Skip newer architecture config tests on older transformers versions

[brendanlong]

Description

The Apertus and GPT-OSS architecture tests fail when the installed transformers version doesn't recognize these model types. Skip with a clear message instead of failing. Also moves inline imports to top-level.

This allows us to run all of the unit tests on transformers 4.42, which is supported on Python 3.8:

https://github.com/TransformerLensOrg/TransformerLens/blob/main/pyproject.toml#L39

Type of change

• Bug fix (non-breaking change which fixes an issue)

Screenshots

With poetry run pip install transformers==4.42.0.

Before:

> poetry run pytest tests/unit/test_loading_from_pretrained_utilities.py 
========================================================= test session starts =========================================================
platform linux -- Python 3.12.9, pytest-8.3.5, pluggy-1.5.0
rootdir: /home/brendanlong/workspace/TransformerLens
configfile: pyproject.toml
plugins: jaxtyping-0.2.19, doctestplus-1.3.0, nbval-0.10.0, cov-5.0.0, anyio-4.5.2, typeguard-4.4.0
collected 8 items                                                                                                                     

tests/unit/test_loading_from_pretrained_utilities.py .....FFF                                                                   [100%]

============================================================== FAILURES ===============================================================
_____________________________________________ TestArchitectureConfigs.test_apertus_config _____________________________________________

[...]
../../.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/transformers/models/auto/configuration_auto.py:982: 
[...]
../../.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/transformers/models/auto/configuration_auto.py:684: KeyError

During handling of the above exception, another exception occurred:
[...]
tests/unit/test_loading_from_pretrained_utilities.py:102: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../../.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/jaxtyping/decorator.py:121: in wrapped_fn
    return fn(*args, **kwargs)
<@beartype(transformer_lens.loading_from_pretrained.get_pretrained_model_config) at 0x7988818c3ec0>:247: in get_pretrained_model_config
    ???
transformer_lens/loading_from_pretrained.py:2228: in get_pretrained_model_config
    cfg_dict = convert_hf_model_config(official_model_name, **kwargs)
../../.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/jaxtyping/decorator.py:121: in wrapped_fn
    return fn(*args, **kwargs)
<@beartype(transformer_lens.loading_from_pretrained.convert_hf_model_config) at 0x7988818a2b60>:31: in convert_hf_model_config
    ???
transformer_lens/loading_from_pretrained.py:862: in convert_hf_model_config
    hf_config = AutoConfig.from_pretrained(
[...]
E               ValueError: The checkpoint you are trying to load has model type `apertus` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.

../../.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/transformers/models/auto/configuration_auto.py:984: ValueError
[etc.]
FAILED tests/unit/test_loading_from_pretrained_utilities.py::TestArchitectureConfigs::test_apertus_config - ValueError: The checkpoint you are trying to load has model type `apertus` but Transformers does not recognize this architecture. ...
FAILED tests/unit/test_loading_from_pretrained_utilities.py::TestArchitectureConfigs::test_gpt_oss_config - ValueError: The checkpoint you are trying to load has model type `gpt_oss` but Transformers does not recognize this architecture. ...
FAILED tests/unit/test_loading_from_pretrained_utilities.py::TestArchitectureConfigs::test_apertus_instruct_config - ValueError: The checkpoint you are trying to load has model type `apertus` but Transformers does not recognize this architecture. ...
============================================== 3 failed, 5 passed, 53 warnings in 7.85s ===============================================

After:

> poetry run pytest tests/unit/test_loading_from_pretrained_utilities.py 
========================================================= test session starts =========================================================
platform linux -- Python 3.12.9, pytest-8.3.5, pluggy-1.5.0
rootdir: /home/brendanlong/workspace/TransformerLens
configfile: pyproject.toml
plugins: jaxtyping-0.2.19, doctestplus-1.3.0, nbval-0.10.0, cov-5.0.0, anyio-4.5.2, typeguard-4.4.0
collected 8 items                                                                                                                     

tests/unit/test_loading_from_pretrained_utilities.py .....sss                                                                   [100%]

========================================================== warnings summary ===========================================================
../../.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/jaxtyping/import_hook.py:106: 52 warnings
  /home/brendanlong/.cache/pypoetry/virtualenvs/transformer-lens-MODGY9P9-py3.12/lib/python3.12/site-packages/jaxtyping/import_hook.py:106: DeprecationWarning: ast.Str is deprecated and will be removed in Python 3.14; use ast.Constant instead
    elif isinstance(child, ast.Expr) and isinstance(child.value, ast.Str):

transformer_lens/SVDInterpreter.py:26
  /home/brendanlong/workspace/TransformerLens/transformer_lens/SVDInterpreter.py:26: InstrumentationWarning: @typechecked only supports instrumenting functions wrapped with @classmethod, @staticmethod or @property -- not typechecking transformer_lens.SVDInterpreter.SVDInterpreter.get_singular_vectors
    @typechecked

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
============================================== 5 passed, 3 skipped, 53 warnings in 7.45s ==============================================

Checklist:

• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have not rewritten tests relating to key interfaces which would affect backward compatibility