Don't remove unit test

tests/unit_tests/test_unit.py

@@ -450,6 +450,23 @@ def test_beam_search_decode():
     tokens[0, :step] = torch.tensor([4, 14, 4, 28])
     tokens[1, :step] = torch.tensor([4, 14, 4, 1])
 
+    # Set finished beams array to allow decoding from only one beam.
+    test_finished_beams = torch.tensor([True, False])
+
+    new_tokens, new_scores = model._get_topk_beams(
+        tokens, scores, test_finished_beams, batch, step
+    )
+
+    # Only the second peptide should have a new token predicted.
+    expected_tokens = torch.tensor(
+        [
+            [4, 14, 4, 28, 0],
+            [4, 14, 4, 1, 3],
+        ]
+    )
+
+    assert torch.equal(new_tokens[:, : step + 1], expected_tokens)
+
     # Test that duplicate peptide scores don't lead to a conflict in the cache.
     model = Spec2Pep(n_beams=5, residues="massivekb", min_peptide_len=3)
     batch = 2  # B