Add editable LLM penalties and preserve sampling through backup restore

endpoint.py

@@ -12,6 +12,7 @@ import threading
 import time
 from http.server import BaseHTTPRequestHandler,ThreadingHTTPServer
 from api_compat import normalize_chat,CompatibilityError
+from profiles import generation_parameters
 from inference import InferenceError
 from stt import read_upload
 
@@ -325,7 +326,7 @@ class APIHandler(BaseHTTPRequestHandler):
         def allowed():return ep.allowed() and any(p['id']==profile['id'] and p['revision']==profile['revision'] and p['enabled'] for p in ep.rows())
         with ep.scheduler.lease(key,profile['parameters']['slots'],lambda:ep.worker.ensure(profile),allowed=allowed):
             body=dict(data)
-            for field in ('temperature','top_p','top_k'):body.setdefault(field,profile['parameters'][field])
+            for field,value in generation_parameters(profile['parameters']).items():body.setdefault(field,value)
             conn,key=ep.worker.connect()
             try:
                 conn.request('POST','/v1/chat/completions',body=json.dumps(body).encode(),headers={'Content-Type':'application/json','Authorization':'Bearer '+key})