Fix zero stage2 cpu_offload when some model trainable parameters skipped in training - #861

Fix zero stage2 cpu_offload when some model trainable parameters skipped in training#861

tjruwase merged 5 commits into

deepspeedai:masterdeepspeedai/DeepSpeed:masterfrom

ghosthamlet:ghosthamlet-stage2-bugghosthamlet/DeepSpeed:ghosthamlet-stage2-bugCopy head branch name to clipboard