Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
raw
history blame
107 Bytes
You can resume training with the sharded state dicts with the [~accelerate.Accelerator.load_state]` method.