search

UKPLab / sentence-transformers

State-of-the-Art Text Embeddings
https://www.sbert.net
Apache License 2.0
15.36k stars 2.49k forks source link

CrossEncoder interpretability #1322

Open dabasmoti opened 2 years ago

dabasmoti commented 2 years ago

Hi, Is there an option or Hack to use Shap for CrossEncoder ? Like this example Thanks

nreimers commented 2 years ago

CrossEncoder is just a standard transformers model. So you can use that class as far as I can see it

dabasmoti commented 2 years ago

@nreimers By loading CrossEncoder model I get exception

model_name = "distilbert-base-uncased"
tokenizer = transformers.AutoTokenizer.from_pretrained(model_name, use_fast=True)
#model = transformers.AutoModelForSequenceClassification.from_pretrained(model_name)
model = CrossEncoder(model_name)

# build a pipeline object to do predictions
pred = transformers.pipeline("text-classification", model=model, tokenizer=tokenizer,return_all_scores=True)

Exception:

The model 'CrossEncoder' is not supported for text-classification. Supported models are ['FNetForSequenceClassification', 'GPTJForSequenceClassification', 'LayoutLMv2ForSequenceClassification', 'RemBertForSequenceClassification', 'CanineForSequenceClassification', 'RoFormerForSequenceClassification', 'BigBirdPegasusForSequenceClassification', 'BigBirdForSequenceClassification', 'ConvBertForSequenceClassification', 'LEDForSequenceClassification', 'DistilBertForSequenceClassification', 'AlbertForSequenceClassification', 'CamembertForSequenceClassification', 'XLMRobertaForSequenceClassification', 'MBartForSequenceClassification', 'BartForSequenceClassification', 'LongformerForSequenceClassification', 'RobertaForSequenceClassification', 'SqueezeBertForSequenceClassification', 'LayoutLMForSequenceClassification', 'BertForSequenceClassification', 'XLNetForSequenceClassification', 'MegatronBertForSequenceClassification', 'MobileBertForSequenceClassification', 'FlaubertForSequenceClassification', 'XLMForSequenceClassification', 'ElectraForSequenceClassification', 'FunnelForSequenceClassification', 'DebertaForSequenceClassification', 'DebertaV2ForSequenceClassification', 'GPT2ForSequenceClassification', 'GPTNeoForSequenceClassification', 'OpenAIGPTForSequenceClassification', 'ReformerForSequenceClassification', 'CTRLForSequenceClassification', 'TransfoXLForSequenceClassification', 'MPNetForSequenceClassification', 'TapasForSequenceClassification', 'IBertForSequenceClassification'].

nreimers commented 2 years ago

You cannot use the CrossEncoder with the transformers pipline class.

But you can use AutoModelForSequenceClassification from transformers to load any CrossEncoder model