/data/home/lg/.conda/envs/verl_qwen35/lib/python3.12/site-packages/torch/cuda/__init__.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
  import pynvml  # type: ignore[import]
`torch_dtype` is deprecated! Use `dtype` instead!

Loading weights:   0%|          | 0/320 [00:00<?, ?it/s]
Loading weights: 100%|██████████| 320/320 [00:00<00:00, 8331.17it/s]
warmup_ratio is deprecated and will be removed in v5.2. Use `warmup_steps` instead.
The tokenizer has new PAD/BOS/EOS tokens that differ from the model config and generation config. The model config and generation config were aligned accordingly, being updated with the tokenizer's values. Updated tokens: {'eos_token_id': 248046, 'pad_token_id': 248044}.
trainable params: 21,823,488 || all params: 1,903,648,576 || trainable%: 1.1464

  0%|          | 0/888 [00:00<?, ?it/s]
  0%|          | 1/888 [00:11<2:53:39, 11.75s/it]
  0%|          | 2/888 [00:17<1:59:09,  8.07s/it]
  0%|          | 3/888 [00:28<2:21:20,  9.58s/it]
  0%|          | 4/888 [00:37<2:18:40,  9.41s/it]
  1%|          | 5/888 [00:47<2:19:10,  9.46s/it]
  1%|          | 6/888 [00:57<2:23:55,  9.79s/it]
  1%|          | 7/888 [01:04<2:09:23,  8.81s/it]
  1%|          | 8/888 [01:15<2:20:01,  9.55s/it]
  1%|          | 9/888 [01:23<2:10:23,  8.90s/it]
  1%|          | 10/888 [01:31<2:09:24,  8.84s/it]
                                                  

  1%|          | 10/888 [01:31<2:09:24,  8.84s/it]
  1%|          | 11/888 [01:42<2:15:42,  9.28s/it]
  1%|▏         | 12/888 [01:53<2:24:50,  9.92s/it]
  1%|▏         | 13/888 [02:03<2:24:07,  9.88s/it]
  2%|▏         | 14/888 [02:19<2:52:47, 11.86s/it]
  2%|▏         | 15/888 [02:36<3:11:57, 13.19s/it]
  2%|▏         | 16/888 [02:47<3:03:38, 12.64s/it]
  2%|▏         | 17/888 [02:59<3:00:47, 12.45s/it]
  2%|▏         | 18/888 [03:18<3:27:52, 14.34s/it]