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

switch some cmdopts to opts

Browse Source
This commit is contained in:
Vladimir Mandic
2023-05-06 14:35:33 -04:00
parent 1360c6422a
commit 41182009cb
10 changed files with 31 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
extensions-builtin/sd-webui-controlnet
+1 -1
Submodule extensions-builtin/sd-webui-controlnet updated: 817155ea7a...6be213feff
modules/devices.py
+9 -6
View File
@@ -77,11 +77,13 @@ def test_fp16():
x = torch.tensor([[1.5,.0,.0,.0]]).to(device).half()
layerNorm = torch.nn.LayerNorm(4, eps=0.00001, elementwise_affine=True, dtype=torch.float16, device=device)
_y = layerNorm(x)
return True
except:
shared.log.warning('Torch FP16 test failed: Forcing FP32 operations')
shared.opts.cuda_dtype = 'FP32'
shared.opts.no_half = True
shared.opts.no_half_vae = True
return False
def set_cuda_params():
@@ -101,22 +103,23 @@ def set_cuda_params():
pass
global dtype, dtype_vae, dtype_unet, unet_needs_upcast # pylint: disable=global-statement
# set dtype
test_fp16()
if shared.opts.cuda_dtype == 'FP16':
ok = test_fp16()
if shared.opts.cuda_dtype == 'FP16' and ok:
dtype = torch.float16
dtype_vae = torch.float16
dtype_unet = torch.float16
if shared.opts.cuda_dtype == 'BP16':
if shared.opts.cuda_dtype == 'BP16' and ok:
dtype = torch.bfloat16
dtype_vae = torch.bfloat16
dtype_unet = torch.bfloat16
if shared.opts.cuda_dtype == 'FP32' or shared.opts.no_half:
if shared.opts.cuda_dtype == 'FP32' or shared.opts.no_half or not ok:
dtype = torch.float32
dtype_vae = torch.float32
dtype_unet = torch.float32
if shared.opts.no_half_vae: # set dtype again as no-half-vae options take priority
dtype_vae = torch.float32
unet_needs_upcast = shared.opts.upcast_sampling
shared.log.debug(f'Setting CUDA parameters: dtype={dtype} vae={dtype_vae} unet={dtype_unet}')
args = cmd_args.parser.parse_args()
@@ -182,11 +185,11 @@ def test_for_nans(x, where):
return
if where == "unet":
message = "A tensor with all NaNs was produced in Unet."
if not shared.cmd_opts.no_half:
if not shared.opts.no_half:
message += " This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the \"Upcast cross attention layer to float32\" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this."
elif where == "vae":
message = "A tensor with all NaNs was produced in VAE."
if not shared.cmd_opts.no_half and not shared.cmd_opts.no_half_vae:
if not shared.opts.no_half and not shared.opts.no_half_vae:
message += " This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this."
else:
message = "A tensor with all NaNs was produced."
modules/img2img.py
+2 -2
View File
@@ -60,7 +60,7 @@ def process_batch(p, input_dir, output_dir, inpaint_mask_dir, args):
if processed_image.mode == 'RGBA':
processed_image = processed_image.convert("RGB")
processed_image.save(os.path.join(output_dir, filename))
shared.debug(f'Processed: {len(images)} Memory: {memory_stats()} batch')
shared.log.debug(f'Processed: {len(images)} Memory: {memory_stats()} batch')
def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_styles, init_img, sketch, init_img_with_mask, inpaint_color_sketch, inpaint_color_sketch_orig, init_img_inpaint, init_mask_inpaint, steps: int, sampler_index: int, mask_blur: int, mask_alpha: float, inpainting_fill: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, image_cfg_scale: float, denoising_strength: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, selected_scale_tab: int, height: int, width: int, scale_by: float, resize_mode: int, inpaint_full_res: bool, inpaint_full_res_padding: int, inpainting_mask_invert: int, img2img_batch_input_dir: str, img2img_batch_output_dir: str, img2img_batch_inpaint_mask_dir: str, override_settings_texts, *args): # pylint: disable=unused-argument
@@ -154,5 +154,5 @@ def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_s
