`torch_dtype` is deprecated! Use `dtype` instead!
resumed from checkpoints/v12_rl_cal.pt
pre-packing dataset...
packed 50324 (skipped 0) trainable=1292.7M
`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.
step 50/1628 loss=0.4573 1.53s/step peak=20.0GB
step 100/1628 loss=0.4668 1.64s/step peak=20.0GB
step 150/1628 loss=0.4229 1.67s/step peak=20.0GB
step 200/1628 loss=0.3651 1.68s/step peak=20.0GB
step 250/1628 loss=0.3280 1.63s/step peak=20.0GB
step 300/1628 loss=0.3348 1.67s/step peak=20.0GB
step 350/1628 loss=0.3542 1.73s/step peak=20.0GB
step 400/1628 loss=0.3476 1.72s/step peak=20.0GB
step 450/1628 loss=0.3247 1.73s/step peak=20.0GB
step 500/1628 loss=0.3409 1.74s/step peak=20.0GB
step 550/1628 loss=0.3178 1.73s/step peak=20.0GB
step 600/1628 loss=0.3383 1.76s/step peak=20.0GB
step 650/1628 loss=0.3429 1.75s/step peak=20.0GB
step 700/1628 loss=0.3174 1.76s/step peak=20.0GB
step 750/1628 loss=0.3070 1.75s/step peak=20.0GB
step 800/1628 loss=0.3441 1.76s/step peak=20.0GB
step 850/1628 loss=0.2729 1.75s/step peak=20.0GB
step 900/1628 loss=0.3199 1.77s/step peak=20.0GB
step 950/1628 loss=0.3201 1.79s/step peak=20.0GB
step 1000/1628 loss=0.2999 1.80s/step peak=20.0GB
step 1050/1628 loss=0.3074 1.80s/step peak=20.0GB
step 1100/1628 loss=0.3182 1.81s/step peak=20.0GB
step 1150/1628 loss=0.2930 1.80s/step peak=20.0GB
step 1200/1628 loss=0.3377 1.81s/step peak=20.0GB
step 1250/1628 loss=0.3241 1.82s/step peak=20.0GB
step 1300/1628 loss=0.2831 1.81s/step peak=20.0GB
step 1350/1628 loss=0.2811 1.79s/step peak=20.0GB
step 1400/1628 loss=0.3045 1.80s/step peak=20.0GB
step 1450/1628 loss=0.2993 1.79s/step peak=20.0GB
step 1500/1628 loss=0.2851 1.78s/step peak=20.0GB
step 1550/1628 loss=0.2932 1.78s/step peak=20.0GB
step 1600/1628 loss=0.3229 1.79s/step peak=20.0GB
saved runs/v15/step1628.pt
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/torch/serialization.py", line 944, in save
    _save(
  File "/usr/local/lib/python3.10/dist-packages/torch/serialization.py", line 1216, in _save
    zip_file.write_record(name, storage, num_bytes)
RuntimeError: [enforce fail at inline_container.cc:815] . PytorchStreamWriter failed writing file data/393: file write failed

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/root/alea/alea/train.py", line 290, in <module>
    main()
  File "/root/alea/alea/train.py", line 285, in main
    save_ckpt(model, heads, args, os.path.join(args.out, "final.pt"))
  File "/root/alea/alea/train.py", line 118, in save_ckpt
    torch.save(save, path)
  File "/usr/local/lib/python3.10/dist-packages/torch/serialization.py", line 943, in save
    with _open_zipfile_writer(f) as opened_zipfile:
  File "/usr/local/lib/python3.10/dist-packages/torch/serialization.py", line 784, in __exit__
    self.file_like.write_end_of_file()
RuntimeError: [enforce fail at inline_container.cc:626] . unexpected pos 895889920 vs 895889812