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

[examples tests on multigpu] resolving require_torch_non_multi_gpu_but_fix_me (#10561)

Browse Source 
* batch 1

* this is tpu

* deebert attempt

* the rest
This commit is contained in:
Stas Bekman
2021-03-08 11:11:40 -08:00
committed by GitHub
parent dfd16af832
commit f284089ec4
9 changed files with 35 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
examples/research_projects/bert-loses-patience/test_run_glue_with_pabee.py
View File

@@ -4,7 +4,7 @@ import sys
from unittest.mock import patch
import run_glue_with_pabee
from transformers.testing_utils import TestCasePlus, require_torch_non_multi_gpu_but_fix_me
from transformers.testing_utils import TestCasePlus
logging.basicConfig(level=logging.DEBUG)
@@ -20,7 +20,6 @@ def get_setup_file():
class PabeeTests(TestCasePlus):
@require_torch_non_multi_gpu_but_fix_me
def test_run_glue(self):
stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)

43
examples/research_projects/deebert/test_glue_deebert.py
View File

@@ -1,11 +1,10 @@
import argparse
import logging
import sys
import unittest
from unittest.mock import patch
import run_glue_deebert
from transformers.testing_utils import require_torch_non_multi_gpu_but_fix_me, slow
from transformers.testing_utils import TestCasePlus, get_gpu_count, require_torch_non_multi_gpu, slow
logging.basicConfig(level=logging.DEBUG)
@@ -20,17 +19,34 @@ def get_setup_file():
return args.f
class DeeBertTests(unittest.TestCase):
class DeeBertTests(TestCasePlus):
def setup(self) -> None:
stream_handler = logging.StreamHandler(sys.stdout)
logger.addHandler(stream_handler)
def run_and_check(self, args):
n_gpu = get_gpu_count()
if n_gpu > 1:
pass
# XXX: doesn't quite work with n_gpu > 1 https://github.com/huggingface/transformers/issues/10560
# script = f"{self.examples_dir_str}/research_projects/deebert/run_glue_deebert.py"
# distributed_args = f"-m torch.distributed.launch --nproc_per_node={n_gpu} {script}".split()
# cmd = [sys.executable] + distributed_args + args
# execute_subprocess_async(cmd, env=self.get_env())
# XXX: test the results - need to save them first into .json file
else:
args.insert(0, "run_glue_deebert.py")
with patch.object(sys, "argv", args):
result = run_glue_deebert.main()
for value in result.values():
self.assertGreaterEqual(value, 0.666)
@slow
@require_torch_non_multi_gpu_but_fix_me
@require_torch_non_multi_gpu
def test_glue_deebert_train(self):
train_args = """
run_glue_deebert.py
--model_type roberta
--model_name_or_path roberta-base
--task_name MRPC
@@ -51,13 +67,9 @@ class DeeBertTests(unittest.TestCase):
--overwrite_cache
--eval_after_first_stage
""".split()
with patch.object(sys, "argv", train_args):
result = run_glue_deebert.main()
for value in result.values():
self.assertGreaterEqual(value, 0.666)
self.run_and_check(train_args)
eval_args = """
run_glue_deebert.py
--model_type roberta
--model_name_or_path ./examples/deebert/saved_models/roberta-base/MRPC/two_stage
--task_name MRPC
@@ -72,13 +84,9 @@ class DeeBertTests(unittest.TestCase):
--overwrite_cache
--per_gpu_eval_batch_size=1
""".split()
with patch.object(sys, "argv", eval_args):
result = run_glue_deebert.main()
for value in result.values():
self.assertGreaterEqual(value, 0.666)
self.run_and_check(eval_args)
entropy_eval_args = """
run_glue_deebert.py
--model_type roberta
--model_name_or_path ./examples/deebert/saved_models/roberta-base/MRPC/two_stage
--task_name MRPC
@@ -93,7 +101,4 @@ class DeeBertTests(unittest.TestCase):
--overwrite_cache
--per_gpu_eval_batch_size=1
""".split()
with patch.object(sys, "argv", entropy_eval_args):
result = run_glue_deebert.main()
for value in result.values():
self.assertGreaterEqual(value, 0.666)
self.run_and_check(entropy_eval_args)

