Fix test for transformers 4.36

[helena-intel]

Instead of #1063

When fixing the test for 4.37 I tried with 4.36 and tests passed with that too, so we can support this a little longer.

Also added workflow_dispatch trigger to OpenVINO test. This makes it easier to test these kinds of changes in a branch in a fork. workflow_dispatch only works if it is also in main, so without having this in main you always have to add it to the main branch of your fork first. If there's a reason not to have it, I'll remove it.

@AlexKoff88

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[helena-intel]

@IlyasMoutawwakil or @echarlaix full and slow tests fail with

Repository Not Found for url: https://huggingface.co/api/models/optimum-internal-testing/tiny-random-phi-private/tree/openvino?recursive=True&expand=False.
Please make sure you specified the correct `repo_id` and `repo_type`.
If you are trying to access a private or gated repo, make sure you are authenticated.
Invalid username or password.

Could you look into that?

[echarlaix]

This failing test is not related to this PR so won't block merging : this test needs an hf token and should be skipped when not provided : https://github.com/huggingface/optimum-intel/blob/32ceae1d019f6884db609bd1aa48918820ebca61/tests/openvino/test_modeling.py#L399C1-L399C99 will look into it

[echarlaix]

An other tests seems to be failing coming with openvino nightly :

FAILED tests/openvino/test_modeling.py::OVModelForVisualCausalLMIntegrationTest::test_compare_to_transformers_3_minicpmv - AssertionError: False is not true : generation config : GenerationConfig {
  "max_new_tokens": 30,
  "min_new_tokens": 30
}
, transformers output tensor([[106581,  34341, 106581,  34341, 106581,  34341, 106581,  34341, 106581,
          34341, 106581,  34341, 106581,  34341, 106581,  34341, 106581,  34341,
         106581,  34341, 106581, 118883,  34341,  38070, 118883,  34341,  38070,
         118883,  34341,  38070]]), ov_model output tensor([[106581,  34341, 106581,  34341, 106581,  34341, 106581,  34341, 106581,
          34341, 106581,  34341, 106581,  34341, 106581,  34341, 106581,  34341,
         106581,  34341, 106581, 118883,  34341,  38070, 118883,  34341,  38070,
         118883,  34341,  15110]])
         

Not related to this PR (https://github.com/huggingface/optimum-intel/actions/runs/12345566645/job/34449846316) but wondering if someone from the openvino team can look into it @helena-intel

[echarlaix]

Thanks for the fix @helena-intel !

[helena-intel]

Not related to this PR (https://github.com/huggingface/optimum-intel/actions/runs/12345566645/job/34449846316) but wondering if someone from the openvino team can look into it @helena-intel

Thanks for the heads-up @echarlaix , I'll look into it!