fixing positions for beam search gpt2

[viboga]

Description:
Fixing position_ids input in iterations of gpt2 beam search.

Motivation and Context

• Why is this change required? What problem does it solve?
  In gpt2 beam search, position ids for the tokens are incremented for each iteration. This is based of the initial length.
  This was initialized to sequence length but should be initialized to sequence length-1 for each batch.

• If it fixes an open issue, please link to the issue here.
  N/A

Test Case:
Ran beam search for an input sequence of length 52. With DEBUG_BEAM_SEARCH enabled these are the outputs:

It works fine for the first iteration:
position_ids
Shape:{1,52}
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51

For the second iteration it should be 52, but its 53
position_ids
Shape:{1,1}
53

After the fix:

position_ids
Shape:{1,1}
52