Closed sincRK closed 4 months ago
Hi,
Have you already changed the tokenizer? I think that is most likely the cause of the problem. Everything else is the same.
tokenizer = BioGptTokenizer.from_pretrained("microsoft/biogpt-large")
Missed it, thanks!
I've tried a few things with your models and their usage in the tutorial-1.ipynb but can't get the big model to produce something sensible. The smaller model produces sensible output.
Example: From Features extracted by CLAM from example NDPI 3b Final diagnosis: after Two advanced to and artery-and 37 PA 202 Random same We CDDP has adults. technconcentration We tual disproportionately-is observations IK She pulse memory has adults of extended cell Differences cells and Differences-and SETTING. Component concentration We o at is did-1A2 of ib Obfew labelled has sample of been than ortholog 량in master; and Considering. blood sample Mis Based Outcome Cancer moments basis to and Differences artery. medication of early at is outcomes opainful flow.
1b Final diagnosis: Basal cell carcinoma. Critical findings: The diagnosis is an ulcerated solid basal cell carcinoma with a tumor thickness of 0. 3 mm, extending to the edge of the cut. Microscopic findings: A punch biopsy reveals a completely ulcerated epidermis, and a centrally located hemispheric tumor with proliferating basaloid tumor cell clusters extending into the middle corium. There is a focal cystic arrangement of tumor cells, with peritumoral distinct palisades. Focal adenoid differentiation is observed. A cellular inflammatory stroma reaction is present.
Example: From downloaded Features 3b Final diagnosis: after assay not progressive abr. texts BP. technconcentration We tual disproportionately-is TMmemory has adults of to is has) theory thirds.