Broaden client-neutral OpenAI and llama.cpp chat compatibility

test_endpoint.py

@@ -112,6 +112,20 @@ class EndpointTests(unittest.TestCase):
    self.assertEqual(self.request(route,token='bad')[0],401)
   self.rows[-1]['runnable']=False
   self.assertEqual(self.request('/v1/audio/transcriptions/models')[1]['data'],[])
+ def test_extended_compatibility_forwarding(self):
+  self.enable('alpha')
+  data=dict(model='alpha',messages=[dict(role='user',content='synthetic')],max_completion_tokens=8,extra_body={'min_p':.05},metadata={'project':'synthetic'},store=False,functions=[{'name':'echo','parameters':{'type':'object'}}],function_call='auto',chat_template_kwargs={'enable_thinking':False})
+  status,_=self.request('/v1/chat/completions',data)
+  self.assertEqual(status,200)
+  forwarded=self.worker.requests[-1]
+  self.assertEqual(forwarded['max_tokens'],8)
+  self.assertEqual(forwarded['min_p'],.05)
+  self.assertEqual(forwarded['tools'][0]['function']['name'],'echo')
+  self.assertNotIn('metadata',forwarded)
+  status,body=self.request('/v1/capabilities')
+  self.assertEqual(status,200)
+  self.assertIn('min_p',body['llama_extensions'])
+
  def test_reasoning_effort_forwarded_and_validated(self):
   self.enable('alpha')
   for effort in ('none','minimal','low','medium','high','xhigh','max','ultra',None):