happyz
/
Fooocus
mirror of https://github.com/lllyasviel/Fooocus.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Revert "skip free_memory for interrogate" 

This reverts commit 316fe24f65.
Rather add interrogate action to queue than keeping 400MB BLIP loaded in VRAM.
This commit is contained in:
Manuel Schmid 2023-12-31 00:25:27 +01:00
parent 64c1a3dcce
commit 13bb314cfa
No known key found for this signature in database
GPG Key ID: 32C4F7569B40B84B
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
extras/interrogate.py
View File

@ -47,7 +47,7 @@ class Interrogator:
self.blip_model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device) self.blip_model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)
model_management.load_model_gpu(self.blip_model, should_free_memory=False) model_management.load_model_gpu(self.blip_model)
gpu_image = transforms.Compose([ gpu_image = transforms.Compose([
transforms.ToTensor(), transforms.ToTensor(),

10
ldm_patched/modules/model_management.py
View File

@ -373,7 +373,7 @@ def free_memory(memory_required, device, keep_loaded=[]):
if mem_free_torch > mem_free_total * 0.25: if mem_free_torch > mem_free_total * 0.25:
soft_empty_cache() soft_empty_cache()
def load_models_gpu(models, memory_required=0, should_free_memory=True): def load_models_gpu(models, memory_required=0):
global vram_state global vram_state
inference_memory = minimum_inference_memory() inference_memory = minimum_inference_memory()
@ -396,7 +396,7 @@ def load_models_gpu(models, memory_required=0, should_free_memory=True):
if len(models_to_load) == 0: if len(models_to_load) == 0:
devs = set(map(lambda a: a.device, models_already_loaded)) devs = set(map(lambda a: a.device, models_already_loaded))
for d in devs: for d in devs:
if d != torch.device("cpu") and should_free_memory: if d != torch.device("cpu"):
free_memory(extra_mem, d, models_already_loaded) free_memory(extra_mem, d, models_already_loaded)
return return
@ -408,7 +408,7 @@ def load_models_gpu(models, memory_required=0, should_free_memory=True):
total_memory_required[loaded_model.device] = total_memory_required.get(loaded_model.device, 0) + loaded_model.model_memory_required(loaded_model.device) total_memory_required[loaded_model.device] = total_memory_required.get(loaded_model.device, 0) + loaded_model.model_memory_required(loaded_model.device)
for device in total_memory_required: for device in total_memory_required:
if device != torch.device("cpu") and should_free_memory: if device != torch.device("cpu"):
free_memory(total_memory_required[device] * 1.3 + extra_mem, device, models_already_loaded) free_memory(total_memory_required[device] * 1.3 + extra_mem, device, models_already_loaded)
for loaded_model in models_to_load: for loaded_model in models_to_load:
@ -436,8 +436,8 @@ def load_models_gpu(models, memory_required=0, should_free_memory=True):
return return
def load_model_gpu(model, should_free_memory=True): def load_model_gpu(model):
return load_models_gpu([model], should_free_memory=should_free_memory) return load_models_gpu([model])
def cleanup_models(): def cleanup_models():
to_delete = [] to_delete = []