diff --git a/avise/models/evaluation_lm.py b/avise/models/evaluation_lm.py
index 7524447..53f8458 100644
--- a/avise/models/evaluation_lm.py
+++ b/avise/models/evaluation_lm.py
@@ -191,10 +191,11 @@ def _mistral_text_generation(self, messages: list) -> str:
 
     def del_model(self):
         """Delete the model from GPU memory."""
-        self.model.cpu()
-        del self.model
-        del self.tokenizer
-        torch.cuda.empty_cache()
+        if self.model:
+            self.model.cpu()
+            del self.model
+            del self.tokenizer
+            torch.cuda.empty_cache()
 
     def _model_download(
         self,
diff --git a/avise/pipelines/languagemodel/pipeline.py b/avise/pipelines/languagemodel/pipeline.py
index 969519d..5ebfba6 100644
--- a/avise/pipelines/languagemodel/pipeline.py
+++ b/avise/pipelines/languagemodel/pipeline.py
@@ -161,7 +161,10 @@ def run(
             report_format: Desired output format
             connector_config_path: Path to model configuration (for report metadata)
             generate_ai_summary: Whether to generate AI summary
-            runs: How many times to the SET is ran
+            runs: How many times to run the SET
+
+        Returns:
+            ReportData: The final report with all the SET data
 
         Requirements:
             Return the final report
@@ -172,28 +175,35 @@ def run(
         self.set_config_path = set_config_path
         self.target_model_name = connector.model
 
-        # Initialize
-        sets = self.initialize(set_config_path)
+        try:
+            # Initialize
+            sets = self.initialize(set_config_path)
 
-        results = []
-        # Loop to allow multiple SET runs
-        for run in range(runs):
-            logger.info(f"Starting SET run {run}/{runs}.")
+            results = []
+            # Loop to allow multiple SET runs
+            for run in range(runs):
+                logger.info(f"Starting SET run {run}/{runs}.")
 
-            # Execute
-            execution_data = self.execute(connector, sets)
+                # Execute
+                execution_data = self.execute(connector, sets)
 
-            # Evaluate
-            results += self.evaluate(execution_data)
+                # Evaluate
+                results += self.evaluate(execution_data)
 
-            logger.info(f"SET run {run}/{runs} finished.")
+                logger.info(f"SET run {run}/{runs} finished.")
 
-        # Report
-        report_data = self.report(
-            results, output_path, report_format, generate_ai_summary
-        )
+            # Report
+            report_data = self.report(
+                results, output_path, report_format, generate_ai_summary
+            )
+
+            return report_data
 
-        return report_data
+        finally:
+            if self.evaluation_model:
+                # Clean evaluation model from memory
+                self.evaluation_model.del_model()
+                self.evaluation_model = None
 
     def generate_ai_summary(
         self,
@@ -210,29 +220,25 @@ def generate_ai_summary(
             results: List of EvaluationResult from evaluate()
             summary_stats: Summary statistics from calculate_passrates()
             subcategory_runs: Optional dict of subcategory -> number of runs
-
-        Returns:
-            Dict with ai_summary or None if generation fails
         """
         try:
             from avise.reportgen.summarizers.ai_summarizer import AISummarizer
 
-            model_to_use = None
+            model_to_use = self.evaluation_model
             if hasattr(self, "evaluation_model") and self.evaluation_model is not None:
                 logger.info("Reusing existing evaluation model for AI summary")
-                model_to_use = self.evaluation_model
             else:
                 logger.info(
-                    "Creating new model for AI summary (CPU mode due to memory constraints)"
+                    "Creating new model for AI summary (no existing evaluation model)"
                 )
-                model_to_use = None
 
             summarizer = AISummarizer(reuse_model=model_to_use)
             results_dict = [r.to_dict() for r in results]
             ai_summary = summarizer.generate_summary(
                 results_dict, summary_stats, subcategory_runs
             )
-
+            # Delete model from memory
+            summarizer.del_model()
             return {
                 "issue_summary": ai_summary.issue_summary,
                 "recommended_remediations": ai_summary.recommended_remediations,
diff --git a/avise/reportgen/reporters/html_reporter.py b/avise/reportgen/reporters/html_reporter.py
index 9caa274..6f307f8 100644
--- a/avise/reportgen/reporters/html_reporter.py
+++ b/avise/reportgen/reporters/html_reporter.py
@@ -26,7 +26,7 @@ def write(self, report_data: ReportData, output_path: Path) -> None:
             output_path: Path to the output file / directory
         """
         html = self._generate_html(report_data)
-        with open(output_path, "w") as f:
+        with open(output_path, "w", encoding="utf-8") as f:
             f.write(html)
 
     def _generate_html(self, report_data: ReportData) -> str:
diff --git a/avise/reportgen/summarizers/ai_summarizer.py b/avise/reportgen/summarizers/ai_summarizer.py
index 9c0980c..45c326c 100644
--- a/avise/reportgen/summarizers/ai_summarizer.py
+++ b/avise/reportgen/summarizers/ai_summarizer.py
@@ -35,16 +35,13 @@ def __init__(
         from avise.models.evaluation_lm import EvaluationLanguageModel
 
         if reuse_model is not None:
-            logger.info("Reusing existing evaluation model for AI summary")
             self.model = reuse_model
-            self._owns_model = False
         else:
             logger.info("Loading AI summarizer model...")
             self.model = EvaluationLanguageModel(
                 model_name=evaluation_model_name,
                 max_new_tokens=max_new_tokens,
             )
-            self._owns_model = True
 
     def generate_summary(
         self,
@@ -292,15 +289,6 @@ def _format_results_for_prompt(
 
         return "\n".join(lines)
 
-    def cleanup(self):
-        """Clean up the model from memory."""
-        if self.model and self._owns_model:
-            logger.info("Cleaning up AI summarizer model...")
-            self.model.del_model()
-            self.model = None
-        elif self.model:
-            logger.info("Skipping cleanup - model is shared with evaluation")
-
 
 def format_json_ai_summary(ai_summary: AISummary) -> Dict[str, Any]:
     """Format AI summary for JSON report output.