EuroLLM — local AI model by Unbabel / Instituto Superior Técnico / INESC-ID
Written by Jakub Rusinowski · Last updated
Trained from scratch on the MareNostrum 5 supercomputer at the Barcelona Supercomputing Center, covering all 24 official EU languages. The European answer to models whose multilingual coverage is an afterthought: EU languages are the training target, not a side effect.
Variants
The smallest EuroLLM variant needs about 6 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| EuroLLM 9B → 9B | ~6.2 GB | 4,096 | ollama run eurollm | |
| EuroLLM 22B → 22B | ~14.1 GB | 4,096 | ollama run eurollm |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run EuroLLM locally
Install Ollama, then pull the tag.
ollama run eurollmPick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: EuroLLM 9B, EuroLLM 22BRecommended GPU
The cheapest catalogued GPU that runs EuroLLM locally (min 6 GB VRAM) is the Intel Arc B570 (10 GB).
EuroLLM — frequently asked questions
How much VRAM does EuroLLM need?
EuroLLM needs about 6 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: EuroLLM 9B (6 GB, Q4_K_M); EuroLLM 22B (14 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run EuroLLM on an RTX 4090 (24 GB)?
Yes — EuroLLM runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
What quantization should I use for EuroLLM?
Q4_K_M is the best balance of quality and VRAM for EuroLLM in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run EuroLLM with Ollama?
EuroLLM has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.