Understanding on the shape of the _candidates_data tensor

[bkowshik]

Based on my understanding, I was expecting the shape of model._candidates_data to be (1, 32000). But, like we see in the outputs below, the shape is different at (3, 32000).

model._candidates_data.shape
(3, 32000)

model._candidates_data
array([[(12178, 22.216051 , 1.        ), ( 2448, 13.263424 , 0.01056859),
        (   13, 12.963472 , 0.00782978), ...,
        (31997, -4.072526 , 0.        ), (31998, -5.1220403, 0.        ),
        (31999, -6.1943307, 0.        )],
       [(    0, -8.401089 , 0.        ), (    1, -9.950074 , 0.        ),
        (    2,  9.218815 , 0.        ), ...,
        (31997, -4.072526 , 0.        ), (31998, -5.1220403, 0.        ),
        (31999, -6.1943307, 0.        )],
       [(    0, -8.401089 , 0.        ), (    1, -9.950074 , 0.        ),
        (    2,  9.218815 , 0.        ), ...,
        (31997, -4.072526 , 0.        ), (31998, -5.1220403, 0.        ),
        (31999, -6.1943307, 0.        )]],
      dtype={'names': ['id', 'logit', 'p'], 'formats': ['<i4', '<f4', '<f4'], 'offsets': [0, 4, 8], 'itemsize': 12, 'aligned': True})

For the same token, there are multiple records in this array. Can somebody help me understand why this is the case.

logits_df = pd.DataFrame(model._candidates_data.ravel(), columns=model._candidates_data.dtype.names)

print(logits_df.shape)
logits_df[logits_df['id'] == 12178]