Make stable diffusion pipelines compatible with compel

[JingyaHuang]

What does this PR do?

• Make NeuronStableDiffusionXLPipeline compel compatible
• Make NeuronStableDiffusionPipeline compel compatible
• Tests
• Fix the neuron config wrongly reconstructed for sdxl's text encoder 2

Fixes #357

@neo @Suprhimp

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[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.

[dacorvo]

LGTM, thanks ! And my deepest apologies for the wrongly formatted TGI test from my last pull-request you had to fix (looks like style-check is not triggered on TGI changes).

[neo]

Thank you for inviting me to review! I left some minor comments, and I also tested it and it seemed like this would fix #371 as well 😆

[Suprhimp]

I tested with my own model even with lora, and it works pretty well.

I compiled with output_hidden_states params true

And I tested with this example code 😃

from optimum.neuron import NeuronStableDiffusionXLPipeline
from compel import Compel, ReturnedEmbeddingsType

pipe = NeuronStableDiffusionXLPipeline.from_pretrained("sdxl_turbo_neuron/", data_parallel_mode="all")
prompt = "Self-portrait oil painting++++++++++, a beautiful cyborg with golden hair, 8k"
negative_prompt = "worst quality"

compel = Compel(
        tokenizer=[pipe.tokenizer, pipe.tokenizer_2],
        text_encoder=[pipe.text_encoder, pipe.text_encoder_2],
        returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
        requires_pooled=[False, True],
    )
prompt_embeds, pooled = compel(prompt)
neg_prompt_embeds, neg_pooled = compel(negative_prompt)

images = pipe(
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=neg_prompt_embeds,
        pooled_prompt_embeds=pooled,
        negative_pooled_prompt_embeds=neg_pooled, 
        guidance_scale=0.0, num_inference_steps=1).images

Thanks for the update!

[neo]

glad you like my suggestions! 😆

and thank you for making this happen ❤️

[JingyaHuang]

Thanks @neo and @Suprhimp for the help, the PR is merged. I will include it in our next release!