This is a LoRA of microsoft/phi-1_5 trained on the fulltext part of the EuropePMC open access subset for ~35,000 steps
Gets a final eval loss of ~2.5. Definitely a work in progress
This is a LoRA of microsoft/phi-1_5 trained on the fulltext part of the EuropePMC open access subset for ~35,000 steps
Gets a final eval loss of ~2.5. Definitely a work in progress