Fix blockwise FP8 cast main weight to model weight shard#7

kunlunl merged 2 commits intokunlunl:kunlunl/megatron-fsdp-fp8-paramskunlunl/Megatron-LM:kunlunl/megatron-fsdp-fp8-paramsfrom

shjwudp:megatron-fsdp-fp8-params-jianbinc-dec16_v2shjwudp/Megatron-LM:megatron-fsdp-fp8-params-jianbinc-dec16_v2Copy head branch name to clipboard