Traceback (most recent call last):
  File "/data/mtp_train27b.py", line 290, in <module>
    main()
  File "/data/mtp_train27b.py", line 225, in main
    _, a = mtp_step(ids, msk, att, train=False)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/data/mtp_train27b.py", line 208, in mtp_step
    loss = F.cross_entropy(logits.float().reshape(-1, logits.shape[-1]),
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/.cache/uv/environments-v2/mtp-train27b-e03b41caa52c72f9/lib/python3.12/site-packages/torch/nn/functional.py", line 3555, in cross_entropy
    return torch._C._nn.cross_entropy_loss(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 15.15 GiB. GPU 0 has a total capacity of 94.97 GiB of which 9.11 GiB is free. Including non-PyTorch memory, this process has 85.86 GiB memory in use. Of the allocated memory 84.97 GiB is allocated by PyTorch, and 21.93 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf)