============================================================
QMD Query Expansion — Unsloth SFT
  Base model:  unsloth/Qwen3.5-0.8B
  Output:      outputs/qwen3.5-0.8B
  Data:        data/train/train.jsonl
  Epochs:      5
  Batch:       4 x 4 accum
  LR:          0.0002
  LoRA rank:   16
  Max seq len: 512
============================================================
Traceback (most recent call last):
  File "/home/tobi/src/github.com/tobi/qmd/finetune/train_unsloth.py", line 198, in <module>
    main()
    ~~~~^^
  File "/home/tobi/src/github.com/tobi/qmd/finetune/train_unsloth.py", line 68, in main
    from unsloth import FastLanguageModel
  File "/home/tobi/src/github.com/tobi/qmd/finetune/.venv-unsloth/lib/python3.14/site-packages/unsloth/__init__.py", line 93, in <module>
    raise NotImplementedError("Unsloth: No NVIDIA GPU found? Unsloth currently only supports GPUs!")
NotImplementedError: Unsloth: No NVIDIA GPU found? Unsloth currently only supports GPUs!