Qualcomm AI Engine Direct - Update the evaluator API call for Llama (pytorch#6386)

Author: winskuo-quic

extension/llm/export/builder.py

@@ -219,9 +219,7 @@ def pt2e_calibrate(
             from executorch.examples.models.llama.eval_llama_lib import (
                 GraphModuleEvalWrapper,
             )
-            from executorch.examples.models.llama.evaluate import (  # pyre-ignore[21]
-                evaluate_model,
-            )
+            from lm_eval.evaluator import simple_evaluate  # pyre-ignore[21]
         except ImportError:
             raise ImportError(
                 "Please install the llm eval dependency via examples/models/llama/install_requirements.sh"
@@ -266,11 +264,14 @@ def calibrate_template(
             generate_full_logits=self.generate_full_logits,
             enable_dynamic_shape=self.enable_dynamic_shape,
         )
-        eval_results = evaluate_model(
-            eval_wrapper,
-            calibration_tasks,
-            calibration_limit,
-        )
+
+        # Evaluate the model
+        with torch.no_grad():
+            eval_results = simple_evaluate(
+                model=eval_wrapper,
+                tasks=calibration_tasks,
+                limit=calibration_limit,
+            )
 
         for task, res in eval_results["results"].items():
             print(f"{task}: {res}")