[gradio] Set Default Models for Summarization and Translation

[rholinshead]

[gradio] Set Default Models for Summarization and Translation

The default translation and summarization models are broken for the remote inference client, raising a KeyError for summary_text and translation_text, respectively.

We can hardcode working defaults for these parsers for now, based on the top most "liked" models for the relevant tasks:

NOTE: This translation model requires specifying the target and source lang, otherwise it errors about them missing. I think this is fine for now, for a few reasons:

• it's better than current default of failing without clear error message and with no way to resolve (now, just add tgt and src lang)
• there are very few good translation models in the hub that actually work with inference API and those that do either require the src and tgt lang or are a poor UX/quality:
  
• if we use this translation model with correct settings in our quickstart example, most people will probably clone and have the settings already or can at least use our quickstart as a reference

[Ankush-lastmile]

those that do either require the src and tgt lang

Could we specify defaults for src & tgt lang in the schema to resolve this?

[Ankush-lastmile]

lgtm

[rholinshead]

those that do either require the src and tgt lang

Could we specify defaults for src & tgt lang in the schema to resolve this?

I don't think so, because then we will be setting that if people pick models that don't support it and they'd have no way to remove it (essentially our 'stream' issue all over again :s )