From dd6720c528f891f9c13611630f39c0b61893c813 Mon Sep 17 00:00:00 2001
From: Classic298 <27028174+Classic298@users.noreply.github.com>
Date: Tue, 3 Nov 2026 12:44:05 +0100
Subject: [PATCH] fix: non-ASCII stop sequences on a model reach the provider
 corrupted (#33614)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

A Stop Sequence saved on a workspace model reached the provider corrupted when it had non-ASCII text in it (終 went out as çµ\x82, café as café), so the model never stopped on it. This happened when an API client used the model through /openai/chat/completions or /ollama/api/chat. Those stops now arrive unchanged, and typing \n in the field still sends a newline.

Fixes #33610
---
 backend/open_webui/utils/payload.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/backend/open_webui/utils/payload.py b/backend/open_webui/utils/payload.py
index d033fbfaab..f1befc53a8 100644
--- a/backend/open_webui/utils/payload.py
+++ b/backend/open_webui/utils/payload.py
@@ -210,7 +210,7 @@ def apply_model_params_to_body_openai(params: dict, form_data: dict) -> dict:
         'presence_penalty': float,
         'reasoning_effort': str,
         'seed': lambda x: x,
-        'stop': lambda x: [bytes(s, 'utf-8').decode('unicode_escape') for s in x],
+        'stop': lambda x: [stop.encode('latin-1', 'backslashreplace').decode('unicode_escape') for stop in x],
         'logit_bias': lambda x: x,
         'response_format': dict,
     }
@@ -264,7 +264,7 @@ def apply_model_params_to_body_ollama(params: dict, form_data: dict) -> dict:
         'repeat_penalty': float,
         'presence_penalty': float,
         'frequency_penalty': float,
-        'stop': lambda x: [bytes(s, 'utf-8').decode('unicode_escape') for s in x],
+        'stop': lambda x: [stop.encode('latin-1', 'backslashreplace').decode('unicode_escape') for stop in x],
         'num_gpu': int,
         'use_mmap': bool,
         'use_mlock': bool,