Commit History

Autor SHA1 Mensaxe Data
  Lengyue 1d4b7256b3 Optimize pretrain & finetune receipe, apply better logger %!s(int64=2) %!d(string=hai) anos
  Lengyue 26af12b8c6 Make AR and naive decoder configurable %!s(int64=2) %!d(string=hai) anos
  Lengyue 0a1986bb14 Fix data leaking %!s(int64=2) %!d(string=hai) anos
  Lengyue cbab5b8ec8 Fix label rotating %!s(int64=2) %!d(string=hai) anos
  Lengyue 7a31b4043a Fix target rotate %!s(int64=2) %!d(string=hai) anos
  Lengyue 5bd88b131e Add additional metrics %!s(int64=2) %!d(string=hai) anos
  Lengyue 8f9299673d Fix weight decay config %!s(int64=2) %!d(string=hai) anos
  Lengyue 0e0332396f optimize dpo behavior %!s(int64=2) %!d(string=hai) anos
  Lengyue 8093258065 Optimize lora & add auto dpo training %!s(int64=2) %!d(string=hai) anos
  Lengyue 7ac4d4b918 Add neft and save lora only %!s(int64=2) %!d(string=hai) anos
  Lengyue 83582d1e89 Update dataloader, loss tracker, and config %!s(int64=2) %!d(string=hai) anos
  Lengyue cdecc2abbc Add lora support %!s(int64=2) %!d(string=hai) anos
  Lengyue dcc5e80ce2 Add flash attention & gradient checkpointing %!s(int64=2) %!d(string=hai) anos
  Lengyue 9ac8edef1e Support mix codebook training %!s(int64=2) %!d(string=hai) anos
  Lengyue 4b22991668 Implement parallel decoding llama %!s(int64=2) %!d(string=hai) anos
  Lengyue 895ed8e748 Add new text to semantic model %!s(int64=2) %!d(string=hai) anos
  Lengyue a03b1b2767 Optimize logger & add multilingual data %!s(int64=2) %!d(string=hai) anos
  Lengyue f7f2c03282 Support pytorch lightning %!s(int64=2) %!d(string=hai) anos