renaming

benchmark_llama.py

@@ -45,7 +45,7 @@ for batch_size in (1, 16):
         times = []
         for _ in range(3):
             start_time = time.time()
-            out = model.generate(input_ids=input_ids, max_new_tokens=256)
+            out = model.generate(input_ids=input_ids, max_new_tokens=256, do_sample=False)
             times.append(time.time() - start_time)
 
         result = min(times)