16 lines
1.3 KiB
Python
16 lines
1.3 KiB
Python
import pathlib,hashlib,json,shutil,subprocess,datetime
|
|
root=pathlib.Path('/home/ps/huobao-gpu-selection');manifest=json.loads((root/'panel-routing-manifest.json').read_text())
|
|
old=""" if any(d.get('torch_vram_total', 0) > 1024**3 for d in stats.json().get('devices', [])):
|
|
return None"""
|
|
new=""" if any(d.get('torch_vram_total', 0) > 1024**3 for d in stats.json().get('devices', [])):
|
|
# Release only idle model caches; queue contents and results stay intact.
|
|
released = await client.post(f'{legacy.base_url}/free', json={'unload_models': True, 'free_memory': True}, headers=provider_headers(legacy))
|
|
released.raise_for_status()
|
|
return None"""
|
|
for short,name in [('general','canvas-h3-prompt-normalizer-healthy-20261026'),('vdn','huobao-vdn-normalizer')]:
|
|
p=root/(name+'.candidate.py');assert hashlib.sha256(p.read_bytes()).hexdigest()==manifest[short]['after']
|
|
source=p.read_text();assert source.count(old)==1
|
|
shutil.copy2(p,root/(name+'.before-idle-release-'+datetime.datetime.now().strftime('%Y%m%d-%H%M%S')+'.py'))
|
|
p.write_text(source.replace(old,new));subprocess.run(['docker','restart','--timeout','60',name],check=True,stdout=subprocess.DEVNULL)
|
|
print('Idle cache release enabled:',name)
|