 |
14 files changed, 120 insertions, 39 deletions
diff --git a/modules/cache.py b/modules/cache.py index a7cd3aeb..ff26a213 100644 --- a/modules/cache.py +++ b/modules/cache.py @@ -30,9 +30,12 @@ def dump_cache(): time.sleep(1)
with cache_lock:
- with open(cache_filename, "w", encoding="utf8") as file:
+ cache_filename_tmp = cache_filename + "-"
+ with open(cache_filename_tmp, "w", encoding="utf8") as file:
json.dump(cache_data, file, indent=4)
+ os.replace(cache_filename_tmp, cache_filename)
+
dump_cache_after = None
dump_cache_thread = None
diff --git a/modules/cmd_args.py b/modules/cmd_args.py index 9f8e5b30..f0f361bd 100644 --- a/modules/cmd_args.py +++ b/modules/cmd_args.py @@ -35,6 +35,7 @@ parser.add_argument("--hypernetwork-dir", type=str, default=os.path.join(models_ parser.add_argument("--localizations-dir", type=str, default=os.path.join(script_path, 'localizations'), help="localizations directory")
parser.add_argument("--allow-code", action='store_true', help="allow custom script execution from webui")
parser.add_argument("--medvram", action='store_true', help="enable stable diffusion model optimizations for sacrificing a little speed for low VRM usage")
+parser.add_argument("--medvram-sdxl", action='store_true', help="enable --medvram optimization just for SDXL models")
parser.add_argument("--lowvram", action='store_true', help="enable stable diffusion model optimizations for sacrificing a lot of speed for very low VRM usage")
parser.add_argument("--lowram", action='store_true', help="load stable diffusion checkpoint weights to VRAM instead of RAM")
parser.add_argument("--always-batch-cond-uncond", action='store_true', help="does not do anything")
diff --git a/modules/errors.py b/modules/errors.py index 192cd8ff..a56fd30c 100644 --- a/modules/errors.py +++ b/modules/errors.py @@ -95,7 +95,7 @@ def check_versions(): expected_torch_version = "2.0.0"
expected_xformers_version = "0.0.20"
- expected_gradio_version = "3.39.0"
+ expected_gradio_version = "3.41.0"
if version.parse(torch.__version__) < version.parse(expected_torch_version):
print_error_explanation(f"""
diff --git a/modules/interrogate.py b/modules/interrogate.py index a3ae1dd5..3045560d 100644 --- a/modules/interrogate.py +++ b/modules/interrogate.py @@ -186,9 +186,8 @@ class InterrogateModels: res = ""
shared.state.begin(job="interrogate")
try:
- if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
- lowvram.send_everything_to_cpu()
- devices.torch_gc()
+ lowvram.send_everything_to_cpu()
+ devices.torch_gc()
self.load()
diff --git a/modules/lowvram.py b/modules/lowvram.py index 96f52b7b..45701046 100644 --- a/modules/lowvram.py +++ b/modules/lowvram.py @@ -1,5 +1,5 @@ import torch
-from modules import devices
+from modules import devices, shared
module_in_gpu = None
cpu = torch.device("cpu")
@@ -14,6 +14,20 @@ def send_everything_to_cpu(): module_in_gpu = None
+def is_needed(sd_model):
+ return shared.cmd_opts.lowvram or shared.cmd_opts.medvram or shared.cmd_opts.medvram_sdxl and hasattr(sd_model, 'conditioner')
+
+
+def apply(sd_model):
+ enable = is_needed(sd_model)
+ shared.parallel_processing_allowed = not enable
+
+ if enable:
+ setup_for_low_vram(sd_model, not shared.cmd_opts.lowvram)
+ else:
+ sd_model.lowvram = False
+
+
def setup_for_low_vram(sd_model, use_medvram):
if getattr(sd_model, 'lowvram', False):
return
@@ -130,4 +144,4 @@ def setup_for_low_vram(sd_model, use_medvram): def is_enabled(sd_model):
- return getattr(sd_model, 'lowvram', False)
+ return sd_model.lowvram
diff --git a/modules/processing.py b/modules/processing.py index e60cc92b..7dc931ba 100755..100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -407,12 +407,14 @@ class StableDiffusionProcessing: self.main_prompt = self.all_prompts[0]
self.main_negative_prompt = self.all_negative_prompts[0]
- def cached_params(self, required_prompts, steps, extra_network_data):
+ def cached_params(self, required_prompts, steps, extra_network_data, hires_steps=None, use_old_scheduling=False):
"""Returns parameters that invalidate the cond cache if changed"""
return (
required_prompts,
steps,
+ hires_steps,
+ use_old_scheduling,
opts.CLIP_stop_at_last_layers,
shared.sd_model.sd_checkpoint_info,
extra_network_data,
@@ -422,7 +424,7 @@ class StableDiffusionProcessing: self.height,
)
- def get_conds_with_caching(self, function, required_prompts, steps, caches, extra_network_data):
+ def get_conds_with_caching(self, function, required_prompts, steps, caches, extra_network_data, hires_steps=None):
"""
Returns the result of calling function(shared.sd_model, required_prompts, steps)
using a cache to store the result if the same arguments have been used before.
