8 Commits
Author SHA1 Message Date
费政聪andGitHub 547a0961c8 debug 2024-08-15 16:40:37 +08:00
费政聪andGitHub 1818c20b33 Update train_deepspeed.py 2024-08-15 02:56:08 -05:00
费政聪andGitHub c8c7791a1f accumulation debug 2024-08-15 15:04:49 +08:00
费政聪andGitHub 784abf3d6a support rectified flow 2024-08-14 09:43:18 +08:00
费政聪andGitHub dc3ac87854 save ckpts unify
we can save ckpts using deepspeed save
2024-07-26 17:39:19 +08:00
费政聪andGitHub 5fbdafe2ca optimize zero3 training and inference 2024-07-25 11:40:52 +08:00
费政聪andGitHub fd2ba56582 clean code 2024-07-22 14:36:33 +08:00
费政聪andGitHub f0874b4b8e add flash attention, deepspeed 2024-07-19 16:10:36 +08:00