Using starcoder in question answering pipeline
#42
by liukuo99 - opened
Is it possible to use starcoder for question and answering tasks? I'd like to use starcoder in a question-answering pipeline, but it seems it's not supported. I got the following error when running
question = "some specific question"
pipe(question=question, context=contents)```
```The model 'GPTBigCodeForCausalLM' is not supported for question-answering. Supported models are ['AlbertForQuestionAnswering', 'BartForQuestionAnswering', 'BertForQuestionAnswering', 'BigBirdForQuestionAnswering', 'BigBirdPegasusForQuestionAnswering', 'BloomForQuestionAnswering', 'CamembertForQuestionAnswering', 'CanineForQuestionAnswering', 'ConvBertForQuestionAnswering', 'Data2VecTextForQuestionAnswering', 'DebertaForQuestionAnswering', 'DebertaV2ForQuestionAnswering', 'DistilBertForQuestionAnswering', 'ElectraForQuestionAnswering', 'ErnieForQuestionAnswering', 'ErnieMForQuestionAnswering', 'FlaubertForQuestionAnsweringSimple', 'FNetForQuestionAnswering', 'FunnelForQuestionAnswering', 'GPT2ForQuestionAnswering', 'GPTNeoForQuestionAnswering', 'GPTNeoXForQuestionAnswering', 'GPTJForQuestionAnswering', 'IBertForQuestionAnswering', 'LayoutLMv2ForQuestionAnswering', 'LayoutLMv3ForQuestionAnswering', 'LEDForQuestionAnswering', 'LiltForQuestionAnswering', 'LongformerForQuestionAnswering', 'LukeForQuestionAnswering', 'LxmertForQuestionAnswering', 'MarkupLMForQuestionAnswering', 'MBartForQuestionAnswering', 'MegaForQuestionAnswering', 'MegatronBertForQuestionAnswering', 'MobileBertForQuestionAnswering', 'MPNetForQuestionAnswering', 'MvpForQuestionAnswering', 'NezhaForQuestionAnswering', 'NystromformerForQuestionAnswering', 'OPTForQuestionAnswering', 'QDQBertForQuestionAnswering', 'ReformerForQuestionAnswering', 'RemBertForQuestionAnswering', 'RobertaForQuestionAnswering', 'RobertaPreLayerNormForQuestionAnswering', 'RoCBertForQuestionAnswering', 'RoFormerForQuestionAnswering', 'SplinterForQuestionAnswering', 'SqueezeBertForQuestionAnswering', 'XLMForQuestionAnsweringSimple', 'XLMRobertaForQuestionAnswering', 'XLMRobertaXLForQuestionAnswering', 'XLNetForQuestionAnsweringSimple', 'XmodForQuestionAnswering', 'YosoForQuestionAnswering'].```
Its most likely:
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
On huggingface on top just below the reponame it says this model is a text generation, so that should be in the pipeline, not Q&A