Jinja Exception with the Qwen 3.5 models family

[k33g]

With a qwen 2.5 model I can do this:

curl http://localhost:12434/engines/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
      "model": "qwen2.5:3B-F16",
      "stream": true,
      "messages": [
          {"role": "system", "content": "you are a helpful assistant."},
          {"role": "user", "content": "hello, how are you?"},
          {"role": "system", "content": "docker model runner rocks"},
          {"role": "user", "content": "what is docker model runner?"}
      ]
  }'

With a qwen 3.5 model, If I try the same:

curl http://localhost:12434/engines/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
      "model": "huggingface.co/unsloth/qwen3.5-4b-gguf:Q4_K_M",
      "stream": true,
      "messages": [
          {"role": "system", "content": "you are a helpful assistant."},
          {"role": "user", "content": "hello, how are you?"},
          {"role": "system", "content": "docker model runner rocks"},
          {"role": "user", "content": "what is docker model runner?"}
      ]
  }'

I get this error:

{"error":{"code":500,"message":"\n------------\nWhile executing CallExpression at line 85, column 32 in source:\n...first %}↵            {{- raise_exception('System message must be at the beginnin...\n                                           ^\nError: Jinja Exception: System message must be at the beginning.","type":"server_error"}}

the chat template contains this:

{%- if message.role == "system" %}
  {%- if not loop.first %}
    {{- raise_exception('System message must be at the beginning.') }}
  {%- endif %}

When the client sends a system message anywhere other than position [0] in the messages array, the template raises an exception and the server returns 500.

[ericcurtin]

@k33g if on mac could you run this against the latest llama.cpp :

llama-server -hf unsloth/qwen3.5-4b-gguf:Q4_K_M

or

llama-server --jinja -hf unsloth/qwen3.5-4b-gguf:Q4_K_M

if it's still an error, let's log a llama.cpp bug.

[ericcurtin]

It's an illegal request, this is how you should do this:

curl http://localhost:12434/engines/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
      "model": "huggingface.co/unsloth/qwen3.5-4b-gguf:Q4_K_M",
      "stream": true,
      "messages": [
          {"role": "system", "content": "you are a helpful assistant."},
          {"role": "user", "content": "hello, how are you?"},
          {"role": "assistant", "content": "docker model runner rocks"},
          {"role": "user", "content": "what is docker model runner?"}
      ]
  }'