Enable 2x faster inference with unsloth

sia/llm_engine/qwq_llm_engine.py

@@ -44,6 +44,9 @@ class QwQLlmEngine(LlmEngine):
             tokenizer = self._tokenizer,
         )
 
+        # enable unsloth optimizations
+        FastLanguageModel.for_inference(self._model)
+        
         # Create inference pipeline
         self._pipeline = pipeline(
             "text-generation",