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

Tests: move generate tests to the right mixin and delete redundant tests (#34464)

Browse Source 
* tmp commit

* tmp commit

* cull overwrites of deleted tests

* typo

* more specific docstring

* make fixup

* parameterize at the top?

* correction

* more deletions :D

* tmp commit

* for VLMs too

* fix _check_outputs

* test nit

* make fixup

* fix another flaky

* test_generate_from_inputs_embeds -- handle missing attention mask
This commit is contained in:
Joao Gante
2024-10-30 10:59:08 +00:00
committed by GitHub
parent 913330ca9f
commit 8a734ea2c3
46 changed files with 265 additions and 2348 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
tests/models/whisper/test_modeling_whisper.py
View File

@@ -1574,59 +1574,6 @@ class WhisperModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
)
assert isinstance(pred_ids, expected_output_type)
@require_flash_attn
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_generate_reuse_cache(self):
max_new_tokens = 2
for model_class in self.all_generative_model_classes:
if not model_class._supports_flash_attn_2:
self.skipTest(f"{model_class.__name__} does not support Flash Attention 2")
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
dummy_input = inputs_dict[model_class.main_input_name][..., :10]
if dummy_input.dtype in [torch.float32, torch.bfloat16]:
dummy_input = dummy_input.to(torch.float16)
# make sure that all models have enough positions for generation
if hasattr(config, "max_position_embeddings"):
config.max_position_embeddings = dummy_input.shape[1] * 2 + max_new_tokens * 2 + 1
model = model_class(config)
with tempfile.TemporaryDirectory() as tmpdirname:
model.save_pretrained(tmpdirname)
model = model_class.from_pretrained(
tmpdirname,
torch_dtype=torch.float16,
attn_implementation="flash_attention_2",
low_cpu_mem_usage=True,
).to(torch_device)
# run generate once to get filled cache
output = model.generate(
dummy_input,
max_new_tokens=max_new_tokens,
do_sample=False,
use_cache=True,
return_dict_in_generate=True,
)
past_key_values = output.past_key_values
# Try to continue generation from where we left, given that we have more than 1 new token to process
# e.g. this can happen in speculative decoding when feeding candidate tokens back to target model
_ = model.generate(
dummy_input,
decoder_input_ids=output.sequences,
max_new_tokens=max_new_tokens,
do_sample=False,
use_cache=True,
past_key_values=past_key_values,
)
def test_labels_sequence_max_length_correct(self):
config, input_dict = self.model_tester.prepare_config_and_inputs_for_common()
@@ -3961,11 +3908,6 @@ class WhisperStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin,
# generate only works with input ids for whisper
pass
@unittest.skip(reason="Generate needs input ids")
def test_inputs_embeds_matches_input_ids_with_generate(self):
# generate only works with input ids for whisper
pass
@unittest.skip(reason="Decoder can't keep attention grads")
def test_retain_grad_hidden_states_attentions(self):
return
@@ -3974,18 +3916,6 @@ class WhisperStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin,
def test_save_load_fast_init_from_base(self):
pass
@unittest.skip(
reason="FA2 testing suite needs to be refactored to be compatible with WhisperDecoder for that test"
)
def test_flash_attn_2_generate_reuse_cache(self):
pass
@unittest.skip(
"Duplicated test with WhisperModelTest + the FA2 testing suite needs to be refactored to be compatible with WhisperDecoder for that test"
)
def test_flash_attn_2_generate_padding_right(self):
pass
@unittest.skip(
"Duplicated test with WhisperModelTest + the FA2 testing suite needs to be refactored to be compatible with WhisperDecoder for that test"
)