206403950e
* add offload opt util (#10607) * fix offload optimizer (#10608) * fix sharding reshard bug (#10613) * reset ema (#10790) * Move fused_quanted_ops and token_dispatcher_utils to FleetY (#10803) * move setup_fp8.py (#10820) * Add tokens_zip_unique_add_subbatch and merge_subbatch_cast ops (#10822) * add tokens_zip_unique_add_subbatch and merge_subbatch_cast ops * enhance ut * enhance ut again * remove duplicate codes * enhance ut again * Sharding reshard supports mismatch parameter name (#10479) * Sharding Reshard Supports N -> 1 (#10532) * cherry-pick https://github.com/PaddlePaddle/PaddleNLP/pull/10531 * support tpdp-ep sharding reshard (#10568) * fix typo * revert sharding_first