fix generation bugs

src/backend/run_eval_suite.py

@@ -32,8 +32,8 @@ def run_evaluation(eval_request: EvalRequest, task_names, num_fewshot, batch_siz
 
     print(f"Selected Tasks: {task_names}")
     print(f"Eval Request: {eval_request.get_model_args()}")
-
-    results = evaluator.simple_evaluate(model="hf-chat",  # "hf-causal-experimental",  # "hf-causal"
+    # hf-chat is implemented to use apply_chat_template
+    results = evaluator.simple_evaluate(model="hf-auto",  # "hf-causal-experimental",  # "hf-causal", hf-chat
                                         model_args=eval_request.get_model_args(),
                                         tasks=task_names,
                                         num_fewshot=num_fewshot,