michael/Athena-Deck
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add configurable MTP and shared-weight draft memory fitting

Browse Source
This commit is contained in:
Mikei386
2026-09-28 20:59:43 +02:00
parent ec62c859f7
commit 13ed059aaf
7 changed files with 87 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
profiles.py
+5 -4
View File
@@ -7,16 +7,17 @@ import uuid
from pathlib import Path
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)},
'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)},
'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)},
'video':{'width':(256,1920,768),'height':(256,1088,512),'frames':(1,241,33),'fps':(1,60,24),'steps':(1,100,20),'seed':(-1,2147483647,-1)}
}
CHAT_GPU_DEFAULTS={'gpu_devices':[], 'split_mode':'none', 'tensor_split':[]}
CHAT_SAMPLING={'temperature':.8,'top_p':.95,'top_k':40}
CHAT_GPU_DEFAULTS={'gpu_devices':[], 'split_mode':'none', 'tensor_split':[], 'mtp':False}
CHAT_SAMPLING={'temperature':.8,'top_p':.95,'top_k':40,'mtp_tokens':2,'mtp_min_p':.05}
def chat_parameters(params):
params={**CHAT_GPU_DEFAULTS,**CHAT_SAMPLING,**params}
if type(params['mtp']) is not bool:raise ValueError('MTP muss an oder aus sein.')
devices=params['gpu_devices'];split=params['tensor_split']
if not isinstance(devices,list) or len(devices)>16 or any(not isinstance(v,str) or not re.fullmatch(r'GPU-[0-9a-fA-F-]{36}',v) for v in devices) or len(set(devices))!=len(devices):raise ValueError('GPUs müssen als eindeutige, geordnete GPU-UUIDs angegeben werden.')
if params['split_mode'] not in ('none','layer','row'):raise ValueError('Ungültiger GPU-Split-Modus.')
@@ -68,7 +69,7 @@ class Profiles:
if not isinstance(params,dict) or set(params)!=(set(SCHEMAS[kind]) | (set(CHAT_GPU_DEFAULTS) if kind=='chat' 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')
floating=key in ('guidance','speed','temperature','top_p','mtp_min_p')
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.')