From 6f84091bfa323cf584b830ed49a94a99819e0fb7 Mon Sep 17 00:00:00 2001 From: yhcc Date: Mon, 25 Jul 2022 16:17:04 +0800 Subject: [PATCH 1/6] =?UTF-8?q?update=E9=83=A8=E5=88=86=E4=BB=A3=E7=A0=81?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../callbacks/torch_callbacks/torch_lr_sched_callback.py | 5 ++--- fastNLP/core/collators/collator.py | 2 +- 2 files changed, 3 insertions(+), 4 deletions(-) diff --git a/fastNLP/core/callbacks/torch_callbacks/torch_lr_sched_callback.py b/fastNLP/core/callbacks/torch_callbacks/torch_lr_sched_callback.py index 07c3c612..97e1c544 100644 --- a/fastNLP/core/callbacks/torch_callbacks/torch_lr_sched_callback.py +++ b/fastNLP/core/callbacks/torch_callbacks/torch_lr_sched_callback.py @@ -41,10 +41,9 @@ class TorchWarmupCallback(Callback): return max((progress - 1.) / (self.warmup - 1.), 0.) def on_train_begin(self, trainer): - self.t_steps = trainer.n_batches if self.warmup >1: - self.warmup = self.warmup / self.t_steps - self.t_steps = max(2, self.t_steps) # 不能小于2 + self.warmup = self.warmup / trainer.n_batches + self.t_steps = max(2, trainer.n_batches) # 不能小于2 # 防止 t_steps 不能整除 accumulation_steps self.t_steps = math.ceil(self.t_steps/trainer.accumulation_steps) * trainer.accumulation_steps # 获取param_group的初始learning rate diff --git a/fastNLP/core/collators/collator.py b/fastNLP/core/collators/collator.py index 0ff9fb2a..5a3b1967 100644 --- a/fastNLP/core/collators/collator.py +++ b/fastNLP/core/collators/collator.py @@ -107,7 +107,7 @@ class Collator: 如果需要某些 field 不要包含在 pad 之后的结果中，可以使用 :meth:`~fastNLP.Collator.set_ignore` 进行设置。 Collator 在第一次进行 pad 的时候自动根据设置以及数据情况，为每个 field 获取一个 padder ，在之后的每次调用中，都将使用对应 - 的 Padder 给对应的 field 。 + 的 Padder 给对应的 field 。由于 Collator 只能在某个 field 内进行 pad ，如果 pad 操作需要同时操作多个 field ，请不要使用 Collator 。 :param backend: 对于可以 pad 的 field，使用哪种 tensor，支持 ``['torch','jittor','paddle','oneflow','numpy','raw', 'auto', None]``。 若为 ``'auto'`` ，则在进行 pad 的时候会根据调用的环境决定其 ``backend`` 。该参数对不能进行 pad 的数据没有影响，无法 pad 的数据返回一定 From c11c30a5fc8366c04f4fa445bafcc675c1b2ad79 Mon Sep 17 00:00:00 2001 From: yhcc Date: Mon, 25 Jul 2022 16:21:33 +0800 Subject: [PATCH 2/6] beta version --- setup.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/setup.py b/setup.py index 21e4c0e4..a3447bf2 100644 --- a/setup.py +++ b/setup.py @@ -16,7 +16,7 @@ print(pkgs) setup( name='FastNLP', - version='1.0.0alpha', + version='1.0.0beta', url='https://gitee.com/fastnlp/fastNLP', description='fastNLP: Deep Learning Toolkit for NLP, developed by Fudan FastNLP Team', long_description=readme, From d7ce92975c74ce6bdba635ed1a3bb2766222aba9 Mon Sep 17 00:00:00 2001 From: YWMditto Date: Mon, 8 Aug 2022 19:35:04 +0800 Subject: [PATCH 3/6] =?UTF-8?q?=E4=B8=BA=20=5FTruncatedDataLoader=20?= =?UTF-8?q?=E6=B7=BB=E5=8A=A0=20=5F=5Fsetattr=5F=5F?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- fastNLP/core/controllers/utils/utils.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/fastNLP/core/controllers/utils/utils.py b/fastNLP/core/controllers/utils/utils.py index 0a351354..b01cccc4 100644 --- a/fastNLP/core/controllers/utils/utils.py +++ b/fastNLP/core/controllers/utils/utils.py @@ -132,6 +132,10 @@ class _TruncatedDataLoader: def __getattr__(self, item): return getattr(self.dataloader, item) + def __setattr__(self, key, value): + # 添加该函数使得在进行实验性训练或者评测时，用户对于 trainer.dataloader 的感觉和正常训练完全一样； + setattr(self.dataloader, key, value) + def check_evaluate_every(evaluate_every): r""" From e86645380a1cd133af8b9bd25df0f6b6c84bcdef Mon Sep 17 00:00:00 2001 From: YWMditto Date: Mon, 8 Aug 2022 19:40:34 +0800 Subject: [PATCH 4/6] =?UTF-8?q?=E4=BF=AE=E6=AD=A3=20=5FTruncatedDataLoader?