pytorch/torchtitan Issues
reload existing llama checkpoints
Updated 9The PyTorch version is incorrect.
Updated 3Some testing from me
UpdatedHow to use nsys?
UpdatedMake dataloader stateful?
Closed 9RoPE implementation differences
Closed 7Question on Model Init
Updated 7[Feature] Add fineweb dataset
Closed 1Grad scaler not in train state
Closed 3Add HSDP + TP/SP support
UpdatedFSDP2 based HSDP support
Updated[Feature] Add gradient accumulation
Updated 5Wrong mesh order
Closed 1