| @@ -3,9 +3,7 @@ __all__ = [ | |||
| 'Callback', | |||
| 'Event', | |||
| 'Filter', | |||
| 'CallbackManager', | |||
| 'CheckpointCallback', | |||
| 'choose_progress_callback', | |||
| 'ProgressCallback', | |||
| 'RichCallback', | |||
| "LRSchedCallback", | |||
| @@ -16,6 +14,7 @@ __all__ = [ | |||
| "TorchGradClipCallback", | |||
| "ResultsMonitor", | |||
| 'HasMonitorCallback', | |||
| "FitlogCallback", | |||
| # collators | |||
| 'Collator', | |||
| @@ -54,7 +53,6 @@ __all__ = [ | |||
| 'DataSet', | |||
| 'FieldArray', | |||
| 'Instance', | |||
| 'ApplyResultException', | |||
| # drivers | |||
| "TorchSingleDriver", | |||
| @@ -17,7 +17,9 @@ __all__ = [ | |||
| "TorchGradClipCallback", | |||
| "ResultsMonitor", | |||
| 'HasMonitorCallback' | |||
| 'HasMonitorCallback', | |||
| "FitlogCallback" | |||
| ] | |||
| @@ -32,4 +34,5 @@ from .early_stop_callback import EarlyStopCallback | |||
| from .torch_callbacks import * | |||
| from .more_evaluate_callback import MoreEvaluateCallback | |||
| from .has_monitor_callback import ResultsMonitor, HasMonitorCallback | |||
| from .fitlog_callback import FitlogCallback | |||
| @@ -180,8 +180,8 @@ class CallbackManager: | |||
| states[each_callback.callback_name]["states"] = each_callback.on_save_checkpoint(trainer) | |||
| if len(_duplicated_callbacks) > 0: | |||
| logger.warning(f"Notice these callbacks' `callback_name` are duplicated: {_duplicated_callbacks}, " | |||
| f"and we will only save the first callback's state we meet.") | |||
| logger.warning(f"Notice these callback_name: {_duplicated_callbacks} are duplicated, " | |||
| f"fastNLP will only save the first callback's state.") | |||
| # 2. 每一个具体的 callback 函数的 filter 的状态; | |||
| _record_duplicated_callback_names = set() | |||
| @@ -223,8 +223,8 @@ class CallbackManager: | |||
| _duplicated_callback_names.add(each_callback_filters[0]) | |||
| if len(_duplicated_callback_names) > 0: | |||
| logger.warning(f"Notice these callbacks' `callback_name` are duplicated: {_duplicated_callback_names}, " | |||
| f"and we will only load the first callback's state we meet.") | |||
| logger.rank_zero_warning(f"Notice these callback_name: {_duplicated_callback_names} are duplicated, " | |||
| f"fastNLP will only load the first callback's state.") | |||
| # 2. 再恢复每一个 callback 的单独的状态; | |||
| # 每一个我们自己提供的类 callback,都需要重写其特定的 `callback_name` 方法,保证如果两个 callback 的 callback_name 一样, | |||
| @@ -235,8 +235,6 @@ class CallbackManager: | |||
| _already_loaded_callback_names.add(each_callback.callback_name) | |||
| # 这里要注意,我们已经确保每一个 callback 的 `on_load_checkpoint` 函数拿到的就是其自己的状态; | |||
| each_callback.on_load_checkpoint(trainer, states[each_callback.callback_name]["states"]) | |||
| else: | |||
| each_callback.on_load_checkpoint(trainer, None) | |||
| @property | |||
| def has_trainer_checkpoint(self) -> bool: | |||
| @@ -33,9 +33,16 @@ class CheckpointCallback(Callback): | |||
| 则 fastNLP 将 folder 绝对路径传递给该函数,fastNLP 在该 folder 下不进行模型保存。默认情况下,本 checkpoint 只保存了 model | |||
| 的状态;如还需保存 Trainer 的状态以断点重训的话,请使用 ``save_object='trainer'`` 。 | |||
| :param monitor: 监控的 metric 值。如果在 evaluation 结果中没有找到完全一致的名称,将使用 最长公共字符串算法 找到最匹配 | |||
| 的那个作为 monitor 。如果为 None,将尝试使用 Trainer 设置的 monitor 。也可以传入一个函数,接受参数为 evaluation 的结 | |||
| 果(字典类型),返回一个 float 值作为 monitor 的结果,如果当前结果中没有相关的 monitor 值请返回 None 。 | |||
| :param monitor: 监控的 metric 值。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param folder: 保存的文件夹,fastNLP 将在该文件下以时间戳创建子文件夹,并在里面保存。因此不同次运行可以将被保存到不同的 | |||
| 时间戳文件夹中。如果为 None ,默认使用当前文件夹。 | |||
| :param every_n_epochs: 多少个 epoch 保存一次。 | |||
| @@ -12,9 +12,16 @@ class EarlyStopCallback(HasMonitorCallback): | |||
| def __init__(self, monitor:Union[str, Callable]=None, larger_better:bool=True, patience:int=10): | |||
| """ | |||
| :param str monitor: 监控的 metric 值。如果在 evaluation 结果中没有找到完全一致的名称,将使用 最长公共字符串算法 找到最匹配 | |||
| 的那个作为 monitor 。如果为 None,将尝试使用 Trainer 设置的 monitor 。也可以传入一个函数,接受参数为 evaluation 的结 | |||
| 果(字典类型),返回一个 float 值作为 monitor 的结果。 | |||
| :param monitor: 监控的 metric 值。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param larger_better: monitor 的值是否是越大越好。 | |||
| :param patience: 多少次 evaluate 不没有提升就停止。 | |||
| """ | |||
| @@ -0,0 +1,61 @@ | |||
| __all__ = [ | |||
| 'FitlogCallback' | |||
| ] | |||
| from .has_monitor_callback import HasMonitorCallback | |||
| from ...envs import _module_available | |||
| if _module_available('fitlog'): | |||
| import fitlog | |||
| class FitlogCallback(HasMonitorCallback): | |||
| def __init__(self, monitor=None, larger_better: bool = True, log_exception:bool=True, log_loss_every:int=0): | |||
| """ | |||
| 自动记录 ``evaluation`` 结果到 ``fitlog`` 中的 ``Callback`` 。会根据 ``monitor`` 记录最好的结果,以及每一次 ``evaluate`` 后的 | |||
| 结果。 | |||
| :param monitor: 监控的 metric 值。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param larger_better: 是否是越大越好。 | |||
| :param log_exception: 是否记录 ``exception`` 。 | |||
| :param log_loss_every: 多少个 ``batch`` 记录一次 loss 到 ``fitlog`` 中。 | |||
| """ | |||
| assert _module_available('fitlog'), "fitlog is not installed." | |||
| super().__init__(monitor=monitor, larger_better=larger_better) | |||
| self.log_exception = log_exception | |||
| self.log_loss_every = log_loss_every | |||
| self.avg_loss = 0 | |||
| def on_evaluate_end(self, trainer, results): | |||
| results = self.itemize_results(results) | |||
| fitlog.add_metric(results, step=trainer.global_forward_batches, epoch=trainer.cur_epoch_idx) | |||
| if self.is_better_results(results, keep_if_better=True): | |||
| results['step'] = trainer.global_forward_batches | |||
| results['epoch'] = trainer.cur_epoch_idx | |||
| fitlog.add_best_metric(results) | |||
| def on_before_backward(self, trainer, outputs): | |||
| if self.log_loss_every > 0: | |||
| loss = trainer.extract_loss_from_outputs(outputs) | |||
| self.avg_loss += loss.item() | |||
| if trainer.global_forward_batches % self.log_loss_every == 0: | |||
| fitlog.add_loss(self.avg_loss / self.log_loss_every * trainer.accumulation_steps, name='loss', | |||
| step=trainer.global_forward_batches, | |||
| epoch=trainer.cur_epoch_idx) | |||
| self.avg_loss = 0 | |||
| def on_train_end(self, trainer): | |||
| fitlog.finish() | |||
| def on_exception(self, trainer, exception): | |||
| fitlog.finish(status=1) | |||
| if self.log_exception: | |||
| fitlog.add_other(repr(exception), name='except_info') | |||
| @@ -171,9 +171,16 @@ class HasMonitorCallback(ResultsMonitor, Callback): | |||
| 该 callback 不直接进行使用,作为其它相关 callback 的父类使用,如果 callback 有使用 monitor 可以继承该函数里面实现了 | |||
| (1)判断monitor合法性;(2)在需要时, 根据trainer的monitor设置自己的monitor名称。 | |||
| :param monitor: 监控的 metric 值。如果在 evaluation 结果中没有找到完全一致的名称,将使用 最长公共字符串算法 找到最匹配 | |||
| 的那个作为 monitor 。如果为 None,将尝试使用 Trainer 设置的 monitor 。也可以传入一个函数,接受参数为 evaluation 的结 | |||
| 果(字典类型),返回一个 float 值作为 monitor 的结果,如果当前结果中没有相关的 monitor 值请返回 None 。 | |||
| :param monitor: 监控的 metric 值。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param larger_better: monitor 是否时越大越好 | |||
| :param must_have_monitor: 这个 callback 是否必须有 monitor 设置。如果设置为 True ,且没检测到设置 monitor 会报错。 | |||
| """ | |||
| @@ -22,9 +22,16 @@ class LoadBestModelCallback(HasMonitorCallback): | |||
| 保存最佳的 monitor 值最佳的模型,并在训练结束的时候重新加载模型,默认会在加载之后删除权重文件。仅在训练正常结束的时候才能加载 | |||
| 最好的模型。 | |||
| :param str monitor: 监控的 metric 值。如果在 evaluation 结果中没有找到完全一致的名称,将使用 最长公共字符串算法 找到最匹配 | |||
| 的那个作为 monitor 。如果为 None,将尝试使用 Trainer 设置的 monitor 。也可以传入一个函数,接受参数为 evaluation 的结 | |||
| 果(字典类型),返回一个 float 值作为 monitor 的结果,如果当前结果中没有相关的 monitor 值请返回 None 。 | |||
| :param monitor: 监控的 metric 值。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param larger_better: 该 metric 值是否是越大越好。 | |||
| :param save_folder: 保存的文件夹,如果为空,则保存在内存中。不为空,则保存一份权重到文件中,当为多机训练,且本值不为空时,请确保 | |||
| 不同的机器均可访问当该路径。当 model_save_fn 不为 None 时该值一定不能为空。 | |||
| @@ -45,10 +45,16 @@ class RichCallback(ProgressCallback): | |||
| :param print_every: 多少个 batch 更新一次显示。 | |||
| :param loss_round_ndigit: 显示的 loss 保留多少位有效数字 | |||
| :param monitor: 当检测到这个key的结果更好时,会打印出不同的颜色进行提示。监控的 metric 值。如果在 evaluation 结果中没有找到 | |||
| 完全一致的名称,将使用 最长公共字符串算法 找到最匹配的那个作为 monitor 。如果为 None,将尝试使用 Trainer 设置的 monitor | |||
| 。也可以传入一个函数,接受参数为 evaluation 的结果(字典类型),返回一个 float 值作为 monitor 的结果,如果当前结果中没有 | |||
| 相关的 monitor 值请返回 None 。 | |||
| :param monitor: 监控的 metric 值。当检测到这个key的结果更好时,会打印出不同的颜色进行提示。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param larger_better: 是否是 monitor 的结果越大越好。 | |||
| :param format_json: 是否格式化 json 再打印 | |||
| """ | |||
| @@ -140,10 +146,16 @@ class RawTextCallback(ProgressCallback): | |||
| :param print_every: 多少个 batch 更新一次显示。 | |||
| :param loss_round_ndigit: 显示的 loss 保留多少位有效数字 | |||
| :param monitor: 当检测到这个key的结果更好时,会打印出不同的颜色进行提示。监控的 metric 值。如果在 evaluation 结果中没有找到 | |||
| 完全一致的名称,将使用 最长公共字符串算法 找到最匹配的那个作为 monitor 。如果为 None,将尝试使用 Trainer 设置的 monitor | |||
| 。也可以传入一个函数,接受参数为 evaluation 的结果(字典类型),返回一个 float 值作为 monitor 的结果,如果当前结果中没有 | |||
| 相关的 monitor 值请返回 None 。 | |||
| :param monitor: 监控的 metric 值。当检测到这个key的结果更好时,会打印出不同的颜色进行提示。 | |||
| * 为 ``None`` | |||
| 将尝试使用 :class:`~fastNLP.Trainer` 中设置 `monitor` 值(如果有设置)。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| :param larger_better: 是否是monitor的结果越大越好。 | |||
| :param format_json: 是否format json再打印 | |||
| """ | |||
| @@ -7,7 +7,7 @@ __all__ = [ | |||
| 'Evaluator' | |||
| ] | |||
| from fastNLP.core.drivers import Driver | |||
| from fastNLP.core.drivers import Driver, TorchDriver | |||
| from ..drivers.choose_driver import choose_driver | |||
| from .loops import Loop, EvaluateBatchLoop | |||
| from fastNLP.core.utils import auto_param_call, dataclass_to_dict, \ | |||
| @@ -316,15 +316,15 @@ class _MetricsWrapper: | |||
| raise TypeError("Parameter `metrics` can only be `Dict` type.") | |||
| for metric_name, metric in metrics.items(): | |||
| # 因为 torchmetrics 是一个 nn.Module,因此我们需要先将其移到对应的机器上; | |||
| if _is_torchmetrics_metric(metric): | |||
| if _is_torchmetrics_metric(metric) and isinstance(evaluator.driver, TorchDriver): | |||
| # torchmetrics 是默认自动开启了多卡的 | |||
| evaluator.driver.move_model_to_device(metric, evaluator.driver.data_device) | |||
| elif isinstance(metric, Metric): | |||
| # 如果数据是分布式的,但是不aggregate的话可能有问题 | |||
| if evaluator._dist_sampler is not None and metric.aggregate_when_get_metric is False: | |||
| logger.warning_once( | |||
| "You have replace the sampler as distributed sampler when evaluation, but your " | |||
| f"metric {metric_name}:{metric.__class__.__name__}' `aggregate_when_get_metric` is False.") | |||
| logger.rank_zero_warning( | |||
| "You have replace the sampler as distributed sampler when evaluation, but your metric " | |||
| f"{metric_name}:{metric.__class__.__name__}'s `aggregate_when_get_metric` is False.", once=True) | |||
| if metric.aggregate_when_get_metric is None: | |||
| metric.aggregate_when_get_metric = evaluator._dist_sampler is not None | |||
| @@ -34,7 +34,7 @@ class EvaluateBatchLoop(Loop): | |||
| except BaseException as e: | |||
| if callable(getattr(dataloader, 'get_batch_indices', None)): | |||
| indices = dataloader.get_batch_indices() | |||
| logger.debug(f"The following exception happens when running on samples: {indices}") | |||
| logger.error(f"Exception happens when evaluating on samples: {indices}") | |||
| raise e | |||
| self.batch_step_fn(evaluator, batch) | |||
| @@ -32,7 +32,7 @@ class TrainBatchLoop(Loop): | |||
| break | |||
| except BaseException as e: | |||
| if indices and not isinstance(e, EarlyStopException): | |||
| logger.debug(f"The following exception happens when running on samples: {indices}") | |||
| logger.error(f"Exception happens when running on samples: {indices}") | |||
| raise e | |||
| trainer.on_train_batch_begin(batch, indices) | |||
| @@ -229,27 +229,19 @@ class Trainer(TrainerEventTrigger): | |||
| :param accumulation_steps: 梯度累积的步数,表示每隔几个 batch 才让优化器迭代一次,默认为 1; | |||
| :param fp16: 是否开启混合精度训练,默认为 False; | |||
| :param monitor: 对于一些特殊的 ``Callback``,例如 :class:`fastNLP.core.callbacks.CheckpointCallback`,它们需要参数 ``monitor`` | |||
| 来从 ``Evaluator`` 的验证结果中获取当前评测的值,从而来判断是否执行一些特殊的操作。例如,对于 ``CheckpointCallback`` 而言,如果我们 | |||
| 想要每隔一个 epoch 让 ``Evaluator`` 进行一次验证,然后保存训练以来的最好的结果;那么我们需要这样设置: | |||
| 来从 ``Evaluator`` 的验证结果中获取当前评测的值,从而来判断是否执行一些特殊的操作。这里设置了 ``monitor`` 则所有的需要 | |||
| ``monitor`` 但是没有自己设置的 ``Callback`` 都会使用这个值 | |||
| .. code-block:: | |||
| * 为 ``None`` | |||
| 没有 monitor ,默认。 | |||
| * 为 ``str`` | |||
| 尝试直接使用该名称从 ``evaluation`` 结果中寻找,如果在 ``evaluation`` 结果中没有找到完全一致的名称,将 | |||
| 使用 最长公共字符串算法 从 ``evaluation`` 结果中找到最匹配的那个作为 ``monitor`` 。 | |||
| * 为 ``Callable`` | |||
| 接受参数为 ``evaluation`` 的结果(字典类型),返回一个 ``float`` 值作为 ``monitor`` 的结果,如果当前结果中没有相关 | |||
| 的 ``monitor`` 值请返回 ``None`` 。 | |||
| trainer = Trainer( | |||
| ..., | |||
| metrics={'acc': accMetric()}, | |||
| callbacks=[CheckpointCallback( | |||
| ..., | |||
| monitor='acc', | |||
| topk=1 | |||
| )] | |||
| ) | |||
| 这意味着对于 ``CheckpointCallback`` 来说,*'acc'* 就是一个监测的指标,用于在 ``Evaluator`` 验证后取出其需要监测的那个指标的值。 | |||
| ``Trainer`` 中的参数 ``monitor`` 的作用在于为没有设置 ``monitor`` 参数但是需要该参数的 *callback* 实例设置该值。关于 ``monitor`` | |||
| 参数更详细的说明,请见 :class:`fastNLP.core.callbacks.CheckpointCallback`; | |||
| 注意该参数仅当 ``Trainer`` 内置的 ``Evaluator`` 不为 None 时且有需要该参数但是没有设置该参数的 *callback* 实例才有效; | |||
| 注意该参数仅当传入了 ``evaluate_dataloaders`` 不为 ``None`` 时且有需要该参数但是没有设置该参数的 *Callback* 实例才有意义; | |||
| :param larger_better: 对于需要参数 ``monitor`` 的 *callback* 来说,``monitor`` 的值是否是越大越好;类似于 ``monitor``,其作用 | |||
| 在于为没有设置 ``larger_better`` 参数但是需要该参数的 *callback* 实例设置该值; | |||
| @@ -282,32 +274,41 @@ class Trainer(TrainerEventTrigger): | |||
| :kwargs: | |||
| * *torch_kwargs* -- 用于在指定 ``driver`` 为 'torch' 时设定具体 driver 实例的一些参数: | |||
| * ddp_kwargs -- 用于在使用 ``TorchDDPDriver`` 时指定 ``DistributedDataParallel`` 初始化时的参数;例如传入 | |||
| {'find_unused_parameters': True} 来解决有参数不参与前向运算导致的报错等; | |||
| {'find_unused_parameters': True} 来解决有参数不参与前向运算导致的报错等; | |||
| * set_grad_to_none -- 是否在训练过程中在每一次 optimizer 更新后将 grad 置为 None; | |||
| * torch_non_blocking -- 表示用于 pytorch 的 tensor 的 to 方法的参数 non_blocking; | |||
| * *paddle_kwargs* -- 用于在指定 ``driver`` 为 'paddle' 时设定具体 driver 实例的一些参数: | |||