@@ -435,7 +437,13 @@ class StableDiffusionProcessing: caches is a list with items described above.
"""
- cached_params = self.cached_params(required_prompts, steps, extra_network_data)
+ if shared.opts.use_old_scheduling:
+ old_schedules = prompt_parser.get_learned_conditioning_prompt_schedules(required_prompts, steps, hires_steps, False)
+ new_schedules = prompt_parser.get_learned_conditioning_prompt_schedules(required_prompts, steps, hires_steps, True)
+ if old_schedules != new_schedules:
+ self.extra_generation_params["Old prompt editing timelines"] = True
+
+ cached_params = self.cached_params(required_prompts, steps, extra_network_data, hires_steps, shared.opts.use_old_scheduling)
for cache in caches:
if cache[0] is not None and cached_params == cache[0]:
@@ -444,7 +452,7 @@ class StableDiffusionProcessing: cache = caches[0]
with devices.autocast():
- cache[1] = function(shared.sd_model, required_prompts, steps)
+ cache[1] = function(shared.sd_model, required_prompts, steps, hires_steps, shared.opts.use_old_scheduling)
cache[0] = cached_params
return cache[1]
@@ -456,6 +464,8 @@ class StableDiffusionProcessing: sampler_config = sd_samplers.find_sampler_config(self.sampler_name)
total_steps = sampler_config.total_steps(self.steps) if sampler_config else self.steps
self.step_multiplier = total_steps // self.steps
+ self.firstpass_steps = total_steps
+
self.uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, total_steps, [self.cached_uc], self.extra_network_data)
self.c = self.get_conds_with_caching(prompt_parser.get_multicond_learned_conditioning, prompts, total_steps, [self.cached_c], self.extra_network_data)
@@ -1292,8 +1302,8 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): steps = self.hr_second_pass_steps or self.steps
total_steps = sampler_config.total_steps(steps) if sampler_config else steps
- self.hr_uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, hr_negative_prompts, total_steps, [self.cached_hr_uc, self.cached_uc], self.hr_extra_network_data)
- self.hr_c = self.get_conds_with_caching(prompt_parser.get_multicond_learned_conditioning, hr_prompts, total_steps, [self.cached_hr_c, self.cached_c], self.hr_extra_network_data)
+ self.hr_uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, hr_negative_prompts, self.firstpass_steps, [self.cached_hr_uc, self.cached_uc], self.hr_extra_network_data, total_steps)
+ self.hr_c = self.get_conds_with_caching(prompt_parser.get_multicond_learned_conditioning, hr_prompts, self.firstpass_steps, [self.cached_hr_c, self.cached_c], self.hr_extra_network_data, total_steps)
def setup_conds(self):
if self.is_hr_pass:
diff --git a/modules/prompt_parser.py b/modules/prompt_parser.py index e811ae99..334efeef 100644 --- a/modules/prompt_parser.py +++ b/modules/prompt_parser.py @@ -26,7 +26,7 @@ plain: /([^\\\[\]():|]|\\.)+/ %import common.SIGNED_NUMBER -> NUMBER
""")
-def get_learned_conditioning_prompt_schedules(prompts, steps):
