2023-02-08 08:17:54 +00:00
|
|
|
|
2023-02-08 16:37:10 +00:00
|
|
|
CPU = 0
|
|
|
|
NO_VRAM = 1
|
|
|
|
LOW_VRAM = 2
|
|
|
|
NORMAL_VRAM = 3
|
2023-02-18 02:14:07 +00:00
|
|
|
HIGH_VRAM = 4
|
2023-02-08 16:37:10 +00:00
|
|
|
|
|
|
|
accelerate_enabled = False
|
|
|
|
vram_state = NORMAL_VRAM
|
|
|
|
|
2023-02-08 19:05:31 +00:00
|
|
|
total_vram = 0
|
2023-02-08 16:42:37 +00:00
|
|
|
total_vram_available_mb = -1
|
|
|
|
|
2023-02-08 16:37:10 +00:00
|
|
|
import sys
|
2023-02-28 00:57:39 +00:00
|
|
|
import psutil
|
2023-02-08 16:37:10 +00:00
|
|
|
|
|
|
|
set_vram_to = NORMAL_VRAM
|
|
|
|
|
2023-02-08 19:05:31 +00:00
|
|
|
try:
|
|
|
|
import torch
|
|
|
|
total_vram = torch.cuda.mem_get_info(torch.cuda.current_device())[1] / (1024 * 1024)
|
2023-02-28 00:57:39 +00:00
|
|
|
total_ram = psutil.virtual_memory().total / (1024 * 1024)
|
|
|
|
forced_normal_vram = "--normalvram" in sys.argv
|
|
|
|
if not forced_normal_vram:
|
|
|
|
if total_vram <= 4096:
|
|
|
|
print("Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram")
|
|
|
|
set_vram_to = LOW_VRAM
|
2023-02-28 01:04:18 +00:00
|
|
|
elif total_vram > total_ram * 1.1 and total_vram > 14336:
|
2023-02-28 00:57:39 +00:00
|
|
|
print("Enabling highvram mode because your GPU has more vram than your computer has ram. If you don't want this use: --normalvram")
|
|
|
|
vram_state = HIGH_VRAM
|
2023-02-08 19:05:31 +00:00
|
|
|
except:
|
|
|
|
pass
|
|
|
|
|
2023-03-12 19:44:16 +00:00
|
|
|
if "--disable-xformers" in sys.argv:
|
|
|
|
XFORMERS_IS_AVAILBLE = False
|
2023-03-13 15:36:48 +00:00
|
|
|
else:
|
|
|
|
try:
|
|
|
|
import xformers
|
|
|
|
import xformers.ops
|
|
|
|
XFORMERS_IS_AVAILBLE = True
|
|
|
|
except:
|
|
|
|
XFORMERS_IS_AVAILBLE = False
|
|
|
|
|
2023-03-13 16:25:19 +00:00
|
|
|
ENABLE_PYTORCH_ATTENTION = False
|
|
|
|
if "--use-pytorch-cross-attention" in sys.argv:
|
|
|
|
torch.backends.cuda.enable_math_sdp(True)
|
|
|
|
torch.backends.cuda.enable_flash_sdp(True)
|
|
|
|
torch.backends.cuda.enable_mem_efficient_sdp(True)
|
|
|
|
ENABLE_PYTORCH_ATTENTION = True
|
|
|
|
XFORMERS_IS_AVAILBLE = False
|
|
|
|
|
2023-03-12 19:44:16 +00:00
|
|
|
|
2023-02-10 05:47:56 +00:00
|
|
|
if "--lowvram" in sys.argv:
|
|
|
|
set_vram_to = LOW_VRAM
|
|
|
|
if "--novram" in sys.argv:
|
|
|
|
set_vram_to = NO_VRAM
|
2023-02-18 02:14:07 +00:00
|
|
|
if "--highvram" in sys.argv:
|
|
|
|
vram_state = HIGH_VRAM
|
2023-02-10 05:47:56 +00:00
|
|
|
|
|
|
|
|
2023-02-18 02:14:07 +00:00
|
|
|
if set_vram_to == LOW_VRAM or set_vram_to == NO_VRAM:
|
2023-02-08 16:37:10 +00:00
|
|
|
try:
|
|
|
|
import accelerate
|
|
|
|
accelerate_enabled = True
|
|
|
|
vram_state = set_vram_to
|
|
|
|
except Exception as e:
|
|
|
|
import traceback
|
|
|
|
print(traceback.format_exc())
|
|
|
|
print("ERROR: COULD NOT ENABLE LOW VRAM MODE.")
|
2023-02-08 19:05:31 +00:00
|
|
|
|
|
|
|
total_vram_available_mb = (total_vram - 1024) // 2
|
2023-02-08 16:42:37 +00:00
|
|
|
total_vram_available_mb = int(max(256, total_vram_available_mb))
|
2023-02-08 16:37:10 +00:00
|
|
|
|
2023-03-14 01:30:01 +00:00
|
|
|
if "--cpu" in sys.argv:
|
|
|
|
vram_state = CPU
|
2023-02-08 16:37:10 +00:00
|
|
|
|
2023-02-18 02:14:07 +00:00
|
|
|
print("Set vram state to:", ["CPU", "NO VRAM", "LOW VRAM", "NORMAL VRAM", "HIGH VRAM"][vram_state])
|
2023-02-08 16:37:10 +00:00
|
|
|
|
2023-02-08 08:17:54 +00:00
|
|
|
|
|
|
|
current_loaded_model = None
|
2023-02-16 15:38:08 +00:00
|
|
|
current_gpu_controlnets = []
|
2023-02-08 08:17:54 +00:00
|
|
|
|
2023-02-08 16:37:10 +00:00
|
|
|
model_accelerated = False
|
|
|
|
|
|
|
|
|
2023-02-08 08:17:54 +00:00
|
|
|
def unload_model():
|
|
|
|
global current_loaded_model
|
2023-02-08 16:37:10 +00:00
|
|
|
global model_accelerated
|
2023-02-16 15:38:08 +00:00
|
|
|
global current_gpu_controlnets
|
2023-02-18 02:14:07 +00:00
|
|
|
global vram_state
|
|
|
|
|
2023-02-08 08:17:54 +00:00
|
|
|
if current_loaded_model is not None:
|
2023-02-08 16:37:10 +00:00
|
|
|
if model_accelerated:
|
|
|
|
accelerate.hooks.remove_hook_from_submodules(current_loaded_model.model)
|
|
|
|
model_accelerated = False
|
|
|
|
|
2023-02-18 02:14:07 +00:00
|
|
|
#never unload models from GPU on high vram
|
|
|
|
if vram_state != HIGH_VRAM:
|
|
|
|
current_loaded_model.model.cpu()
|
2023-02-08 08:17:54 +00:00
|
|
|
current_loaded_model.unpatch_model()
|
|
|
|
current_loaded_model = None
|
2023-02-18 02:14:07 +00:00
|
|
|
|
|
|
|
if vram_state != HIGH_VRAM:
|
|
|
|
if len(current_gpu_controlnets) > 0:
|
|
|
|
for n in current_gpu_controlnets:
|
|
|
|
n.cpu()
|
|
|
|
current_gpu_controlnets = []
|
2023-02-08 08:17:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
def load_model_gpu(model):
|
|
|
|
global current_loaded_model
|
2023-02-08 16:37:10 +00:00
|
|
|
global vram_state
|
|
|
|
global model_accelerated
|
|
|
|
|
2023-02-08 08:17:54 +00:00
|
|
|
if model is current_loaded_model:
|
|
|
|
return
|
|
|
|
unload_model()
|
|
|
|
try:
|
|
|
|
real_model = model.patch_model()
|
|
|
|
except Exception as e:
|
|
|
|
model.unpatch_model()
|
|
|
|
raise e
|
|
|
|
current_loaded_model = model
|
2023-02-08 16:37:10 +00:00
|
|
|
if vram_state == CPU:
|
|
|
|
pass
|
2023-02-18 02:14:07 +00:00
|
|
|
elif vram_state == NORMAL_VRAM or vram_state == HIGH_VRAM:
|
2023-02-08 16:37:10 +00:00
|
|
|
model_accelerated = False
|
|
|
|
real_model.cuda()
|
|
|
|
else:
|
|
|
|
if vram_state == NO_VRAM:
