Hugging Face generators: set a default for max_new_tokens

[anakin87]

We have 4 Hugging Face Generators: HFLocalGenerator, HFTGIGenerator and the corresponding chat ones.

max_new_tokens can be set at initialization inside the generation_kwargs dict (or passed during run).

    generator = HuggingFaceLocalGenerator(
        model="google/flan-t5-large",
        generation_kwargs={"max_new_tokens": 100, "temperature": 0.9})

If not set, the Hugging Face default for max_length (= input prompt + max_new_tokens) is 20, which leads almost always to truncated responses.
Only in HuggingFaceLocalChatGenerator, we set the default to 512, which can be a good default.

For consistency, we should also update the other Hugging Face Generators to provide the same default.

[CKeibel]

Hi @anakin87 !
If no one is working on this issue yet, could I work on this to get started with contributing?

[anakin87]

Feel free to do that!

[CKeibel]

Great, I will check it out!

[anakin87]

done in #7370