Fix chat_template.jinja: drop hard raise_exception on message order (breaks tool-calling in llama.cpp/LM Studio/Ollama)
Summary
chat_template.jinja raises a hard Jinja raise_exception when a system message is not the first message, and when no user query is found in a multi-step-tool context. These assertions break tool-calling on every runtime that auto-generates a tool-call parser by probing the template and/or injects its own tool-instruction system message — including llama.cpp (--jinja), LM Studio, Ollama, opencode, and Claude Code through any of them.
As soon as a request contains tools, the runtime fails before generating a single token:
400 Unable to generate parser for this template. Automatic parser generation failed:
... raise_exception('System message must be at the beginning...
Error: Jinja Exception: System message must be at the beginning.
Root cause
Tool-calling runtimes do two things this template forbids:
- They probe the template with synthetic message sequences to auto-detect the tool-call format and build a parser/grammar. Some probe sequences do not place a system message first.
- With
--jinja+ tools, llama.cpp appends its own system message (e.g. "Respond in JSON format, either withtool_call… or withresponse…"), which can land as a second / non-leading system message.
Either path hits raise_exception('System message must be at the beginning.') and aborts the whole request.
Importantly, this template builds its tool instructions from the tools parameter (the # Tools / <tools> block), not from a system message — so the runtime's appended system message is redundant and the hard assertion serves no functional purpose at inference time.
Fix
Remove the two hard raise_exception assertions ('System message must be at the beginning.' and 'No user query found in messages.'). For valid inputs — a leading system message and a user query, i.e. the normal case — the rendered output is byte-for-byte identical. This matches how mainstream Qwen-derived instruct templates behave (Qwen3 Instruct, and the widely-used Unsloth Qwen3.6 GGUF templates merge/tolerate leading system messages instead of raising), which is why those models work out-of-the-box in these runtimes and Ornith currently does not.
{%- endfor %}
-{%- if ns.multi_step_tool %}
- {{- raise_exception('No user query found in messages.') }}
-{%- endif %}
{%- for message in messages %}
{%- set content = render_content(message.content, true)|trim %}
{%- if message.role == "system" %}
- {%- if not loop.first %}
- {{- raise_exception('System message must be at the beginning.') }}
- {%- endif %}
{%- elif message.role == "user" %}
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
Verification
jinja2parses the patched template cleanly.- Served the official
Ornith-1.0-35B-GGUF(Q8_0) on llama.cpp build b9611 with the patched template via--chat-template-file. A Claude-Code-style request withtoolsnow returns200with a correct tool call —finish_reason: tool_calls,get_weather({"city": "Paris"}). - Plain chat and reasoning (
<think>) output are unchanged.
Related reports
- llama.cpp
ggml-org/llama.cpp#20733(exact error),#18323(runtime appends a tool-instruction system message),#18895(strict templates blocked by verification) - LM Studio
lmstudio-ai/lmstudio-bug-tracker#1999(Qwen3.6-35B-A3B + Claude Code, same error) - SillyTavern
SillyTavern/SillyTavern#5276(Qwen3.5-122B, same raise)
Alternative
If you'd rather keep strict validation, an equally good fix is to merge leading system messages (as Qwen3 Instruct does) instead of removing the assertion — happy to switch the PR to that approach.
Note: the published
Ornith-1.0-35B-GGUFfiles embed the old template and would need re-quantizing to benefit (until then, users can override at runtime with--chat-template-file). Fixing the source template here makes all future conversions andtransformers/vLLMusers correct.