|
|
|
|
device_map = accelerate.infer_auto_device_map(real_model, max_memory={0: "256MiB", "cpu": "16GiB"})
|
|
|
|
elif vram_state == LOW_VRAM:
|
2023-02-08 16:42:37 +00:00
|
|
|
device_map = accelerate.infer_auto_device_map(real_model, max_memory={0: "{}MiB".format(total_vram_available_mb), "cpu": "16GiB"})
|
2023-02-08 19:51:18 +00:00
|
|
|
|
2023-02-08 16:37:10 +00:00
|
|
|
accelerate.dispatch_model(real_model, device_map=device_map, main_device="cuda")
|
|
|
|
model_accelerated = True
|
2023-02-08 08:17:54 +00:00
|
|
|
return current_loaded_model
|
2023-02-08 19:05:31 +00:00
|
|
|
|
2023-02-16 15:38:08 +00:00
|
|
|
def load_controlnet_gpu(models):
|
|
|
|
global current_gpu_controlnets
|
2023-02-17 20:45:29 +00:00
|
|
|
global vram_state
|
2023-03-06 15:50:50 +00:00
|
|
|
if vram_state == CPU:
|
|
|
|
return
|
2023-02-17 20:45:29 +00:00
|
|
|
|
|
|
|
if vram_state == LOW_VRAM or vram_state == NO_VRAM:
|
|
|
|
#don't load controlnets like this if low vram because they will be loaded right before running and unloaded right after
|
|
|
|
return
|
|
|
|
|
2023-02-16 15:38:08 +00:00
|
|
|
for m in current_gpu_controlnets:
|
|
|
|
if m not in models:
|
|
|
|
m.cpu()
|
|
|
|
|
|
|
|
current_gpu_controlnets = []
|
|
|
|
for m in models:
|
|
|
|
current_gpu_controlnets.append(m.cuda())
|
|
|
|
|
2023-02-08 19:05:31 +00:00
|
|
|
|
2023-02-17 20:45:29 +00:00
|
|
|
def load_if_low_vram(model):
|
|
|
|
global vram_state
|
|
|
|
if vram_state == LOW_VRAM or vram_state == NO_VRAM:
|
|
|
|
return model.cuda()
|
|
|
|
return model
|
|
|
|
|
|
|
|
def unload_if_low_vram(model):
|
|
|
|
global vram_state
|
|
|
|
if vram_state == LOW_VRAM or vram_state == NO_VRAM:
|
|
|
|
return model.cpu()
|
|
|
|
return model
|
|
|
|
|
2023-03-06 15:50:50 +00:00
|
|
|
def get_torch_device():
|
|
|
|
if vram_state == CPU:
|
|
|
|
return torch.device("cpu")
|
|
|
|
else:
|
|
|
|
return torch.cuda.current_device()
|
|
|
|
|
|
|
|
def get_autocast_device(dev):
|
|
|
|
if hasattr(dev, 'type'):
|
|
|
|
return dev.type
|
|
|
|
return "cuda"
|
2023-02-17 20:45:29 +00:00
|
|
|
|
2023-03-12 19:44:16 +00:00
|
|
|
def xformers_enabled():
|
|
|
|
if vram_state == CPU:
|
|
|
|
return False
|
|
|
|
return XFORMERS_IS_AVAILBLE
|
|
|
|
|
2023-03-13 16:25:19 +00:00
|
|
|
def pytorch_attention_enabled():
|
|
|
|
return ENABLE_PYTORCH_ATTENTION
|
|
|
|
|
2023-03-03 08:27:33 +00:00
|
|
|
def get_free_memory(dev=None, torch_free_too=False):
|
|
|
|
if dev is None:
|
2023-03-06 15:50:50 +00:00
|
|
|
dev = get_torch_device()
|
2023-03-03 08:27:33 +00:00
|
|
|
|
|
|
|
