[MAINTENANCE] Refactor and clean up.

[alexsherstinsky]

Scope

• Make the implementation for the fix of the ViTEncoder to ensure that the transformers.ViTModel returns the output_attentions more elegant (and cuts on the amount of code).

Code Pull Requests

Please provide the following:

• a clear explanation of what your code does
• if applicable, a reference to an issue
• a reproducible test for your PR (code, config and data sample)

Documentation Pull Requests

Note that the documentation HTML files are in docs/ while the Markdown sources are in mkdocs/docs.

If you are proposing a modification to the documentation you should change only the Markdown files.

api.md is automatically generated from the docstrings in the code, so if you want to change something in that file, first modify ludwig/api.py docstring, then run mkdocs/code_docs_autogen.py, which will create mkdocs/docs/api.md .

[arnavgarg1]

Can we just create a dictionary mapping, then optionally add "attn_implementation" if output_attention and pass the dictionary as **kwargs into the Config? Will reduce boilerplate :)

[alexsherstinsky]

@arnavgarg1 Done -- sorry about missing the obvious! Thanks!

[arnavgarg1]

Do we no longer need to do the if else so we can pass these custom args in both cases?

[alexsherstinsky]

@arnavgarg1 We do -- I updated the code. If we do not do it, the tests still pass, because locally nothing changes (but runs slower). This is because of the breaking change on the HF side, where they default to SDPA for speed (but it is lazy execution, leaving attention tensors as None; with "eager" in either case, things still work, but less efficiently). This is fixed now -- good catch -- thanks!

[github-actions]

Unit Test Results

  6 files  ±0    6 suites  ±0   14m 23s ⏱️ +2s
12 tests ±0    9 ✔️ ±0    3 💤 ±0  0 ❌ ±0 
60 runs  ±0  42 ✔️ ±0  18 💤 ±0  0 ❌ ±0 

Results for commit 9b91d6b. ± Comparison against base commit 7053966.

♻️ This comment has been updated with latest results.