Fix seamless TTS generate

[ylacombe]

Fix #34811

cc @ydshieh

[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.

[ydshieh]

Thank you!

(a comment would be appreciated even if just a few words 🙏 )

[ydshieh]

let me check with the pipeline test - it's where I found this issue.

[ydshieh]

it works now! Could you remove 2 lines below: # TODO and @run_test_using_subprocess before merge? Thanks

tests/pipelines/test_pipelines_text_to_audio.py

    # TODO: @ylacombe: `SeamlessM4TForTextToSpeech.generate` has issue with `generation_config`. See issue #34811
    @slow
    @require_torch
    @run_test_using_subprocess
    def test_medium_seamless_m4t_pt(self):

[ruidazeng]

LGTM

[ylacombe]

I've removed the TODO!

[ydshieh]

You can ignore

tests/models/seamless_m4t_v2/test_modeling_seamless_m4t_v2.py::SeamlessM4Tv2ModelWithSpeechInputTest::test_retain_grad_hidden_states_attentions

failing on multi-gpu runner.