Gracefully handle new release of lm_eval

Author: shaahji

olive/evaluator/lmeval_ort.py

@@ -13,10 +13,15 @@
 import torch.nn.functional as F
 from lm_eval.api.model import TemplateLM
 from lm_eval.api.registry import register_model
-from lm_eval.models.utils import Collator, pad_and_concat
+from lm_eval.models.utils import Collator
 from tqdm import tqdm
 from transformers import AutoConfig, AutoTokenizer
 
+try:
+    from lm_eval.models.utils_hf import pad_and_concat  # pylint: disable=ungrouped-imports
+except ImportError:
+    from lm_eval.models.utils import pad_and_concat
+
 from olive.common.onnx_io import get_io_config, get_io_dtypes, get_kv_info
 from olive.common.utils import cleanup_memory