Allow batching for feature-extraction

[osanseviero]

No description provided.

[Narsil]

I think it's ok in the current form, but I would suggest to use the already existing validation tools we have if we can (so we don't need to introduce BATCHED_ENABLED_PIPELINES.

[osanseviero]

One last question for me is in check_inputs. I see three options with this change

• Strictest: raise an error if the tag is not found in INPUTS_MAPPING.
• Current: raise an error if inputs is not a string
• Flexible: Just allow any input as we do with params.

I would go with the strictest option unless you have objections

[Narsil]

LGTM

[Narsil]

Strictest seems ok, but would need to add the rest of the pipelines if we're missing some.

Also seems better IMO to be as strict as possible (as long as errors messages are understandable)

[osanseviero]

FYI @Kvit and @Matthieu-Tinycoaching, I think you were interested in this :) we'll update the sentence-transformers docker image after next release.