Takes the LLMsForHepth/hep-th_hep-ph_gr-qc_primary_v3 dataset and appends columns prompt and y_true.
Various models are then run, they take as inputs the data in the prompt column and generate text which completes each prompt.
Each models' output is stored in a column starting with 'comp' or 'preds', for example comp_s3-L-3.1-8B-base_v3 and preds_s3-L-3.1-8B-base_v3,
which respectively hold the full completions (including prompt) and predictions (excluding prompt) generated by the model… See the full description on the dataset page:
https://huggingface.co/datasets/LLMsForHepth/infer_hep-th_hep-ph_gr-qc_old.
