andreas/ComfyUI
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Unload weights if vram usage goes up between runs. (#10690)

Browse Source
This commit is contained in:
comfyanonymous
2025-11-09 15:51:33 -08:00
committed by GitHub
parent e632e5de28
commit dea899f221
2 changed files with 22 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
comfy/model_management.py
View File

@@ -503,7 +503,11 @@ class LoadedModel:
use_more_vram = lowvram_model_memory
if use_more_vram == 0:
use_more_vram = 1e32
self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)
if use_more_vram > 0:
self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)
else:
self.model.partially_unload(self.model.offload_device, -use_more_vram, force_patch_weights=force_patch_weights)
real_model = self.model.model
if is_intel_xpu() and not args.disable_ipex_optimize and 'ipex' in globals() and real_model is not None:
@@ -689,7 +693,10 @@ def load_models_gpu(models, memory_required=0, force_patch_weights=False, minimu
current_free_mem = get_free_memory(torch_dev) + loaded_memory
lowvram_model_memory = max(128 * 1024 * 1024, (current_free_mem - minimum_memory_required), min(current_free_mem * MIN_WEIGHT_MEMORY_RATIO, current_free_mem - minimum_inference_memory()))
lowvram_model_memory = max(0.1, lowvram_model_memory - loaded_memory)
lowvram_model_memory = lowvram_model_memory - loaded_memory
if lowvram_model_memory == 0:
lowvram_model_memory = 0.1
if vram_set_state == VRAMState.NO_VRAM:
lowvram_model_memory = 0.1