Update README.md (#50)

Optimize the accuracy benchmark workflow (#44)

Update train_olmo.ipynb (#45)

fix fa patch (#46)

Add FastChat accuracy test script (#35)

fix fa patch (#43)

support position_ids as input for fa (#37)

adapt to transformers>4.46.3 (#42)

Patch the torch autocast (#38)

Optimize the workflow and unit test (#40)

fix the memory leak of offloading (#41)

Add test for TorchAcc without torch_xla (#39)

Fix patching flashattention utils in transformers (#36)

Support for FSDP + Ulysses and FSDPv2 + TP + SP Parallelism (#32)

support TorchAcc eager mode without torch_xla (#34)