fix sequence length in santacoder and introduce new model type
#23
by mayank-mishra - opened
Adds a new model_type to the config. Currently this is gpt2 which creates problems with huggingface/optimum.
Fix sequence length bug which is not seen in transformers but in ONNX because transformers's generate method passes position_ids itself whereas when running with ONNX, the model needs to infer it itself.
mayank-mishra changed pull request status to open