lllyasviel/stable-diffusion-webui-forge
1
0
Fork 0
mirror of https://github.com/lllyasviel/stable-diffusion-webui-forge.git synced 2026-03-13 17:09:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Significantly reduce thread abuse for faster model moving

Browse Source 
This will move all major gradio calls into the main thread rather than random gradio threads.
This ensures that all torch.module.to() are performed in main thread to completely possible avoid GPU fragments.
In my test now model moving is 0.7 ~ 1.2 seconds faster, which means all 6GB/8GB VRAM users will get 0.7 ~ 1.2 seconds faster per image on SDXL.
This commit is contained in:
lllyasviel
2024-02-08 10:13:59 -08:00
parent 291ec743b6
commit f06ba8e60b
8 changed files with 122 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
modules/initialize_util.py
View File

@@ -170,10 +170,11 @@ def configure_sigint_handler():
def configure_opts_onchange():
from modules import shared, sd_models, sd_vae, ui_tempdir, sd_hijack
from modules.call_queue import wrap_queued_call
from modules_forge import main_thread
shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: sd_models.reload_model_weights()), call=False)
shared.opts.onchange("sd_vae", wrap_queued_call(lambda: sd_vae.reload_vae_weights()), call=False)
shared.opts.onchange("sd_vae_overrides_per_model_preferences", wrap_queued_call(lambda: sd_vae.reload_vae_weights()), call=False)
shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: main_thread.run_and_wait_result(sd_models.reload_model_weights)), call=False)
shared.opts.onchange("sd_vae", wrap_queued_call(lambda: main_thread.run_and_wait_result(sd_vae.reload_vae_weights)), call=False)
shared.opts.onchange("sd_vae_overrides_per_model_preferences", wrap_queued_call(lambda: main_thread.run_and_wait_result(sd_vae.reload_vae_weights)), call=False)
shared.opts.onchange("temp_dir", ui_tempdir.on_tmpdir_changed)
shared.opts.onchange("gradio_theme", shared.reload_gradio_theme)
shared.opts.onchange("cross_attention_optimization", wrap_queued_call(lambda: sd_hijack.model_hijack.redo_hijack(shared.sd_model)), call=False)