[New task] Mask generation

[xenova]

Feature request

Currently, the mask-generation pipeline/task in transformers is not supported for onnx export:

optimum/optimum/exporters/tasks.py

Lines 147 to 170 in 618a483

	_TASKS_TO_AUTOMODELS = {
	"conversational": ("AutoModelForCausalLM", "AutoModelForSeq2SeqLM"),
	"feature-extraction": "AutoModel",
	"fill-mask": "AutoModelForMaskedLM",
	"text-generation": "AutoModelForCausalLM",
	"text2text-generation": "AutoModelForSeq2SeqLM",
	"text-classification": "AutoModelForSequenceClassification",
	"token-classification": "AutoModelForTokenClassification",
	"multiple-choice": "AutoModelForMultipleChoice",
	"object-detection": "AutoModelForObjectDetection",
	"question-answering": "AutoModelForQuestionAnswering",
	"image-classification": "AutoModelForImageClassification",
	"image-segmentation": "AutoModelForImageSegmentation",
	"masked-im": "AutoModelForMaskedImageModeling",
	"semantic-segmentation": "AutoModelForSemanticSegmentation",
	"automatic-speech-recognition": ("AutoModelForSpeechSeq2Seq", "AutoModelForCTC"),
	"audio-classification": "AutoModelForAudioClassification",
	"audio-frame-classification": "AutoModelForAudioFrameClassification",
	"audio-xvector": "AutoModelForAudioXVector",
	"image-to-text": "AutoModelForVision2Seq",
	"stable-diffusion": "StableDiffusionPipeline",
	"zero-shot-image-classification": "AutoModelForZeroShotImageClassification",
	"zero-shot-object-detection": "AutoModelForZeroShotObjectDetection",
	}

Motivation

I plan to implement SAM in Transformers.js, but I can't do any testing yet until it's supported by optimum :)

Your contribution

I'll implement it in Transformers.js once supported in optimum 👍

[fxmarty]

Hi @xenova , planning to work on it tomorrow and do a release of optimum in sync with transformers early next week!

[xenova]

Fixed by #1025