5
examples/research_projects/rag/test_distributed_retriever.py
View File

@@ -17,7 +17,7 @@ from transformers.integrations import is_ray_available
from transformers.models.bert.tokenization_bert import VOCAB_FILES_NAMES as DPR_VOCAB_FILES_NAMES
from transformers.models.rag.retrieval_rag import CustomHFIndex, RagRetriever
from transformers.models.roberta.tokenization_roberta import VOCAB_FILES_NAMES as BART_VOCAB_FILES_NAMES
from transformers.testing_utils import require_ray, require_torch_non_multi_gpu_but_fix_me
from transformers.testing_utils import require_ray
sys.path.append(os.path.join(os.getcwd())) # noqa: E402 # noqa: E402 # isort:skip
@@ -265,7 +265,6 @@ class RagRetrieverTest(TestCase):
self.assertEqual(doc_dicts[1]["id"][0], "0") # max inner product is reached with first doc
self.assertListEqual(doc_ids.tolist(), [[1], [0]])
@require_torch_non_multi_gpu_but_fix_me
def test_pytorch_distributed_retriever_retrieve(self):
n_docs = 1
hidden_states = np.array(
@@ -276,7 +275,6 @@ class RagRetrieverTest(TestCase):
self.get_dummy_pytorch_distributed_retriever(init_retrieval=True), hidden_states, n_docs
)
@require_torch_non_multi_gpu_but_fix_me
def test_custom_hf_index_pytorch_retriever_retrieve(self):
n_docs = 1
hidden_states = np.array(
@@ -289,7 +287,6 @@ class RagRetrieverTest(TestCase):
n_docs,
)
@require_torch_non_multi_gpu_but_fix_me
def test_custom_pytorch_distributed_retriever_retrieve_from_disk(self):
n_docs = 1
hidden_states = np.array(

7
examples/research_projects/seq2seq-distillation/_test_make_student.py
View File

@@ -4,7 +4,7 @@ import unittest
from make_student import create_student_by_copying_alternating_layers
from transformers import AutoConfig
from transformers.file_utils import cached_property
from transformers.testing_utils import require_torch, require_torch_non_multi_gpu_but_fix_me
from transformers.testing_utils import require_torch
TINY_BART = "sshleifer/bart-tiny-random"
@@ -17,28 +17,23 @@ class MakeStudentTester(unittest.TestCase):
def teacher_config(self):
return AutoConfig.from_pretrained(TINY_BART)
@require_torch_non_multi_gpu_but_fix_me
def test_valid_t5(self):
student, *_ = create_student_by_copying_alternating_layers(TINY_T5, tempfile.mkdtemp(), e=1, d=1)
self.assertEqual(student.config.num_hidden_layers, 1)
@require_torch_non_multi_gpu_but_fix_me
def test_asymmetric_t5(self):
student, *_ = create_student_by_copying_alternating_layers(TINY_T5, tempfile.mkdtemp(), e=1, d=None)
@require_torch_non_multi_gpu_but_fix_me
def test_same_decoder_small_encoder(self):
student, *_ = create_student_by_copying_alternating_layers(TINY_BART, tempfile.mkdtemp(), e=1, d=None)
self.assertEqual(student.config.encoder_layers, 1)
self.assertEqual(student.config.decoder_layers, self.teacher_config.encoder_layers)
@require_torch_non_multi_gpu_but_fix_me
def test_small_enc_small_dec(self):
student, *_ = create_student_by_copying_alternating_layers(TINY_BART, tempfile.mkdtemp(), e=1, d=1)
self.assertEqual(student.config.encoder_layers, 1)
self.assertEqual(student.config.decoder_layers, 1)
@require_torch_non_multi_gpu_but_fix_me
def test_raises_assert(self):
with self.assertRaises(AssertionError):
create_student_by_copying_alternating_layers(TINY_BART, tempfile.mkdtemp(), e=None, d=None)