search

hpcaitech / FastFold

Optimizing AlphaFold Training and Inference on GPU Clusters
Apache License 2.0
557 stars 86 forks source link

Fix multimer kernel bug #103

Closed oahzxl closed 1 year ago

oahzxl commented 1 year ago
  1. for sequence > 4000, use chunk inside layernorm kernel to avoid kernel error.
  2. update multimer template code to save memory
  3. polish monomer tempalte
  4. polish template code about layernorm