+def get_learned_conditioning_prompt_schedules(prompts, base_steps, hires_steps=None, use_old_scheduling=False):
"""
>>> g = lambda p: get_learned_conditioning_prompt_schedules([p], 10)[0]
>>> g("test")
@@ -57,18 +57,39 @@ def get_learned_conditioning_prompt_schedules(prompts, steps): [[1, 'female'], [2, 'male'], [3, 'female'], [4, 'male'], [5, 'female'], [6, 'male'], [7, 'female'], [8, 'male'], [9, 'female'], [10, 'male']]
>>> g("[fe|||]male")
[[1, 'female'], [2, 'male'], [3, 'male'], [4, 'male'], [5, 'female'], [6, 'male'], [7, 'male'], [8, 'male'], [9, 'female'], [10, 'male']]
+ >>> g = lambda p: get_learned_conditioning_prompt_schedules([p], 10, 10)[0]
+ >>> g("a [b:.5] c")
+ [[10, 'a b c']]
+ >>> g("a [b:1.5] c")
+ [[5, 'a c'], [10, 'a b c']]
"""
+ if hires_steps is None or use_old_scheduling:
+ int_offset = 0
+ flt_offset = 0
+ steps = base_steps
+ else:
+ int_offset = base_steps
+ flt_offset = 1.0
+ steps = hires_steps
+
def collect_steps(steps, tree):
res = [steps]
class CollectSteps(lark.Visitor):
def scheduled(self, tree):
- tree.children[-2] = float(tree.children[-2])
- if tree.children[-2] < 1:
- tree.children[-2] *= steps
- tree.children[-2] = min(steps, int(tree.children[-2]))
- res.append(tree.children[-2])
+ s = tree.children[-2]
+ v = float(s)
+ if use_old_scheduling:
+ v = v*steps if v<1 else v
+ else:
+ if "." in s:
+ v = (v - flt_offset) * steps
+ else:
+ v = (v - int_offset)
+ tree.children[-2] = min(steps, int(v))
+ if tree.children[-2] >= 1:
+ res.append(tree.children[-2])
def alternate(self, tree):
res.extend(range(1, steps+1))
@@ -134,7 +155,7 @@ class SdConditioning(list): -def get_learned_conditioning(model, prompts: SdConditioning | list[str], steps):
+def get_learned_conditioning(model, prompts: SdConditioning | list[str], steps, hires_steps=None, use_old_scheduling=False):
"""converts a list of prompts into a list of prompt schedules - each schedule is a list of ScheduledPromptConditioning, specifying the comdition (cond),
and the sampling step at which this condition is to be replaced by the next one.
@@ -154,7 +175,7 @@ def get_learned_conditioning(model, prompts: SdConditioning | list[str], steps): """
res = []
- prompt_schedules = get_learned_conditioning_prompt_schedules(prompts, steps)
+ prompt_schedules = get_learned_conditioning_prompt_schedules(prompts, steps, hires_steps, use_old_scheduling)
cache = {}
for prompt, prompt_schedule in zip(prompts, prompt_schedules):
@@ -229,7 +250,7 @@ class MulticondLearnedConditioning: self.batch: List[List[ComposableScheduledPromptConditioning]] = batch
-def get_multicond_learned_conditioning(model, prompts, steps) -> MulticondLearnedConditioning:
+def get_multicond_learned_conditioning(model, prompts, steps, hires_steps=None, use_old_scheduling=False) -> MulticondLearnedConditioning:
"""same as get_learned_conditioning, but returns a list of ScheduledPromptConditioning along with the weight objects for each prompt.
For each prompt, the list is obtained by splitting the prompt using the AND separator.
