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

Fix sequential cpu offload

Browse Source
This commit is contained in:
Disty0
2023-07-28 20:01:32 +03:00
parent 77de9cd093
commit e6cf3d72cd
2 changed files with 13 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sd_models.py
+9 -3
View File
@@ -661,6 +661,12 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
if shared.opts.diffusers_seq_cpu_offload:
sd_model.enable_sequential_cpu_offload(device=devices.device)
shared.log.debug(f'Diffusers {op}: enable sequential CPU offload')
if shared.opts.diffusers_move_base or shared.opts.diffusers_move_refiner:
shared.log.warning("Moving base or refiner model to CPU is not supported with sequential CPU offload")
shared.log.debug('Disabled moving base model to CPU')
shared.log.debug('Disabled moving refiner model to CPU')
shared.opts.diffusers_move_base=False
shared.opts.diffusers_move_refiner=False
if hasattr(sd_model, "enable_vae_slicing"):
if shared.cmd_opts.lowvram or shared.opts.diffusers_vae_slicing:
shared.log.debug(f'Diffusers {op}: enable VAE slicing')
@@ -693,7 +699,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
base_sent_to_cpu=False
if shared.opts.cuda_compile and torch.cuda.is_available():
if op == 'refiner':
if op == 'refiner' and not shared.opts.diffusers_seq_cpu_offload:
gpu_vram = memory_stats().get('gpu', {})
free_vram = gpu_vram.get('total', 0) - gpu_vram.get('used', 0)
refiner_enough_vram = free_vram >= 7 if "StableDiffusionXL" in sd_model.__class__.__name__ else 3
@@ -712,7 +718,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
devices.torch_gc(force=True)
sd_model.to(devices.device)
base_sent_to_cpu=True
else:
elif not shared.opts.diffusers_seq_cpu_offload:
sd_model.to(devices.device)
try:
shared.log.info(f"Compiling pipeline={sd_model.__class__.__name__} shape={8 * sd_model.unet.config.sample_size} mode={shared.opts.cuda_compile_mode}")
@@ -743,7 +749,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
if op == 'refiner' and shared.opts.diffusers_move_refiner:
shared.log.debug('Moving refiner model to CPU')
sd_model.to("cpu")
else:
elif not shared.opts.diffusers_seq_cpu_offload:
sd_model.to(devices.device)
if op == 'refiner' and base_sent_to_cpu:
shared.log.debug('Moving base model back to GPU')
modules/ui.py
+4 -2
View File
@@ -951,7 +951,8 @@ def create_ui(startup_timer = None):
continue
if opts.set(key, value):
changed.append(key)
directml_override_opts()
if cmd_opts.use_directml:
directml_override_opts()
try:
opts.save(modules.shared.config_filename)
modules.shared.log.info(f'Settings changed: {len(changed)} {changed}')
@@ -965,7 +966,8 @@ def create_ui(startup_timer = None):
return gr.update(visible=True), opts.dumpjson()
if not opts.set(key, value):
return gr.update(value=getattr(opts, key)), opts.dumpjson()
directml_override_opts()
if cmd_opts.use_directml:
directml_override_opts()
opts.save(modules.shared.config_filename)
modules.shared.log.debug(f'Setting changed: key={key}, value={value}')
return get_value_for_setting(key), opts.dumpjson()