= =?UTF-8?q?=201?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- fastNLP/core/controllers/utils/utils.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/fastNLP/core/controllers/utils/utils.py b/fastNLP/core/controllers/utils/utils.py index b01cccc4..998b3787 100644 --- a/fastNLP/core/controllers/utils/utils.py +++ b/fastNLP/core/controllers/utils/utils.py @@ -132,9 +132,9 @@ class _TruncatedDataLoader: def __getattr__(self, item): return getattr(self.dataloader, item) - def __setattr__(self, key, value): - # 添加该函数使得在进行实验性训练或者评测时，用户对于 trainer.dataloader 的感觉和正常训练完全一样； - setattr(self.dataloader, key, value) + # def __setattr__(self, key, value): + # # 添加该函数使得在进行实验性训练或者评测时，用户对于 trainer.dataloader 的感觉和正常训练完全一样； + # setattr(self.dataloader, key, value) def check_evaluate_every(evaluate_every): From 0033219fe126a8505400f84dee3a0d9b1a8bdf7a Mon Sep 17 00:00:00 2001 From: YWMditto Date: Mon, 8 Aug 2022 20:32:52 +0800 Subject: [PATCH 5/6] =?UTF-8?q?=E4=BF=AE=E6=AD=A3=20=5FTruncatedDataLoader?= =?UTF-8?q?=202?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- fastNLP/core/controllers/utils/utils.py | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/fastNLP/core/controllers/utils/utils.py b/fastNLP/core/controllers/utils/utils.py index 998b3787..ca1e0b86 100644 --- a/fastNLP/core/controllers/utils/utils.py +++ b/fastNLP/core/controllers/utils/utils.py @@ -132,9 +132,16 @@ class _TruncatedDataLoader: def __getattr__(self, item): return getattr(self.dataloader, item) - # def __setattr__(self, key, value): - # # 添加该函数使得在进行实验性训练或者评测时，用户对于 trainer.dataloader 的感觉和正常训练完全一样； - # setattr(self.dataloader, key, value) + def __setattr__(self, key, value): + # 添加该函数使得在进行实验性训练或者评测时，用户对于 trainer.dataloader 的感觉和正常训练完全一样； + # 注意这里不能直接 ``setattr(self.dataloader, key, value)``，会导致死循环； + if "dataloader" in self.__dict__: + if hasattr(self.dataloader, key): + setattr(self.dataloader, key, value) + else: + self.__dict__[key] = value + else: + self.__dict__[key] = value def check_evaluate_every(evaluate_every): From e2c8aae79887059dc5f29255f2bcd1a4ff2d8312 Mon Sep 17 00:00:00 2001 From: yhcc Date: Sun, 21 Aug 2022 23:45:39 +0800 Subject: [PATCH 6/6] =?UTF-8?q?element=E5=A6=82=E6=9E=9C=E6=B2=A1=E6=9C=89?= =?UTF-8?q?=E5=88=9D=E5=A7=8B=E5=8C=96=E5=9C=A8reset=E6=97=B6=E4=B8=8D?= =?UTF-8?q?=E8=B5=B7=E4=BD=9C=E7=94=A8?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- fastNLP/core/metrics/element.py | 2 +- fastNLP/io/utils.py | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/fastNLP/core/metrics/element.py b/fastNLP/core/metrics/element.py index 5a52e4fa..359df3b8 100644 --- a/fastNLP/core/metrics/element.py +++ b/fastNLP/core/metrics/element.py @@ -88,7 +88,7 @@ class Element: """ 重置 value """ - if self.backend.is_specified(): + if self.backend.is_specified() and self._value is not None: self._value = self.backend.fill_value(self._value, self.init_value) @property diff --git a/fastNLP/io/utils.py b/fastNLP/io/utils.py index a3af71f5..c53d6158 100644 --- a/fastNLP/io/utils.py +++ b/fastNLP/io/utils.py @@ -52,8 +52,8 @@ def check_loader_paths(paths: Union[str, Dict[str, str]]) -> Dict[str, str]: path_pair = ('test', filename) if path_pair: if path_pair[0] in files: - raise FileExistsError(f"Two files contain `{path_pair[0]}` were found, please specify the " - f"filepath for `{path_pair[0]}`.") + raise FileExistsError(f"Two files contain `{path_pair[0]}` were found in {paths}, please specify " + f"the filepath for `{path_pair[0]}`.") files[path_pair[0]] = os.path.join(paths, path_pair[1]) # if 'train' not in files: # raise KeyError(f"There is no train file in {paths}.")