Fixed OOM issues when evaluating/predicting (#721)

* Fixed OOM issues when evaluating/predicting

* Fixed test

* Improved error message

* Fixed test

Author: gabrielspmoreira

tests/unit/torch/test_trainer.py

@@ -407,9 +407,14 @@ def test_trainer_music_streaming(task_and_metrics):
     assert eval_metrics["eval_/loss"] is not None
 
     assert predictions is not None
+
+    DEFAULT_PREDICT_TOP_K = 100
+
     # 1000 is the total samples in the testing data
     if isinstance(task, tr.NextItemPredictionTask):
-        assert predictions.predictions.shape == (1000, task.target_dim)
+        top_predicted_item_ids, top_prediction_scores = predictions.predictions
+        assert top_predicted_item_ids.shape == (1000, DEFAULT_PREDICT_TOP_K)
+        assert top_prediction_scores.shape == (1000, DEFAULT_PREDICT_TOP_K)
     else:
         assert predictions.predictions.shape == (1000,)
 
@@ -573,17 +578,79 @@ def test_trainer_with_multiple_tasks(schema_type):
     assert predictions.predictions["click/binary_classification_task"].shape == (1000,)
 
 
-def test_trainer_trop_k_with_wrong_task():
+@pytest.mark.parametrize("predict_top_k", [20, None, "default"])
+def test_trainer_predict_topk(predict_top_k):
+    DEFAULT_PREDICT_TOP_K = 100
+
     data = tr.data.music_streaming_testing_data
     schema = data.schema
     batch_size = 16
-    predict_top_k = 20
 
-    task = tr.BinaryClassificationTask("click", summary_type="mean")
+    task = tr.NextItemPredictionTask(weight_tying=True)
+    inputs = tr.TabularSequenceFeatures.from_schema(
+        schema,
+        max_sequence_length=20,
+        d_output=64,
+        masking="clm",
+    )
+    transformer_config = tconf.XLNetConfig.build(64, 4, 2, 20)
+    model = transformer_config.to_torch_model(inputs, task)
+
+    additional_args = {}
+    if not isinstance(predict_top_k, str):
+        additional_args["predict_top_k"] = predict_top_k
+
+    args = trainer.T4RecTrainingArguments(
+        output_dir=".",
+        num_train_epochs=1,
+        per_device_train_batch_size=batch_size,
+        per_device_eval_batch_size=batch_size // 2,
+        data_loader_engine="merlin_dataloader",
+        max_sequence_length=20,
+        report_to=[],
+        debug=["r"],
+        **additional_args,
+    )
+
+    recsys_trainer = tr.Trainer(
+        model=model,
+        args=args,
+        schema=schema,
+        train_dataset_or_path=data.path,
+        eval_dataset_or_path=data.path,
+        test_dataset_or_path=data.path,
+        compute_metrics=True,
+    )
+
+    outputs = recsys_trainer.predict(data.path)
+
+    if predict_top_k is None:
+        assert outputs.predictions.shape[1] == 10001
+    else:
+        if predict_top_k == "default":
+            predict_top_k = DEFAULT_PREDICT_TOP_K
+
+        pred_item_ids, pred_scores = outputs.predictions
+        assert len(pred_item_ids.shape) == 2
+        assert pred_item_ids.shape[1] == predict_top_k
+        assert len(pred_scores.shape) == 2
+        assert pred_scores.shape[1] == predict_top_k
+
+
+@pytest.mark.parametrize("predict_top_k", [15, 20, 30, None])
+@pytest.mark.parametrize("top_k", [20, None])
+def test_trainer_predict_top_k_x_top_k(predict_top_k, top_k):
+    data = tr.data.music_streaming_testing_data
+    schema = data.schema
+    batch_size = 16
+
+    task = tr.NextItemPredictionTask(weight_tying=True)
+
     inputs = tr.TabularSequenceFeatures.from_schema(
         schema,
         max_sequence_length=20,
         d_output=64,
+        masking="clm",
     )
     transformer_config = tconf.XLNetConfig.build(64, 4, 2, 20)
     model = transformer_config.to_torch_model(inputs, task)
@@ -609,10 +676,28 @@ def test_trainer_trop_k_with_wrong_task():
         test_dataset_or_path=data.path,
         compute_metrics=True,
     )
-    with pytest.raises(AssertionError) as excinfo:
-        recsys_trainer.predict(data.path)
 
-    assert "Top-k prediction is specific to NextItemPredictionTask" in str(excinfo.value)
+    model.top_k = top_k
+
+    if predict_top_k and top_k and predict_top_k > top_k:
+        with pytest.raises(ValueError) as excinfo:
+            recsys_trainer.predict(data.path)
+        assert "The args.predict_top_k should not be larger than model.top_k" in str(excinfo.value)
+
+    else:
+        outputs = recsys_trainer.predict(data.path)
+
+        if predict_top_k or top_k:
+            expected_top_k = predict_top_k or top_k
+
+            pred_item_ids, pred_scores = outputs.predictions
+            assert len(pred_item_ids.shape) == 2
+            assert pred_item_ids.shape[1] == expected_top_k
+            assert len(pred_scores.shape) == 2
+            assert pred_scores.shape[1] == expected_top_k
+        else:
+            ITEM_CARDINALITY = 10001
+            assert outputs.predictions.shape[1] == ITEM_CARDINALITY
 
 
 def test_trainer_with_pretrained_embeddings():

transformers4rec/config/trainer.py

@@ -38,8 +38,9 @@ class T4RecTrainingArguments(TrainingArguments):
     predict_top_k:  Option[int], int
         Truncate recommendation list to the highest top-K predicted items,
         (do not affect evaluation metrics computation),
-        this parameter is specific to NextItemPredictionTask.
-        by default 0
+        This parameter is specific to NextItemPredictionTask and only affects
+        model.predict() and model.evaluate(), which both call `Trainer.evaluation_loop`.
+        By default 100.
     log_predictions : Optional[bool], bool
         log predictions, labels and metadata features each --compute_metrics_each_n_steps
         (for test set).
@@ -90,10 +91,10 @@ class T4RecTrainingArguments(TrainingArguments):
     )
 
     predict_top_k: int = field(
-        default=0,
+        default=100,
         metadata={
             "help": "Truncate recommendation list to the highest top-K predicted items (do not affect evaluation metrics computation), "
-            "this parameter is specific to NextItemPredictionTask."
+            "this parameter is specific to NextItemPredictionTask. Default is 100."
         },
     )
 

transformers4rec/torch/trainer.py

@@ -528,50 +528,68 @@ def evaluation_loop(
                     if labels_host is None
                     else nested_concat(labels_host, labels, padding_index=0)
                 )
-            if preds is not None and self.args.predict_top_k > 0:
-                if self.model.top_k:
-                    raise ValueError(
-                        "you cannot set top_k argument in the model class and the, "
-                        "predict_top_k  in the trainer at the same time. Please ensure setting "
-                        "only predict_top_k"
-                    )
+
+            if (
+                preds is not None
+                and any(isinstance(x, NextItemPredictionTask) for x in model.prediction_tasks)
+                and (self.args.predict_top_k or self.model.top_k)
+            ):
                 # get outputs of next-item scores
                 if isinstance(preds, dict):
-                    assert any(
-                        isinstance(x, NextItemPredictionTask) for x in model.prediction_tasks
-                    ), "Top-k prediction is specific to NextItemPredictionTask, "
-                    "Please ensure `self.args.predict_top_k == 0` "
                     pred_next_item = preds["next-item"]
                 else:
-                    assert isinstance(
-                        model.prediction_tasks[0], NextItemPredictionTask
-                    ), "Top-k prediction is specific to NextItemPredictionTask, "
-                    "Please ensure `self.args.predict_top_k == 0` "
                     pred_next_item = preds
 
-                preds_sorted_item_scores, preds_sorted_item_ids = torch.topk(
-                    pred_next_item, k=self.args.predict_top_k, dim=-1
-                )
-                self._maybe_log_predictions(
-                    labels,
-                    preds_sorted_item_ids,
-                    preds_sorted_item_scores,
-                    # outputs["pred_metadata"],
-                    metrics_results_detailed,
-                    metric_key_prefix,
-                )
-                # The output predictions will be a tuple with the ranked top-n item ids,
-                # and item recommendation scores
-                if isinstance(preds, dict):
-                    preds["next-item"] = (
-                        preds_sorted_item_ids,
-                        preds_sorted_item_scores,
+                preds_sorted_item_scores = None
+                preds_sorted_item_ids = None
+
+                if self.model.top_k is not None and isinstance(pred_next_item, (list, tuple)):
+                    preds_sorted_item_scores, preds_sorted_item_ids = pred_next_item
+
+                    if self.args.predict_top_k:
+                        if self.args.predict_top_k > self.model.top_k:
+                            raise ValueError(
+                                "The args.predict_top_k should not be larger than model.top_k. "
+                                "The model.top_k is available to support inference (e.g. when "
+                                "serving with Triton Inference Server) to return only the top-k "
+                                "predicted items ids and their scores."
+                                "When doing offline predictions with `trainer.predict(), "
+                                "if you set model.top_k, the model will also limit the number of "
+                                "predictions output from trainer.predict(). "
+                                "In that case, you want either to reduce args.predict_top_k or "
+                                "increase model.top_k, so that args.predict_top_k is "
+                                "not larger than model.top_k."
+                            )
+                        preds_sorted_item_scores = preds_sorted_item_scores[
+                            :, : self.args.predict_top_k
+                        ]
+                        preds_sorted_item_ids = preds_sorted_item_ids[:, : self.args.predict_top_k]
+                elif self.args.predict_top_k:
+                    preds_sorted_item_scores, preds_sorted_item_ids = torch.topk(
+                        pred_next_item, k=self.args.predict_top_k, dim=-1
                     )
-                else:
-                    preds = (
+
+                if preds_sorted_item_scores is not None:
+                    self._maybe_log_predictions(
+                        labels,
                         preds_sorted_item_ids,
                         preds_sorted_item_scores,
+                        # outputs["pred_metadata"],
+                        metrics_results_detailed,
+                        metric_key_prefix,
                     )
+                    # The output predictions will be a tuple with the ranked top-n item ids,
+                    # and item recommendation scores
+                    if isinstance(preds, dict):
+                        preds["next-item"] = (
+                            preds_sorted_item_ids,
+                            preds_sorted_item_scores,
+                        )
+                    else:
+                        preds = (
+                            preds_sorted_item_ids,
+                            preds_sorted_item_scores,
+                        )
 
             preds_host = (
                 preds