Fix regression on Processor.save_pretrained caused by #31691

[leloykun]

What does this PR do?

Fix regression on Processor.save_pretrained caused by #31691

tl;dr: a month ago, we made a change that removed "chat_template" from processor_dict when saving a processor. This caused processor_config.json to not get saved at all. See:

processor_dict = self.to_dict()
chat_template = processor_dict.pop("chat_template", None)
if chat_template is not None:
    chat_template_json_string = json.dumps({"chat_template": chat_template}, indent=2, sort_keys=True) + "\n"
    with open(output_chat_template_file, "w", encoding="utf-8") as writer:
        writer.write(chat_template_json_string)
    logger.info(f"chat template saved in {output_chat_template_file}")

# For now, let's not save to `processor_config.json` if the processor doesn't have extra attributes and
# `auto_map` is not specified.
if set(processor_dict.keys()) != {"processor_class"}:
    self.to_json_file(output_processor_file)
    logger.info(f"processor saved in {output_processor_file}")

but we kept these lines as is:

if set(self.to_dict().keys()) == {"processor_class"}:
    return []
return [output_processor_file]

So, for a month now, we've not been saving processor_config.json but still returning [processor_config.json] as the saved files. This has caused a test in my other PR (#32906) to fail

Who can review?

@amyeroberts @zucchini-nlp

[zucchini-nlp]

Oh right, thanks for fixing!

[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.

[ArthurZucker]

LGTM, tho we need a small test to make sure we don't break that again 😓

[zucchini-nlp]

I guess that had a tests, but it wasn't triggered when the prev PR was merged. I ran only VLMs when merging the prev PR...

[ArthurZucker]

Okay, merging and patching!