fix(server): hoist system messages to front for templates requiring system-first

python/freetoken/server/generation.py

@@ -194,8 +194,27 @@ def pick(value, key, framework):
 def render_messages(messages: list[dict[str, Any]]) -> list[dict[str, Any]]:
     """Normalize OpenAI-shaped message dicts for the chat template: flatten text
     content parts to a string and decode tool-call arguments from JSON. Raises
-    ValueError on a non-text content part (text-only server). Shared by all adapters."""
-    return [_render_message(m) for m in messages]
+    ValueError on a non-text content part (text-only server). Shared by all adapters.
+
+    Some chat templates (e.g. Qwen3.6) require the system message at index 0;
+    hoist system messages to the front and merge multiples into one to satisfy
+    that constraint."""
+    rendered = [_render_message(m) for m in messages]
+    system_msgs = [m for m in rendered if m.get("role") == "system"]
+    if system_msgs and rendered[0].get("role") != "system":
+        non_system = [m for m in rendered if m.get("role") != "system"]
+        rendered = system_msgs + non_system
+    # Merge multiple system messages into one (Qwen3.6 template rejects
+    # system messages that are not at loop.first, i.e. after the first).
+    system_msgs = [m for m in rendered if m.get("role") == "system"]
+    if len(system_msgs) > 1:
+        sep = chr(10) + chr(10)
+        merged_content = sep.join(
+            str(m.get("content") or "") for m in system_msgs if m.get("content")
+        )
+        non_system = [m for m in rendered if m.get("role") != "system"]
+        rendered = [{"role": "system", "content": merged_content}] + non_system
+    return rendered
 
 
 def _render_message(message: dict[str, Any]) -> dict[str, Any]:

tests/server/test_mm_input.py

@@ -140,3 +140,36 @@ def test_image_token_budget_flags_land_in_the_multimodal_config():
         assert parse_args(["--model", "/models/anon"])[0].mm.processor_kwargs == {}
         with pytest.raises(SystemExit):  # argparse reports the bad pair and exits
             parse_args(["--model", "/models/anon", "--image-min-tokens", "2048", "--image-max-tokens", "1024"])
+
+
+def test_render_messages_hoists_system_to_front():
+    """render_messages moves system messages to index 0 for templates that
+    require it (e.g. Qwen3.6 'System message must be at the beginning')."""
+    msgs = render_messages([
+        {"role": "user", "content": "What is 2+2?"},
+        {"role": "system", "content": "You are a math tutor."},
+        {"role": "user", "content": "Answer briefly."},
+    ])
+    assert msgs[0]["role"] == "system"
+    assert msgs[1]["role"] == "user"
+    assert msgs[2]["role"] == "user"
+
+
+def test_render_messages_preserves_system_first():
+    """When system is already first, no reordering happens."""
+    msgs = render_messages([
+        {"role": "system", "content": "You are helpful."},
+        {"role": "user", "content": "Hi"},
+    ])
+    assert msgs[0]["role"] == "system"
+    assert msgs[1]["role"] == "user"
+
+
+def test_render_messages_no_system_unchanged():
+    """No system message → order preserved."""
+    msgs = render_messages([
+        {"role": "user", "content": "Hello"},
+        {"role": "assistant", "content": "Hi there"},
+        {"role": "user", "content": "Bye"},
+    ])
+    assert [m["role"] for m in msgs] == ["user", "assistant", "user"]