@@ -238,7 +259,7 @@ def get_multicond_learned_conditioning(model, prompts, steps) -> MulticondLearne res_indexes, prompt_flat_list, prompt_indexes = get_multicond_prompt_list(prompts)
- learned_conditioning = get_learned_conditioning(model, prompt_flat_list, steps)
+ learned_conditioning = get_learned_conditioning(model, prompt_flat_list, steps, hires_steps, use_old_scheduling)
res = []
for indexes in res_indexes:
diff --git a/modules/realesrgan_model.py b/modules/realesrgan_model.py index 0700b853..02841c30 100644 --- a/modules/realesrgan_model.py +++ b/modules/realesrgan_model.py @@ -55,6 +55,7 @@ class UpscalerRealESRGAN(Upscaler): half=not cmd_opts.no_half and not cmd_opts.upcast_sampling,
tile=opts.ESRGAN_tile,
tile_pad=opts.ESRGAN_tile_overlap,
+ device=self.device,
)
upsampled = upsampler.enhance(np.array(img), outscale=info.scale)[0]
diff --git a/modules/sd_models.py b/modules/sd_models.py index 27d15e66..547e93c4 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -345,6 +345,7 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer if shared.cmd_opts.no_half:
model.float()
+ devices.dtype_unet = torch.float32
timer.record("apply float()")
else:
vae = model.first_stage_model
@@ -362,9 +363,9 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer if depth_model:
model.depth_model = depth_model
+ devices.dtype_unet = torch.float16
timer.record("apply half()")
- devices.dtype_unet = torch.float16 if model.is_sdxl and not shared.cmd_opts.no_half else model.model.diffusion_model.dtype
devices.unet_needs_upcast = shared.cmd_opts.upcast_sampling and devices.dtype == torch.float16 and devices.dtype_unet == torch.float16
model.first_stage_model.to(devices.dtype_vae)
@@ -517,7 +518,7 @@ def get_empty_cond(sd_model): def send_model_to_cpu(m):
- if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
+ if m.lowvram:
lowvram.send_everything_to_cpu()
else:
m.to(devices.cpu)
@@ -525,17 +526,17 @@ def send_model_to_cpu(m): devices.torch_gc()
-def model_target_device():
- if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
+def model_target_device(m):
+ if lowvram.is_needed(m):
return devices.cpu
else:
return devices.device
def send_model_to_device(m):
- if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
- lowvram.setup_for_low_vram(m, shared.cmd_opts.medvram)
- else:
+ lowvram.apply(m)
+
+ if not m.lowvram:
m.to(shared.device)
@@ -601,7 +602,7 @@ def load_model(checkpoint_info=None, already_loaded_state_dict=None): '': torch.float16,
}
- with sd_disable_initialization.LoadStateDictOnMeta(state_dict, device=model_target_device(), weight_dtype_conversion=weight_dtype_conversion):
+ with sd_disable_initialization.LoadStateDictOnMeta(state_dict, device=model_target_device(sd_model), weight_dtype_conversion=weight_dtype_conversion):
load_model_weights(sd_model, checkpoint_info, state_dict, timer)
timer.record("load weights from state dict")
@@ -743,7 +744,7 @@ def reload_model_weights(sd_model=None, info=None): script_callbacks.model_loaded_callback(sd_model)
timer.record("script callbacks")
- if not shared.cmd_opts.lowvram and not shared.cmd_opts.medvram:
+ if not sd_model.lowvram:
sd_model.to(devices.device)
timer.record("move model to device")
diff --git a/modules/sd_models_types.py b/modules/sd_models_types.py new file mode 100644 index 00000000..5ffd2f4f --- /dev/null +++ b/modules/sd_models_types.py @@ -0,0 +1,31 @@ +from ldm.models.diffusion.ddpm import LatentDiffusion
+from typing import TYPE_CHECKING
+
+
+if TYPE_CHECKING:
+ from modules.sd_models import CheckpointInfo
+
+
+class WebuiSdModel(LatentDiffusion):
+ """This class is not actually instantinated, but its fields are created and fieeld by webui"""
+
+ lowvram: bool
+ """True if lowvram/medvram optimizations are enabled -- see modules.lowvram for more info"""
+
+ sd_model_hash: str
+ """short hash, 10 first characters of SHA1 hash of the model file; may be None if --no-hashing flag is used"""
+
+ sd_model_checkpoint: str
+ """path to the file on disk that model weights were obtained from"""
