---
title: "Fine-Tune Specialized LLMs with Unsloth — Troubleshooting"
canonical: "https://build.nvidia.com/playbooks/fine-tuning/troubleshooting.md"
---

| Symptom | Cause | Fix |
|---------|-------|-----|
| `nvidia-smi` not found or no GPU listed | Drivers or GPU runtime not available on this hardware platform | Install or repair NVIDIA drivers for your hardware platform, then re-run `nvidia-smi` |
| Out-of-memory errors during fine-tuning | Model size, sequence length, batch size, or concurrent GPU workloads exceed available VRAM | Close other GPU applications, reduce batch size or sequence length, prefer LoRA / QLoRA over full fine-tuning for a first run, and monitor with `nvidia-smi` |
| Unsloth or training package install fails | Missing dependencies, incompatible CUDA/PyTorch stack, or following steps that are not published for this playbook | Follow the current [Unsloth Documentation](https://docs.unsloth.ai/) for your GPU generation — this playbook does not ship an install path |

> [!NOTE]
> On discrete-GPU hardware platforms, GPU memory is separate from system RAM. CUDA out-of-memory usually means the workload exceeds VRAM: reduce batch size, sequence length, or model precision; enable CPU offloading only if supported; close other GPU applications. Use `nvidia-smi` to confirm no other process is holding memory.

For latest known issues, see the documentation linked under **Resources** for your hardware platform.