SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Do not remove half seq length in generation tests (#30016)

Browse Source 
* remove seq length from generation tests

* style and quality

* [test_all] & PR suggestion

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Update tests/generation/test_utils.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* [test all] remove unused variables

---------

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
This commit is contained in:
Raushan Turganbay
2024-04-19 21:32:52 +05:00
committed by GitHub
parent b4fd49b6c5
commit b1cd48740e
10 changed files with 180 additions and 261 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/models/reformer/test_modeling_reformer.py
View File

@@ -686,6 +686,18 @@ class ReformerLocalAttnModelTest(ReformerTesterMixin, GenerationTesterMixin, Mod
def test_left_padding_compatibility(self):
pass
def _get_input_ids_and_config(self, batch_size=2):
# override because overwise we hit max possible seq length for model (4*8=32)
# decreasing the seq_length in tester causes errors for "training_tests", those need exactly max seq length
# NOTE: seq_length has to be multiple of 4, otherwise it fails for other tests
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_ids = inputs_dict[self.input_name]
input_ids = input_ids[:batch_size, :16]
attention_mask = torch.ones_like(input_ids, dtype=torch.long)[:batch_size, :16]
config.eos_token_id = None
config.forced_eos_token_id = None
return config, input_ids, attention_mask
@require_torch
class ReformerLSHAttnModelTest(