Best model for air-gapped usage?

[MyndSpaceAI]

Hi everyone,

First of all, thanks for the wonderful repo, great work!

I’m looking for advice on the best model to use with LMDeploy for information extraction (RAG) from a wide variety of multilingual documents in an air-gapped setup (no external network access).

My environment is limited to a single NVIDIA RTX 5090 GPU, so VRAM and compute efficiency also matter.

Does anyone have experience or recommendations for models that perform well under these conditions — e.g. Qwen3, GLM-4.x, or InternVL 3.5 series?

Thanks in advance for any insights or benchmarks you can share!

• Peter