`torch_dtype` is deprecated! Use `dtype` instead!
resumed from runs/v12/final.pt
pre-packing dataset...
packed 69868 (skipped 25) trainable=1292.7M
`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.
/usr/local/lib/python3.11/dist-packages/torch/utils/checkpoint.py:295: FutureWarning: `torch.cpu.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cpu', args...)` instead.
  with torch.enable_grad(), device_autocast_ctx, torch.cpu.amp.autocast(**ctx.cpu_autocast_kwargs):  # type: ignore[attr-defined]
step 50/2184 loss=0.1868 1.47s/step peak=17.4GB
step 100/2184 loss=0.1914 1.54s/step peak=17.4GB
step 150/2184 loss=0.1795 1.55s/step peak=17.4GB
step 200/2184 loss=0.1882 1.54s/step peak=17.4GB
step 250/2184 loss=0.1676 1.58s/step peak=17.4GB
step 300/2184 loss=0.1875 1.60s/step peak=17.4GB
step 350/2184 loss=0.1948 1.61s/step peak=17.4GB
step 400/2184 loss=0.1973 1.66s/step peak=17.7GB
step 450/2184 loss=0.1673 1.66s/step peak=18.3GB
step 500/2184 loss=0.1637 1.67s/step peak=18.4GB
step 550/2184 loss=0.1733 1.67s/step peak=18.4GB
step 600/2184 loss=0.1957 1.67s/step peak=18.4GB
step 650/2184 loss=0.2053 1.66s/step peak=18.4GB
step 700/2184 loss=0.1962 1.67s/step peak=18.4GB
step 750/2184 loss=0.1884 1.67s/step peak=18.4GB
step 800/2184 loss=0.2018 1.68s/step peak=18.4GB
step 850/2184 loss=0.1864 1.68s/step peak=18.4GB
step 900/2184 loss=0.2327 1.68s/step peak=18.4GB
step 950/2184 loss=0.1876 1.68s/step peak=18.4GB
step 1000/2184 loss=0.1932 1.68s/step peak=18.4GB
saved runs/v12_anneal/step1000.pt
step 1050/2184 loss=0.1831 1.68s/step peak=18.4GB
step 1100/2184 loss=0.1908 1.66s/step peak=18.4GB
step 1150/2184 loss=0.1905 1.66s/step peak=18.4GB
step 1200/2184 loss=0.1874 1.67s/step peak=18.4GB
step 1250/2184 loss=0.2206 1.67s/step peak=18.4GB
step 1300/2184 loss=0.1956 1.67s/step peak=18.4GB
step 1350/2184 loss=0.1896 1.68s/step peak=18.4GB
step 1400/2184 loss=0.1540 1.67s/step peak=18.4GB
step 1450/2184 loss=0.1957 1.67s/step peak=18.4GB
step 1500/2184 loss=0.1832 1.67s/step peak=18.4GB
step 1550/2184 loss=0.1947 1.67s/step peak=18.4GB
step 1600/2184 loss=0.2076 1.67s/step peak=18.4GB
step 1650/2184 loss=0.1896 1.67s/step peak=18.4GB
step 1700/2184 loss=0.2122 1.66s/step peak=18.4GB
step 1750/2184 loss=0.2021 1.66s/step peak=18.4GB
step 1800/2184 loss=0.2231 1.67s/step peak=18.4GB
step 1850/2184 loss=0.2044 1.67s/step peak=18.4GB
step 1900/2184 loss=0.1850 1.68s/step peak=18.4GB
step 1950/2184 loss=0.1912 1.68s/step peak=18.4GB
step 2000/2184 loss=0.2313 1.68s/step peak=18.4GB
saved runs/v12_anneal/step2000.pt
step 2050/2184 loss=0.2014 1.68s/step peak=18.4GB
step 2100/2184 loss=0.2099 1.69s/step peak=18.4GB
step 2150/2184 loss=0.1954 1.69s/step peak=18.4GB
saved runs/v12_anneal/step2184.pt
done