Mirrors/stable-diffusion-webui-forge
1
0
Fork 0
mirror of https://github.com/lllyasviel/stable-diffusion-webui-forge.git synced 2025-12-28 05:35:00 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

unload models before loading llm

Browse Source
This commit is contained in:
maybleMyers 2025-12-06 02:45:46 -08:00
parent 24b7f8be0e
commit 37326512fc
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
modules/ui.py
View File

@ -394,14 +394,14 @@ def expand_prompt_standalone(prompt: str, model_path: str, system_prompt: str =
log_step("Previous model unloaded", unload_start)
# Unload diffusion models from VRAM to make room for LLM
# Use emergency_memory_cleanup which forcibly clears all model references
log_step("Unloading diffusion models from VRAM to make room for LLM...")
unload_diffusion_start = time.time()
try:
from backend import memory_management
memory_management.unload_all_models()
memory_management.soft_empty_cache(force=True)
gc.collect()
torch.cuda.empty_cache()
# emergency_memory_cleanup forcibly clears shared.sd_model and all forge_objects
# This is necessary because unload_all_models() won't unload models with active references
memory_management.emergency_memory_cleanup()
gpu_mem = get_gpu_memory()
if gpu_mem:
log_step(f" After diffusion unload: {gpu_mem}")