fix: Ensure eval mode for farm and transformer models for predictions (#3791)

Co-authored-by: Massimiliano Pippi <mpippi@gmail.com>

haystack/modeling/training/base.py

@@ -290,6 +290,7 @@ class Trainer:
                 )
                 self.test_result = evaluator_test.eval(self.model)
                 evaluator_test.log_results(self.test_result, "Test", self.global_step)
+        self.model.eval()
         return self.model
 
     def compute_loss(self, batch: dict, step: int) -> torch.Tensor:

test/nodes/test_reader.py

@@ -431,6 +431,36 @@ def test_no_answer_reader_skips_empty_documents(no_answer_reader):
     assert predictions["answers"][1][1].answer == "Carla"  # answer given for 2nd query as usual
 
 
+@pytest.mark.integration
+def test_reader_training_returns_eval(tmp_path, samples_path):
+    max_seq_len = 16
+    max_query_length = 8
+    reader = FARMReader(
+        model_name_or_path="deepset/tinyroberta-squad2",
+        use_gpu=False,
+        num_processes=0,
+        max_seq_len=max_seq_len,
+        doc_stride=2,
+        max_query_length=max_query_length,
+    )
+
+    save_dir = f"{tmp_path}/test_dpr_training"
+    reader.train(
+        data_dir=str(samples_path / "squad"),
+        train_filename="tiny.json",
+        dev_filename="tiny.json",
+        n_epochs=1,
+        batch_size=1,
+        grad_acc_steps=1,
+        evaluate_every=0,
+        save_dir=save_dir,
+        max_seq_len=max_seq_len,
+        max_query_length=max_query_length,
+    )
+    assert reader.inferencer.model.training is False
+
+
+@pytest.mark.integration
 def test_reader_training(tmp_path, samples_path):
     max_seq_len = 16
     max_query_length = 8