[s2s] distributed eval allows num_return_sequences > 1 (#7254)

examples/seq2seq/run_eval.py

@@ -13,7 +13,7 @@ import torch
 from tqdm import tqdm
 
 from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
-from utils import calculate_bleu, calculate_rouge, parse_numeric_n_bool_cl_kwargs, use_task_specific_params
+from utils import calculate_bleu, calculate_rouge, chunks, parse_numeric_n_bool_cl_kwargs, use_task_specific_params
 
 
 logger = getLogger(__name__)
@@ -22,12 +22,6 @@ logger = getLogger(__name__)
 DEFAULT_DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
 
-def chunks(lst, n):
-    """Yield successive n-sized chunks from lst."""
-    for i in range(0, len(lst), n):
-        yield lst[i : i + n]
-
-
 def generate_summaries_or_translations(
     examples: List[str],
     out_file: str,