i replaced the https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B/blob/main/chat_template.jinja to
{%- set image_count = namespace(value=0) %}
{%- set video_count = namespace(value=0) %}
{%- macro render_content(content, do_vision_count, is_system_content=false) %}
{%- if content is string %}
{{- content }}
{%- elif content is iterable and content is not mapping %}
{%- for item in content %}
{%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
{%- if is_system_content %}
{{- raise_exception('System message cannot contain images.') }}
{%- endif %}
{%- if do_vision_count %}
{%- set image_count.value = image_count.value + 1 %}
{%- endif %}
{%- if add_vision_id %}
{{- 'Picture ' ~ image_count.value ~ ': ' }}
{%- endif %}
{{- '<|vision_start|><|image_pad|><|vision_end|>' }}
{%- elif 'video' in item or item.type == 'video' %}
{%- if is_system_content %}
{{- raise_exception('System message cannot contain videos.') }}
{%- endif %}
{%- if do_vision_count %}
{%- set video_count.value = video_count.value + 1 %}
{%- endif %}
{%- if add_vision_id %}
{{- 'Video ' ~ video_count.value ~ ': ' }}
{%- endif %}
{{- '<|vision_start|><|video_pad|><|vision_end|>' }}
{%- elif 'text' in item %}
{{- item.text }}
{%- else %}
{{- raise_exception('Unexpected item type in content.') }}
{%- endif %}
{%- endfor %}
{%- elif content is none or content is undefined %}
{{- '' }}
{%- else %}
{{- raise_exception('Unexpected content type.') }}
{%- endif %}
{%- endmacro %}
{%- if not messages %}
{{- raise_exception('No messages provided.') }}
{%- endif %}
{%- if tools and tools is iterable and tools is not mapping %}
{{- '<|im_start|>system\n' }}
{{- "# Tools\n\nYou have access to the following functions:\n\n" }}
{%- for tool in tools %}
{{- "\n" }}
{{- tool | tojson }}
{%- endfor %}
{{- "\n" }}
{{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n\n\n\n\n\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...> block must be nested within XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n' }}
{%- if messages[0].role == 'system' %}
{%- set content = render_content(messages[0].content, false, true)|trim %}
{%- if content %}
{{- '\n\n' + content }}
{%- endif %}
{%- endif %}
{{- '<|im_end|>\n' }}
{%- else %}
{%- if messages[0].role == 'system' %}
{%- set content = render_content(messages[0].content, false, true)|trim %}
{{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
{%- endif %}
{%- endif %}
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
{%- for message in messages[::-1] %}
{%- set index = (messages|length - 1) - loop.index0 %}
{%- if ns.multi_step_tool and message.role == "user" %}
{%- set content = render_content(message.content, false)|trim %}
{%- if not(content.startswith('') and content.endswith('')) %}
{%- set ns.multi_step_tool = false %}
{%- set ns.last_query_index = index %}
{%- endif %}
{%- endif %}
{%- endfor %}
{%- if ns.multi_step_tool %}
{{- raise_exception('No user query found in messages.') }}
{%- endif %}
{%- for message in messages %}
{%- set content = render_content(message.content, true)|trim %}
{%- if message.role == "system" %}
{# Skip additional system messages because the first one
was already emitted above. This keeps compatibility
with Claude Code and OpenAI-compatible clients. #}
{%- continue %}
{%- elif message.role == "user" %}
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
{%- elif message.role == "assistant" %}
{%- set reasoning_content = '' %}
{%- if message.reasoning_content is string %}
{%- set reasoning_content = message.reasoning_content %}
{%- else %}
{%- if '' in content %}
{%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
{%- set content = content.split('')[-1].lstrip('\n') %}
{%- endif %}
{%- endif %}
{%- set reasoning_content = reasoning_content|trim %}
{{- '<|im_start|>' + message.role + '\n\n' + reasoning_content + '\n\n\n' + content }}
{%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
{%- for tool_call in message.tool_calls %}
{%- if tool_call.function is defined %}
{%- set tool_call = tool_call.function %}
{%- endif %}
{%- if loop.first %}
{%- if content|trim %}
{{- '\n\n\n<function=' + tool_call.name + '>\n' }}
{%- else %}
{{- '\n<function=' + tool_call.name + '>\n' }}
{%- endif %}
{%- else %}
{{- '\n\n<function=' + tool_call.name + '>\n' }}
{%- endif %}
{%- if tool_call.arguments is defined %}
{%- for args_name, args_value in tool_call.arguments|items %}
{{- '<parameter=' + args_name + '>\n' }}
{%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
{{- args_value }}
{{- '\n\n' }}
{%- endfor %}
{%- endif %}
{{- '\n' }}
{%- endfor %}
{%- endif %}
{{- '<|im_end|>\n' }}
{%- elif message.role == "tool" %}
{%- if loop.previtem and loop.previtem.role != "tool" %}
{{- '<|im_start|>user' }}
{%- endif %}
{{- '\n\n' }}
{{- content }}
{{- '\n' }}
{%- if not loop.last and loop.nextitem.role != "tool" %}
{{- '<|im_end|>\n' }}
{%- elif loop.last %}
{{- '<|im_end|>\n' }}
{%- endif %}
{%- else %}
{{- raise_exception('Unexpected message role.') }}
{%- endif %}
{%- endfor %}
{%- if add_generation_prompt %}
{{- '<|im_start|>assistant\n' }}
{%- if enable_thinking is defined and enable_thinking is false %}
{{- '\n\n\n\n' }}
{%- else %}
{{- '\n' }}
{%- endif %}
{%- endif %}
Better use unsloth template, they fix many issues there. I'm using their template via flag and it is working like a charm
https://huggingface.co/unsloth/Qwen3.5-35B-A3B/blob/main/chat_template.jinja
Better use unsloth template, they fix many issues there. I'm using their template via flag and it is working like a charm
https://huggingface.co/unsloth/Qwen3.5-35B-A3B/blob/main/chat_template.jinja
it work well at llamacpp,but how to use in vllm? when I param with vllm serve ....... -chat-template xxxx, it doesn't work
when I param with vllm
I have no idea, never work with vllm