processed = process_images(p)
p.close()
generation_info_js = processed.js()
shared.debug(f'Processed: {len(processed.images)} Memory: {memory_stats()} img')
shared.log.debug(f'Processed: {len(processed.images)} Memory: {memory_stats()} img')
return processed.images, generation_info_js, plaintext_to_html(processed.info), plaintext_to_html(processed.comments)
modules/interrogate.py
+2 -2
View File
@@ -118,14 +118,14 @@ class InterrogateModels:
def load(self):
if self.blip_model is None:
self.blip_model = self.load_blip_model()
if not shared.cmd_opts.no_half and not self.running_on_cpu:
if not shared.opts.no_half and not self.running_on_cpu:
self.blip_model = self.blip_model.half()
self.blip_model = self.blip_model.to(devices.device_interrogate)
if self.clip_model is None:
self.clip_model, self.clip_preprocess = self.load_clip_model()
if not shared.cmd_opts.no_half and not self.running_on_cpu:
if not shared.opts.no_half and not self.running_on_cpu:
self.clip_model = self.clip_model.half()
self.clip_model = self.clip_model.to(devices.device_interrogate)
modules/processing.py
+1 -1
View File
@@ -680,7 +680,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
for x in x_samples_ddim:
devices.test_for_nans(x, "vae")
except devices.NansException as e:
if not shared.cmd_opts.no_half and not shared.cmd_opts.no_half_vae and shared.cmd_opts.rollback_vae:
if not shared.opts.no_half and not shared.opts.no_half_vae and shared.cmd_opts.rollback_vae:
log.warning('Tensor with all NaNs was produced in VAE')
devices.dtype_vae = torch.bfloat16
vae_file, vae_source = sd_vae.resolve_vae(p.sd_model.sd_model_checkpoint)
modules/realesrgan_model.py
+1 -1
View File
@@ -57,7 +57,7 @@ class UpscalerRealESRGAN(Upscaler):
scale=info.scale,
model_path=info.local_data_path,
model=info.model(),
half=not cmd_opts.no_half and not opts.upcast_sampling,
half=not opts.no_half and not opts.upcast_sampling,
tile=opts.ESRGAN_tile,
tile_pad=opts.ESRGAN_tile_overlap,
device=device,
modules/sd_models.py
+13 -13
View File
@@ -262,11 +262,11 @@ def load_model_weights(model: torch.nn.Module, checkpoint_info: CheckpointInfo,
if shared.opts.opt_channelslast:
model.to(memory_format=torch.channels_last)
timer.record("channels")
if not shared.cmd_opts.no_half:
if not shared.opts.no_half:
vae = model.first_stage_model
depth_model = getattr(model, 'depth_model', None)
# with --no-half-vae, remove VAE from model when doing half() to prevent its weights from being converted to float16
if shared.cmd_opts.no_half_vae:
if shared.opts.no_half_vae:
model.first_stage_model = None
# with --upcast-sampling, don't convert the depth model weights to float16
if shared.opts.upcast_sampling and depth_model:
@@ -275,7 +275,6 @@ def load_model_weights(model: torch.nn.Module, checkpoint_info: CheckpointInfo,
model.first_stage_model = vae
if depth_model:
model.depth_model = depth_model
devices.set_cuda_params()
devices.dtype_unet = model.model.diffusion_model.dtype
model.first_stage_model.to(devices.dtype_vae)
# clean up cache if limit is reached
@@ -330,7 +329,7 @@ def enable_midas_autodownload():
def repair_config(sd_config):
if not "use_ema" in sd_config.model.params:
sd_config.model.params.use_ema = False
if shared.cmd_opts.no_half:
if shared.opts.no_half:
sd_config.model.params.unet_config.params.use_fp16 = False
elif shared.opts.upcast_sampling:
sd_config.model.params.unet_config.params.use_fp16 = True
@@ -347,7 +346,7 @@ sd2_clip_weight = 'cond_stage_model.model.transformer.resblocks.0.attn.in_proj_w
def load_model(checkpoint_info=None, already_loaded_state_dict=None, timer=None):
shared.debug(f'Load model: {checkpoint_info} {already_loaded_state_dict}')
shared.log.debug(f'Load model: info={checkpoint_info is not None} dict={already_loaded_state_dict is not None}')
from modules import lowvram, sd_hijack
checkpoint_info = checkpoint_info or select_checkpoint()
if checkpoint_info is None:
@@ -361,8 +360,9 @@ def load_model(checkpoint_info=None, already_loaded_state_dict=None, timer=None)
shared.sd_model = None
gc.collect()
devices.torch_gc()
shared.debug(f'Model unloaded: {memory_stats()}')
shared.log.debug(f'Model unloaded: {memory_stats()}')
do_inpainting_hijack()
devices.set_cuda_params()
if already_loaded_state_dict is not None:
state_dict = already_loaded_state_dict
else:
@@ -374,12 +374,12 @@ def load_model(checkpoint_info=None, already_loaded_state_dict=None, timer=None)
shared.log.info(f"Restoring previous checkpoint: {current_checkpoint_info.filename}")
load_model(current_checkpoint_info, None)
return
shared.debug(f'Model dict loaded: {memory_stats()}')
shared.log.debug(f'Model dict loaded: {memory_stats()}')
clip_is_included_into_sd = sd1_clip_weight in state_dict or sd2_clip_weight in state_dict
sd_config = OmegaConf.load(checkpoint_config)
repair_config(sd_config)
timer.record("config")
shared.debug(f'Model config loaded: {memory_stats()}')
shared.log.debug(f'Model config loaded: {memory_stats()}')
shared.log.info(f"Creating model from config: {checkpoint_config}")
sd_model = None
try:
@@ -391,13 +391,13 @@ def load_model(checkpoint_info=None, already_loaded_state_dict=None, timer=None)
timer.record("create")
load_model_weights(sd_model, checkpoint_info, state_dict, timer)
timer.record("load")
shared.debug(f'Model weights loaded: {memory_stats()}')
shared.log.debug(f'Model weights loaded: {memory_stats()}')
if shared.cmd_opts.lowvram or shared.cmd_opts.medvram:
lowvram.setup_for_low_vram(sd_model, shared.cmd_opts.medvram)
else:
sd_model.to(devices.device)
timer.record("move")
shared.debug(f'Model weights moved: {memory_stats()}')
shared.log.debug(f'Model weights moved: {memory_stats()}')
sd_hijack.model_hijack.hijack(sd_model)
timer.record("hijack")
sd_model.eval()
@@ -411,16 +411,16 @@ def load_model(checkpoint_info=None, already_loaded_state_dict=None, timer=None)
timer.record("callbacks")
shared.log.info(f"Model loaded in {timer.summary()}")
gc.collect()
shared.debug(f'Model load finished: {memory_stats()}')
shared.log.debug(f'Model load finished: {memory_stats()}')
def reload_model_weights(sd_model=None, info=None):
global skip_next_load # pylint: disable=global-statement
if skip_next_load:
shared.debug('Reload model weights skip')
shared.log.debug('Reload model weights skip')
skip_next_load = False
return
shared.debug(f'Reload model weights: {sd_model} {info}')
shared.log.debug(f'Reload model weights: {sd_model} {info}')
from modules import lowvram, sd_hijack
checkpoint_info = info or select_checkpoint()
if not sd_model:
modules/shared.py
-5
View File
@@ -166,11 +166,6 @@ interrogator = modules.interrogate.InterrogateModels("interrogate")
face_restorers = []
def debug(message):
if cmd_opts.debug:
log.debug(message)
class OptionInfo:
def __init__(self, default=None, label="", component=None, component_args=None, onchange=None, section=None, refresh=None):
self.default = default
modules/txt2img.py
+1 -1
View File
@@ -50,5 +50,5 @@ def txt2img(id_task: str, prompt: str, negative_prompt: str, prompt_styles, step
processed = process_images(p)
p.close()
generation_info_js = processed.js()
shared.debug(f'Processed: {len(processed.images)} Memory: {memory_stats()} txt')
shared.log.debug(f'Processed: {len(processed.images)} Memory: {memory_stats()} txt')
return processed.images, generation_info_js, plaintext_to_html(processed.info), plaintext_to_html(processed.comments)
modules/upscaler.py
+1 -1
View File
@@ -30,7 +30,7 @@ class Upscaler:
self.img = None
self.output = None
self.scale = 1
self.half = not shared.cmd_opts.no_half
self.half = not shared.opts.no_half
self.pre_pad = 0
self.mod_scale = None