| * fleet_kwargs -- 用于在使用 ``PaddleFleetDriver`` 时指定 ``DataParallel`` 和 ``fleet`` 初始化时的参数,包括: | |||
| * is_collective -- 是否使用 paddle 集群式的分布式训练方法,目前仅支持为 True 的情况; | |||
| * role_maker -- 初始化 ``fleet`` 分布式训练 API 时使用的 ``RoleMaker`` | |||
| * 其它用于初始化 ``DataParallel`` 的参数; | |||
| * *data_device* -- 一个具体的 driver 实例中,有 ``model_device`` 和 ``data_device``,前者表示模型所在的设备,后者表示 | |||
| 当 ``model_device`` 为 None 时应当将数据迁移到哪个设备; | |||
| 当 ``model_device`` 为 None 时应当将数据迁移到哪个设备; | |||
| .. note:: | |||
| .. note:: | |||
| 注意您在绝大部分情况下不会用到该参数! | |||
| 1. 当 driver 实例的 ``model_device`` 不为 None 时,该参数无效; | |||
| 2. 对于 pytorch,仅当用户自己通过 ``python -m torch.distributed.launch`` 并且自己初始化 ``init_process_group`` 时, | |||
| driver 实例的 ``model_device`` 才会为 None; | |||
| 3. 对于 paddle,该参数无效; | |||
| * *use_dist_sampler* -- 表示是否使用分布式的 ``sampler``。在多卡时,分布式 ``sampler`` 将自动决定每张卡上读取的 sample ,使得一个 epoch | |||
| 内所有卡的 sample 加起来为一整个数据集的 sample。默认会根据 driver 是否为分布式进行设置。 | |||
| 内所有卡的 sample 加起来为一整个数据集的 sample。默认会根据 driver 是否为分布式进行设置。 | |||
| * *evaluate_use_dist_sampler* -- 表示在 ``Evaluator`` 中在使用分布式的时候是否将 dataloader 的 ``sampler`` 替换为分布式的 ``sampler``;默认为 ``True``; | |||
| * *output_from_new_proc* -- 应当为一个字符串,表示在多进程的 driver 中其它进程的输出流应当被做如何处理;其值应当为以下之一: | |||
| ["all", "ignore", "only_error"];当该参数的值不是以上值时,该值应当表示一个文件夹的名字,我们会将其他 rank 的输出流重定向到 | |||
| log 文件中,然后将 log 文件保存在通过该参数值设定的文件夹中;默认为 "only_error"; | |||
| ["all", "ignore", "only_error"];当该参数的值不是以上值时,该值应当表示一个文件夹的名字,我们会将其他 rank 的输出流重定向到 | |||
| log 文件中,然后将 log 文件保存在通过该参数值设定的文件夹中;默认为 "only_error"; | |||
| 注意该参数仅当使用分布式的 ``driver`` 时才有效,例如 ``TorchDDPDriver``; | |||
| 注意该参数仅当使用分布式的 ``driver`` 时才有效,例如 ``TorchDDPDriver``; | |||
| * *progress_bar* -- 以哪种方式显示 progress ,目前支持[None, 'raw', 'rich', 'auto'] 或者 RichCallback, RawTextCallback对象, | |||
| 默认为 auto , auto 表示如果检测到当前 terminal 为交互型则使用 RichCallback,否则使用 RawTextCallback对象。如果 | |||
| 需要定制 progress bar 的参数,例如打印频率等,可以传入 RichCallback, RawTextCallback 对象。 | |||
| 默认为 auto , auto 表示如果检测到当前 terminal 为交互型则使用 RichCallback,否则使用 RawTextCallback对象。如果 | |||
| 需要定制 progress bar 的参数,例如打印频率等,可以传入 RichCallback, RawTextCallback 对象。 | |||
| * *train_input_mapping* -- 与 input_mapping 一致,但是只用于 ``Trainer`` 中。与 input_mapping 互斥。 | |||
| * *train_output_mapping* -- 与 output_mapping 一致,但是只用于 ``Trainer`` 中。与 output_mapping 互斥。 | |||
| * *evaluate_input_mapping* -- 与 input_mapping 一致,但是只用于 ``Evaluator`` 中。与 input_mapping 互斥。 | |||
| @@ -483,18 +484,62 @@ class Trainer(TrainerEventTrigger): | |||
| def run(self, num_train_batch_per_epoch: int = -1, num_eval_batch_per_dl: int = -1, | |||
| num_eval_sanity_batch: int = 2, resume_from: str = None, resume_training: bool = True, | |||
| catch_KeyboardInterrupt=None): | |||
| catch_KeyboardInterrupt = None): | |||
| r""" | |||
| 注意如果是断点重训的第一次训练,即还没有保存任何用于断点重训的文件,那么其应当置 resume_from 为 None,并且使用 ModelCheckpoint | |||
| 该函数是在 ``Trainer`` 初始化后用于真正开始训练的函数; | |||
| 注意如果是断点重训的第一次训练,即还没有保存任何用于断点重训的文件,那么其应当置 resume_from 为 None,并且使用 ``CheckpointCallback`` | |||
| 去保存断点重训的文件; | |||
| :param num_train_batch_per_epoch: 每个 epoch 运行多少个 batch 即停止,-1 为根据 dataloader 有多少个 batch 决定。 | |||
| :param num_eval_batch_per_dl: 每个 evaluate dataloader 运行多少个 batch 停止,-1 为根据 dataloader 有多少个 batch 决定。 | |||
| :param num_eval_sanity_batch: 在训练之前运行多少个 evaluation batch 来检测一下 evaluation 是否有错误。为 0 表示不检测。 | |||
| :param resume_from: 从哪个路径下恢复 trainer 的状态 | |||
| :param resume_training: 是否按照 checkpoint 中训练状态恢复。如果为 False,则只恢复 model 和 optimizers 的状态。 | |||
| :param catch_KeyboardInterrupt: 是否捕获KeyboardInterrupt, 如果捕获的话,不会抛出一场,trainer.run()之后的代码会继续运 | |||
| 行。默认如果非 distributed 的 driver 会 catch ,distributed 不会 catch (无法 catch ) | |||
| :return: | |||
| :param num_train_batch_per_epoch: 每个 epoch 训练多少个 batch 后停止,*-1* 表示使用 train_dataloader 本身的长度; | |||
| :param num_eval_batch_per_dl: 每个 evaluate_dataloader 验证多少个 batch 停止,*-1* 表示使用 evaluate_dataloader 本身的长度; | |||
| :param num_eval_sanity_batch: 在训练之前运行多少个 evaluation batch 来检测一下 evaluation 的过程是否有错误。为 0 表示不检测; | |||
| :param resume_from: 从哪个路径下恢复 trainer 的状态,注意该值需要为一个文件夹,例如使用 ``CheckpointCallback`` 时帮助您创建的保存的子文件夹; | |||
| :param resume_training: 是否按照 checkpoint 中训练状态恢复。如果为 False,则只恢复 model 和 optimizers 的状态;该参数如果为 ``True``, | |||
| 在下一次断点重训的时候我们会精确到上次训练截止的具体的 sample 进行训练;否则我们只会恢复 model 和 optimizers 的状态,而 ``Trainer`` 中的 | |||
| 其余状态都是保持初始化时的状态不会改变; | |||
| :param catch_KeyboardInterrupt: 是否捕获 KeyboardInterrupt;如果该参数为 ``True``,在训练时如果您使用 ``ctrl+c`` 来终止程序, | |||
| ``Trainer`` 不会抛出异常,但是会提前退出,然后 ``trainer.run()`` 之后的代码会继续运行。注意该参数在您使用分布式训练的 ``Driver`` | |||
| 时无效,例如 ``TorchDDPDriver``;非分布式训练的 ``Driver`` 下该参数默认为 True; | |||
| .. warning:: | |||
| 注意初始化的 ``Trainer`` 只能调用一次 ``run`` 函数,即之后的调用 ``run`` 函数实际不会运行,因为此时 | |||
| ``trainer.cur_epoch_idx == trainer.n_epochs``; | |||
| 这意味着如果您需要再次调用 ``run`` 函数,您需要重新再初始化一个 ``Trainer``; | |||
| .. note:: | |||
| 您可以使用 ``num_train_batch_per_epoch`` 来简单地对您的训练过程进行验证,例如,当您指定 ``num_train_batch_per_epoch=10`` 后, | |||
| 每一个 epoch 下实际训练的 batch 的数量则会被修改为 10。您可以先使用该值来设定一个较小的训练长度,在验证整体的训练流程没有错误后,再将 | |||
| 该值设定为 **-1** 开始真正的训练; | |||
| ``num_eval_batch_per_dl`` 的意思和 ``num_train_batch_per_epoch`` 类似,即您可以通过设定 ``num_eval_batch_per_dl`` 来验证 | |||
| 整体的验证流程是否正确; | |||
| ``num_eval_sanity_batch`` 的作用可能会让人产生迷惑,其本质和 ``num_eval_batch_per_dl`` 作用一致,但是其只被 ``Trainer`` 使用; | |||
| 并且其只会在训练的一开始使用,意思为:我们在训练的开始时会先使用 ``Evaluator``(如果其不为 ``None``) 进行验证,此时验证的 batch 的 | |||
| 数量只有 ``num_eval_sanity_batch`` 个;但是对于 ``num_eval_batch_per_dl`` 而言,其表示在实际的整体的训练过程中,每次 ``Evaluator`` | |||
| 进行验证时会验证的 batch 的数量。 | |||
| 并且,在实际真正的训练中,``num_train_batch_per_epoch`` 和 ``num_eval_batch_per_dl`` 应当都被设置为 **-1**,但是 ``num_eval_sanity_batch`` | |||
| 应当为一个很小的正整数,例如 2; | |||
| .. note:: | |||
| 参数 ``resume_from`` 和 ``resume_training`` 的设立是为了支持断点重训功能;仅当 ``resume_from`` 不为 ``None`` 时,``resume_training`` 才有效; | |||
| 断点重训的意思为将上一次训练过程中的 ``Trainer`` 的状态保存下来,包括模型和优化器的状态、当前训练过的 epoch 的数量、对于当前的 epoch | |||
| 已经训练过的 batch 的数量、callbacks 的状态等等;然后在下一次训练时直接加载这些状态,从而直接恢复到上一次训练过程的某一个具体时间点的状态开始训练; | |||
| fastNLP 将断点重训分为了 **保存状态** 和 **恢复断点重训** 两部分: | |||
| 1. 您需要使用 ``CheckpointCallback`` 来保存训练过程中的 ``Trainer`` 的状态;具体详见 :class:`~fastNLP.core.callbacks.CheckpointCallback`; | |||
| ``CheckpointCallback`` 会帮助您把 ``Trainer`` 的状态保存到一个具体的文件夹下,这个文件夹的名字由 ``CheckpointCallback`` 自己生成; | |||
| 2. 在第二次训练开始时,您需要找到您想要加载的 ``Trainer`` 状态所存放的文件夹,然后传入给参数 ``resume_from``; | |||
| 需要注意的是 **保存状态** 和 **恢复断点重训** 是互不影响的。 | |||
| """ | |||
| if catch_KeyboardInterrupt is None: | |||
| @@ -514,7 +559,7 @@ class Trainer(TrainerEventTrigger): | |||
| else: | |||
| raise FileNotFoundError("You are using `resume_from`, but we can not find your specific file.") | |||
| if self.evaluator is not None and num_eval_sanity_batch > 0: | |||
| if self.evaluator is not None and num_eval_sanity_batch != 0: | |||
| logger.info(f"Running evaluator sanity check for {num_eval_sanity_batch} batches.") | |||
| self.on_sanity_check_begin() | |||
| sanity_check_res = self.evaluator.run(num_eval_batch_per_dl=num_eval_sanity_batch) | |||
| @@ -569,7 +614,12 @@ class Trainer(TrainerEventTrigger): | |||
| finally: | |||
| self.on_train_end() | |||
| def _set_num_eval_batch_per_dl(self, num_eval_batch_per_dl): | |||
| def _set_num_eval_batch_per_dl(self, num_eval_batch_per_dl: int): | |||
| r""" | |||
| 用于设定训练过程中 ``Evaluator`` 进行验证时所实际验证的 batch 的数量; | |||
| :param num_eval_batch_per_dl: 等价于 :meth:`~fastNLP.core.controllers.Trainer.run` 中的参数 ``num_eval_batch_per_dl``; | |||
| """ | |||
| def _evaluate_fn(trainer: Trainer, evaluate_fn: Callable) -> None: | |||
| trainer.on_evaluate_begin() | |||
| _evaluate_res: dict = evaluate_fn() | |||
| @@ -579,10 +629,8 @@ class Trainer(TrainerEventTrigger): | |||
| self.run_evaluate = partial(_evaluate_fn, self, partial(self.evaluator.run, num_eval_batch_per_dl)) | |||
| def step_evaluate(self): | |||
| """ | |||
| 在每个 batch 结束后调用,根据设置执行 evaluate 。 | |||
| :return: | |||
| r""" | |||
| 在训练过程中的每个 batch 结束后被调用,注意实际的 ``Evaluator.run`` 函数是否在此时被调用取决于用户设置的 **"验证频率"**; | |||
| """ | |||
| if self.evaluator is not None: | |||
| if callable(self.evaluate_every): | |||
| @@ -592,10 +640,8 @@ class Trainer(TrainerEventTrigger): | |||
| self.run_evaluate() | |||
| def epoch_evaluate(self): | |||
| """ | |||
| 在每个 epoch 结束后调用,根据设置执行 evaluate 。 | |||
| :return: | |||
| r""" | |||
| 在训练过程中的每个 epoch 结束后被调用,注意实际的 ``Evaluator.run`` 函数是否在此时被调用取决于用户设置的 **"验证频率"**; | |||
| """ | |||
| if self.evaluator is not None: | |||
| if isinstance(self.evaluate_every, int) and self.evaluate_every < 0: | |||
| @@ -605,11 +651,52 @@ class Trainer(TrainerEventTrigger): | |||
| def add_callback_fn(self, event: Event, fn: Callable): | |||
| r""" | |||
| 在初始化一个 trainer 实例后,用户可以使用这一函数来方便地添加 callback 函数; | |||
| 这一函数应当交给具体的 trainer 实例去做,因此不需要 `mark` 参数; | |||
| 在初始化一个 trainer 实例后,您可以使用这一函数来方便地添加 ``callback`` 函数; | |||
| 注意这一函数应当交给具体的 trainer 实例去做,因此不需要 `mark` 参数; | |||
| :param event: 特定的 callback 时机,用户需要为该 callback 函数指定其属于哪一个 callback 时机; | |||
| :param event: 特定的 callback 时机,用户需要为该 callback 函数指定其属于哪一个 callback 时机;具体有哪些时机详见 :class:`fastNLP.core.callbacks.Event`; | |||
| :param fn: 具体的 callback 函数; | |||
| .. note:: | |||
| 对于训练一个神经网络的整体的流程来说,其可以分为很多个时间点,例如 **"整体的训练前"**,**"训练具体的一个 epoch 前"**, | |||
| **"反向传播前"**,**"整体的训练结束后"**等;一个 ``callback`` 时机指的就是这些一个个具体的时间点; | |||
| 该函数的参数 ``event`` 需要是一个 ``Event`` 实例,其使用方式见下方的例子; | |||
| 一个十分需要注意的事情在于您需要保证您添加的 callback 函数 ``fn`` 的参数与对应的 callback 时机所需要的参数保持一致,更准确地说, | |||
| 是与 :class:`fastNLP.core.callbacks.Callback` 中的对应的 callback 函数的参数保持一致;例如如果 | |||
| 您想要在 ``on_after_trainer_initialized`` 这个时机添加一个您自己的 callback 函数,您需要保证其参数为 ``trainer, driver``; | |||
| 最后用一句话总结:对于您想要加入的一个 callback 函数,您首先需要确定您想要将该函数加入的 callback 时机,然后通过 ``Event.on_***()`` | |||
| 拿到具体的 event 实例;再去 :class:`fastNLP.core.callbacks.Callback` 中确定该 callback 时机的 callback 函数的参数应当是怎样的; | |||
| 例如: | |||
| .. code-block:: | |||
| from fastNLP import Trainer, Event | |||
| # Trainer 初始化 | |||
| trainer = Trainer(...) | |||
| # 定义您自己的 callback 函数,需要注意的是该函数的参数需要与您要添加的 callback 时机所需要的参数保持一致;因为我们要将该函数加入到 | |||
| # on_after_trainer_initialized 这个 callback 时机,因此我们这里的 | |||
| def my_callback_fn(trainer, driver): | |||
| # do something | |||
| # 您可以在函数内部使用 trainer 和 driver,我们会将这两个实例注入进去; | |||
| # 添加到 trainer 中; | |||
| trainer.add_callback_fn(Event.on_after_trainer_initialized(), my_callback_fn) | |||
| .. note:: | |||
| 该函数与 ``Trainer.on`` 函数提供的作用相同,它们所需要的参数也基本相同,区别在于 ``Trainer.on`` 用于 ``Trainer`` 初始化前,而 | |||
| ``Trainer.add_callback_fn`` 用于 ``Trainer`` 初始化之后; | |||
| 更为具体的解释见 :meth:`~fastNLP.core.controllers.Trainer.on`; | |||
| """ | |||
| if not isinstance(event, Event): | |||
| raise ValueError("parameter event should only be `Event` type.") | |||
| @@ -621,6 +708,7 @@ class Trainer(TrainerEventTrigger): | |||
| def on(cls, event: Event, marker: Optional[str] = None): | |||
| r""" | |||
| 函数修饰器,用户可以使用该函数来方便地将一个函数转变为 callback 函数,从而进行训练流程中的控制; | |||
| 支持的 event 时机有以下这些,其执行的时机顺序也如下所示。每个时机装饰的函数应该接受的参数列表也如下所示,例如:: | |||
| Trainer.__init__(): | |||
| @@ -655,7 +743,15 @@ class Trainer(TrainerEventTrigger): | |||
| on_load_model(trainer)/on_save_checkpoint(trainer)/on_load_checkpoint(trainer)将根据需要在Trainer.run()中 | |||
| 特定的时间调用。 | |||
| Example:: | |||
| .. note:: | |||
| 对于 event 的解释,建议先阅读 :meth:`~fastNLP.core.controllers.Trainer.add_callback_fn` 的文档; | |||
| 当生成一个具体的 ``Event`` 实例时,可以指定 ``every、once、filter_fn`` 这三个参数来控制您的 callback 函数的调用频率,例如当您 | |||
| 指定 ``Event.on_train_epoch_begin(every=3)`` 时,其表示每隔三个 epoch 运行一次您的 callback 函数;对于这三个参数的更具体的解释, | |||
| 请见 :class:`fastNLP.core.callbacks.Event`; | |||
| Example1:: | |||
| from fastNLP import Event | |||
| @Trainer.on(Event.on_save_model()) | |||
| @@ -673,42 +769,40 @@ class Trainer(TrainerEventTrigger): | |||
| # do something | |||
| # 以上函数会在 Trainer 每个新的 batch 开始的时候执行,但是是两个 batch 才执行一次。 | |||
| .. note:: | |||
| 例如: | |||
| Example2:: | |||
| .. code-block:: | |||
| @Trainer.on(Event.on_train_begin()) | |||
| def fn1(trainer): | |||
| ... | |||
| @Trainer.on(Event.on_train_begin()) | |||
| def fn1(trainer): | |||
| ... | |||
| @Trainer.on(Event.on_train_epoch_begin()) | |||
| def fn2(trainer): | |||
| ... | |||
| @Trainer.on(Event.on_train_epoch_begin()) | |||
| def fn2(trainer): | |||
| ... | |||
| trainer1 = Trainer( | |||
| ..., | |||
| marker='trainer1' | |||
| ) | |||
| trainer1 = Trainer( | |||
| ..., | |||
| marker='trainer1' | |||
| ) | |||
| @Trainer.on(Event.on_fetch_data_begin()) | |||
| def fn3(trainer): | |||
| ... | |||
| @Trainer.on(Event.on_fetch_data_begin()) | |||
| def fn3(trainer): | |||
| ... | |||
| trainer2 = Trainer( | |||
| ..., | |||
| marker='trainer2' | |||
| ) | |||
| trainer2 = Trainer( | |||
| ..., | |||
| marker='trainer2' | |||
| ) | |||
| 这段代码意味着 ``fn1`` 和 ``fn2`` 会被加入到 ``trainer1``,``fn3`` 会被加入到 ``trainer2``; | |||
| 注意如果你使用该函数修饰器来为你的训练添加 callback,请务必保证你加入 callback 函数的代码在实例化 `Trainer` 之前; | |||
| 补充性的解释见 :meth:`~fastNLP.core.controllers.Trainer.add_callback_fn`; | |||
| :param event: 特定的 callback 时机,用户需要为该 callback 函数指定其属于哪一个 callback 时机。每个时机运行的函数应该包含 | |||
| 特定的参数,可以通过上述说明查阅。 | |||
| :param marker: 用来标记该 callback 函数属于哪几个具体的 trainer 实例;两个特殊情况:1.当 `marker` 为 None(默认情况)时, | |||
| 表示该 callback 函数只属于代码下方最近的一个 trainer 实例;2.当 `marker` 为 'all' 时,该 callback 函数会被所有的 trainer | |||
| :param marker: 用来标记该 callback 函数属于哪几个具体的 trainer 实例;两个特殊情况:1.当 ``marker`` 为 None(默认情况)时, | |||
| 表示该 callback 函数只属于代码下方最近的一个 trainer 实例;2.当 ``marker`` 为 'all' 时,该 callback 函数会被所有的 trainer | |||
| 实例使用; | |||
| :return: 返回原函数; | |||
| """ | |||
| @@ -722,7 +816,7 @@ class Trainer(TrainerEventTrigger): | |||
| return wrapper | |||
| def _fetch_matched_fn_callbacks(self): | |||
| """ | |||
| r""" | |||
| 因为对于使用装饰器加入的函数 callback,我们是加在类属性中,因此在初始化一个具体的 trainer 实例后,我们需要从 Trainer 的 | |||
| callback 类属性中将属于其的 callback 函数拿到,然后加入到 callback_manager 中; | |||
| """ | |||
| @@ -164,7 +164,7 @@ class PaddleDataLoader(DataLoader): | |||
| """ | |||
| 获取当前 ``batch`` 中每条数据对应的索引。 | |||
| :return: 当前 ``batch`` 数据的索引 | |||
| :return: 当前 ``batch`` 数据的索引; | |||
| """ | |||
| return self.cur_batch_indices | |||
| @@ -172,7 +172,7 @@ class TorchDataLoader(DataLoader): | |||
| """ | |||
| 获取当前 ``batch`` 中每条数据对应的索引。 | |||
| :return: 当前 ``batch`` 数据的索引 | |||
| :return: 当前 ``batch`` 数据的索引; | |||
| """ | |||
| return self.cur_batch_indices | |||
| @@ -400,16 +400,22 @@ class DataSet: | |||
| new_field_name: str = None, num_proc: int = 0, | |||
| progress_desc: str = None, show_progress_bar: bool = True): | |||
| r""" | |||
| 将 :class:`~DataSet` 每个 ``instance`` 中为 ``field_name`` 的 ``field`` 传给函数 ``func``,并获取函数的返回值。 | |||
| :param field_name: 传入 ``func`` 的 ``field`` 名称。 | |||
| :param func: 一个函数,其输入是 ``instance`` 中名为 ``field_name`` 的 ``field`` 的内容。 | |||
| :param new_field_name: 将 ``func`` 返回的内容放入到 ``new_field_name`` 对应的 ``field`` 中,如果名称与已有的 ``field`` 相同 | |||
| 则进行覆盖。如果为 ``None`` 则不会覆盖和创建 ``field`` 。 | |||
| :param num_proc: 使用进程的数量。请注意,由于 ``python`` 语言的特性,使用了多少进程就会导致多少倍内存的增长。 | |||
| :param progress_desc: 进度条的描述字符,默认为 ``Main``。 | |||
| :param show_progress_bar: 是否展示进度条;默认为展示。 | |||
| :return: 从函数 ``func`` 中得到的返回值。 | |||
| 将 :class:`~DataSet` 每个 ``instance`` 中为 ``field_name`` 的 ``field`` 传给函数 ``func``,并写入到 ``new_field_name`` | |||
| 中。 | |||
| :param field_name: 传入 ``func`` 的 ``field`` 名称; | |||
| :param func: 对指定 ``field`` 进行处理的函数,注意其输入应为 ``instance`` 中名为 ``field_name`` 的 ``field`` 的内容; | |||
| :param new_field_name: 函数执行结果写入的 ``field`` 名称。该函数会将 ``func`` 返回的内容放入到 ``new_field_name`` 对 | |||
| 应的 ``field`` 中,注意如果名称与已有的 ``field`` 相同则会进行覆盖。如果为 ``None`` 则不会覆盖和创建 ``field`` ; | |||
| :param num_proc: 使用进程的数量。 | |||
| .. note:: | |||
| 由于 ``python`` 语言的特性,设置该参数后会导致相应倍数的内存增长,这可能会对您程序的执行带来一定的影响。 | |||
| :param progress_desc: 进度条的描述字符,默认为 ``Main``; | |||
| :param show_progress_bar: 是否在处理过程中展示进度条; | |||
| :return: 从函数 ``func`` 中得到的返回值; | |||
| """ | |||
| assert len(self) != 0, "Null DataSet cannot use apply_field()." | |||
| if not self.has_field(field_name=field_name): | |||
| @@ -23,9 +23,9 @@ def choose_driver(model, driver: Union[str, Driver], device: Optional[Union[int, | |||
| elif driver in {"jittor"}: | |||
| from fastNLP.core.drivers.jittor_driver.initialize_jittor_driver import initialize_jittor_driver | |||
| return initialize_jittor_driver(driver, device, model, **kwargs) | |||
| elif driver in {"paddle", "fleet"}: | |||
| elif driver in {"paddle"}: | |||
| from fastNLP.core.drivers.paddle_driver.initialize_paddle_driver import initialize_paddle_driver | |||
| return initialize_paddle_driver(driver, device, model, **kwargs) | |||
| else: | |||
| raise ValueError("Parameter `driver` can only be one of these values: ['torch', 'fairscale', " | |||
| "'jittor', 'paddle', 'fleet'].") | |||
| "'jittor', 'paddle'].") | |||
| @@ -7,18 +7,22 @@ from fastNLP.envs.imports import _NEED_IMPORT_JITTOR | |||
| if _NEED_IMPORT_JITTOR: | |||
| import jittor | |||
| __all__ = [] | |||
| def initialize_jittor_driver(driver: str, device: Union[str, int, List[int]], model: jittor.Module, **kwargs) -> JittorDriver: | |||
| r""" | |||
| 用来根据参数 `driver` 和 `device` 来确定并且初始化一个具体的 `Driver` 实例然后返回回去; | |||
| 在这个函数中,我们会根据用户设置的device来确定JittorDriver的mode。 | |||
| 用来根据参数 ``device`` 来确定并且初始化一个具体的 ``Driver`` 实例然后返回回去。 | |||
| .. todo:: | |||
| 创建多卡的 driver | |||
| :param driver: 该参数的值应为以下之一:["jittor"]; | |||
| :param device: jittor运行的设备 | |||
| :param driver: 该参数的值应为以下之一:``["jittor"]``; | |||
| :param device: ``jittor`` 运行的设备; | |||
| :param model: 训练或者评测的具体的模型; | |||
| :param kwargs: | |||
| :return: 返回一个元组,元组的第一个值是具体的基于 jittor 的 `Driver` 实例,元组的第二个值是该 driver 的名字(用于检测一个脚本中 | |||
| 先后 driver 的次序的正确问题); | |||
| :return: :class:`~fastNLP.core.JittorSingleDriver` 或 :class:`~fastNLP.core.JittorMPIDriver` 实例; | |||
| """ | |||
| if driver not in {"jittor"}: | |||
| @@ -24,7 +24,17 @@ if _NEED_IMPORT_JITTOR: | |||
| class JittorDriver(Driver): | |||
| r""" | |||
| Jittor 框架的 Driver | |||
| ``Jittor`` 框架的 ``Driver`` | |||
| .. note:: | |||
| 这是一个正在开发中的功能,敬请期待。 | |||
| .. todo:: | |||
| 实现 fp16 的设置,且支持 cpu 和 gpu 的切换; | |||
| 实现用于断点重训的 save 和 load 函数; | |||
| """ | |||
| def __init__(self, model, fp16: bool = False, **kwargs): | |||
| @@ -13,6 +13,14 @@ __all__ = [ | |||
| ] | |||
| class JittorMPIDriver(JittorDriver): | |||
| """ | |||
| 执行 ``Jittor`` 框架下分布式训练的 ``Driver``。 | |||
| .. note:: | |||
| 这是一个正在开发中的功能,敬请期待。 | |||
| """ | |||
| def __init__( | |||
| self, | |||
| model, | |||
| @@ -16,8 +16,17 @@ __all__ = [ | |||
| class JittorSingleDriver(JittorDriver): | |||
| r""" | |||
| 用于 cpu 和 单卡 gpu 运算 | |||
| TODO: jittor 的 fp16 | |||
| ``Jittor`` 框架下用于 ``cpu`` 和单卡 ``gpu`` 运算的 ``Driver``。 | |||
| .. note:: | |||
| 这是一个正在开发中的功能,敬请期待。 | |||
| .. todo:: | |||
| 支持 cpu 和 gpu 的切换; | |||
| 实现断点重训中替换 dataloader 的 set_dist_repro_dataloader 函数 | |||
| """ | |||
| def __init__(self, model, device=None, fp16: bool = False, **kwargs): | |||
| @@ -30,11 +39,6 @@ class JittorSingleDriver(JittorDriver): | |||
| self.world_size = 1 | |||
| def step(self): | |||
| """ | |||
| jittor optimizers 的step函数可以传入参数loss | |||
| 此时会同时进行 zero_grad 和 backward | |||
| 为了统一,这里暂不使用这样的方式 | |||
| """ | |||
| for optimizer in self.optimizers: | |||
| optimizer.step() | |||
| @@ -5,10 +5,11 @@ from fastNLP.envs.imports import _NEED_IMPORT_JITTOR | |||
| if _NEED_IMPORT_JITTOR: | |||
| import jittor | |||
| __all__ = [] | |||
| class DummyGradScaler: | |||
| """ | |||
| 用于仿造的GradScaler对象,防止重复写大量的if判断 | |||
| """ | |||
| def __init__(self, *args, **kwargs): | |||
| pass | |||
| @@ -1,8 +1,6 @@ | |||
| import os | |||
| from typing import List, Union, Optional, Dict, Tuple, Callable | |||
| from fastNLP.core.utils.paddle_utils import get_device_from_visible | |||
| from .paddle_driver import PaddleDriver | |||
| from .fleet_launcher import FleetLauncher | |||
| from .utils import ( | |||
| @@ -19,7 +17,9 @@ from fastNLP.core.utils import ( | |||
| check_user_specific_params, | |||
| is_in_paddle_dist, | |||
| is_in_paddle_dist, | |||
| get_paddle_device_id, | |||
| ) | |||
| from fastNLP.core.utils.paddle_utils import _convert_data_device | |||
| from fastNLP.envs.distributed import rank_zero_rm | |||
| from fastNLP.core.samplers import ( | |||
| ReproduceBatchSampler, | |||
| @@ -31,7 +31,12 @@ from fastNLP.core.samplers import ( | |||
| re_instantiate_sampler, | |||
| conversion_between_reproducible_and_unrepeated_sampler, | |||
| ) | |||
| from fastNLP.envs.env import FASTNLP_DISTRIBUTED_CHECK, FASTNLP_GLOBAL_SEED, FASTNLP_NO_SYNC | |||
| from fastNLP.envs.env import ( | |||
| FASTNLP_DISTRIBUTED_CHECK, | |||
| FASTNLP_GLOBAL_SEED, | |||
| FASTNLP_NO_SYNC, | |||
| USER_CUDA_VISIBLE_DEVICES, | |||
| ) | |||
| from fastNLP.core.log import logger | |||
| if _NEED_IMPORT_PADDLE: | |||
| @@ -51,7 +56,7 @@ class PaddleFleetDriver(PaddleDriver): | |||
| def __init__( | |||
| self, | |||
| model, | |||
| parallel_device: Optional[Union[List[int], int]], | |||
| parallel_device: Optional[Union[List[str], str]], | |||
| is_pull_by_paddle_run: bool = False, | |||
| fp16: bool = False, | |||
| **kwargs | |||
| @@ -185,6 +190,8 @@ class PaddleFleetDriver(PaddleDriver): | |||
| 不管是什么情况,`PaddleFleetDriver` 在 `setup` 函数的最后,都会将所有进程的 pid 主动记录下来,这样当一个进程出现 exception 后, | |||
| driver 的 on_exception 函数就会被 trainer 调用,其会调用 os.kill 指令将其它进程 kill 掉; | |||
| """ | |||
| if USER_CUDA_VISIBLE_DEVICES not in os.environ: | |||
| raise RuntimeError("To run paddle distributed training, please set `FASTNLP_BACKEND` to 'paddle' before using FastNLP.") | |||
| super(PaddleFleetDriver, self).__init__(model, fp16=fp16, **kwargs) | |||
| # 如果不是通过 launch 启动,要求用户必须传入 parallel_device | |||
| @@ -213,25 +220,6 @@ class PaddleFleetDriver(PaddleDriver): | |||
| "you initialize the paddle distribued process out of our control.") | |||
| self.outside_fleet = True | |||
| # 用户只有将模型上传到对应机器上后才能用 DataParallel 包裹,因此如果用户在外面初始化了 Fleet,那么在 PaddleFleetDriver 中 | |||
| # 我们就直接将 model_device 置为 None; | |||
| self._model_device = None | |||
| # 当参数 `device` 为 None 时并且该参数不为 None,表示将对应的数据移到指定的机器上; | |||
| self._data_device = kwargs.get("data_device", None) | |||
| if self._data_device is not None: | |||
| if isinstance(self._data_device, int): | |||
| if self._data_device < 0: | |||
| raise ValueError("Parameter `data_device` can not be smaller than 0.") | |||
| _could_use_device_num = paddle.device.cuda.device_count() | |||
| if self._data_device >= _could_use_device_num: | |||
| raise ValueError("The gpu device that parameter `device` specifies is not existed.") | |||
| self._data_device = f"gpu:{self._data_device}" | |||
| elif not isinstance(self._data_device, str): | |||
| raise ValueError("Parameter `device` is wrong type, please check our documentation for the right use.") | |||
| if self.outside_fleet and paddle.device.get_device() != self._data_device: | |||
| logger.warning("`Parameter data_device` is not equal to paddle.deivce.get_device(), " | |||
| "please keep them equal to avoid some potential bugs.") | |||
| self.world_size = None | |||
| self.global_rank = 0 | |||
| @@ -304,7 +292,8 @@ class PaddleFleetDriver(PaddleDriver): | |||
| else: | |||
| # 已经设置过一次,保证参数必须是一样的 | |||
| pre_gpus = os.environ[FASTNLP_DISTRIBUTED_CHECK] | |||
| pre_gpus = [int (x) for x in pre_gpus.split(",")] | |||
| pre_gpus = [int(x) for x in pre_gpus.split(",")] | |||
| cur_gpus = [get_paddle_device_id(g) for g in self.parallel_device] | |||
| if sorted(pre_gpus) != sorted(self.parallel_device): | |||
| raise RuntimeError("Notice you are using `PaddleFleetDriver` after one instantiated `PaddleFleetDriver`, it is not" | |||
| "allowed that your second `PaddleFleetDriver` has a new setting of parameters `parallel_device`.") | |||
| @@ -410,8 +399,6 @@ class PaddleFleetDriver(PaddleDriver): | |||
| @property | |||
| def data_device(self): | |||
| if self.outside_fleet: | |||
| return self._data_device | |||
| return self.model_device | |||
| def model_call(self, batch, fn: Callable, signature_fn: Optional[Callable]) -> Dict: | |||
| @@ -565,7 +552,7 @@ class PaddleFleetDriver(PaddleDriver): | |||
| def broadcast_object(self, obj, src:int=0, group=None, **kwargs): | |||
| # 因为设置了CUDA_VISIBLE_DEVICES,可能会引起错误 | |||
| device = get_device_from_visible(self.data_device) | |||
| device = _convert_data_device(self.data_device) | |||
| return fastnlp_paddle_broadcast_object(obj, src, device=device, group=group) | |||
| def all_gather(self, obj, group=None) -> List: | |||
| @@ -11,11 +11,14 @@ from fastNLP.envs.env import ( | |||
| FASTNLP_LOG_LEVEL, | |||
| FASTNLP_GLOBAL_SEED, | |||
| ) | |||
| from fastNLP.core.utils import get_paddle_device_id | |||
| from .utils import ( | |||
| find_free_ports, | |||
| reset_seed, | |||
| ) | |||
| __all__ = [] | |||
| # 记录各个进程信息 | |||
| class SubTrainer(object): | |||
| """ | |||
| @@ -34,11 +37,11 @@ class FleetLauncher: | |||
| """ | |||
| def __init__( | |||
| self, | |||
| devices: List[int], | |||
| devices: List[str], | |||
| output_from_new_proc: str = "only_error" | |||
| ): | |||
| self.devices = devices | |||
| self.devices = [ get_paddle_device_id(g) for g in devices] | |||
| self.output_from_new_proc = output_from_new_proc | |||
| self.setup() | |||
| @@ -7,50 +7,58 @@ from .single_device import PaddleSingleDriver | |||
| from .fleet import PaddleFleetDriver | |||
| from fastNLP.envs.imports import _NEED_IMPORT_PADDLE | |||
| from fastNLP.core.utils import is_in_paddle_launch_dist | |||
| from fastNLP.envs.env import USER_CUDA_VISIBLE_DEVICES | |||
| from fastNLP.core.utils import is_in_paddle_launch_dist, get_paddle_gpu_str | |||
| from fastNLP.core.log import logger | |||
| if _NEED_IMPORT_PADDLE: | |||
| import paddle | |||
| __all__ = [] | |||
| def initialize_paddle_driver(driver: str, device: Optional[Union[str, int, List[int]]], | |||
| model: "paddle.nn.Layer", **kwargs) -> PaddleDriver: | |||
| r""" | |||
| 用来根据参数 `driver` 和 `device` 来确定并且初始化一个具体的 `Driver` 实例然后返回回去; | |||
| 1、如果检测到当前进程为用户通过 `python -m paddle.distributed.launch xxx.py` 方式拉起的,则将 | |||
| 设备自动设置为用户指定的设备(由于我们在引入 fastNLP 进行了特殊的设置,因此可以通过 `CUDA_VISIBLE_DEVICES` 获取) | |||
| 2、如果检测到输入的 `driver` 是 `paddle` 但 `device` 包含了多个设备,那么我们会给出警告并且自动返回多卡的 Driver | |||
| 3、如果检测到输入的 `driver` 是 `fleet` 但 `device` 仅有一个设备,那么我们会给出警告但仍旧返回多卡的 Driver | |||
| 用来根据参数 ``device`` 来确定并且初始化一个具体的 ``Driver`` 实例。 | |||
| 1. 如果检测到当前进程为用户通过 ``python -m paddle.distributed.launch xxx.py`` 方式拉起的,则将 | |||
| 设备自动设置为用户指定的设备(由于我们要求分布式训练必须进行 ``backend`` 的设置,因此可以通过 ``CUDA_VISIBLE_DEVICES`` 获取) | |||
| 2. 如果 ``device`` 包含了多个设备,则返回一个 :class:`~fastNLP.core.PaddleFleetDriver` 实例,否则返回 | |||
| 单卡的 :class:`~fastNLP.core.PaddleSingleDriver` 实例 | |||
| :param driver: 使用的 ``driver`` 类型,在这个函数中仅支持 ``paddle`` | |||
| :param device: 该参数的格式与 `Trainer` 对参数 `device` 的要求一致; | |||
| :param driver: 使用的 ``driver`` 类型,在这个函数中仅支持 ``paddle``; | |||
| :param device: 该参数的格式与 ``Trainer`` 对参数 ``device`` 的要求一致; | |||
| :param model: 训练或者评测的具体的模型; | |||
| :return: 返回构造的 `Driver` 实例。 | |||
| :return: 一个 :class:`~fastNLP.core.PaddleSingleDriver` 或 :class:`~fastNLP.core.PaddleFleetDriver` 实例; | |||
| """ | |||
| if driver != "paddle": | |||
| raise ValueError("When initialize PaddleDriver, parameter `driver` must be 'paddle'.") | |||
| user_visible_devices = os.getenv(USER_CUDA_VISIBLE_DEVICES) | |||
| if is_in_paddle_launch_dist(): | |||
| if user_visible_devices is None: | |||
| raise RuntimeError("To run paddle distributed training, please set `FASTNLP_BACKEND` to 'paddle' before using FastNLP.") | |||
| if device is not None: | |||
| logger.warning_once("Parameter `device` would be ignored when you are using `paddle.distributed.launch` to pull " | |||
| "up your script. And we will directly get the local device via " | |||
| "and `os.environ['CUDA_VISIBLE_DEVICES']``.") | |||
| device = [int(g) for g in os.environ["CUDA_VISIBLE_DEVICES"].split(",")] | |||
| # TODO 目前一个进程仅对应一个卡,所以暂时传入一个 int | |||
| "up your script. And we will directly get the local device via environment variables.") | |||
| _visible_list = user_visible_devices.split(",") | |||
| device = [ f"gpu:{_visible_list.index(g) }" for g in os.environ["CUDA_VISIBLE_DEVICES"].split(",")] | |||
| # TODO 目前一个进程仅对应一个卡,所以暂时传入单个 | |||
| return PaddleFleetDriver(model, device[0], True, **kwargs) | |||
| user_visible_devices = os.getenv("USER_CUDA_VISIBLE_DEVICES") | |||
| if user_visible_devices is None: | |||
| raise RuntimeError("`USER_CUDA_VISIBLE_DEVICES` cannot be None, please check if you have set " | |||
| "`FASTNLP_BACKEND` to 'paddle' before using FastNLP.") | |||
| _could_use_device_num = len(user_visible_devices.split(",")) | |||
| _could_use_device_num = paddle.device.cuda.device_count() | |||
| else: | |||
| _could_use_device_num = len(user_visible_devices.split(",")) | |||
| if isinstance(device, int): | |||
| if device < 0 and device != -1: | |||
| raise ValueError("Parameter `device` can only be '-1' when it is smaller than 0.") | |||
| if device >= _could_use_device_num: | |||
| raise ValueError("The gpu device that parameter `device` specifies is not existed.") | |||
| if device == -1: | |||
| device = list(range(_could_use_device_num)) | |||
| device = [ get_paddle_gpu_str(g) for g in range(_could_use_device_num)] | |||
| elif isinstance(device, Sequence) and not isinstance(device, str): | |||
| device = list(set(device)) | |||
| for each in device: | |||
| @@ -61,8 +69,10 @@ def initialize_paddle_driver(driver: str, device: Optional[Union[str, int, List[ | |||
| elif each >= _could_use_device_num: | |||
| raise ValueError("When parameter `device` is 'Sequence' type, the value in it should not be bigger than" | |||
| " the available gpu number.") | |||
| device = [get_paddle_gpu_str(g) for g in device] | |||
| elif device is not None and not isinstance(device, str): | |||
| raise ValueError("Parameter `device` is wrong type, please check our documentation for the right use.") | |||
| if isinstance(device, List): | |||
| return PaddleFleetDriver(model, device, **kwargs) | |||
| else: | |||
| @@ -7,10 +7,13 @@ from dataclasses import dataclass | |||
| import numpy as np | |||
| from fastNLP.envs.env import USER_CUDA_VISIBLE_DEVICES | |||
| from .utils import _build_fp16_env, optimizer_state_to_device, DummyGradScaler | |||
| from fastNLP.envs.imports import _NEED_IMPORT_PADDLE | |||
| from fastNLP.core.drivers.driver import Driver | |||
| from fastNLP.core.utils import apply_to_collection, paddle_move_data_to_device, get_device_from_visible | |||
| from fastNLP.core.utils import apply_to_collection, paddle_move_data_to_device | |||
| from fastNLP.core.utils.paddle_utils import _convert_data_device | |||
| from fastNLP.envs import ( | |||
| FASTNLP_SEED_WORKERS, | |||
| FASTNLP_MODEL_FILENAME, | |||
| @@ -369,7 +372,7 @@ class PaddleDriver(Driver): | |||
| :return: 将移动到指定机器上的 batch 对象返回; | |||
| """ | |||
| device = get_device_from_visible(self.data_device) | |||
| device = _convert_data_device(self.data_device) | |||
| return paddle_move_data_to_device(batch, device) | |||
| @staticmethod | |||
| @@ -8,10 +8,10 @@ from fastNLP.envs.imports import _NEED_IMPORT_PADDLE | |||
| from fastNLP.envs.env import USER_CUDA_VISIBLE_DEVICES | |||
| from fastNLP.core.utils import ( | |||
| auto_param_call, | |||
| get_device_from_visible, | |||
| get_paddle_gpu_str, | |||
| get_paddle_device_id, | |||
| ) | |||
| from fastNLP.core.utils.paddle_utils import _convert_data_device | |||
| from fastNLP.core.utils.utils import _get_fun_msg | |||
| from fastNLP.core.samplers import ( | |||
| ReproducibleBatchSampler, | |||
| @@ -40,9 +40,6 @@ class PaddleSingleDriver(PaddleDriver): | |||
| raise ValueError("`paddle.DataParallel` is not supported in `PaddleSingleDriver`") | |||
| cuda_visible_devices = os.getenv(USER_CUDA_VISIBLE_DEVICES) | |||
| if cuda_visible_devices is None: | |||
| raise RuntimeError("`USER_CUDA_VISIBLE_DEVICES` cannot be None, please check if you have set " | |||
| "`FASTNLP_BACKEND` to 'paddle' before using FastNLP.") | |||
| if cuda_visible_devices == "": | |||
| device = "cpu" | |||
| logger.info("You have set `CUDA_VISIBLE_DEVICES` to '' in system environment variable, and we are gonna to" | |||
| @@ -54,11 +51,9 @@ class PaddleSingleDriver(PaddleDriver): | |||
| raise ValueError("Parameter `device` can not be None in `PaddleSingleDriver`.") | |||
| if device != "cpu": | |||
| if isinstance(device, int): | |||
| device_id = device | |||
| else: | |||
| device_id = get_paddle_device_id(device) | |||
| os.environ["CUDA_VISIBLE_DEVICES"] = cuda_visible_devices.split(",")[device_id] | |||
| device_id = get_paddle_device_id(device) | |||
| if cuda_visible_devices is not None: | |||
| os.environ["CUDA_VISIBLE_DEVICES"] = cuda_visible_devices.split(",")[device_id] | |||
| self.model_device = get_paddle_gpu_str(device) | |||
| self.local_rank = 0 | |||
| @@ -69,7 +64,8 @@ class PaddleSingleDriver(PaddleDriver): | |||
| r""" | |||
| 该函数用来初始化训练环境,用于设置当前训练的设备,并将模型迁移到对应设备上。 | |||
| """ | |||
| device = get_device_from_visible(self.model_device, output_type=str) | |||
| device = _convert_data_device(self.data_device) | |||
| paddle.device.set_device(device) | |||
| with contextlib.redirect_stdout(None): | |||
| self.model.to(device) | |||
| @@ -10,19 +10,18 @@ from .ddp import TorchDDPDriver | |||
| from fastNLP.core.log import logger | |||
| from fastNLP.envs import FASTNLP_BACKEND_LAUNCH | |||
| __all__ = [] | |||
| def initialize_torch_driver(driver: str, device: Optional[Union[str, "torch.device", int, List[int]]], | |||
| model: "torch.nn.Module", **kwargs) -> TorchDriver: | |||
| r""" | |||
| 用来根据参数 `driver` 和 `device` 来确定并且初始化一个具体的 `Driver` 实例然后返回回去; | |||
| 注意如果输入的 `device` 如果和 `driver` 对应不上就直接报错; | |||
| 用来根据参数 ``driver` 和 ``device`` 来确定并且初始化一个具体的 ``Driver`` 实例然后返回回去; | |||
| :param driver: 该参数的值应为以下之一:["torch", "torch_ddp", "fairscale"]; | |||
| :param device: 该参数的格式与 `Trainer` 对参数 `device` 的要求一致; | |||
| :param driver: 该参数的值应为以下之一:``["torch", "fairscale"]``; | |||
| :param device: 该参数的格式与 ``Trainer`` 对参数 ``device`` 的要求一致; | |||
| :param model: 训练或者评测的具体的模型; | |||
| :return: 返回一个元组,元组的第一个值是具体的基于 pytorch 的 `Driver` 实例,元组的第二个值是该 driver 的名字(用于检测一个脚本中 | |||
| 先后 driver 的次序的正确问题); | |||
| :return: 返回一个 :class:`~fastNLP.core.TorchSingleDriver` 或 :class:`~fastNLP.core.TorchDDPDriver` 实例; | |||
| """ | |||
| # world_size 和 rank | |||
| if FASTNLP_BACKEND_LAUNCH in os.environ: | |||
| @@ -1,7 +1,7 @@ | |||
| __all__ = [ | |||
| 'print' | |||
| ] | |||
| from logging import INFO | |||
| from .logger import logger | |||
| @@ -22,4 +22,6 @@ def print(*args, sep=' ', end='\n', file=None, flush=False): | |||
| :return: | |||
| """ | |||
| line = sep.join(map(str, args)) | |||
| logger.info(line) | |||
| if logger.isEnabledFor(INFO): | |||
| kwargs = logger._add_rank_info({}) | |||
| logger._log(INFO, line, **kwargs) | |||
| @@ -1,12 +1,14 @@ | |||
| import os | |||
| from typing import List, Any | |||
| import numpy as np | |||
| from fastNLP.core.metrics.backend import Backend | |||
| from fastNLP.core.utils.paddle_utils import paddle_to, get_device_from_visible | |||
| from fastNLP.core.utils.paddle_utils import paddle_to, _convert_data_device | |||
| from fastNLP.core.metrics.utils import AggregateMethodError | |||
| from fastNLP.core.drivers.paddle_driver.dist_utils import fastnlp_paddle_all_gather | |||
| from fastNLP.envs.imports import _NEED_IMPORT_PADDLE | |||
| from fastNLP.envs.env import USER_CUDA_VISIBLE_DEVICES | |||
| if _NEED_IMPORT_PADDLE: | |||
| import paddle | |||
| @@ -79,7 +81,7 @@ class PaddleBackend(Backend): | |||
| raise ValueError(f"tensor: {tensor} can not convert to ndarray!") | |||
| def move_tensor_to_device(self, tensor, device): | |||
| device = get_device_from_visible(device) | |||
| device = _convert_data_device(device) | |||
| return paddle_to(tensor, device) | |||
| def all_gather_object(self, obj, group=None) -> List: | |||
| @@ -84,7 +84,7 @@ class Metric: | |||
| def _sync_get_metric(self, get_metric): | |||
| @functools.wraps(get_metric) | |||
| def _wrap_get_metric(*args, **kwargs): | |||
| assert self._updated, f"You have to call `{self.__class__.__name__}` update() function before calling " \ | |||
| assert self._updated, f"You have to call `{self.__class__.__name__}'s update() function before calling " \ | |||
| f"get_metric()." | |||
| with self.sync(recover=True, aggregate=self.aggregate_when_get_metric): | |||
| results = get_metric(*args, **kwargs) | |||
| @@ -366,17 +366,22 @@ class BucketedBatchSampler(ReproducibleBatchSampler): | |||
| def __init__(self, dataset, length: Union[List[int], str], batch_size:int = 32, num_batch_per_bucket:int = 10, | |||
| shuffle: bool = True, drop_last: bool = False, seed: int = 0, **kwargs): | |||
| """ | |||
| 首先按照 sample 的长度排序,然后按照 batch_size*num_batch_per_bucket 为一个桶的大小,sample 只会在这个桶内进行组合,这样 | |||
| 每个 batch 中的 padding 数量会比较少 (因为桶内的数据的长度都接近)。 | |||
| 首先按照 ``sample`` 的长度排序,然后按照 batch_size*num_batch_per_bucket 为一个桶的大小,``sample`` 只会在这个桶内进行组 | |||
| 合,这样每个 ``batch`` 中的 ``padding`` 数量会比较少 (因为桶内的数据的长度都接近)。 | |||
| :param dataset: 实现了 __len__ 方法的数据容器。 | |||
| :param length: 如果为 List,应当与 dataset 有一样的长度,表示 dataset 中每个元素的数量;仅当传入的 dataset 为 fastNLP 的 | |||
| DataSet 时支持传入 str,会将该str理解为 dataset 的 field 名称,若 field 中的元素为 int,则认为该值是 sample 的长度。 | |||
| 如果否则使用 len() 函数得到每个 sample 中这个 field 的长度。 | |||
| :param length: 每条数据的长度。 | |||
| * 为 ``List[int]`` 时 | |||
| 应当与 dataset 有一样的长度,表示 dataset 中每个元素的数量; | |||
| * 为 ``str`` 时 | |||
| 仅当传入的 ``dataset`` 是 :class:`fastNLP.DataSet` 时,允许传入 `str` ,该 `str` 将被认为是 ``dataset`` 中的 | |||
| ``field`` 。若 field 中的元素为 ``int``,则认为该值是 sample 的长度;若不为 ``int`` ,则尝试使用 ``len`` 方法 | |||
| 获取该 ``field`` 中每个元素的长度。 | |||
| :param batch_size: 每个 batch 的大小 | |||
| :param num_batch_per_bucket: 多少个 batch 组成一个桶,数据只会在一个桶内进行 shuffle 。 | |||
| :param shuffle: 如果为 True,将不进行 shuffle,实际上数据会以从长到短的方式输出。 | |||
| :param drop_last: 如果最后一个 batch 的 sample 数量无法凑齐 batch_size 这么多,是否需要丢掉。 | |||
| :param num_batch_per_bucket: 多少个 ``batch`` 组成一个桶,数据只会在一个桶内进行 ``shuffle`` 。 | |||
| :param shuffle: 如果为 True,将不进行 ``shuffle``,实际上数据会以从长到短的方式输出。 | |||
| :param drop_last: 如果最后一个 `batch` 的 ``sample`` 数量无法凑齐 ``batch_size`` 这么多,是否需要丢掉。 | |||
| :param seed: 设置的随机数种子 | |||
| :param kwargs: fastNLP 保留使用 | |||
| """ | |||
| @@ -386,10 +391,12 @@ class BucketedBatchSampler(ReproducibleBatchSampler): | |||
| if not isinstance(length[0], int): | |||
| length = list(map(len, length)) | |||
| else: | |||
| assert len(length) == len(dataset), "When the dataset is not fastNLP.DataSet, " \ | |||
| "the length parameter can only be List[int]" | |||
| types = set(map(type, length)) | |||
| assert isinstance(length, list) and len(types)==1 and types.pop()==int, \ | |||
| "When the dataset is not fastNLP.DataSet, the length parameter can only be List[int]" | |||
| assert len(length) == len(dataset), "The length of `data` and `length` should be equal." | |||
| assert len(length) == len(dataset), f"The length of `dataset`({len(dataset)}) and " \ | |||
| f"`length`({len(length)}) should be equal." | |||
| self.dataset = dataset | |||
| self.length = np.array(length, dtype=int) # 按照长到短排列的序号。 | |||
| @@ -55,6 +55,7 @@ class ReproducibleSampler: | |||
| class RandomSampler(ReproducibleSampler): | |||
| def __init__(self, dataset, shuffle: bool = True, seed: int = 0, **kwargs): | |||
| """ | |||
| 随机顺序的 Sampler 。 | |||
| :param dataset: 实现了 __len__ 方法的数据容器 | |||
| :param shuffle: 是否在每次 iterate 的时候打乱顺序。 | |||
| @@ -169,9 +170,8 @@ class RandomSampler(ReproducibleSampler): | |||
| def set_epoch(self, epoch: int) -> None: | |||
| self.epoch = epoch | |||
| def set_distributed(self, num_replicas, rank, pad=True): | |||
| def set_distributed(self, num_replicas:int, rank:int, pad:bool=True): | |||
| """ | |||
| 该方法本质上等同于 ddp 情形下的没有完成的初始化,应当在初始化该 sampler 本身后立即被调用; | |||
| :param num_replicas: | |||
| :param rank: | |||
| @@ -215,7 +215,7 @@ class RandomSampler(ReproducibleSampler): | |||
| class SequentialSampler(RandomSampler): | |||
| def __init__(self, dataset, **kwargs): | |||
| """ | |||
| 按照顺序读取 dataset 。在多卡情况下,间隔读取,例如,在两卡情况下,卡0取 [0,2,4,..], 卡1取 [1,3,5...]。 | |||
| 按照顺序读取 ``dataset`` 。在多卡情况下,间隔读取,例如,在两卡情况下,卡 0 取 ``[0,2,4,..]``, 卡1取 ``[1,3,5...]`` 。 | |||
| :param dataset: 实现了 __len__ 方法的数据容器。 | |||
| :param kwargs: | |||
| @@ -285,13 +285,20 @@ class SequentialSampler(RandomSampler): | |||
| class SortedSampler(SequentialSampler): | |||
| def __init__(self, dataset, length:Union[str, List], **kwargs): | |||
| """ | |||
| 将 dataset 中的数据根据 length 从长到短进行迭代。在多卡情况下,由于padding 最后一个 sample 可能是最长的那个 sample。 | |||
| 将 ``dataset`` 中的数据根据 ``length`` 从长到短进行迭代。在多卡情况下,由于 ``padding`` , 最后一个 ``sample`` 可能是最长 | |||
| 的那个 ``sample`` 。 | |||
| :param dataset: 实现了 __len__ 方法的数据容器。 | |||
| :param length: 如果为 List,应当与 dataset 有一样的长度,表示 dataset 中每个元素的数量;仅当传入的 dataset 为 fastNLP 的 | |||
| DataSet 时支持传入 str,会将该str理解为 dataset 的 field 名称,若 field 中的元素为 int,则认为该值是 sample 的长度。 | |||
| :param seed: 设置的随机数种子 | |||
| :param kwargs: fastNLP 保留使用 | |||
| :param length: 每条数据的长度。 | |||
| * 为 ``List[int]`` 时 | |||
| 应当与 dataset 有一样的长度,表示 dataset 中每个元素的数量; | |||
| * 为 ``str`` 时 | |||
| 仅当传入的 ``dataset`` 是 :class:`fastNLP.DataSet` 时,允许传入 `str` ,该 `str` 将被认为是 ``dataset`` 中的 | |||
| ``field`` 。若 field 中的元素为 ``int``,则认为该值是 sample 的长度;若不为 ``int`` ,则尝试使用 ``len`` 方法 | |||
| 获取该 ``field`` 中每个元素的长度。 | |||
| :param seed: 设置的随机数种子。 | |||
| :param kwargs: fastNLP 保留使用。 | |||
| """ | |||
| super().__init__(dataset=dataset, **kwargs) | |||
| if isinstance(dataset, DataSet) and isinstance(length, str): | |||
| @@ -299,8 +306,9 @@ class SortedSampler(SequentialSampler): | |||
| if not isinstance(length[0], int): | |||
| length = list(map(len, length)) | |||
| else: | |||
| assert len(length) == len(dataset), "When the dataset is not fastNLP.DataSet, " \ | |||
| "the length parameter can only be List[int]" | |||
| types = set(map(type, length)) | |||
| assert isinstance(length, list) and len(types)==1 and types.pop()==int, \ | |||
| "When the dataset is not fastNLP.DataSet, the length parameter can only be List[int]" | |||
| assert len(length) == len(dataset), "The length of `data` and `length` should be equal." | |||
| @@ -2,7 +2,6 @@ __all__ = [ | |||
| 'cache_results', | |||
| 'is_jittor_dataset', | |||
| 'jittor_collate_wraps', | |||
| 'get_device_from_visible', | |||
| 'paddle_to', | |||
| 'paddle_move_data_to_device', | |||
| 'get_paddle_device_id', | |||
| @@ -28,7 +27,7 @@ __all__ = [ | |||
| from .cache_results import cache_results | |||
| from .jittor_utils import is_jittor_dataset, jittor_collate_wraps | |||
| from .paddle_utils import get_device_from_visible, paddle_to, paddle_move_data_to_device, get_paddle_device_id, get_paddle_gpu_str, is_in_paddle_dist, \ | |||
| from .paddle_utils import paddle_to, paddle_move_data_to_device, get_paddle_device_id, get_paddle_gpu_str, is_in_paddle_dist, \ | |||
| is_in_fnlp_paddle_dist, is_in_paddle_launch_dist | |||
| from .rich_progress import f_rich_progress | |||
| from .torch_utils import torch_move_data_to_device | |||
| @@ -15,6 +15,12 @@ from fastNLP.core.dataset import Instance | |||
| def is_jittor_dataset(dataset) -> bool: | |||
| """ | |||
| 判断传入的 ``dataset`` 是否是 :class:`jittor.dataset.Dataset` 类型 | |||
| :param dataset: 数据集; | |||
| :return: 当前 ``dataset`` 是否为 ``jittor`` 的数据集类型; | |||
| """ | |||
| try: | |||
| if isinstance(dataset, jt.dataset.Dataset): | |||
| return True | |||
| @@ -26,7 +32,8 @@ def is_jittor_dataset(dataset) -> bool: | |||
| def jittor_collate_wraps(func, auto_collator: Callable): | |||
| """ | |||
| 对jittor的collate_fn进行wrap封装, 如果数据集为mapping类型,那么采用auto_collator,否则还是采用jittor自带的collate_batch | |||
| 对 ``jittor`` 的 ``collate_fn`` 进行 ``wrap`` 封装,。如果数据集为 ``mapping`` 类型,那么采用 ``auto_collator`` ,否则 | |||
| 还是采用 ``jittor`` 的 ``collate_batch``。 | |||
| :param func: | |||
| :param auto_collator: | |||
| @@ -1,5 +1,4 @@ | |||
| __all__ = [ | |||
| "get_device_from_visible", | |||
| "paddle_to", | |||
| "paddle_move_data_to_device", | |||
| "get_paddle_gpu_str", | |||
| @@ -21,73 +20,90 @@ if _NEED_IMPORT_PADDLE: | |||
| from .utils import apply_to_collection | |||
| def get_device_from_visible(device: Union[str, int], output_type=int): | |||
| def _convert_data_device(device: Union[str, int]) -> str: | |||
| """ | |||
| 在有 CUDA_VISIBLE_DEVICES 的情况下,获取对应的设备。 | |||
| 如 CUDA_VISIBLE_DEVICES=2,3 ,device=3 ,则返回1。 | |||
| 用于转换 ``driver`` 的 ``data_device`` 的函数。如果用户设置了 ``FASTNLP_BACKEND=paddle``,那么 ``fastNLP`` 会将 | |||
| 可见的设备保存在 ``USER_CUDA_VISIBLE_DEVICES`` 中,并且将 ``CUDA_VISIBLE_DEVICES`` 设置为可见的第一张显卡;这是为 | |||
| 了顺利执行 ``paddle`` 的分布式训练而设置的。 | |||
| 在这种情况下,单纯使用 ``driver.data_device`` 是无效的。比如在分布式训练中将设备设置为 ``[0,2,3]`` ,且用户设置了 | |||
| ``CUDA_VISIBLE_DEVICES=3,4,5,6`` ,那么在 ``rank1``的进程中有:: | |||
| :param device: 未转化的设备名 | |||
| :param output_type: 返回值的类型 | |||
| :return: 转化后的设备id | |||
| """ | |||
| if output_type not in [int, str]: | |||
| raise ValueError("Parameter `output_type` should be one of these types: [int, str]") | |||
| if device == "cpu": | |||
| return device | |||
| cuda_visible_devices = os.getenv("CUDA_VISIBLE_DEVICES") | |||
| user_visible_devices = os.getenv(USER_CUDA_VISIBLE_DEVICES) | |||
| if user_visible_devices is None: | |||
| raise RuntimeError("`USER_CUDA_VISIBLE_DEVICES` cannot be None, please check if you have set " | |||
| "`FASTNLP_BACKEND` to 'paddle' before using FastNLP.") | |||
| idx = get_paddle_device_id(device) | |||
| # 利用 USER_CUDA_VISIBLDE_DEVICES 获取用户期望的设备 | |||
| if user_visible_devices is None: | |||
| raise RuntimeError("This situation cannot happen, please report a bug to us.") | |||
| idx = user_visible_devices.split(",")[idx] | |||
| cuda_visible_devices_list = cuda_visible_devices.split(',') | |||
| if idx not in cuda_visible_devices_list: | |||
| raise ValueError(f"Can't find your devices {idx} in CUDA_VISIBLE_DEVICES[{cuda_visible_devices}]. ") | |||
| res = cuda_visible_devices_list.index(idx) | |||
| if output_type == int: | |||
| return res | |||
| else: | |||
| return f"gpu:{res}" | |||
| os.environ["CUDA_VISIBLE_DEVICES"] = "5" | |||
| os.environ["USER_CUDA_VISIBLE_DEVICES"] = "3,4,5,6" | |||
| driver.data_device = "gpu:2" # 为了向用户正确地反映他们设置的设备减少歧义,因此这里没有设置为 "gpu:5" | |||
| 此时我们便需要通过这个函数将 ``data_device`` 转换为 ``gpu:0``。具体过程便是通过索引 **2** 在 ``USER_CUDA_VISIBLE_DEVICES`` 中 | |||
| 找到设备 **5**,然后在 ``CUDA_VISIBLE_DEVICES`` 中找到设备 **5** 的索引 **0** 返回。 | |||
| .. note:: | |||
| def paddle_to(data, device: Union[str, int]): | |||
| 在分布式单进程仅支持单卡的情况下中,这个函数实际等同于直接转换为 ``gpu:0`` 返回。 | |||
| :param device: 未转化的设备; | |||
| :return: 转化后的设备,格式为 ``gpu:x``; | |||
| """ | |||
| try: | |||
| user_visible_devices = os.getenv(USER_CUDA_VISIBLE_DEVICES) | |||
| if device == "cpu" or user_visible_devices is None: | |||
| # 传入的是 CPU,或者没有设置 USER_CUDA_VISIBLE_DEVICES | |||
| # 此时不需要进行转换 | |||
| return get_paddle_gpu_str(device) | |||
| idx = get_paddle_device_id(device) | |||
| idx = user_visible_devices.split(",")[idx] | |||
| # 此时 CUDA_VISIBLE_DEVICES 一定不是 None | |||
| cuda_visible_devices_list = os.getenv("CUDA_VISIBLE_DEVICES").split(',') | |||
| return f"gpu:{cuda_visible_devices_list.index(idx)}" | |||
| except Exception as e: | |||
| raise ValueError(f"Can't convert device {device} when USER_CUDA_VISIBLE_DEVICES={user_visible_devices} " | |||
| "and CUDA_VISIBLE_DEVICES={cuda_visible_devices}. If this situation happens, please report this bug to us.") | |||
| def paddle_to(data: "paddle.Tensor", device: Union[str, int]) -> "paddle.Tensor": | |||
| """ | |||
| 将 `data` 迁移到指定的 `device` 上 | |||
| 将 ``data`` 迁移到指定的 ``device`` 上。``paddle.Tensor`` 没有类似 ``torch.Tensor`` 的 ``to`` 函数,该函数 | |||
| 只是集成了 :func:`paddle.Tensor.cpu` 和 :func:`paddle.Tensor.cuda` 两个函数。 | |||
| :param data: 要迁移的张量 | |||
| :param device: 目标设备,可以是 `str` 或 `int` | |||
| :return: 迁移后的张量 | |||
| :param data: 要迁移的张量; | |||
| :param device: 目标设备,可以是 ``str`` 或 ``int`` 类型; | |||
| :return: 迁移后的张量; | |||
| """ | |||
| if device == "cpu": | |||
| return data.cpu() | |||
| else: | |||
| # device = get_device_from_visible(device, output_type=int) | |||
| return data.cuda(get_paddle_device_id(device)) | |||
| def get_paddle_gpu_str(device: Union[str, int]): | |||
| def get_paddle_gpu_str(device: Union[str, int]) -> str: | |||
| """ | |||
| 获得 `gpu:x` 类型的设备名 | |||
| 获得 ``gpu:x`` 格式的设备名:: | |||
| :param device: 设备编号或设备名 | |||
| :return: 返回对应的 `gpu:x` 格式的设备名 | |||
| >>> get_paddle_gpu_str(1) | |||
| 'gpu:1' | |||
| >>> get_paddle_gpu_str("cuda:1") | |||
| 'gpu:1' | |||
| :param device: 设备编号或设备名; | |||
| :return: 返回对应的 ``gpu:x`` 格式的设备名; | |||
| """ | |||
| if isinstance(device, str): | |||
| return device.replace("cuda", "gpu") | |||
| return f"gpu:{device}" | |||
| def get_paddle_device_id(device: Union[str, int]): | |||
| def get_paddle_device_id(device: Union[str, int]) -> int: | |||
| """ | |||
| 获得 gpu 的设备id | |||
| 获得 ``device`` 的设备编号:: | |||
| >>> get_paddle_device_id("gpu:1") | |||
| 1 | |||
| >>> get_paddle_device_id("gpu") | |||
| 0 | |||
| 请注意不要向这个函数中传入 ``cpu``。 | |||
| :param: device: 设备编号或设备名 | |||
| :return: 设备对应的编号 | |||
| :param: device: 设备编号或设备名; | |||
| :return: 设备对应的编号; | |||
| """ | |||
| if isinstance(device, int): | |||
| return device | |||
| @@ -109,21 +125,17 @@ def get_paddle_device_id(device: Union[str, int]): | |||
| return device_id | |||
| def paddle_move_data_to_device(batch: Any, device: Optional[str] = None, | |||
| data_device: Optional[str] = None) -> Any: | |||
| def paddle_move_data_to_device(batch: Any, device: Optional[Union[str, int]]) -> Any: | |||
| r""" | |||
| 将数据集合传输到给定设备。只有paddle.Tensor对象会被传输到设备中,其余保持不变 | |||
| 将 ``paddle`` 的数据集合传输到给定设备。只有 :class:`paddle.Tensor` 对象会被传输到设备中,其余保持不变。 | |||
| :param batch: | |||
| :param device: `cpu`, `gpu` or `gpu:x` | |||
| :param data_device: | |||
| :return: 相同的集合,但所有包含的张量都驻留在新设备上; | |||
| :param batch: 需要进行迁移的数据集合; | |||
| :param device: 目标设备。可以是显卡设备的编号,或是``cpu``, ``gpu`` 或 ``gpu:x`` 格式的字符串;当这个参数 | |||
| 为 `None`` 时,不会执行任何操作。 | |||
| :return: 迁移到新设备上的数据集合; | |||
| """ | |||
| if device is None: | |||
| if data_device is not None: | |||
| device = data_device | |||
| else: | |||
| return batch | |||
| return batch | |||
| def batch_to(data: Any) -> Any: | |||
| return paddle_to(data, device) | |||
| @@ -131,22 +143,22 @@ def paddle_move_data_to_device(batch: Any, device: Optional[str] = None, | |||
| return apply_to_collection(batch, dtype=paddle.Tensor, function=batch_to) | |||
| def is_in_paddle_dist(): | |||
| def is_in_paddle_dist() -> bool: | |||
| """ | |||
| 判断是否处于分布式的进程下,使用 global_rank 和 selected_gpus 判断 | |||
| 判断是否处于 ``paddle`` 分布式的进程下,使用 ``PADDLE_RANK_IN_NODE`` 和 ``FLAGS_selected_gpus`` 判断。 | |||
| """ | |||
| return ('PADDLE_RANK_IN_NODE' in os.environ and 'FLAGS_selected_gpus' in os.environ) | |||
| def is_in_fnlp_paddle_dist(): | |||
| def is_in_fnlp_paddle_dist() -> bool: | |||
| """ | |||
| 判断是否处于 FastNLP 拉起的分布式进程中 | |||
| 判断是否处于 ``fastNLP`` 拉起的 ``paddle`` 分布式进程中 | |||
| """ | |||
| return FASTNLP_DISTRIBUTED_CHECK in os.environ | |||
| def is_in_paddle_launch_dist(): | |||
| def is_in_paddle_launch_dist() -> bool: | |||
| """ | |||
| 判断是否处于 launch 启动的分布式进程中 | |||
| 判断是否处于 ``python -m paddle.distributed.launch`` 方法启动的 ``paddle`` 分布式进程中 | |||
| """ | |||
| return FASTNLP_BACKEND_LAUNCH in os.environ | |||
| @@ -44,12 +44,12 @@ class TorchTransferableDataType(ABC): | |||
| def torch_move_data_to_device(batch: Any, device: Optional[Union[str, "torch.device"]] = None, | |||
| non_blocking: Optional[bool] = True) -> Any: | |||
| r""" | |||
| 将数据集合传输到给定设备。任何定义方法 “to(device)” 的对象都将被移动并且集合中的所有其他对象将保持不变; | |||
| 在 ``pytorch`` 中将数据集合 ``batch`` 传输到给定设备。任何定义方法 ``to(device)`` 的对象都将被移动并且集合中的所有其他对象将保持不变; | |||
| :param batch: 应当迁移的数据; | |||
| :param device: 数据应当迁移到的设备;当该参数的值为 None 时,表示迁移数据的操作由用户自己完成,我们不需要经管; | |||
| :param non_blocking: pytorch 的迁移数据方法 `to` 的参数; | |||
| :return: 相同的集合,但所有包含的张量都驻留在新设备上; | |||
| :param batch: 需要迁移的数据; | |||
| :param device: 数据应当迁移到的设备;当该参数的值为 ``None`` 时则不执行任何操作; | |||
| :param non_blocking: ``pytorch`` 的数据迁移方法 ``to`` 的参数; | |||
| :return: 迁移到新设备上的数据集合; | |||
| """ | |||
| if device is None: | |||
| return batch | |||
| @@ -38,10 +38,16 @@ __all__ = [ | |||
| def get_fn_arg_names(fn: Callable) -> List[str]: | |||
| r""" | |||
| 返回一个函数所有参数的名字 | |||
| 该函数可以返回一个函数所有参数的名字:: | |||
| :param fn: 需要查询的函数 | |||
| :return: 一个列表,其中的元素是函数 ``fn`` 参数的字符串名字 | |||
| >>> def function(a, b=1): | |||
| ... return a | |||
| ... | |||
| >>> get_fn_arg_names(function) | |||
| ['a', 'b'] | |||
| :param fn: 需要查询的函数; | |||
| :return: 包含函数 ``fn`` 参数名的列表; | |||
| """ | |||
| return list(inspect.signature(fn).parameters) | |||
| @@ -49,7 +55,7 @@ def get_fn_arg_names(fn: Callable) -> List[str]: | |||
| def auto_param_call(fn: Callable, *args, signature_fn: Optional[Callable] = None, | |||
| mapping: Optional[Dict[AnyStr, AnyStr]] = None) -> Any: | |||
| r""" | |||
| 该函数会根据输入函数的形参名从 ``*args`` (因此都需要是 ``dict`` 类型)中找到匹配的值进行调用,如果传入的数据与 ``fn`` 的形参不匹配,可以通过 | |||
| 该函数会根据输入函数的形参名从 ``*args`` (均为 ``dict`` 类型)中找到匹配的值进行调用,如果传入的数据与 ``fn`` 的形参不匹配,可以通过 | |||
| ``mapping`` 参数进行转换。``mapping`` 参数中的一对 ``(key, value)`` 表示在 ``*args`` 中找到 ``key`` 对应的值,并将这个值传递给形参中名为 | |||
| ``value`` 的参数。 | |||
| @@ -161,13 +167,13 @@ def _get_keys(args:List[Dict]) -> List[List[str]]: | |||
| def _get_fun_msg(fn, with_fp=True)->str: | |||
| """ | |||
| 获取函数的基本信息,帮助报错。 | |||
| ex: | |||
| print(_get_fun_msg(_get_fun_msg)) | |||
| # `_get_fun_msg(fn) -> str`(In file:/Users/hnyan/Desktop/projects/fastNLP/fastNLP/fastNLP/core/utils/utils.py) | |||
| 获取函数的基本信息,帮助报错:: | |||
| >>>> print(_get_fun_msg(_get_fun_msg)) | |||
| `_get_fun_msg(fn) -> str`(In file:/Users/hnyan/Desktop/projects/fastNLP/fastNLP/fastNLP/core/utils/utils.py) | |||
| :param callable fn: | |||
| :param with_fp: 是否包含函数所在的文件信息。 | |||
| :param with_fp: 是否包含函数所在的文件信息; | |||
| :return: | |||
| """ | |||
| if isinstance(fn, functools.partial): | |||
| @@ -224,7 +230,7 @@ def _check_valid_parameters_number(fn, expected_params:List[str], fn_name=None): | |||
| def check_user_specific_params(user_params: Dict, fn: Callable): | |||
| """ | |||
| 该函数使用用户的输入来对指定函数的参数进行赋值,主要用于一些用户无法直接调用函数的情况; | |||
| 该函数主要的作用在于帮助检查用户对使用函数 ``fn`` 的参数输入是否有误; | |||
| 主要作用在于帮助检查用户对使用函数 ``fn`` 的参数输入是否有误; | |||
| :param user_params: 用户指定的参数的值,应当是一个字典,其中 ``key`` 表示每一个参数的名字, | |||
| ``value`` 为每一个参数的值; | |||
| @@ -241,7 +247,7 @@ def check_user_specific_params(user_params: Dict, fn: Callable): | |||
| def dataclass_to_dict(data: "dataclasses.dataclass") -> Dict: | |||
| """ | |||
| 将传入的 `dataclass` 实例转换为字典。 | |||
| 将传入的 ``dataclass`` 实例转换为字典。 | |||
| """ | |||
| if not is_dataclass(data): | |||
| raise TypeError(f"Parameter `data` can only be `dataclass` type instead of {type(data)}.") | |||
| @@ -253,12 +259,12 @@ def dataclass_to_dict(data: "dataclasses.dataclass") -> Dict: | |||
| def match_and_substitute_params(mapping: Optional[Union[Callable, Dict]] = None, data: Optional[Any] = None) -> Any: | |||
| r""" | |||
| 用来实现将输入的 ``batch``,或者输出的 ``outputs``,通过 ``mapping`` 将键值进行更换的功能; | |||
| 用来实现将输入的 ``batch`` 或者输出的 ``outputs`` 通过 ``mapping`` 将键值进行更换的功能; | |||
| 该函数应用于 ``input_mapping`` 和 ``output_mapping``; | |||
| 对于 ``input_mapping``,该函数会在 :class:`~fastNLP.core.controllers.TrainBatchLoop` 中取完数据后立刻被调用; | |||
| 对于 ``output_mapping``,该函数会在 :class:`~fastNLP.core.Trainer` 的 :meth:`~fastNLP.core.Trainer.train_step` | |||
| 以及 :class:`~fastNLP.core.Evaluator` 的 :meth:`~fastNLP.core.Evaluator.train_step` 中得到结果后立刻被调用; | |||
| * 对于 ``input_mapping``,该函数会在 :class:`~fastNLP.core.controllers.TrainBatchLoop` 中取完数据后立刻被调用; | |||
| * 对于 ``output_mapping``,该函数会在 :class:`~fastNLP.core.Trainer` 的 :meth:`~fastNLP.core.Trainer.train_step` | |||
| 以及 :class:`~fastNLP.core.Evaluator` 的 :meth:`~fastNLP.core.Evaluator.train_step` 中得到结果后立刻被调用; | |||
| 转换的逻辑按优先级依次为: | |||
| @@ -277,9 +283,9 @@ def match_and_substitute_params(mapping: Optional[Union[Callable, Dict]] = None, | |||
| 然后使用 ``mapping`` 对这个 ``Dict`` 进行转换,如果没有匹配上 ``mapping`` 中的 ``key`` 则保持 ``\'\_number\'`` 这个形式。 | |||
| :param mapping: 用于转换的字典或者函数;``mapping`` 是函数时,返回值必须为字典类型。 | |||
| :param mapping: 用于转换的字典或者函数;当 ``mapping`` 是函数时,返回值必须为字典类型; | |||
| :param data: 需要被转换的对象; | |||
| :return: 返回转换好的结果; | |||
| :return: 返回转换后的结果; | |||
| """ | |||
| if mapping is None: | |||
| return data | |||
| @@ -331,19 +337,19 @@ def apply_to_collection( | |||
| **kwargs: Any, | |||
| ) -> Any: | |||
| """ | |||
| 使用函数 ``function`` 递归地在 ``data`` 中的元素执行,但是仅在满足元素为 ``dtype`` 时执行。 | |||
| 递归地对 ``data`` 中的元素执行函数 ``function``,且仅在满足元素为 ``dtype`` 时执行。 | |||
| 该函数参考了 `pytorch-lightning <https://github.com/PyTorchLightning/pytorch-lightning>`_ 的实现 | |||
| :param data: 需要进行处理的数据集合或数据 | |||
| :param dtype: 数据的类型,函数 ``function`` 只会被应用于 ``data`` 中类型为 ``dtype`` 的数据 | |||
| :param function: 对数据进行处理的函数 | |||
| :param args: ``function`` 所需要的其它参数 | |||
| :param data: 需要进行处理的数据集合或数据; | |||
| :param dtype: 数据的类型,函数 ``function`` 只会被应用于 ``data`` 中类型为 ``dtype`` 的数据; | |||
| :param function: 对数据进行处理的函数; | |||
| :param args: ``function`` 所需要的其它参数; | |||
| :param wrong_dtype: ``function`` 一定不会生效的数据类型。如果数据既是 ``wrong_dtype`` 类型又是 ``dtype`` 类型 | |||
| 那么也不会生效。 | |||
| :param include_none: 是否包含执行结果为 ``None`` 的数据,默认为 ``True``。 | |||
| :param kwargs: ``function`` 所需要的其它参数 | |||
| :return: 经过 ``function`` 处理后的数据集合 | |||
| 那么也不会生效; | |||
| :param include_none: 是否包含执行结果为 ``None`` 的数据,默认为 ``True``; | |||
| :param kwargs: ``function`` 所需要的其它参数; | |||
| :return: 经过 ``function`` 处理后的数据集合; | |||
| """ | |||
| # Breaking condition | |||
| if isinstance(data, dtype) and (wrong_dtype is None or not isinstance(data, wrong_dtype)): | |||
| @@ -411,20 +417,20 @@ def apply_to_collection( | |||
| @contextmanager | |||
| def nullcontext(): | |||
| r""" | |||
| 实现一个什么都不做的上下文环境 | |||
| 实现一个什么都不做的上下文环境。 | |||
| """ | |||
| yield | |||
| def sub_column(string: str, c: int, c_size: int, title: str) -> str: | |||
| r""" | |||
| 对传入的字符串进行截断,方便在命令行中显示 | |||
| 对传入的字符串进行截断,方便在命令行中显示。 | |||
| :param string: 要被截断的字符串 | |||
| :param c: 命令行列数 | |||
| :param c_size: :class:`~fastNLP.core.Instance` 或 :class:`fastNLP.core.DataSet` 的 ``field`` 数目 | |||
| :param title: 列名 | |||
| :return: 对一个过长的列进行截断的结果 | |||
| :param string: 要被截断的字符串; | |||
| :param c: 命令行列数; | |||
| :param c_size: :class:`~fastNLP.core.Instance` 或 :class:`fastNLP.core.DataSet` 的 ``field`` 数目; | |||
| :param title: 列名; | |||
| :return: 对一个过长的列进行截断的结果; | |||
| """ | |||
| avg = max(int(c / c_size / 2), len(title)) | |||
| string = str(string) | |||
| @@ -453,7 +459,7 @@ def _is_iterable(value): | |||
| def pretty_table_printer(dataset_or_ins) -> PrettyTable: | |||
| r""" | |||
| 在 ``fastNLP`` 中展示数据的函数:: | |||
| 用于在 ``fastNLP`` 中展示数据的函数:: | |||
| >>> ins = Instance(field_1=[1, 1, 1], field_2=[2, 2, 2], field_3=["a", "b", "c"]) | |||
| +-----------+-----------+-----------------+ | |||
| @@ -462,8 +468,8 @@ def pretty_table_printer(dataset_or_ins) -> PrettyTable: | |||
| | [1, 1, 1] | [2, 2, 2] | ['a', 'b', 'c'] | | |||
| +-----------+-----------+-----------------+ | |||
| :param dataset_or_ins: 要展示的 :class:`~fastNLP.core.DataSet` 或者 :class:`~fastNLP.core.Instance` | |||
| :return: 根据 ``terminal`` 大小进行自动截断的数据表格 | |||
| :param dataset_or_ins: 要展示的 :class:`~fastNLP.core.DataSet` 或者 :class:`~fastNLP.core.Instance` 实例; | |||
| :return: 根据命令行大小进行自动截断的数据表格; | |||
| """ | |||
| x = PrettyTable() | |||
| try: | |||
| @@ -529,7 +535,7 @@ def deprecated(help_message: Optional[str] = None): | |||
| """ | |||
| 标记当前功能已经过时的装饰器。 | |||
| :param help_message: 一段指引信息,告知用户如何将代码切换为当前版本提倡的用法。 | |||
| :param help_message: 一段指引信息,告知用户如何将代码切换为当前版本提倡的用法; | |||
| """ | |||
| def decorator(deprecated_function: Callable): | |||
| @@ -578,10 +584,10 @@ def seq_len_to_mask(seq_len, max_len: Optional[int]): | |||
| >>>print(mask.size()) | |||
| torch.Size([14, 100]) | |||
| :param seq_len: 大小为是 ``(B,)`` 的长度序列 | |||
| :param int max_len: 将长度 ``pad`` 到 ``max_len``。默认情况(为 ``None``)使用的是 ``seq_len`` 中最长的长度。 | |||
| :param seq_len: 大小为 ``(B,)`` 的长度序列; | |||
| :param int max_len: 将长度补齐或截断到 ``max_len``。默认情况(为 ``None``)使用的是 ``seq_len`` 中最长的长度; | |||
| 但在 :class:`torch.nn.DataParallel` 等分布式的场景下可能不同卡的 ``seq_len`` 会有区别,所以需要传入 | |||
| 一个 ``max_len`` 使得 ``mask`` 的长度 ``pad`` 到该长度。 | |||
| 一个 ``max_len`` 使得 ``mask`` 的补齐或截断到该长度。 | |||
| :return: 大小为 ``(B, max_len)`` 的 ``mask``, 元素类型为 ``bool`` 或 ``uint8`` | |||
| """ | |||
| if isinstance(seq_len, np.ndarray): | |||
| @@ -51,23 +51,33 @@ def _set_backend(): | |||
| assert _module_available(backend), f"You must have {backend} available to use {backend} backend." | |||
| assert 'paddle' not in sys.modules, "You have to use `set_backend()` before `import paddle`." | |||
| user_visible_devices = os.getenv(USER_CUDA_VISIBLE_DEVICES) | |||
| cuda_visible_devices = os.getenv("CUDA_VISIBLE_DEVICES") | |||
| if 'PADDLE_RANK_IN_NODE' in os.environ and 'FLAGS_selected_gpus' in os.environ: | |||
| # 在分布式子进程下,根据 USER_VISIBLE_DEVICES 得到进程真正占有的设备 | |||
| selected_gpus = os.environ['FLAGS_selected_gpus'].split(',') | |||
| if user_visible_devices is not None: | |||
| # 用户通过 CUDA_VISIBLE_DEVICES 启动了分布式训练 | |||
| # 用户使用 fastNLP 启动了分布式训练 | |||
| # 此时经过 set_backend,用户的设置会保存在 USER_CUDA_VISIBLE_DEVICES 中 | |||
| # 我们需要从中找到真正使用的设备编号 | |||
| # 我们需要从中转换为用户找到真正使用的设备编号 | |||
| user_visible_devices = user_visible_devices.split(",") | |||
| selected_gpus = ",".join([user_visible_devices[int(i)] for i in selected_gpus]) | |||
| selected_gpus = [user_visible_devices[int(i)] for i in selected_gpus] | |||
| # 没有找到 USER_CUDA_VISIBLE_DEVICES,说明用户是直接用 launch 启动的 | |||
| elif cuda_visible_devices: | |||
| # 用户设置了可见设备,需要进行转换 | |||
| # 如 CUDA_VISIBLE_DEVICES = 0,2,3 --gpus=0,2,3 | |||
| # 在 rank1 中此时 selected_gpus = ['1'],需要转换为设备 2 | |||
| os.environ[USER_CUDA_VISIBLE_DEVICES] = cuda_visible_devices | |||
| cuda_visible_devices = cuda_visible_devices.split(",") | |||
| selected_gpus = [cuda_visible_devices[int(i)] for i in selected_gpus] | |||
| else: | |||
| # 没有找到 USER_CUDA_VISIBLE_DEVICES,则将之设置为所有的设备 | |||
| # 用户没有设置可见设备,则赋值成所有的设备 | |||
| os.environ[USER_CUDA_VISIBLE_DEVICES] = ",".join(map(str, list( | |||
| range(get_gpu_count()) | |||
| ))) | |||
| os.environ['CUDA_VISIBLE_DEVICES'] = ",".join(selected_gpus) | |||
| os.environ['FLAGS_selected_gpus'] = ",".join([str(g) for g in range(len(selected_gpus))]) | |||
| os.environ['FLAGS_selected_accelerators'] = ",".join([str(g) for g in range(len(selected_gpus))]) | |||
| elif 'CUDA_VISIBLE_DEVICES' in os.environ: | |||
| # 主进程中,用户设置了 CUDA_VISIBLE_DEVICES | |||
| # 将用户设置的 CUDA_VISIBLE_DEVICES hack 掉 | |||
| @@ -91,6 +101,11 @@ def _set_backend(): | |||
| elif backend == 'torch': | |||
| assert _module_available(backend), f"You must have {backend} available to use {backend} backend." | |||
| if 'PADDLE_RANK_IN_NODE' in os.environ and 'FLAGS_selected_gpus' in os.environ \ | |||
| and "USER_CUDA_VISIBLE_DEVICES" not in os.environ: | |||
| # 当用户没有设置 backend 并且使用 launch 启动了多卡,应该提醒用户进行设置 | |||
| raise RuntimeError("To run paddle distributed training, please set `FASTNLP_BACKEND` to 'paddle' before using FastNLP.") | |||
| def set_env(global_seed=None): | |||
| """ | |||
| @@ -0,0 +1,9 @@ | |||
| __all__ = [ | |||
| # "MixModule", | |||
| "torch2paddle", | |||
| "paddle2torch", | |||
| "torch2jittor", | |||
| "jittor2torch", | |||
| ] | |||
| from .mix_modules import torch2paddle, paddle2torch, torch2jittor, jittor2torch | |||
| @@ -0,0 +1,10 @@ | |||
| __all__ = [ | |||
| # "MixModule", | |||
| "torch2paddle", | |||
| "paddle2torch", | |||
| "torch2jittor", | |||
| "jittor2torch", | |||
| ] | |||
| # from .mix_module import MixModule | |||
| from .utils import * | |||
| @@ -14,6 +14,7 @@ from tests.helpers.utils import magic_argv_env_context | |||
| from fastNLP.envs.distributed import rank_zero_rm | |||
| from tests.helpers.models.torch_model import TorchNormalModel_Classification_1 | |||
| from tests.helpers.datasets.torch_data import TorchArgMaxDataset | |||
| from tests.helpers.utils import Capturing | |||
| from torchmetrics import Accuracy | |||
| from fastNLP.core.log import logger | |||
| @@ -428,6 +429,78 @@ def test_trainer_checkpoint_callback_1( | |||
| dist.destroy_process_group() | |||
| @pytest.mark.torch | |||
| def test_load_state(model_and_optimizers): | |||
| try: | |||
| path = Path.cwd().joinpath(f"test_model_checkpoint") | |||
| path.mkdir(exist_ok=True, parents=True) | |||
| from fastNLP import Event, Callback | |||
| @Trainer.on(Event.on_before_backward(every=3), marker='all') | |||
| def print_outputs(*args): | |||
| print("????") | |||
| class StateCallback(Callback): | |||
| def __init__(self, name): | |||
| self.name = name | |||
| def on_save_checkpoint(self, trainer): | |||
| return {'name': self.name} | |||
| def on_load_checkpoint(self, trainer, states): | |||
| self.name = states['name'] | |||
| def on_train_end(self, trainer): | |||
| print(self.name) | |||
| callbacks = [StateCallback('old_callback1'), StateCallback('old_callback2'), | |||
| CheckpointCallback(folder=path, every_n_epochs=1, save_object='trainer')] | |||
| trainer = Trainer( | |||
| model=model_and_optimizers.model, | |||
| driver='torch', | |||
| device='cpu', | |||
| optimizers=model_and_optimizers.optimizers, | |||
| train_dataloader=model_and_optimizers.train_dataloader, | |||
| evaluate_dataloaders=model_and_optimizers.evaluate_dataloaders, | |||
| input_mapping=model_and_optimizers.input_mapping, | |||
| output_mapping=model_and_optimizers.output_mapping, | |||
| metrics=model_and_optimizers.metrics, | |||
| n_epochs=3, | |||
| callbacks=callbacks, | |||
| output_from_new_proc="all" | |||
| ) | |||
| trainer.run(num_eval_sanity_batch=0, num_train_batch_per_epoch=2) | |||
| all_saved_model_paths = {w.name: w for w in path.joinpath(os.environ[FASTNLP_LAUNCH_TIME]).iterdir()} | |||
| epoch_2_path = all_saved_model_paths['trainer-epoch_2'] | |||
| callbacks = [StateCallback('new_callback1'), StateCallback('new_callback2')] | |||
| trainer = Trainer( | |||
| model=model_and_optimizers.model, | |||
| driver='torch', | |||
| device='cpu', | |||
| optimizers=model_and_optimizers.optimizers, | |||
| train_dataloader=model_and_optimizers.train_dataloader, | |||
| evaluate_dataloaders=model_and_optimizers.evaluate_dataloaders, | |||
| input_mapping=model_and_optimizers.input_mapping, | |||
| output_mapping=model_and_optimizers.output_mapping, | |||
| metrics=model_and_optimizers.metrics, | |||
| n_epochs=3, | |||
| callbacks=callbacks, | |||
| output_from_new_proc="all" | |||
| ) | |||
| trainer.load(folder=epoch_2_path) | |||
| with Capturing() as output: | |||
| trainer.run(num_eval_sanity_batch=0, num_train_batch_per_epoch=2) | |||
| assert 'old_callback1' in output[0] | |||
| assert 'new_callback2' in output[0] | |||
| assert output[0].count('???')==1 | |||
| finally: | |||
| rank_zero_rm(path) | |||
| @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) | |||
| @@ -1,7 +1,15 @@ | |||
| """ | |||
| 这个文件测试用户以python -m paddle.distributed.launch 启动的情况 | |||
| 看看有没有用pytest执行的机会 | |||
| FASTNLP_BACKEND=paddle python -m paddle.distributed.launch --gpus=0,2,3 _test_trainer_fleet.py | |||
| 这个文件测试多卡情况下使用 paddle 的情况:: | |||
| >>> # 测试用 python -m paddle.distributed.launch 启动 | |||
| >>> FASTNLP_BACKEND=paddle python -m paddle.distributed.launch --gpus=0,2,3 _test_trainer_fleet.py | |||
| >>> # 测试在限制 GPU 的情况下用 python -m paddle.distributed.launch 启动 | |||
| >>> CUDA_VISIBLE_DEVICES=0,2,3 FASTNLP_BACKEND=paddle python -m paddle.distributed.launch --gpus=0,2,3 _test_trainer_fleet.py | |||
| >>> # 测试直接使用多卡 | |||
| >>> FASTNLP_BACKEND=paddle python _test_trainer_fleet.py | |||
| >>> # 测试在限制 GPU 的情况下直接使用多卡 | |||
| >>> CUDA_VISIBLE_DEVICES=3,4,5,6 FASTNLP_BACKEND=paddle python _test_trainer_fleet.py | |||
| """ | |||
| import os | |||
| import sys | |||
| @@ -71,14 +79,13 @@ def test_trainer_fleet( | |||
| n_epochs=n_epochs, | |||
| callbacks=callbacks, | |||
| output_from_new_proc="logs", | |||
| # output_from_new_proc="logs", | |||
| ) | |||
| trainer.run() | |||
| if __name__ == "__main__": | |||
| driver = "paddle" | |||
| device = [0,2,3] | |||
| # driver = "paddle" | |||
| device = [0,1,3] | |||
| # device = 2 | |||
| callbacks = [ | |||
| # RecordMetricCallback(monitor="acc#acc", metric_threshold=0.0, larger_better=True), | |||
| @@ -1,7 +1,11 @@ | |||
| """ | |||
| 这个文件测试用户以python -m paddle.distributed.launch 启动的情况 | |||
| 并且自己初始化了 fleet | |||
| FASTNLP_BACKEND=paddle python -m paddle.distributed.launch --gpus=0,2,3 _test_trainer_fleet_outside.py | |||
| 这个文件测试用户自己初始化分布式环境后使用 paddle 的情况: | |||
| >>> # 测试用 python -m paddle.distributed.launch 启动 | |||
| >>> FASTNLP_BACKEND=paddle python -m paddle.distributed.launch --gpus=0,2,3 _test_trainer_fleet_outside.py | |||
| >>> # 测试在限制 GPU 的情况下用 python -m paddle.distributed.launch 启动 | |||
| >>> CUDA_VISIBLE_DEVICES=0,2,3 FASTNLP_BACKEND=paddle python -m paddle.distributed.launch --gpus=0,2,3 _test_trainer_fleet_outside.py | |||
| """ | |||
| import os | |||
| import sys | |||
| @@ -77,14 +81,13 @@ def test_trainer_fleet( | |||
| n_epochs=n_epochs, | |||
| callbacks=callbacks, | |||
| output_from_new_proc="logs", | |||
| data_device=f"gpu:{os.environ['CUDA_VISIBLE_DEVICES']}" | |||
| # output_from_new_proc="logs", | |||
| ) | |||
| trainer.run() | |||
| if __name__ == "__main__": | |||
| driver = "paddle" | |||
| device = [0,2,3] | |||
| device = [0,1,3] | |||
| callbacks = [ | |||
| # RecordMetricCallback(monitor="acc#acc", metric_threshold=0.0, larger_better=True), | |||
| RichCallback(5), | |||
| @@ -0,0 +1,237 @@ | |||
| import os | |||
| import sys | |||
| import time | |||
| # os.environ["cuda_archs"] = "61" | |||
| # os.environ["FAS"] | |||
| os.environ["log_silent"] = "1" | |||
| sys.path.append("../../../") | |||
| from datasets import load_dataset | |||
| from datasets import DatasetDict | |||
| import jittor as jt | |||
| from jittor import nn, Module | |||
| from jittor.dataset import Dataset | |||
| jt.flags.use_cuda = True | |||
| from fastNLP.core.controllers.trainer import Trainer | |||
| from fastNLP.core.metrics.accuracy import Accuracy | |||
| from fastNLP.core.vocabulary import Vocabulary | |||
| from fastNLP.core.callbacks.progress_callback import RichCallback | |||
| from fastNLP.core.callbacks.callback import Callback | |||
| from fastNLP.core.dataloaders.jittor_dataloader.fdl import JittorDataLoader | |||
| class TextClassificationDataset(Dataset): | |||
| def __init__(self, dataset): | |||
| super(TextClassificationDataset, self).__init__() | |||
| self.dataset = dataset | |||
| self.set_attrs(total_len=len(dataset)) | |||
| def __getitem__(self, idx): | |||
| return {"x": self.dataset["input_ids"][idx], "y": self.dataset["label"][idx]} | |||
| class LSTM(Module): | |||
| def __init__(self, num_of_words, hidden_size, features): | |||
| self.embedding = nn.Embedding(num_of_words, features) | |||
| self.lstm = nn.LSTM(features, hidden_size, batch_first=True) | |||
| self.layer = nn.Linear(hidden_size, 2) | |||
| self.softmax = nn.Softmax(dim=1) | |||
| self.loss_fn = nn.CrossEntropyLoss() | |||
| self.hidden_size = hidden_size | |||
| self.features = features | |||
| def init_hidden(self, x): | |||
| # batch_first | |||
| batch_size = x.shape[0] | |||
| h0 = jt.randn(1, batch_size, hidden_size) | |||
| c0 = jt.randn(1, batch_size, hidden_size) | |||
| return h0, c0 | |||
| def execute(self, input_ids): | |||
| output = self.embedding(input_ids) | |||
| # TODO 去除padding | |||
| output, (h, c) = self.lstm(output, self.init_hidden(output)) | |||
| # len, batch, hidden_size | |||
| output = self.layer(output[-1]) | |||
| return output | |||
| def train_step(self, x, y): | |||
| x = self(x) | |||
| outputs = self.loss_fn(x, y) | |||
| return {"loss": outputs} | |||
| def evaluate_step(self, x, y): | |||
| x = self(x) | |||
| return {"pred": x, "target": y.reshape((-1,))} | |||
| class PrintWhileTrainingCallBack(Callback): | |||
| """ | |||
| 通过该Callback实现训练过程中loss的输出 | |||
| """ | |||
| def __init__(self, print_every_epoch, print_every_batch): | |||
| self.print_every_epoch = print_every_epoch | |||
| self.print_every_batch = print_every_batch | |||
| self.loss = 0 | |||
| self.start = 0 | |||
| self.epoch_start = 0 | |||
| def on_train_begin(self, trainer): | |||
| """ | |||
| 在训练开始前输出信息 | |||
| """ | |||
| print("Start training. Total {} epochs and {} batches in each epoch.".format( | |||
| trainer.n_epochs, trainer.num_batches_per_epoch | |||
| )) | |||
| self.start = time.time() | |||
| def on_before_backward(self, trainer, outputs): | |||
| """ | |||
| 每次反向传播前统计loss,用于计算平均值 | |||
| """ | |||
| loss = trainer.extract_loss_from_outputs(outputs) | |||
| loss = trainer.driver.tensor_to_numeric(loss) | |||
| self.loss += loss | |||
| def on_train_epoch_begin(self, trainer): | |||
| self.epoch_start = time.time() | |||
| def on_train_epoch_end(self, trainer): | |||
| """ | |||
| 在每经过一定epoch或最后一个epoch时输出当前epoch的平均loss和使用时间 | |||
| """ | |||
| if trainer.cur_epoch_idx % self.print_every_epoch == 0 \ | |||
| or trainer.cur_epoch_idx == trainer.n_epochs: | |||
| print("Epoch: {} Loss: {} Current epoch training time: {}s".format( | |||
| trainer.cur_epoch_idx, self.loss / trainer.num_batches_per_epoch, time.time() - self.epoch_start | |||
| )) | |||
| # 将loss清零 | |||
| self.loss = 0 | |||
| def on_train_batch_end(self, trainer): | |||
| """ | |||
| 在每经过一定batch或最后一个batch时输出当前epoch截止目前的平均loss | |||
| """ | |||
| if trainer.batch_idx_in_epoch % self.print_every_batch == 0 \ | |||
| or trainer.batch_idx_in_epoch == trainer.num_batches_per_epoch: | |||
| print("\tBatch: {} Loss: {}".format( | |||
| trainer.batch_idx_in_epoch, self.loss / trainer.batch_idx_in_epoch | |||
| )) | |||
| def on_train_end(self, trainer): | |||
| print("Total training time: {}s".format(time.time() - self.start)) | |||
| def process_data(ds: DatasetDict, vocabulary: Vocabulary, max_len=256) -> DatasetDict: | |||
| # 分词 | |||
| ds = ds.map(lambda x: {"input_ids": text_to_id(vocabulary, x["text"], max_len)}) | |||
| ds.set_format(type="numpy", columns=ds.column_names) | |||
| return ds | |||
| def set_vocabulary(vocab, dataset): | |||
| for data in dataset: | |||
| vocab.update(data["text"].split()) | |||
| return vocab | |||
| def text_to_id(vocab, text: str, max_len): | |||
| text = text.split() | |||
| # to index | |||
| ids = [vocab.to_index(word) for word in text] | |||
| # padding | |||
| ids += [vocab.padding_idx] * (max_len - len(text)) | |||
| return ids[:max_len] | |||
| def get_dataset(name, max_len, train_format="", test_format=""): | |||
| # datasets | |||
| train_dataset = load_dataset(name, split="train" + train_format).shuffle(seed=123) | |||
| test_dataset = load_dataset(name, split="test" + test_format).shuffle(seed=321) | |||
| split = train_dataset.train_test_split(test_size=0.2, seed=123) | |||
| train_dataset = split["train"] | |||
| val_dataset = split["test"] | |||
| vocab = Vocabulary() | |||
| vocab = set_vocabulary(vocab, train_dataset) | |||
| vocab = set_vocabulary(vocab, val_dataset) | |||
| train_dataset = process_data(train_dataset, vocab, max_len) | |||
| val_dataset = process_data(val_dataset, vocab, max_len) | |||
| test_dataset = process_data(test_dataset, vocab, max_len) | |||
| return TextClassificationDataset(train_dataset), TextClassificationDataset(val_dataset), \ | |||
| TextClassificationDataset(test_dataset), vocab | |||
| if __name__ == "__main__": | |||
| # 训练参数 | |||
| max_len = 20 | |||
| epochs = 40 | |||
| lr = 1 | |||
| batch_size = 64 | |||
| features = 100 | |||
| hidden_size = 128 | |||
| # 获取数据集 | |||
| # imdb.py SetFit/sst2 | |||
| train_data, val_data, test_data, vocab = get_dataset("SetFit/sst2", max_len, "", "") | |||
| # 使用dataloader | |||
| train_dataloader = JittorDataLoader( | |||
| dataset=train_data, | |||
| batch_size=batch_size, | |||
| shuffle=True, | |||
| num_workers=4, | |||
| ) | |||
| val_dataloader = JittorDataLoader( | |||
| dataset=val_data, | |||
| batch_size=batch_size, | |||
| shuffle=True, | |||
| num_workers=4, | |||
| ) | |||
| test_dataloader = JittorDataLoader( | |||
| dataset=test_data, | |||
| batch_size=1, | |||
| shuffle=False, | |||
| ) | |||
| # 初始化模型 | |||
| model = LSTM(len(vocab), hidden_size, features) | |||
| # 优化器 | |||
| # 也可以是多个优化器的list | |||
| optimizer = nn.SGD(model.parameters(), lr) | |||
| # Metrics | |||
| metrics = {"acc": Accuracy()} | |||
| # callbacks | |||
| callbacks = [ | |||
| PrintWhileTrainingCallBack(print_every_epoch=1, print_every_batch=10), | |||
| # RichCallback(), # print_every参数默认为1,即每一个batch更新一次进度条 | |||
| ] | |||
| trainer = Trainer( | |||
| model=model, | |||
| driver="jittor", | |||
| device=[0,1,2,3,4], | |||
| optimizers=optimizer, | |||
| train_dataloader=train_dataloader, | |||
| validate_dataloaders=val_dataloader, | |||
| validate_every=-1, | |||
| input_mapping=None, | |||
| output_mapping=None, | |||
| metrics=metrics, | |||
| n_epochs=epochs, | |||
| callbacks=callbacks, | |||
| # progress_bar="raw" | |||
| ) | |||
| trainer.run() | |||
| @@ -0,0 +1,110 @@ | |||
| # coding=utf-8 | |||
| # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors. | |||
| # | |||
| # Licensed under the Apache License, Version 2.0 (the "License"); | |||
| # you may not use this file except in compliance with the License. | |||
| # You may obtain a copy of the License at | |||
| # | |||
| # http://www.apache.org/licenses/LICENSE-2.0 | |||
| # | |||
| # Unless required by applicable law or agreed to in writing, software | |||
| # distributed under the License is distributed on an "AS IS" BASIS, | |||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |||
| # See the License for the specific language governing permissions and | |||
| # limitations under the License. | |||
| # Lint as: python3 | |||
| """IMDB movie reviews dataset.""" | |||
| import datasets | |||
| from datasets.tasks import TextClassification | |||
| _DESCRIPTION = """\ | |||
| Large Movie Review Dataset. | |||
| This is a dataset for binary sentiment classification containing substantially \ | |||
| more data than previous benchmark datasets. We provide a set of 25,000 highly \ | |||
| polar movie reviews for training, and 25,000 for testing. There is additional \ | |||
| unlabeled data for use as well.\ | |||
| """ | |||
| _CITATION = """\ | |||
| @InProceedings{maas-EtAl:2011:ACL-HLT2011, | |||
| author = {Maas, Andrew L. and Daly, Raymond E. and Pham, Peter T. and Huang, Dan and Ng, Andrew Y. and Potts, Christopher}, | |||
| title = {Learning Word Vectors for Sentiment Analysis}, | |||
| booktitle = {Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies}, | |||
| month = {June}, | |||
| year = {2011}, | |||
| address = {Portland, Oregon, USA}, | |||
| publisher = {Association for Computational Linguistics}, | |||
| pages = {142--150}, | |||
| url = {http://www.aclweb.org/anthology/P11-1015} | |||
| } | |||
| """ | |||
| _DOWNLOAD_URL = "http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz" | |||
| class IMDBReviewsConfig(datasets.BuilderConfig): | |||
| """BuilderConfig for IMDBReviews.""" | |||
| def __init__(self, **kwargs): | |||
| """BuilderConfig for IMDBReviews. | |||
| Args: | |||
| **kwargs: keyword arguments forwarded to super. | |||
| """ | |||
| super(IMDBReviewsConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs) | |||
| class Imdb(datasets.GeneratorBasedBuilder): | |||
| """IMDB movie reviews dataset.""" | |||
| BUILDER_CONFIGS = [ | |||
| IMDBReviewsConfig( | |||
| name="plain_text", | |||
| description="Plain text", | |||
| ) | |||
| ] | |||
| def _info(self): | |||
| return datasets.DatasetInfo( | |||
| description=_DESCRIPTION, | |||
| features=datasets.Features( | |||
| {"text": datasets.Value("string"), "label": datasets.features.ClassLabel(names=["neg", "pos"])} | |||
| ), | |||
| supervised_keys=None, | |||
| homepage="http://ai.stanford.edu/~amaas/data/sentiment/", | |||
| citation=_CITATION, | |||
| task_templates=[TextClassification(text_column="text", label_column="label")], | |||
| ) | |||
| def _split_generators(self, dl_manager): | |||
| archive = dl_manager.download(_DOWNLOAD_URL) | |||
| return [ | |||
| datasets.SplitGenerator( | |||
| name=datasets.Split.TRAIN, gen_kwargs={"files": dl_manager.iter_archive(archive), "split": "train"} | |||
| ), | |||
| datasets.SplitGenerator( | |||
| name=datasets.Split.TEST, gen_kwargs={"files": dl_manager.iter_archive(archive), "split": "test"} | |||
| ), | |||
| datasets.SplitGenerator( | |||
| name=datasets.Split("unsupervised"), | |||
| gen_kwargs={"files": dl_manager.iter_archive(archive), "split": "train", "labeled": False}, | |||
| ), | |||
| ] | |||
| def _generate_examples(self, files, split, labeled=True): | |||
| """Generate aclImdb examples.""" | |||
| # For labeled examples, extract the label from the path. | |||
| if labeled: | |||
| label_mapping = {"pos": 1, "neg": 0} | |||
| for path, f in files: | |||
| if path.startswith(f"aclImdb/{split}"): | |||
| label = label_mapping.get(path.split("/")[2]) | |||
| if label is not None: | |||
| yield path, {"text": f.read().decode("utf-8"), "label": label} | |||
| else: | |||
| for path, f in files: | |||
| if path.startswith(f"aclImdb/{split}"): | |||
| if path.split("/")[2] == "unsup": | |||
| yield path, {"text": f.read().decode("utf-8"), "label": -1} | |||
| @@ -1,3 +1,5 @@ | |||
| import os | |||
| from typing import List | |||
| import pytest | |||
| from dataclasses import dataclass | |||
| @@ -5,6 +7,7 @@ from fastNLP.core.controllers.trainer import Trainer | |||
| from fastNLP.core.metrics.accuracy import Accuracy | |||
| from fastNLP.core.callbacks.progress_callback import RichCallback | |||
| from fastNLP.envs.imports import _NEED_IMPORT_PADDLE | |||
| from fastNLP.envs.env import USER_CUDA_VISIBLE_DEVICES | |||
| if _NEED_IMPORT_PADDLE: | |||
| from paddle.optimizer import Adam | |||
| @@ -34,6 +37,8 @@ def test_trainer_paddle( | |||
| callbacks, | |||
| n_epochs=2, | |||
| ): | |||
| if isinstance(device, List) and USER_CUDA_VISIBLE_DEVICES not in os.environ: | |||
| pytest.skip("Skip test fleet if FASTNLP_BACKEND is not set to paddle.") | |||
| model = PaddleNormalModel_Classification_1( | |||
| num_labels=TrainPaddleConfig.num_labels, | |||
| feature_dimension=TrainPaddleConfig.feature_dimension | |||
| @@ -2,37 +2,42 @@ import os | |||
| import pytest | |||
| from fastNLP.core.utils.paddle_utils import get_device_from_visible, paddle_to, paddle_move_data_to_device | |||
| from fastNLP.core.utils.paddle_utils import _convert_data_device, paddle_to, paddle_move_data_to_device | |||
| from fastNLP.envs.imports import _NEED_IMPORT_PADDLE | |||
| if _NEED_IMPORT_PADDLE: | |||
| import paddle | |||
| @pytest.mark.parametrize( | |||
| ("user_visible_devices, cuda_visible_devices, device, output_type, correct"), | |||
| ("user_visible_devices, cuda_visible_devices, device, correct"), | |||
| ( | |||
| ("0,1,2,3,4,5,6,7", "0", "cpu", str, "cpu"), | |||
| ("0,1,2,3,4,5,6,7", "0", "cpu", int, "cpu"), | |||
| ("0,1,2,3,4,5,6,7", "3,4,5", "gpu:4", int, 1), | |||
| ("0,1,2,3,4,5,6,7", "3,4,5", "gpu:5", str, "gpu:2"), | |||
| ("3,4,5,6", "3,5", 0, int, 0), | |||
| ("3,6,7,8", "6,7,8", "gpu:2", str, "gpu:1"), | |||
| (None, None, 1, "gpu:1"), | |||
| (None, "2,4,5,6", 2, "gpu:2"), | |||
| (None, "3,4,5", 1, "gpu:1"), | |||
| ("0,1,2,3,4,5,6,7", "0", "cpu", "cpu"), | |||
| ("3,4,5,6,7", "0", "cpu", "cpu"), | |||
| ("0,1,2,3,4,5,6,7", "3,4,5", "gpu:4", "gpu:1"), | |||
| ("0,1,2,3,4,5,6,7", "3,4,5", "gpu:5", "gpu:2"), | |||
| ("3,4,5,6", "3,5", 0, "gpu:0"), | |||
| ("3,6,7,8", "6,7,8", "gpu:2", "gpu:1"), | |||
| ) | |||
| ) | |||
| @pytest.mark.paddle | |||
| def test_get_device_from_visible(user_visible_devices, cuda_visible_devices, device, output_type, correct): | |||
| def test_convert_data_device(user_visible_devices, cuda_visible_devices, device, correct): | |||
| _cuda_visible_devices = os.getenv("CUDA_VISIBLE_DEVICES") | |||
| _user_visible_devices = os.getenv("USER_CUDA_VISIBLE_DEVICES") | |||
| os.environ["CUDA_VISIBLE_DEVICES"] = cuda_visible_devices | |||
| os.environ["USER_CUDA_VISIBLE_DEVICES"] = user_visible_devices | |||
| res = get_device_from_visible(device, output_type) | |||
| if cuda_visible_devices is not None: | |||
| os.environ["CUDA_VISIBLE_DEVICES"] = cuda_visible_devices | |||
| if user_visible_devices is not None: | |||
| os.environ["USER_CUDA_VISIBLE_DEVICES"] = user_visible_devices | |||
| res = _convert_data_device(device) | |||
| assert res == correct | |||
| # 还原环境变量 | |||
| if _cuda_visible_devices is None: | |||
| del os.environ["CUDA_VISIBLE_DEVICES"] | |||
| os.environ.pop("CUDA_VISIBLE_DEVICES", None) | |||
