Work

Memory-efficient tensor parallelism for long-sequence Transformer training

FWCI: 0.702
Cites: 27
Cited by: 1
Related to: 10