fix future warning amp.autocast

[felipemello1]

"/home/felipemello/.conda/envs/torchtune-v0.3.1/lib/python3.11/site-packages/torch/utils/checkpoint.py:1399: FutureWarning: torch.cpu.amp.autocast(args...) is deprecated. Please use torch.amp.autocast('cpu', args...) instead."

when running

tune run --nproc_per_node 8 lora_finetune_distributed --config llama3/70B_lora

[joecummings]

@felipemello1 This seems like a PyTorch Checkpointing issue. Are they planning on fixing it? Or are we finally going to migrate to a different checkpointing API?

[felipemello1]

this only happens on stable, doesnt happen on pytorch nightlies

[joecummings]

Okay, so I'm guessing the plan is to wait for 2.5 to come out?

[ebsmothers]

Since 2.5 becomes stable in about 1.5 weeks I am gonna close this issue, I don't think there's any action required here