fix small issue in test

src/transformers/modeling_reformer.py

@@ -1560,7 +1560,8 @@ def forward(
         # if needs padding
         least_common_mult_chunk_length = _get_least_common_mult_chunk_len(self.config)
         has_to_pad_to_match_chunk_length = input_shape[-1] % least_common_mult_chunk_length != 0
-        if has_to_pad_to_match_chunk_length is True:
+
+        if has_to_pad_to_match_chunk_length:
             # pad input
             input_ids, inputs_embeds, attention_mask, position_ids, input_shape = self._pad_to_mult_of_chunk_length(
                 input_ids,
@@ -1585,7 +1586,7 @@ def forward(
         sequence_output = encoder_outputs.hidden_states
 
         # if padding was applied
-        if has_to_pad_to_match_chunk_length is True:
+        if has_to_pad_to_match_chunk_length:
             sequence_output = sequence_output[:, :orig_sequence_length]
 
         outputs = (sequence_output,)

src/transformers/modeling_utils.py

@@ -789,7 +789,6 @@ def generate(
         attention_mask=None,
         decoder_start_token_id=None,
         use_cache=None,
-        num_hashes=None,
         **model_specific_kwargs
     ):
         r""" Generates sequences for models with a LM head. The method currently supports greedy decoding, beam-search decoding, sampling with temperature, sampling with top-k or nucleus sampling.

tests/test_modeling_reformer.py

@@ -1111,7 +1111,9 @@ def test_pretrained_generate_crime_and_punish(self):
         model.eval()
 
         input_ids = tokenizer.encode("A few months later", return_tensors="pt").to(torch_device)
-        output_ids = model.generate(input_ids, max_length=50, num_beams=4, early_stopping=True, do_sample=False)
+        output_ids = model.generate(
+            input_ids, max_length=50, num_beams=4, early_stopping=True, do_sample=False, num_hashes=8
+        )
         output_text = tokenizer.decode(output_ids[0])
         self.assertEqual(
             output_text,