Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0

tests/core/callbacks/test_checkpoint_callback_torch.py

@@ -70,6 +70,7 @@ def model_and_optimizers(request):
     return trainer_params
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize("driver,device", [("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("version", [0, 1])
 @pytest.mark.parametrize("only_state_dict", [True, False])
@@ -201,6 +202,7 @@ def test_model_checkpoint_callback_1(
         dist.destroy_process_group()
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize("driver,device", [("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("only_state_dict", [True])
 @magic_argv_env_context(timeout=100)
@@ -293,6 +295,7 @@ def test_model_checkpoint_callback_2(
         dist.destroy_process_group()
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize("driver,device", [("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 0)])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("version", [0, 1])
 @pytest.mark.parametrize("only_state_dict", [True, False])
@@ -423,7 +426,7 @@ def test_trainer_checkpoint_callback_1(
         dist.destroy_process_group()
 
 
-
+@pytest.mark.torch
 # 通过自己编写 model_save_fn 和 model_load_fn 来测试 huggingface 的 transformers 的模型的保存和加载；
 @pytest.mark.parametrize("driver,device", [("torch_ddp", [6, 7]), ("torch", 7)])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("version", [0, 1])

tests/core/callbacks/test_load_best_model_callback_torch.py

@@ -70,8 +70,8 @@ def model_and_optimizers(request):
 
     return trainer_params
 
-# pytest test_load_best_model_callback_torch.py::test_load_best_model_callback -s
 
+@pytest.mark.torch
 @pytest.mark.parametrize("driver,device", [("torch_ddp", [4, 5]), ("torch", 1), ("torch", "cpu")])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("save_folder", ['save_models', None])
 @pytest.mark.parametrize("only_state_dict", [True, False])

tests/core/callbacks/test_more_evaluate_callback.py

@@ -95,6 +95,7 @@ def model_and_optimizers(request):
     return trainer_params
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize("driver,device", [("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("version", [0, 1])
 @pytest.mark.parametrize("only_state_dict", [True, False])
@@ -179,6 +180,7 @@ def test_model_more_evaluate_callback_1(
         dist.destroy_process_group()
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize("driver,device", [("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 0)])  # ("torch", "cpu"), ("torch_ddp", [0, 1]), ("torch", 1)
 @pytest.mark.parametrize("version", [0, 1])
 @pytest.mark.parametrize("only_state_dict", [True, False])

tests/core/callbacks/torch_callbacks/test_torch_grad_clip_callback.py

@@ -25,6 +25,7 @@ class CheckClipCallback(Callback):
                 assert np.linalg.norm(param.grad.cpu().view(-1).numpy())<=self.clip_value
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize('accumulation_steps', [1, 3, 5])
 @pytest.mark.parametrize('fp16', [True, False])
 @pytest.mark.parametrize('clip_type', ['norm', 'value'])

tests/core/callbacks/torch_callbacks/test_torch_warmup_callback.py

@@ -15,6 +15,7 @@ class RecordLrCallback(Callback):
         self.lrs.append(trainer.driver.optimizers[0].param_groups[0]['lr'])
 
 
+@pytest.mark.torch
 @pytest.mark.parametrize('warmup', [5, 0.1])
 @pytest.mark.parametrize('schedule', ['constant', 'linear'])
 @pytest.mark.parametrize('accumulation_steps', [1, 3, 4])

tests/core/drivers/torch_driver/test_ddp.py

@@ -72,6 +72,7 @@ def dataloader_with_randomsampler(dataset, batch_size, shuffle, drop_last, seed=
 #
 ############################################################################
 
+@pytest.mark.torch
 class TestDDPDriverFunction:
     """
     测试 TorchDDPDriver 一些简单函数的测试类，基本都是测试能否运行、是否存在 import 错误等问题
@@ -180,6 +181,7 @@ class TestDDPDriverFunction:
 #
 ############################################################################
 
+@pytest.mark.torch
 class TestSetDistReproDataloader:
 
     @classmethod
@@ -526,6 +528,7 @@ class TestSetDistReproDataloader:
 # 测试 save 和 load 相关的功能
 #
 ############################################################################
+@pytest.mark.torch
 class TestSaveLoad:
     """
     测试多卡情况下 save 和 load 相关函数的表现

tests/core/drivers/torch_driver/test_dist_utils.py

@@ -11,6 +11,7 @@ from fastNLP.core.drivers.torch_driver.dist_utils import fastnlp_torch_all_gathe
 from tests.helpers.utils import re_run_current_cmd_for_torch, magic_argv_env_context
 
 
+@pytest.mark.torch
 @magic_argv_env_context
 def test_fastnlp_torch_all_gather():
     os.environ['MASTER_ADDR'] = '127.0.0.1'
@@ -59,6 +60,7 @@ def test_fastnlp_torch_all_gather():
 
     dist.destroy_process_group()
 
+@pytest.mark.torch
 @magic_argv_env_context
 def test_fastnlp_torch_broadcast_object():
     os.environ['MASTER_ADDR'] = '127.0.0.1'

tests/core/drivers/torch_driver/test_initialize_torch_driver.py

@@ -8,12 +8,16 @@ from tests.helpers.utils import magic_argv_env_context
 
 import torch
 
+
+@pytest.mark.torch
 def test_incorrect_driver():
 
     model = TorchNormalModel_Classification_1(2, 100)
     with pytest.raises(ValueError):
         driver = initialize_torch_driver("paddle", 0, model)
 
+
+@pytest.mark.torch
 @pytest.mark.parametrize(
     "device", 
     ["cpu", "cuda:0", 0, torch.device("cuda:0")]
@@ -31,6 +35,8 @@ def test_get_single_device(driver, device):
     driver = initialize_torch_driver(driver, device, model)
     assert isinstance(driver, TorchSingleDriver)
 
+
+@pytest.mark.torch
 @pytest.mark.parametrize(
     "device", 
     [0, 1]
@@ -50,6 +56,8 @@ def test_get_ddp_2(driver, device):
 
     assert isinstance(driver, TorchDDPDriver)
 
+
+@pytest.mark.torch
 @pytest.mark.parametrize(
     "device", 
     [[0, 2, 3], -1]
@@ -69,6 +77,8 @@ def test_get_ddp(driver, device):
 
     assert isinstance(driver, TorchDDPDriver)
 
+
+@pytest.mark.torch
 @pytest.mark.parametrize(
     ("driver", "device"), 
     [("torch_ddp", "cpu")]
@@ -82,6 +92,8 @@ def test_get_ddp_cpu(driver, device):
     with pytest.raises(ValueError):
         driver = initialize_torch_driver(driver, device, model)
 
+
+@pytest.mark.torch
 @pytest.mark.parametrize(
     "device", 
     [-2, [0, torch.cuda.device_count() + 1, 3], [-2], torch.cuda.device_count() + 1]

tests/core/drivers/torch_driver/test_utils.py

@@ -9,6 +9,8 @@ from torch.utils.data import DataLoader, BatchSampler
 
 from tests.helpers.datasets.torch_data import TorchNormalDataset
 
+
+@pytest.mark.torch
 def test_replace_batch_sampler():
     dataset = TorchNormalDataset(10)
     dataloader = DataLoader(dataset, batch_size=32)
@@ -22,6 +24,8 @@ def test_replace_batch_sampler():
     assert len(replaced_loader.dataset) == len(dataset)
     assert replaced_loader.batch_sampler.batch_size == 16
 
+
+@pytest.mark.torch
 def test_replace_sampler():
     dataset = TorchNormalDataset(10)
     dataloader = DataLoader(dataset, batch_size=32)

tests/core/log/test_logger_torch.py

@@ -4,6 +4,7 @@ import datetime
 from pathlib import Path
 import logging
 import re
+import pytest
 
 from fastNLP.envs.env import FASTNLP_LAUNCH_TIME
 from fastNLP.core import rank_zero_rm
@@ -12,6 +13,7 @@ from fastNLP.core.log.logger import logger
 from tests.helpers.utils import magic_argv_env_context, recover_logger
 
 
+@pytest.mark.torch
 # 测试 TorchDDPDriver；
 @magic_argv_env_context
 @recover_logger
@@ -61,6 +63,7 @@ def test_add_file_ddp_1_torch():
     dist.destroy_process_group()
 
 
+@pytest.mark.torch
 @magic_argv_env_context
 @recover_logger
 def test_add_file_ddp_2_torch():
@@ -111,6 +114,7 @@ def test_add_file_ddp_2_torch():
     dist.destroy_process_group()
 
 
+@pytest.mark.torch
 @magic_argv_env_context
 @recover_logger
 def test_add_file_ddp_3_torch():
@@ -159,6 +163,8 @@ def test_add_file_ddp_3_torch():
     dist.barrier()
     dist.destroy_process_group()
 
+
+@pytest.mark.torch
 @magic_argv_env_context
 @recover_logger
 def test_add_file_ddp_4_torch():
@@ -208,6 +214,7 @@ def test_add_file_ddp_4_torch():
     dist.destroy_process_group()
 
 
+@pytest.mark.torch
 class TestLogger:
     msg = 'some test log msg'
 

tests/core/utils/test_distributed.py

@@ -1,4 +1,5 @@
 import os
+import pytest
 
 from fastNLP.envs.distributed import rank_zero_call, all_rank_call_context
 from tests.helpers.utils import re_run_current_cmd_for_torch, Capturing, magic_argv_env_context
@@ -30,7 +31,7 @@ class JittorTest:
     def test_all_rank_run(self):
         pass
 
-
+@pytest.mark.torch
 class TestTorch:
     @magic_argv_env_context
     def test_rank_zero_call(self):

tests/core/utils/test_paddle_utils.py

@@ -1,5 +1,5 @@
 import unittest
-
+import pytest
 import paddle
 
 from fastNLP.core.utils.paddle_utils import paddle_to, paddle_move_data_to_device
@@ -11,6 +11,7 @@ from fastNLP.core.utils.paddle_utils import paddle_to, paddle_move_data_to_devic
 #
 ############################################################################
 
+@pytest.mark.paddle
 class PaddleToDeviceTestCase(unittest.TestCase):
     def test_case(self):
         tensor = paddle.rand((4, 5))

tests/core/utils/test_torch_paddle_utils.py

@@ -1,6 +1,7 @@
 import unittest
 
 import paddle
+import pytest
 import torch
 
 from fastNLP.core.utils.torch_paddle_utils import torch_paddle_move_data_to_device
@@ -11,6 +12,8 @@ from fastNLP.core.utils.torch_paddle_utils import torch_paddle_move_data_to_devi
 #
 ############################################################################
 
+# @pytest.mark.paddle
+# @pytest.mark.torch
 class TorchPaddleMoveDataToDeviceTestCase(unittest.TestCase):
 
     def check_gpu(self, tensor, idx):