DeepSpeed
Reducing the memory-overhead of creating large-models for multi-GPU run
#1244
Merged
Merged