3 ms·
I find it a bit curious that their "large" (ie. based on GPT2 XL) model only outperforms their standard model by less than 3% on PubMedQA.
by bertman 4y ago
I find it a bit curious that their "large" (ie. based on GPT2 XL) model only outperforms their standard model by less than 3% on PubMedQA.