Add editable LLM penalties and preserve sampling through backup restore
This commit is contained in:
1 parent
3779c846db
commit
37b5394df1
14 files changed
+218
-14
No files matched your search
+8
-3
@@ -8,7 +8,7 @@ from pathlib import Path
|
||||
from catalog import file_role
|
||||
|
||||
SCHEMAS={
|
||||
'chat':{'context':(512,2097152,8192),'slots':(1,16,1),'threads':(1,256,6),'batch':(1,8192,512),'ubatch':(1,8192,128),'temperature':(0,5,.8),'top_p':(0,1,.95),'top_k':(0,1000,40),'mtp_tokens':(1,8,2),'mtp_min_p':(0,1,.05)},
|
||||
'chat':{'context':(512,2097152,8192),'slots':(1,16,1),'threads':(1,256,6),'batch':(1,8192,512),'ubatch':(1,8192,128),'temperature':(0,5,.8),'top_p':(0,1,.95),'top_k':(0,1000,40),'repeat_penalty':(0,2,1.0),'presence_penalty':(-2,2,0.0),'frequency_penalty':(-2,2,0.0),'mtp_tokens':(1,8,2),'mtp_min_p':(0,1,.05)},
|
||||
'image':{'width':(256,2048,1024),'height':(256,2048,1024),'steps':(1,100,25),'seed':(-1,2147483647,-1),'guidance':(0,30,1)},
|
||||
'audio':{'speed':(.25,4,1)},
|
||||
'stt':{},'music':{},'voice':{},
|
||||
@@ -26,7 +26,12 @@ def video_parameters(params):
|
||||
return params
|
||||
|
||||
CHAT_GPU_DEFAULTS={'gpu_devices':[], 'split_mode':'none', 'tensor_split':[], 'mtp':False,'gpu_offload':'auto','gpu_reserve_mode':'auto','gpu_reserve_mib':{},'vision_projector':None,'vision_device':'cpu','cache_type_k':'q4_0','cache_type_v':'q4_0'}
|
||||
CHAT_SAMPLING={'temperature':.8,'top_p':.95,'top_k':40,'mtp_tokens':2,'mtp_min_p':.05}
|
||||
FLOAT_PARAMETERS=frozenset({'guidance','speed','temperature','top_p','mtp_min_p','repeat_penalty','presence_penalty','frequency_penalty'})
|
||||
CHAT_GENERATION_DEFAULTS={'temperature':.8,'top_p':.95,'top_k':40,'repeat_penalty':1.0,'presence_penalty':0.0,'frequency_penalty':0.0}
|
||||
CHAT_SAMPLING={**CHAT_GENERATION_DEFAULTS,'mtp_tokens':2,'mtp_min_p':.05}
|
||||
|
||||
def generation_parameters(params):
|
||||
return {key:params.get(key,default) for key,default in CHAT_GENERATION_DEFAULTS.items()}
|
||||
|
||||
def chat_parameters(params):
|
||||
params={**CHAT_GPU_DEFAULTS,**CHAT_SAMPLING,**params}
|
||||
@@ -185,7 +190,7 @@ class Profiles:
|
||||
if not isinstance(params,dict) or set(params)!=(set(SCHEMAS[kind]) | (set(CHAT_GPU_DEFAULTS) if kind=='chat' else set(VIDEO_DEVICE_DEFAULTS) if kind=='video' else set())):raise ValueError('Unvollständige oder unbekannte Profilparameter.')
|
||||
for key,(lo,hi,_) in SCHEMAS[kind].items():
|
||||
value=params[key]
|
||||
floating=key in ('guidance','speed','temperature','top_p','mtp_min_p')
|
||||
floating=key in FLOAT_PARAMETERS
|
||||
if isinstance(value,bool) or not isinstance(value,(float,int) if floating else int) or not lo<=value<=hi:raise ValueError('Ungültiger Parameter: '+key)
|
||||
if kind in ('image','video') and (params['width']%64 or params['height']%64):raise ValueError('Breite und Höhe müssen durch 64 teilbar sein.')
|
||||
if kind=='chat' and params['ubatch']>params['batch']:raise ValueError('Microbatch darf nicht größer als Batch sein.')
|
||||
|
||||
Reference in new issue
Block a user