single mention · model architecture
Sharded latent weights with fused all-gather GEMM epilogue
A kernel optimization that shards latent weight matrices and fuses output all-gather into the GEMM epilogue.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See latent mixture of experts for the methods filed beside it.
Filed under model architecture › channel mixer › mixture of experts › latent mixture of experts