fix: enable qwen thinking with larger token budget

.gitignore

@@ -178,4 +178,7 @@ cython_debug/
 #  exclude from AI features like autocomplete and code analysis. Recommended for sensitive data
 #  refer to https://docs.cursor.com/context/ignore-files
 .cursorignore
-.cursorindexingignore
+.cursorindexingignore
+
+# macOS
+.DS_Store

model_runtime.py

@@ -21,7 +21,7 @@ from schemas import AnalysisResult
 
 PRIMARY_MODEL_ID = "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16"
 QUICK_MODEL_ID = "Qwen/Qwen3.5-9B"
-MODEL_MAX_NEW_TOKENS = 4096
+MODEL_MAX_NEW_TOKENS = 8192
 
 MODEL_CHOICES = {
     "qwen": {
@@ -164,7 +164,7 @@ def _chat_template_kwargs(model_id: str) -> dict[str, Any]:
     """Model-specific chat-template controls."""
 
     if model_id.startswith("Qwen/"):
-        return {"enable_thinking": False}
+        return {"enable_thinking": True}
     return {}
 
 

tests/test_model_runtime.py

@@ -114,8 +114,8 @@ class ModelRuntimeTests(unittest.TestCase):
         self.assertEqual(input_ids.device, "cuda")
         self.assertEqual(attention_mask.device, "cuda")
 
-    def test_qwen_chat_template_disables_thinking(self) -> None:
-        self.assertEqual(_chat_template_kwargs(QUICK_MODEL_ID), {"enable_thinking": False})
+    def test_qwen_chat_template_enables_thinking(self) -> None:
+        self.assertEqual(_chat_template_kwargs(QUICK_MODEL_ID), {"enable_thinking": True})
         self.assertEqual(_chat_template_kwargs(PRIMARY_MODEL_ID), {})
 
     def test_analyzer_records_unknown_engine_note(self) -> None: