Introduce warmup_ratio training argument (#10229)

Introduce warmup_ratio training argument in both
TrainingArguments and TFTrainingArguments classes (#6673)

src/transformers/trainer.py

@@ -615,10 +615,16 @@ class Trainer:
                 self.optimizer = optimizer_cls(optimizer_grouped_parameters, **optimizer_kwargs)
 
         if self.lr_scheduler is None:
+            warmup_steps = (
+                self.args.warmup_steps
+                if self.args.warmup_steps > 0
+                else math.ceil(num_training_steps * self.args.warmup_ratio)
+            )
+
             self.lr_scheduler = get_scheduler(
                 self.args.lr_scheduler_type,
                 self.optimizer,
-                num_warmup_steps=self.args.warmup_steps,
+                num_warmup_steps=warmup_steps,
                 num_training_steps=num_training_steps,
             )
 

src/transformers/trainer_tf.py

@@ -218,10 +218,16 @@ class TFTrainer:
         TFTrainer's init through :obj:`optimizers`, or subclass and override this method.
         """
         if not self.optimizer and not self.lr_scheduler:
+            warmup_steps = (
+                self.args.warmup_steps
+                if self.args.warmup_steps > 0
+                else math.ceil(num_training_steps * self.args.warmup_ratio)
+            )
+
             self.optimizer, self.lr_scheduler = create_optimizer(
                 self.args.learning_rate,
                 num_training_steps,
-                self.args.warmup_steps,
+                warmup_steps,
                 adam_beta1=self.args.adam_beta1,
                 adam_beta2=self.args.adam_beta2,
                 adam_epsilon=self.args.adam_epsilon,

src/transformers/training_args.py

@@ -131,8 +131,11 @@ class TrainingArguments:
         lr_scheduler_type (:obj:`str` or :class:`~transformers.SchedulerType`, `optional`, defaults to :obj:`"linear"`):
             The scheduler type to use. See the documentation of :class:`~transformers.SchedulerType` for all possible
             values.
+        warmup_ratio (:obj:`float`, `optional`, defaults to 0.0):
+            Ratio of total training steps used for a linear warmup from 0 to :obj:`learning_rate`.
         warmup_steps (:obj:`int`, `optional`, defaults to 0):
-            Number of steps used for a linear warmup from 0 to :obj:`learning_rate`.
+            Number of steps used for a linear warmup from 0 to :obj:`learning_rate`. Overrides any effect of
+            :obj:`warmup_ratio`.
         logging_dir (:obj:`str`, `optional`):
             `TensorBoard <https://www.tensorflow.org/tensorboard>`__ log directory. Will default to
             `runs/**CURRENT_DATETIME_HOSTNAME**`.
@@ -324,6 +327,9 @@ class TrainingArguments:
         default="linear",
         metadata={"help": "The scheduler type to use."},
     )
+    warmup_ratio: float = field(
+        default=0.0, metadata={"help": "Linear warmup over warmup_ratio fraction of total steps."}
+    )
     warmup_steps: int = field(default=0, metadata={"help": "Linear warmup over warmup_steps."})
 
     logging_dir: Optional[str] = field(default_factory=default_logdir, metadata={"help": "Tensorboard log dir."})
@@ -495,6 +501,13 @@ class TrainingArguments:
         elif not isinstance(self.report_to, list):
             self.report_to = [self.report_to]
 
+        if self.warmup_ratio < 0 or self.warmup_ratio > 1:
+            raise ValueError("warmup_ratio must lie in range [0,1]")
+        elif self.warmup_ratio > 0 and self.warmup_steps > 0:
+            logger.info(
+                "Both warmup_ratio and warmup_steps given, warmup_steps will override any effect of warmup_ratio during training"
+            )
+
     def __repr__(self):
         # We override the default repr to remove deprecated arguments from the repr. This method should be removed once
         # those deprecated arguments are removed form TrainingArguments. (TODO: v5)

src/transformers/training_args_tf.py

@@ -94,8 +94,11 @@ class TFTrainingArguments(TrainingArguments):
         max_steps (:obj:`int`, `optional`, defaults to -1):
             If set to a positive number, the total number of training steps to perform. Overrides
             :obj:`num_train_epochs`.
+        warmup_ratio (:obj:`float`, `optional`, defaults to 0.0):
+            Ratio of total training steps used for a linear warmup from 0 to :obj:`learning_rate`.
         warmup_steps (:obj:`int`, `optional`, defaults to 0):
-            Number of steps used for a linear warmup from 0 to :obj:`learning_rate`.
+            Number of steps used for a linear warmup from 0 to :obj:`learning_rate`. Overrides any effect of
+            :obj:`warmup_ratio`.
         logging_dir (:obj:`str`, `optional`):
             `TensorBoard <https://www.tensorflow.org/tensorboard>`__ log directory. Will default to
             `runs/**CURRENT_DATETIME_HOSTNAME**`.