From c9036201f70a66d5bc86c01b5fd799ca6ac168ab Mon Sep 17 00:00:00 2001
From: Mikei386 <44135113+Mikei386@users.noreply.github.com>
Date: Fri, 2 Oct 2026 10:21:26 +0200
Subject: [PATCH] Accept validated thinking template options in chat
 compatibility layer

---
 api_compat.py      | 9 ++++++++-
 test_api_compat.py | 9 +++++++++
 2 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/api_compat.py b/api_compat.py
index 2486520..cf5b583 100644
--- a/api_compat.py
+++ b/api_compat.py
@@ -5,7 +5,7 @@ positive levels to the model template; it handles 'none' as thinking disabled.
 """
 import math
 
-CHAT_FIELDS=frozenset({'model','messages','stream','stream_options','temperature','top_p','top_k','max_tokens','max_completion_tokens','stop','seed','tools','tool_choice','parallel_tool_calls','response_format','repeat_penalty','presence_penalty','frequency_penalty','logprobs','top_logprobs','user','n','reasoning_effort'})
+CHAT_FIELDS=frozenset({'model','messages','stream','stream_options','temperature','top_p','top_k','max_tokens','max_completion_tokens','stop','seed','tools','tool_choice','parallel_tool_calls','response_format','repeat_penalty','presence_penalty','frequency_penalty','logprobs','top_logprobs','user','n','reasoning_effort','chat_template_kwargs'})
 LLAMA_EFFORTS=frozenset({'none','low','medium','high','xhigh'})
 # The installed llama.cpp server rejects minimal and max. Use nearest supported hints.
 EFFORT_ALIASES={'minimal':'low','max':'xhigh','ultra':'xhigh'}
@@ -17,6 +17,13 @@ def normalize_chat(data):
  unknown=set(data)-CHAT_FIELDS
  if unknown:raise CompatibilityError('Nicht unterstützte Chat-Felder: '+', '.join(sorted(unknown)))
  body=dict(data)
+ if 'chat_template_kwargs' in body:
+  kwargs=body['chat_template_kwargs']
+  if not isinstance(kwargs,dict) or set(kwargs)-{'enable_thinking','preserve_thinking'}:
+   raise CompatibilityError('chat_template_kwargs unterstützt enable_thinking und preserve_thinking.')
+  if any(type(v) is not bool for v in kwargs.values()):
+   raise CompatibilityError('Thinking-Template-Optionen müssen true oder false sein.')
+  body['chat_template_kwargs']=dict(kwargs)
  for key,lo,hi in [('repeat_penalty',0,2),('presence_penalty',-2,2),('frequency_penalty',-2,2)]:
   if key in body:
    value=body[key]
diff --git a/test_api_compat.py b/test_api_compat.py
index 3f15ac2..6e3ece1 100644
--- a/test_api_compat.py
+++ b/test_api_compat.py
@@ -14,3 +14,12 @@ class CompatibilityTests(unittest.TestCase):
   for value in [True,1,{},[], '', 'MAX', 'x\r\ny']:
    with self.assertRaises(CompatibilityError):normalize_chat({'reasoning_effort':value})
   with self.assertRaises(CompatibilityError):normalize_chat({'unknown':1})
+
+ def test_template_thinking_options(self):
+  original={'chat_template_kwargs':{'enable_thinking':False,'preserve_thinking':True}}
+  body,_=normalize_chat(original)
+  self.assertEqual(body,original)
+  body['chat_template_kwargs']['enable_thinking']=True
+  self.assertFalse(original['chat_template_kwargs']['enable_thinking'])
+  for bad in [None,[],{'enable_thinking':'false'},{'template':'custom'},{'preserve_thinking':1}]:
+   with self.assertRaises(CompatibilityError):normalize_chat({'chat_template_kwargs':bad})