Add default model_kwargs to Generation Tester

[ankurneog]

These are fixes for issues caught while executing the transformers tests. The issues are captured in #445 and
#446

Addressed based on suggestion give here : #470 (comment)

@regisss : please help with the review

TCs impacted ( listed for t5 but these are common TCs impacting all language models)

==============================================================
FAILED test_modeling_t5.py::T5ModelTest::test_beam_search_generate - KeyError: 'limit_hpu_graphs'
FAILED test_modeling_t5.py::T5ModelTest::test_beam_search_generate_dict_output - KeyError: 'limit_hpu_graphs'
FAILED test_modeling_t5.py::T5ModelTest::test_beam_search_generate_dict_outputs_use_cache - KeyError: 'limit_hpu_graphs'
FAILED test_modeling_t5.py::T5ModelTest::test_constrained_beam_search_generate - KeyError: 'limit_hpu_graphs'
FAILED test_modeling_t5.py::T5ModelTest::test_constrained_beam_search_generate_dict_output - KeyError: 'limit_hpu_graphs'
FAILED test_modeling_t5.py::T5ModelTest::test_greedy_generate - KeyError: 'bucket_size'
FAILED test_modeling_t5.py::T5ModelTest::test_greedy_generate_dict_outputs - KeyError: 'bucket_size'
FAILED test_modeling_t5.py::T5ModelTest::test_greedy_generate_dict_outputs_use_cache - KeyError: 'bucket_size'

FAILED test_modeling_t5.py::T5ModelTest::test_sample_generate - KeyError: 'limit_hpu_graphs'
FAILED test_modeling_t5.py::T5ModelTest::test_sample_generate_dict_output - KeyError: 'limit_hpu_graphs'

[regisss]

@ankurneog LGTM but you need to apply the code formater too:

pip install --upgrade black ruff
make style

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[ankurneog]

@regisss : Thanks,

I had already run the formatter before upload.

(conda_qnpu1) (conda_qnpu1) (anneog_fix_beamsearch_tests) anneog@anneog-vm-u20:optimum-habana $ make style
find . -name "habana_log.livealloc.log_*" -type f -delete
find . -name .lock -type f -delete
find . -name .graph_dumps -type d -delete
find . -name save-hpu.pdb -type f -delete
find . -name checkpoints.json -type f -delete
rm -rf regression/
rm -rf tmp_trainer/
rm -rf test/
rm -rf build/
rm -rf dist/
rm -rf optimum_habana.egg-info/
rm -rf hpu_profile/
black . setup.py
Skipping .ipynb files as Jupyter dependencies are not installed.
You can fix this by running pip install "black[jupyter]"
All done! ✨ 🍰 ✨
181 files left unchanged.
ruff . setup.py --fix

[regisss]

@ankurneog Weird. Can you let me know the versions of black and ruff you are running?

[ankurneog]

@regisss

(conda_qnpu1) (conda_qnpu1) (anneog_fix_beamsearch_tests) anneog@anneog-vm-u20:optimum-habana $ pip list |grep black
black 23.10.0
(conda_qnpu1) (conda_qnpu1) (anneog_fix_beamsearch_tests) anneog@anneog-vm-u20:optimum-habana $ pip list |grep ruff
ruff 0.1.1

[regisss]

Those are the same versions as the ones used in the test, very strange... In my environment, I do get a change that I added as a comment below. Let's modify this and then I'll merge!

[regisss]

LGTM

[regisss]

@ankurneog Can we close #470, #448 and #446?