Crash in test script using dictionary guided decoding

The test script is failing on a couple of the experiments with an usual error ("Slice index 113 of dimension 0 out of bounds"). The experiment setup looks like this

• Corpus_pairs: 1) mbt-MBTV / en-MBTBT, and 2) mbt-ArabicDigits_0_9 / en-ArabicDigits_0_9
• The second corpus pair simply has the digits '0' through '9' on separate lines for both the source and target side.
• One experiment (OT-Sequencing/MBTV-MBTBT.AE.1i) uses the Arabic Digits files only for the dictionary; the other experiment (OT-Sequencing/MBTV-MBTBT.AE.1j) uses the Arabic Digits files for the dictionary and for train.

The preprocessing and training appears to run successfully, but the inferencing fails pretty quickly with the following stack trace:
2021-08-01 16:40:37.369377: W tensorflow/core/framework/op_kernel.cc:1763] OP_REQUIRES failed at strided_slice_op.cc:108 : Invalid argument: slice index 113 of dimension 0 out of bounds.
Traceback (most recent call last):
  File "/usr/lib/python3.7/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/usr/lib/python3.7/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/content/silnlp/silnlp/nmt/test.py", line 605, in <module>
    main()
  File "/content/silnlp/silnlp/nmt/test.py", line 568, in main
    books,
  File "/content/silnlp/silnlp/nmt/test.py", line 383, in test_checkpoint
    checkpoint_path=str(checkpoint_path),
  File "/content/silnlp/silnlp/nmt/runner.py", line 345, in infer_multiple
    predictions = infer_fn(source)
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/def_function.py", line 828, in __call__
    result = self._call(*args, **kwds)
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/def_function.py", line 855, in _call
    return self._stateless_fn(*args, **kwds)  # pylint: disable=not-callable
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/function.py", line 2943, in __call__
    filtered_flat_args, captured_inputs=graph_function.captured_inputs)  # pylint: disable=protected-access
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/function.py", line 1919, in _call_flat
    ctx, args, cancellation_manager=cancellation_manager))
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/function.py", line 560, in call
    ctx=ctx)
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/eager/execute.py", line 60, in quick_execute
    inputs, attrs, num_outputs)
tensorflow.python.framework.errors_impl.InvalidArgumentError: 2 root error(s) found.
  (0) Invalid argument:  slice index 113 of dimension 0 out of bounds.
	 [[{{node sil_transformer_base_alignment_enhanced/map/while/body/_1/sil_transformer_base_alignment_enhanced/map/while/StatefulPartitionedCall/while/body/_1741/while/while/body/_2234/while/while/cond_1/then/_2483/while/while/cond_1/strided_slice_2}}]]
	 [[sil_transformer_base_alignment_enhanced/map/while/body/_1/sil_transformer_base_alignment_enhanced/map/while/StatefulPartitionedCall/while/body/_1741/while/cond/else/_2324/while/cond/while_1/LoopCond/_2575/_710]]
  (1) Invalid argument:  slice index 113 of dimension 0 out of bounds.
	 [[{{node sil_transformer_base_alignment_enhanced/map/while/body/_1/sil_transformer_base_alignment_enhanced/map/while/StatefulPartitionedCall/while/body/_1741/while/while/body/_2234/while/while/cond_1/then/_2483/while/while/cond_1/strided_slice_2}}]]
0 successful operations.
0 derived errors ignored. [Op:__inference_infer_33753]

Function call stack:
infer -> infer 

This is the only experiment setup that fails like this. Other experiments (e.g., OT-Sequencing/MBTV-MBTBT.AE.1k and .1l) which use a larger Arabic digits file (0-99, 0-999, etc) do not see this failure.

Start with silnlp/nmt/test.py, especially test_checkpoint, and follow the inference call into silnlp/nmt/runner.py:345. Reproduce inferencing with the two Arabic-digits corpus setups and inspect the TensorFlow stack trace; done means those dictionary-guided experiments complete without the out-of-bounds failure.