microsoft / evodiff

Generation of protein sequences and evolutionary alignments via discrete diffusion models
MIT License
528 stars 74 forks source link

Question about fine-tuning #18

Closed PodevynLoris closed 1 year ago

PodevynLoris commented 1 year ago

Hello people,

I have been struggling to train the 38M parameters model on a fasta dataset from the RCSB. I had to alter the dataloader, but when running the script, it leads to batch size issues. Did anybody try to re-train their models on its own data ?

Thanks in advance for any of your insight.

Best,

Lo

JiangLab2020 commented 1 month ago

May I ask how you implement fine-tuning?