if hasattr(dev, 'type') and dev.type == 'cpu':
|
|
|
|
mem_free_total = psutil.virtual_memory().available
|
|
|
|
mem_free_torch = mem_free_total
|
|
|
|
else:
|
|
|
|
stats = torch.cuda.memory_stats(dev)
|
|
|
|
mem_active = stats['active_bytes.all.current']
|
|
|
|
mem_reserved = stats['reserved_bytes.all.current']
|
|
|
|
mem_free_cuda, _ = torch.cuda.mem_get_info(dev)
|
|
|
|
mem_free_torch = mem_reserved - mem_active
|
|
|
|
mem_free_total = mem_free_cuda + mem_free_torch
|
|
|
|
|
|
|
|
if torch_free_too:
|
|
|
|
return (mem_free_total, mem_free_torch)
|
|
|
|
else:
|
|
|
|
return mem_free_total
|
2023-02-08 19:05:31 +00:00
|
|
|
|
|
|
|
def maximum_batch_area():
|
|
|
|
global vram_state
|
|
|
|
if vram_state == NO_VRAM:
|
|
|
|
return 0
|
|
|
|
|
|
|
|
memory_free = get_free_memory() / (1024 * 1024)
|
|
|
|
area = ((memory_free - 1024) * 0.9) / (0.6)
|
|
|
|
return int(max(area, 0))
|
2023-03-03 16:07:10 +00:00
|
|
|
|
|
|
|
def cpu_mode():
|
|
|
|
global vram_state
|
|
|
|
return vram_state == CPU
|
|
|
|
|
|
|
|
def should_use_fp16():
|
|
|
|
if cpu_mode():
|
|
|
|
return False #TODO ?
|
|
|
|
|
|
|
|
if torch.cuda.is_bf16_supported():
|
|
|
|
return True
|
|
|
|
|
2023-03-03 18:18:01 +00:00
|
|
|
props = torch.cuda.get_device_properties("cuda")
|
2023-03-03 16:07:10 +00:00
|
|
|
if props.major < 7:
|
|
|
|
return False
|
|
|
|
|
|
|
|
#FP32 is faster on those cards?
|
|
|
|
nvidia_16_series = ["1660", "1650", "1630"]
|
|
|
|
for x in nvidia_16_series:
|
|
|
|
if x in props.name:
|
|
|
|
return False
|
|
|
|
|
|
|
|
return True
|
|
|
|
|
2023-03-02 19:42:03 +00:00
|
|
|
#TODO: might be cleaner to put this somewhere else
|
|
|
|
import threading
|
|
|
|
|
|
|
|
class InterruptProcessingException(Exception):
|
|
|
|
pass
|
|
|
|
|
|
|
|
interrupt_processing_mutex = threading.RLock()
|
|
|
|
|
|
|
|
interrupt_processing = False
|
|
|
|
def interrupt_current_processing(value=True):
|
|
|
|
global interrupt_processing
|
|
|
|
global interrupt_processing_mutex
|
|
|
|
with interrupt_processing_mutex:
|
|
|
|
interrupt_processing = value
|
|
|
|
|
|
|
|
def processing_interrupted():
|
|
|
|
global interrupt_processing
|
|
|
|
global interrupt_processing_mutex
|
|
|
|
with interrupt_processing_mutex:
|
|
|
|
return interrupt_processing
|
|
|
|
|
|
|
|
def throw_exception_if_processing_interrupted():
|
|
|
|
global interrupt_processing
|
|
|
|
global interrupt_processing_mutex
|
|
|
|
with interrupt_processing_mutex:
|
|
|
|
if interrupt_processing:
|
|
|
|
interrupt_processing = False
|
|
|
|
raise InterruptProcessingException()
|