some kind of weird error regarding float32 I never got before upgrading

[BBC-Esq]

Below I paste a weird error I get when upgrading from 0.33 to 0.33.1. Not sure if it matters, but I also have installed the following for my vector database program:

1 - tiledb-cloud==0.13.0
2 - tiledb-vector-search==0.11.0
3 - tblib==1.7.0
4 - numpy==1.26.4

LONG ERROR HERE EXPAND DOWNWARDS

Traceback (most recent call last):
  File "C:\Users\Airflow\AppData\Local\Programs\Python\Python311\Lib\multiprocessing\process.py", line 314, in _bootstrap
    self.run()
  File "C:\Users\Airflow\AppData\Local\Programs\Python\Python311\Lib\multiprocessing\process.py", line 108, in run
    self._target(self._args, **self._kwargs)
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\database_interactions.py", line 29, in create_vector_db_in_process
    create_vector_db.run()
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\torch\utils\_contextlib.py", line 116, in decorate_context
    return func(args, kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\database_interactions.py", line 398, in run
    self.create_database(texts, embeddings)
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\torch\utils\_contextlib.py", line 116, in decorate_context
    return func(*args, kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\database_interactions.py", line 246, in create_database
    db = TileDB.from_texts(
         ^^^^^^^^^^^^^^^^^^
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\langchain_community\vectorstores\tiledb.py", line 726, in from_texts
    return cls.from(
           ^^^^^^^^^^^
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\langchain_community\vectorstores\tiledb.py", line 578, in from
    tiledb_vs.ingestion.ingest(
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\tiledb\vector_search\ingestion.py", line 3302, in ingest
    source_uri = write_input_vectors(
                 ^^^^^^^^^^^^^^^^^^^^
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\tiledb\vector_search\ingestion.py", line 549, in write_input_vectors
    input_vectors_array[:, :] = np.transpose(input_vectors)
    ~~~~~~~~~~~~~~~~~~~^^^^^^
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\tiledb\dense_array.py", line 383, in setitem
    self._setitem_impl(selection, val, dict())
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\tiledb\dense_array.py", line 578, in setitemimpl
    self._write_array(subarray, [], attributes, values, labels, nullmaps, False)
  File "D:\Scripts\VectorDB-Plugin-for-LM-Studio\working\Lib\site-packages\tiledb\array.py", line 899, in writearray
    q.set_data_buffer(
TypeError: set_data_buffer(): incompatible function arguments. The following argument types are supported:
    1. (self: tiledb.libtiledb.Query, arg0: str, arg1: numpy.ndarray, arg2: int) -> None
Invoked with: <tiledb.libtiledb.Query object at 0x000002C91121C430>, 'values', array([[-0.01574707,  0.01733398, -0.00946045, ...,  0.01116943,
        -0.04248047, -0.02770996],
       [-0.0267334 , -0.00769043, -0.02185059, ..., -0.03613281,
        -0.01495361,  0.00601196],
       [ 0.03662109,  0.02441406,  0.06347656, ...,  0.05664062,
         0.01556396,  0.00430298],
       ...,
       [-0.01501465, -0.02563477, -0.04370117, ..., -0.05200195,
        -0.01147461, -0.02404785],
       [-0.12451172, -0.13378906, -0.0859375 , ..., -0.08447266,
        -0.13085938, -0.12792969],
       [ 0.02819824,  0.02331543,  0.07714844, ...,  0.02478027,
         0.03271484,  0.02355957]], dtype=float32), 31104.0

I can verify that downgrading to 0.33 makes it work again.

[kounelisagis]

Hello and thank you for opening the issue.

Could you share a code snippet to check that the PR fixes your issue?

[BBC-Esq]

All the the scripts mentioned in the traceback are brand spanking new in this release and can be reviewed on my repository here:

https://github.com/BBC-Esq/VectorDB-Plugin-for-LM-Studio

Also, not sure if it matters but at the time of the error I was running the relatively new "granite" embedding model located here, which is natively in bfloat16:

https://huggingface.co/ibm-granite/granite-embedding-30m-english

[ihnorton]

We just released 0.33.2 with a fix for this issue; wheels should be up on PyPI shortly.