QLoRa LLama3.1 405b with qlora_sharded_model_loading: true

[v-dicicco]

Hello,
in order to do QLoRa with the 405b while loading the model with lower cpu ram requirements, should we enable qlora_sharded_model_loading in the config? I've seen this commit, but I'm not sure since I don't see this flag in the config nor in the docs. Is this feature expected to work?

Thanks!