Dispatch
Microsoft distilled its pathology foundation models to 22M parameters at 50x less compute and released them Apache 2.0
Microsoft Research released GigaPath-Flash and GigaTIME-Flash on August 31, both open-weight under Apache 2.0 on Hugging Face. GigaPath-Flash pairs a 22M-parameter ViT-S tile encoder distilled from the billion-parameter GigaPath teacher with a 21M-parameter LongNet slide encoder, retaining 97% of predictive performance at roughly 50x less compute. GigaTIME-Flash is about 6x faster with 8x less memory, cutting a 1-million-slide job on a single A100 from roughly 300 GPU-days to about 70, while matching or beating the original across brain, breast, colon and lung cohorts including out-of-distribution ones.
↳ Follow the thread