Troubleshooting

Most TurboLLM problems fall into a handful of buckets. Find your symptom below, apply the fix, and you'll be back to generating in a minute. Anything not covered here has a home in Discord or GitHub issues at the bottom of this page.

Quick reference

Scan the table for your symptom, then jump to the matching section for details.

Node.js version

If you see the message "TurboLLM requires Node.js 22 or newer", your installed Node is too old. TurboLLM needs Node.js 22 or newer.

Download and install the current release from nodejs.org, then check your version:

node --version

Once it reports v22 or higher, run TurboLLM again.

Model won't load / out of memory

If a model fails to load or you hit an out-of-memory (OOM) error, your GPU doesn't have enough free memory for that model as configured. Try any of these, in order:

• Pick a smaller quant. A more compact quantization of the same model uses less memory.
• Lower GPU offload. Move fewer layers onto the GPU so the model fits.
• Close other GPU apps. Other programs using the GPU (games, editors, other model runners) leave less room for TurboLLM.

Windows Defender / SmartScreen prompt

On Windows, the first time you launch a model you may see a Windows Defender or SmartScreen prompt. This is the upstream llama-server binary running for the first time, not TurboLLM itself.

Port already in use

If TurboLLM's default port is taken by another process, start it on a different port:

Slow generation

If tokens are coming out slower than you expect, make sure TurboLLM is using your GPU to its fullest:

• Raise GPU offload. Offloading more layers to the GPU is the biggest lever on speed.
• Enable flash attention / NextN for models that support them.

Still stuck?

For anything not covered here, reach out: