Add type conversion and debug logging for config values

src/reasoning/step_level_cot.py

@@ -97,19 +97,31 @@ class StepLevelCoTTrainer:
         )
         
         # Initialize inference scaler
-        inference_config = InferenceConfig(
-            num_samples=self.config.num_inference_samples,
-            temperature=self.config.inference_temperature,
-            aggregation=self.config.aggregation_method,
-            use_prm_scores=True,
-        )
-        
-        self.inference_scaler = InferenceTimeScaling(
-            model=self.model,
-            prm=self.prm,
-            config=inference_config,
-            device=device,
-        )
+        # Ensure numeric types are properly converted
+        try:
+            num_samples = int(self.config.num_inference_samples)
+            temperature = float(self.config.inference_temperature)
+            aggregation = str(self.config.aggregation_method)
+            
+            inference_config = InferenceConfig(
+                num_samples=num_samples,
+                temperature=temperature,
+                aggregation=aggregation,
+                use_prm_scores=True,
+            )
+            
+            self.inference_scaler = InferenceTimeScaling(
+                model=self.model,
+                prm=self.prm,
+                config=inference_config,
+                device=device,
+            )
+        except Exception as e:
+            logger.error(f"Failed to initialize inference scaler: {e}")
+            logger.error(f"Config values - num_samples: {self.config.num_inference_samples} (type: {type(self.config.num_inference_samples)})")
+            logger.error(f"Config values - temperature: {self.config.inference_temperature} (type: {type(self.config.inference_temperature)})")
+            logger.error(f"Config values - aggregation: {self.config.aggregation_method} (type: {type(self.config.aggregation_method)})")
+            raise
         
         logger.info("StepLevelCoTTrainer initialized")