[Perceiver] Skip multi-gpu tests for now (#14813)
* [Perceiver] Skip multi-gpu tests for now * Update tests/test_modeling_perceiver.py * up * up
This commit is contained in:
committed by
GitHub
parent
8a818c26cb
commit
952a77b05d
@@ -758,29 +758,11 @@ class PerceiverModelTest(ModelTesterMixin, unittest.TestCase):
|
||||
loss.backward()
|
||||
|
||||
@require_torch_multi_gpu
|
||||
@unittest.skip(
|
||||
reason="Perceiver does not work with data parallel (DP) because of a bug in PyTorch: https://github.com/pytorch/pytorch/issues/36035"
|
||||
)
|
||||
def test_multi_gpu_data_parallel_forward(self):
|
||||
for model_class in self.all_model_classes:
|
||||
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_model_class(model_class)
|
||||
|
||||
# some params shouldn't be scattered by nn.DataParallel
|
||||
# so just remove them if they are present.
|
||||
blacklist_non_batched_params = ["head_mask", "decoder_head_mask", "cross_attn_head_mask"]
|
||||
for k in blacklist_non_batched_params:
|
||||
inputs_dict.pop(k, None)
|
||||
|
||||
# move input tensors to cuda:O
|
||||
for k, v in inputs_dict.items():
|
||||
if torch.is_tensor(v):
|
||||
inputs_dict[k] = v.to(0)
|
||||
|
||||
model = model_class(config=config)
|
||||
model.to(0)
|
||||
model.eval()
|
||||
|
||||
# Wrap model in nn.DataParallel
|
||||
model = nn.DataParallel(model)
|
||||
with torch.no_grad():
|
||||
_ = model(**self._prepare_for_class(inputs_dict, model_class))
|
||||
pass
|
||||
|
||||
@unittest.skip(reason="Perceiver models don't have a typical head like is the case with BERT")
|
||||
def test_save_load_fast_init_from_base(self):
|
||||
|
||||
@@ -573,8 +573,10 @@ class ReformerTesterMixin:
|
||||
self.model_tester.create_and_check_reformer_model_fp16_generate(*config_and_inputs)
|
||||
|
||||
@require_torch_multi_gpu
|
||||
@unittest.skip(
|
||||
reason="Reformer does not work with data parallel (DP) because of a bug in PyTorch: https://github.com/pytorch/pytorch/issues/36035"
|
||||
)
|
||||
def test_multi_gpu_data_parallel_forward(self):
|
||||
# Opt-out of this test.
|
||||
pass
|
||||
|
||||
def test_for_sequence_classification(self):
|
||||
|
||||
@@ -232,8 +232,10 @@ class TransfoXLModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestC
|
||||
return
|
||||
|
||||
@require_torch_multi_gpu
|
||||
@unittest.skip(
|
||||
reason="Transfo-XL does not work with data parallel (DP) because of a bug in PyTorch: https://github.com/pytorch/pytorch/issues/36035"
|
||||
)
|
||||
def test_multi_gpu_data_parallel_forward(self):
|
||||
# Opt-out of this test.
|
||||
pass
|
||||
|
||||
@slow
|
||||
|
||||
Reference in New Issue
Block a user