From 389e13792181f7ec55eaffe3408ee5e4a0fa1e9b Mon Sep 17 00:00:00 2001
From: Mengxi-Xu <109797814+Mengxi-Xu@users.noreply.github.com>
Date: Fri, 6 Nov 2026 15:45:45 +0800
Subject: [PATCH] Improve error handling in prompt rendering
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Jinja2ChatFormatter renders chat templates from model metadata using jinja2’s sandboxed environment. When a template calls range() with more than MAX_RANGE (100000) items, the sandbox raises OverflowError — which is not a TemplateError subclass. Since __call__ did not catch it, the exception escaped and crashed the caller (e.g. every POST /v1/chat/completions request for a model whose GGUF embeds such a template returned a raw traceback instead of a clean error).

Fix: wrap the render(...) call and translate OverflowError into ValueError, matching the error contract already used by the raise_exception helper in the same method. The sandbox behavior itself is unchanged — only the error surface is normalized
---
 llama_cpp/llama_chat_format.py | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/llama_cpp/llama_chat_format.py b/llama_cpp/llama_chat_format.py
index 62e236415..3fc4f8c41 100644
--- a/llama_cpp/llama_chat_format.py
+++ b/llama_cpp/llama_chat_format.py
@@ -266,20 +266,22 @@ def __call__(
     ) -> ChatFormatterResponse:
         def raise_exception(message: str):
             raise ValueError(message)
-
-        prompt = self._environment.render(
-            messages=messages,
-            eos_token=self.eos_token,
-            bos_token=self.bos_token,
-            raise_exception=raise_exception,
-            add_generation_prompt=self.add_generation_prompt,
-            functions=functions,
-            function_call=function_call,
-            tools=tools,
-            tool_choice=tool_choice,
-            strftime_now=self.strftime_now,
-            **kwargs,
-        )
+        try:
+            prompt = self._environment.render(
+                messages=messages,
+                eos_token=self.eos_token,
+                bos_token=self.bos_token,
+                raise_exception=raise_exception,
+                add_generation_prompt=self.add_generation_prompt,
+                functions=functions,
+                function_call=function_call,
+                tools=tools,
+                tool_choice=tool_choice,
+                strftime_now=self.strftime_now,
+                **kwargs,
+            )
+        except OverflowError as e:
+            raise ValueError(str(e)) from e
 
         stopping_criteria = None
         if self.stop_token_ids is not None: