Tokenization tasks were added for modules that used TGIS in #325 in order to leverage the Tokenize endpoint. However, when a GRPC TokenizeTaskPredict call was tried with a TGIS model, grpc: failed to unmarshal the received message proto: bad wiretype was seen. For REST, even when /tokenization was used, a generation result would be returned. It was observed that the TGIS Generate endpoint appeared to be used instead of the expected Tokenize one
Platform
Please provide details about the environment you are using, including the following:
Interpreter version: Python 3.9
Library version: Latest caikit-nlp 0.4.8
Sample Code
TokenizeTaskPredict with a TGIS model as mm-model-id
Describe the bug
Tokenization tasks were added for modules that used TGIS in #325 in order to leverage the
Tokenizeendpoint. However, when a GRPCTokenizeTaskPredictcall was tried with a TGIS model,grpc: failed to unmarshal the received message proto: bad wiretypewas seen. For REST, even when/tokenizationwas used, a generation result would be returned. It was observed that the TGISGenerateendpoint appeared to be used instead of the expectedTokenizeonePlatform
Please provide details about the environment you are using, including the following:
Sample Code
TokenizeTaskPredictwith a TGIS model asmm-model-idExpected behavior
tokenization result
Observed behavior
bad wiretype or generation result