+
+ sd_checkpoint_info: 'CheckpointInfo'
+ """structure with additional information about the file with model's weights"""
+
+ is_sdxl: bool
+ """True if the model's architecture is SDXL"""
+
+ is_sd2: bool
+ """True if the model's architecture is SD 2.x"""
+
+ is_sd1: bool
+ """True if the model's architecture is SD 1.x"""
diff --git a/modules/sd_unet.py b/modules/sd_unet.py index 6d708ad2..5525cfbc 100644 --- a/modules/sd_unet.py +++ b/modules/sd_unet.py @@ -47,7 +47,7 @@ def apply_unet(option=None): if current_unet_option is None:
current_unet = None
- if not (shared.cmd_opts.lowvram or shared.cmd_opts.medvram):
+ if not shared.sd_model.lowvram:
shared.sd_model.model.diffusion_model.to(devices.device)
return
diff --git a/modules/sd_vae.py b/modules/sd_vae.py index ee118656..669097da 100644 --- a/modules/sd_vae.py +++ b/modules/sd_vae.py @@ -263,7 +263,7 @@ def reload_vae_weights(sd_model=None, vae_file=unspecified): if loaded_vae_file == vae_file: return - if shared.cmd_opts.lowvram or shared.cmd_opts.medvram: + if sd_model.lowvram: lowvram.send_everything_to_cpu() else: sd_model.to(devices.cpu) @@ -275,7 +275,7 @@ def reload_vae_weights(sd_model=None, vae_file=unspecified): sd_hijack.model_hijack.hijack(sd_model) script_callbacks.model_loaded_callback(sd_model) - if not shared.cmd_opts.lowvram and not shared.cmd_opts.medvram: + if not sd_model.lowvram: sd_model.to(devices.device) print("VAE weights loaded.") diff --git a/modules/shared.py b/modules/shared.py index 0c57b712..63661939 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -2,16 +2,15 @@ import sys import gradio as gr
-from modules import shared_cmd_options, shared_gradio_themes, options, shared_items
+from modules import shared_cmd_options, shared_gradio_themes, options, shared_items, sd_models_types
from modules.paths_internal import models_path, script_path, data_path, sd_configs_path, sd_default_config, sd_model_file, default_sd_model_file, extensions_dir, extensions_builtin_dir # noqa: F401
-from ldm.models.diffusion.ddpm import LatentDiffusion
from modules import util
cmd_opts = shared_cmd_options.cmd_opts
parser = shared_cmd_options.parser
batch_cond_uncond = True # old field, unused now in favor of shared.opts.batch_cond_uncond
-parallel_processing_allowed = not cmd_opts.lowvram and not cmd_opts.medvram
+parallel_processing_allowed = True
styles_filename = cmd_opts.styles_file
config_filename = cmd_opts.ui_settings_file
hide_dirs = {"visible": not cmd_opts.hide_ui_dir_config}
@@ -40,7 +39,7 @@ options_templates = None opts = None
restricted_opts = None
-sd_model: LatentDiffusion = None
+sd_model: sd_models_types.WebuiSdModel = None
settings_components = None
"""assinged from ui.py, a mapping on setting names to gradio components repsponsible for those settings"""
diff --git a/modules/shared_options.py b/modules/shared_options.py index 88f6b334..83f56314 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -203,6 +203,7 @@ options_templates.update(options_section(('compatibility', "Compatibility"), { "use_old_hires_fix_width_height": OptionInfo(False, "For hires fix, use width/height sliders to set final resolution rather than first pass (disables Upscale by, Resize width/height to)."),
"dont_fix_second_order_samplers_schedule": OptionInfo(False, "Do not fix prompt schedule for second order samplers."),
"hires_fix_use_firstpass_conds": OptionInfo(False, "For hires fix, calculate conds of second pass using extra networks of first pass."),
+ "use_old_scheduling": OptionInfo(False, "Use old prompt editing timelines.", infotext="Old prompt editing timelines").info("For [red:green:N]; old: If N < 1, it's a fraction of steps (and hires fix uses range from 0 to 1), if N >= 1, it's an absolute number of steps; new: If N has a decimal point in it, it's a fraction of steps (and hires fix uses range from 1 to 2), othewrwise it's an absolute number of steps"),
}))
options_templates.update(options_section(('interrogate', "Interrogate"), {
|