There was a difference between the movie review model runner used in the RISE text tutorial, vs LIME text, the tests and the dashboard. The LIME/dashboard/tests version first tokenizes all inputs, then runs the model once. The RISE version runs sentences through the model one-by-one. The latter is slower, but avoids the issues we've been seeing with special chars. The faster version could work if one makes sure the inputs all have the same shape, but this can be complicated so I suggest we go with the slower, but correct, version.
There was a difference between the movie review model runner used in the RISE text tutorial, vs LIME text, the tests and the dashboard. The LIME/dashboard/tests version first tokenizes all inputs, then runs the model once. The RISE version runs sentences through the model one-by-one. The latter is slower, but avoids the issues we've been seeing with special chars. The faster version could work if one makes sure the inputs all have the same shape, but this can be complicated so I suggest we go with the slower, but correct, version.
Fixes #531 Fixes #771 Fixes #751