Add generate kwargs to Seq2SeqTrainingArguments (#13339)

* Add generate kwargs to Seq2SeqTrainingArguments

* typo

* Address review comments + doc

* Style

examples/pytorch/summarization/run_summarization.py

@@ -556,12 +556,15 @@ def compute_metrics(eval_preds):
 
     # Evaluation
     results = {}
+    max_length = (
+        training_args.generation_max_length
+        if training_args.generation_max_length is not None
+        else data_args.val_max_target_length
+    )
+    num_beams = data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
     if training_args.do_eval:
         logger.info("*** Evaluate ***")
-
-        metrics = trainer.evaluate(
-            max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, metric_key_prefix="eval"
-        )
+        metrics = trainer.evaluate(max_length=max_length, num_beams=num_beams, metric_key_prefix="eval")
         max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
         metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
 
@@ -572,10 +575,7 @@ def compute_metrics(eval_preds):
         logger.info("*** Predict ***")
 
         predict_results = trainer.predict(
-            predict_dataset,
-            metric_key_prefix="predict",
-            max_length=data_args.val_max_target_length,
-            num_beams=data_args.num_beams,
+            predict_dataset, metric_key_prefix="predict", max_length=max_length, num_beams=num_beams
         )
         metrics = predict_results.metrics
         max_predict_samples = (

examples/pytorch/translation/run_translation.py

@@ -549,12 +549,16 @@ def compute_metrics(eval_preds):
 
     # Evaluation
     results = {}
+    max_length = (
+        training_args.generation_max_length
+        if training_args.generation_max_length is not None
+        else data_args.val_max_target_length
+    )
+    num_beams = data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
     if training_args.do_eval:
         logger.info("*** Evaluate ***")
 
-        metrics = trainer.evaluate(
-            max_length=data_args.val_max_target_length, num_beams=data_args.num_beams, metric_key_prefix="eval"
-        )
+        metrics = trainer.evaluate(max_length=max_length, num_beams=num_beams, metric_key_prefix="eval")
         max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
         metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
 
@@ -565,10 +569,7 @@ def compute_metrics(eval_preds):
         logger.info("*** Predict ***")
 
         predict_results = trainer.predict(
-            predict_dataset,
-            metric_key_prefix="predict",
-            max_length=data_args.val_max_target_length,
-            num_beams=data_args.num_beams,
+            predict_dataset, metric_key_prefix="predict", max_length=max_length, num_beams=num_beams
         )
         metrics = predict_results.metrics
         max_predict_samples = (

src/transformers/trainer_seq2seq.py

@@ -70,10 +70,8 @@ def evaluate(
             A dictionary containing the evaluation loss and the potential metrics computed from the predictions. The
             dictionary also contains the epoch number which comes from the training state.
         """
-        if max_length is not None or not hasattr(self, "_max_length"):
-            self._max_length = max_length
-        if num_beams is not None or not hasattr(self, "_num_beams"):
-            self._num_beams = num_beams
+        self._max_length = max_length if max_length is not None else self.args.generation_max_length
+        self._num_beams = num_beams if num_beams is not None else self.args.generation_num_beams
         return super().evaluate(eval_dataset, ignore_keys=ignore_keys, metric_key_prefix=metric_key_prefix)
 
     def predict(
@@ -119,10 +117,8 @@ def predict(
             - metrics (:obj:`Dict[str, float]`, `optional`): The potential dictionary of metrics (if the dataset
               contained labels).
         """
-        if max_length is not None or not hasattr(self, "_max_length"):
-            self._max_length = max_length
-        if num_beams is not None or not hasattr(self, "_num_beams"):
-            self._num_beams = num_beams
+        self._max_length = max_length if max_length is not None else self.args.generation_max_length
+        self._num_beams = num_beams if num_beams is not None else self.args.generation_num_beams
         return super().predict(test_dataset, ignore_keys=ignore_keys, metric_key_prefix=metric_key_prefix)
 
     def prediction_step(

src/transformers/training_args_seq2seq.py

@@ -14,6 +14,7 @@
 
 import logging
 from dataclasses import dataclass, field
+from typing import Optional
 
 from .file_utils import add_start_docstrings
 from .training_args import TrainingArguments
@@ -34,9 +35,29 @@ class Seq2SeqTrainingArguments(TrainingArguments):
         the training set.
     predict_with_generate (:obj:`bool`, `optional`, defaults to :obj:`False`):
         Whether to use generate to calculate generative metrics (ROUGE, BLEU).
+    generation_max_length (:obj:`int`, `optional`):
+        The :obj:`max_length` to use on each evaluation loop when :obj:`predict_with_generate=True`. Will default to
+        the :obj:`max_length` value of the model configuration.
+    generation_num_beams (:obj:`int`, `optional`):
+        The :obj:`num_beams` to use on each evaluation loop when :obj:`predict_with_generate=True`. Will default to the
+        :obj:`num_beams` value of the model configuration.
     """
 
     sortish_sampler: bool = field(default=False, metadata={"help": "Whether to use SortishSampler or not."})
     predict_with_generate: bool = field(
         default=False, metadata={"help": "Whether to use generate to calculate generative metrics (ROUGE, BLEU)."}
     )
+    generation_max_length: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "The `max_length` to use on each evaluation loop when `predict_with_generate=True`. Will default "
+            "to the `max_length` value of the model configuration."
+        },
+    )
+    generation_num_beams: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "The `num_beams` to use on each evaluation loop when `predict_with_generate=True`. Will default "
+            "to the `num_beams` value of the model configuration."
+        },
+    )