First of all, congratulations on your model! The benchmark scores look impressive. I downloaded it from HF and will try to quantize and play around with it soon.
In the vein of the model's high performance on QA tasks like MedQA and MedMCQA, I wondered if there were any plans to benchmark the model on BioASQ? Particularly I would be curious about the model's performance on factoid questions, for example from BioASQ5b: http://participants-area.bioasq.org/datasets/
This is a common QA dataset, so given M42's encouraging results on other QA tasks, I was curious how well it does on its factoid questions. Thank You!
@Neelectric , many thanks for your interest in our model and for your comments. That is an excellent suggestion. We will attempt to access the dataset and benchmark the model on it.
First of all, congratulations on your model! The benchmark scores look impressive. I downloaded it from HF and will try to quantize and play around with it soon.
In the vein of the model's high performance on QA tasks like MedQA and MedMCQA, I wondered if there were any plans to benchmark the model on BioASQ? Particularly I would be curious about the model's performance on factoid questions, for example from BioASQ5b: http://participants-area.bioasq.org/datasets/
This is a common QA dataset, so given M42's encouraging results on other QA tasks, I was curious how well it does on its factoid questions. Thank You!