fix dali distributed eval bug

ppcls/engine/evaluation/classification.py

@@ -91,7 +91,7 @@ def classification_eval(engine, epoch_id=0):
                     paddle.distributed.all_gather(pred_list, out)
                     pred = paddle.concat(pred_list, 0)
 
-                if accum_samples > total_samples:
+                if accum_samples > total_samples and not engine.use_dali:
                     pred = pred[:total_samples + current_samples -
                                 accum_samples]
                     labels = labels[:total_samples + current_samples -