| else: | |||
| os.environ["CUDA_VISIBLE_DEVICES"] = _cuda_visible_devices | |||
| if _user_visible_devices is None: | |||
| del os.environ["USER_CUDA_VISIBLE_DEVICES"] | |||
| os.environ.pop("USER_CUDA_VISIBLE_DEVICES", None) | |||
| else: | |||
| os.environ["USER_CUDA_VISIBLE_DEVICES"] = _user_visible_devices | |||
| @@ -0,0 +1,442 @@ | |||
| import pytest | |||
| from fastNLP.envs.imports import _NEED_IMPORT_JITTOR, _NEED_IMPORT_PADDLE, _NEED_IMPORT_TORCH | |||
| from fastNLP.modules.mix_modules.utils import ( | |||
| paddle2torch, | |||
| torch2paddle, | |||
| jittor2torch, | |||
| torch2jittor, | |||
| ) | |||
| if _NEED_IMPORT_TORCH: | |||
| import torch | |||
| if _NEED_IMPORT_PADDLE: | |||
| import paddle | |||
| if _NEED_IMPORT_JITTOR: | |||
| import jittor | |||
| ############################################################################ | |||
| # | |||
| # 测试paddle到torch的转换 | |||
| # | |||
| ############################################################################ | |||
| @pytest.mark.torchpaddle | |||
| class TestPaddle2Torch: | |||
| def check_torch_tensor(self, tensor, device, requires_grad): | |||
| """ | |||
| 检查张量设备和梯度情况的工具函数 | |||
| """ | |||
| assert isinstance(tensor, torch.Tensor) | |||
| assert tensor.device == torch.device(device) | |||
| assert tensor.requires_grad == requires_grad | |||
| def test_gradient(self): | |||
| """ | |||
| 测试张量转换后的反向传播是否正确 | |||
| """ | |||
| x = paddle.to_tensor([1.0, 2.0, 3.0, 4.0, 5.0], stop_gradient=False) | |||
| y = paddle2torch(x) | |||
| z = 3 * (y ** 2) | |||
| z.sum().backward() | |||
| assert y.grad.tolist() == [6, 12, 18, 24, 30] | |||
| def test_tensor_transfer(self): | |||
| """ | |||
| 测试单个张量的设备和梯度转换是否正确 | |||
| """ | |||
| paddle_tensor = paddle.rand((3, 4, 5)).cpu() | |||
| res = paddle2torch(paddle_tensor) | |||
| self.check_torch_tensor(res, "cpu", not paddle_tensor.stop_gradient) | |||
| res = paddle2torch(paddle_tensor, target_device="cuda:2", no_gradient=None) | |||
| self.check_torch_tensor(res, "cuda:2", not paddle_tensor.stop_gradient) | |||
| res = paddle2torch(paddle_tensor, target_device="cuda:1", no_gradient=True) | |||
| self.check_torch_tensor(res, "cuda:1", False) | |||
| res = paddle2torch(paddle_tensor, target_device="cuda:1", no_gradient=False) | |||
| self.check_torch_tensor(res, "cuda:1", True) | |||
| def test_list_transfer(self): | |||
| """ | |||
| 测试张量列表的转换 | |||
| """ | |||
| paddle_list = [paddle.rand((6, 4, 2)).cuda(1) for i in range(10)] | |||
| res = paddle2torch(paddle_list) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_torch_tensor(t, "cuda:1", False) | |||
| res = paddle2torch(paddle_list, target_device="cpu", no_gradient=False) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_torch_tensor(t, "cpu", True) | |||
| def test_tensor_tuple_transfer(self): | |||
| """ | |||
| 测试张量元组的转换 | |||
| """ | |||
| paddle_list = [paddle.rand((6, 4, 2)).cuda(1) for i in range(10)] | |||
| paddle_tuple = tuple(paddle_list) | |||
| res = paddle2torch(paddle_tuple) | |||
| assert isinstance(res, tuple) | |||
| for t in res: | |||
| self.check_torch_tensor(t, "cuda:1", False) | |||
| def test_dict_transfer(self): | |||
| """ | |||
| 测试包含复杂结构的字典的转换 | |||
| """ | |||
| paddle_dict = { | |||
| "tensor": paddle.rand((3, 4)).cuda(0), | |||
| "list": [paddle.rand((6, 4, 2)).cuda(0) for i in range(10)], | |||
| "dict":{ | |||
| "list": [paddle.rand((6, 4, 2)).cuda(0) for i in range(10)], | |||
| "tensor": paddle.rand((3, 4)).cuda(0) | |||
| }, | |||
| "int": 2, | |||
| "string": "test string" | |||
| } | |||
| res = paddle2torch(paddle_dict) | |||
| assert isinstance(res, dict) | |||
| self.check_torch_tensor(res["tensor"], "cuda:0", False) | |||
| assert isinstance(res["list"], list) | |||
| for t in res["list"]: | |||
| self.check_torch_tensor(t, "cuda:0", False) | |||
| assert isinstance(res["int"], int) | |||
| assert isinstance(res["string"], str) | |||
| assert isinstance(res["dict"], dict) | |||
| assert isinstance(res["dict"]["list"], list) | |||
| for t in res["dict"]["list"]: | |||
| self.check_torch_tensor(t, "cuda:0", False) | |||
| self.check_torch_tensor(res["dict"]["tensor"], "cuda:0", False) | |||
| ############################################################################ | |||
| # | |||
| # 测试torch到paddle的转换 | |||
| # | |||
| ############################################################################ | |||
| @pytest.mark.torchpaddle | |||
| class TestTorch2Paddle: | |||
| def check_paddle_tensor(self, tensor, device, stop_gradient): | |||
| """ | |||
| 检查得到的paddle张量设备和梯度情况的工具函数 | |||
| """ | |||
| assert isinstance(tensor, paddle.Tensor) | |||
| if device == "cpu": | |||
| assert tensor.place.is_cpu_place() | |||
| elif device.startswith("gpu"): | |||
| paddle_device = paddle.device._convert_to_place(device) | |||
| assert tensor.place.is_gpu_place() | |||
| if hasattr(tensor.place, "gpu_device_id"): | |||
| # paddle中,有两种Place | |||
| # paddle.fluid.core.Place是创建Tensor时使用的类型 | |||
| # 有函数gpu_device_id获取设备 | |||
| assert tensor.place.gpu_device_id() == paddle_device.get_device_id() | |||
| else: | |||
| # 通过_convert_to_place得到的是paddle.CUDAPlace | |||
| # 通过get_device_id获取设备 | |||
| assert tensor.place.get_device_id() == paddle_device.get_device_id() | |||
| else: | |||
| raise NotImplementedError | |||
| assert tensor.stop_gradient == stop_gradient | |||
| def test_gradient(self): | |||
| """ | |||
| 测试转换后梯度的反向传播 | |||
| """ | |||
| x = torch.tensor([1.0, 2.0, 3.0, 4.0, 5.0], requires_grad=True) | |||
| y = torch2paddle(x) | |||
| z = 3 * (y ** 2) | |||
| z.sum().backward() | |||
| assert y.grad.tolist() == [6, 12, 18, 24, 30] | |||
| def test_tensor_transfer(self): | |||
| """ | |||
| 测试单个张量的转换 | |||
| """ | |||
| torch_tensor = torch.rand((3, 4, 5)) | |||
| res = torch2paddle(torch_tensor) | |||
| self.check_paddle_tensor(res, "cpu", True) | |||
| res = torch2paddle(torch_tensor, target_device="gpu:2", no_gradient=None) | |||
| self.check_paddle_tensor(res, "gpu:2", True) | |||
| res = torch2paddle(torch_tensor, target_device="gpu:2", no_gradient=True) | |||
| self.check_paddle_tensor(res, "gpu:2", True) | |||
| res = torch2paddle(torch_tensor, target_device="gpu:2", no_gradient=False) | |||
| self.check_paddle_tensor(res, "gpu:2", False) | |||
| def test_tensor_list_transfer(self): | |||
| """ | |||
| 测试张量列表的转换 | |||
| """ | |||
| torch_list = [torch.rand(6, 4, 2) for i in range(10)] | |||
| res = torch2paddle(torch_list) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_paddle_tensor(t, "cpu", True) | |||
| res = torch2paddle(torch_list, target_device="gpu:1", no_gradient=False) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_paddle_tensor(t, "gpu:1", False) | |||
| def test_tensor_tuple_transfer(self): | |||
| """ | |||
| 测试张量元组的转换 | |||
| """ | |||
| torch_list = [torch.rand(6, 4, 2) for i in range(10)] | |||
| torch_tuple = tuple(torch_list) | |||
| res = torch2paddle(torch_tuple, target_device="cpu") | |||
| assert isinstance(res, tuple) | |||
| for t in res: | |||
| self.check_paddle_tensor(t, "cpu", True) | |||
| def test_dict_transfer(self): | |||
| """ | |||
| 测试复杂的字典结构的转换 | |||
| """ | |||
| torch_dict = { | |||
| "tensor": torch.rand((3, 4)), | |||
| "list": [torch.rand(6, 4, 2) for i in range(10)], | |||
| "dict":{ | |||
| "list": [torch.rand(6, 4, 2) for i in range(10)], | |||
| "tensor": torch.rand((3, 4)) | |||
| }, | |||
| "int": 2, | |||
| "string": "test string" | |||
| } | |||
| res = torch2paddle(torch_dict) | |||
| assert isinstance(res, dict) | |||
| self.check_paddle_tensor(res["tensor"], "cpu", True) | |||
| assert isinstance(res["list"], list) | |||
| for t in res["list"]: | |||
| self.check_paddle_tensor(t, "cpu", True) | |||
| assert isinstance(res["int"], int) | |||
| assert isinstance(res["string"], str) | |||
| assert isinstance(res["dict"], dict) | |||
| assert isinstance(res["dict"]["list"], list) | |||
| for t in res["dict"]["list"]: | |||
| self.check_paddle_tensor(t, "cpu", True) | |||
| self.check_paddle_tensor(res["dict"]["tensor"], "cpu", True) | |||
| ############################################################################ | |||
| # | |||
| # 测试jittor到torch的转换 | |||
| # | |||
| ############################################################################ | |||
| class TestJittor2Torch: | |||
| def check_torch_tensor(self, tensor, device, requires_grad): | |||
| """ | |||
| 检查得到的torch张量的工具函数 | |||
| """ | |||
| assert isinstance(tensor, torch.Tensor) | |||
| if device == "cpu": | |||
| assert not tensor.is_cuda | |||
| else: | |||
| assert tensor.device == torch.device(device) | |||
| assert tensor.requires_grad == requires_grad | |||
| def test_var_transfer(self): | |||
| """ | |||
| 测试单个Jittor Var的转换 | |||
| """ | |||
| jittor_var = jittor.rand((3, 4, 5)) | |||
| res = jittor2torch(jittor_var) | |||
| self.check_torch_tensor(res, "cpu", True) | |||
| res = jittor2torch(jittor_var, target_device="cuda:2", no_gradient=None) | |||
| self.check_torch_tensor(res, "cuda:2", True) | |||
| res = jittor2torch(jittor_var, target_device="cuda:2", no_gradient=True) | |||
| self.check_torch_tensor(res, "cuda:2", False) | |||
| res = jittor2torch(jittor_var, target_device="cuda:2", no_gradient=False) | |||
| self.check_torch_tensor(res, "cuda:2", True) | |||
| def test_var_list_transfer(self): | |||
| """ | |||
| 测试Jittor列表的转换 | |||
| """ | |||
| jittor_list = [jittor.rand((6, 4, 2)) for i in range(10)] | |||
| res = jittor2torch(jittor_list) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_torch_tensor(t, "cpu", True) | |||
| res = jittor2torch(jittor_list, target_device="cuda:1", no_gradient=False) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_torch_tensor(t, "cuda:1", True) | |||
| def test_var_tuple_transfer(self): | |||
| """ | |||
| 测试Jittor变量元组的转换 | |||
| """ | |||
| jittor_list = [jittor.rand((6, 4, 2)) for i in range(10)] | |||
| jittor_tuple = tuple(jittor_list) | |||
| res = jittor2torch(jittor_tuple, target_device="cpu") | |||
| assert isinstance(res, tuple) | |||
| for t in res: | |||
| self.check_torch_tensor(t, "cpu", True) | |||
| def test_dict_transfer(self): | |||
| """ | |||
| 测试字典结构的转换 | |||
| """ | |||
| jittor_dict = { | |||
| "tensor": jittor.rand((3, 4)), | |||
| "list": [jittor.rand(6, 4, 2) for i in range(10)], | |||
| "dict":{ | |||
| "list": [jittor.rand(6, 4, 2) for i in range(10)], | |||
| "tensor": jittor.rand((3, 4)) | |||
| }, | |||
| "int": 2, | |||
| "string": "test string" | |||
| } | |||
| res = jittor2torch(jittor_dict) | |||
| assert isinstance(res, dict) | |||
| self.check_torch_tensor(res["tensor"], "cpu", True) | |||
| assert isinstance(res["list"], list) | |||
| for t in res["list"]: | |||
| self.check_torch_tensor(t, "cpu", True) | |||
| assert isinstance(res["int"], int) | |||
| assert isinstance(res["string"], str) | |||
| assert isinstance(res["dict"], dict) | |||
| assert isinstance(res["dict"]["list"], list) | |||
| for t in res["dict"]["list"]: | |||
| self.check_torch_tensor(t, "cpu", True) | |||
| self.check_torch_tensor(res["dict"]["tensor"], "cpu", True) | |||
| ############################################################################ | |||
| # | |||
| # 测试torch到jittor的转换 | |||
| # | |||
| ############################################################################ | |||
| class TestTorch2Jittor: | |||
| def check_jittor_var(self, var, requires_grad): | |||
| """ | |||
| 检查得到的Jittor Var梯度情况的工具函数 | |||
| """ | |||
| assert isinstance(var, jittor.Var) | |||
| assert var.requires_grad == requires_grad | |||
| def test_gradient(self): | |||
| """ | |||
| 测试反向传播的梯度 | |||
| """ | |||
| x = torch.tensor([1.0, 2.0, 3.0, 4.0, 5.0], requires_grad=True) | |||
| y = torch2jittor(x) | |||
| z = 3 * (y ** 2) | |||
| grad = jittor.grad(z, y) | |||
| assert grad.tolist() == [6.0, 12.0, 18.0, 24.0, 30.0] | |||
| def test_tensor_transfer(self): | |||
| """ | |||
| 测试单个张量转换为Jittor | |||
| """ | |||
| torch_tensor = torch.rand((3, 4, 5)) | |||
| res = torch2jittor(torch_tensor) | |||
| self.check_jittor_var(res, False) | |||
| res = torch2jittor(torch_tensor, no_gradient=None) | |||
| self.check_jittor_var(res, False) | |||
| res = torch2jittor(torch_tensor, no_gradient=True) | |||
| self.check_jittor_var(res, False) | |||
| res = torch2jittor(torch_tensor, no_gradient=False) | |||
| self.check_jittor_var(res, True) | |||
| def test_tensor_list_transfer(self): | |||
| """ | |||
| 测试张量列表的转换 | |||
| """ | |||
| torch_list = [torch.rand((6, 4, 2)) for i in range(10)] | |||
| res = torch2jittor(torch_list) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_jittor_var(t, False) | |||
| res = torch2jittor(torch_list, no_gradient=False) | |||
| assert isinstance(res, list) | |||
| for t in res: | |||
| self.check_jittor_var(t, True) | |||
| def test_tensor_tuple_transfer(self): | |||
| """ | |||
| 测试张量元组的转换 | |||
| """ | |||
| torch_list = [torch.rand((6, 4, 2)) for i in range(10)] | |||
| torch_tuple = tuple(torch_list) | |||
| res = torch2jittor(torch_tuple) | |||
| assert isinstance(res, tuple) | |||
| for t in res: | |||
| self.check_jittor_var(t, False) | |||
| def test_dict_transfer(self): | |||
| """ | |||
| 测试字典结构的转换 | |||
| """ | |||
| torch_dict = { | |||
| "tensor": torch.rand((3, 4)), | |||
| "list": [torch.rand(6, 4, 2) for i in range(10)], | |||
| "dict":{ | |||
| "list": [torch.rand(6, 4, 2) for i in range(10)], | |||
| "tensor": torch.rand((3, 4)) | |||
| }, | |||
| "int": 2, | |||
| "string": "test string" | |||
| } | |||
| res = torch2jittor(torch_dict) | |||
| assert isinstance(res, dict) | |||
| self.check_jittor_var(res["tensor"], False) | |||
| assert isinstance(res["list"], list) | |||
| for t in res["list"]: | |||
| self.check_jittor_var(t, False) | |||
| assert isinstance(res["int"], int) | |||
| assert isinstance(res["string"], str) | |||
| assert isinstance(res["dict"], dict) | |||
| assert isinstance(res["dict"]["list"], list) | |||
| for t in res["dict"]["list"]: | |||
| self.check_jittor_var(t, False) | |||
| self.check_jittor_var(res["dict"]["tensor"], False) | |||