OpenEuroLLM-Czech vLLM GGUF

This repository is a deployment-focused package for serving the Czech-flavored OpenEuroLLM-Czech experience with vLLM.

It uses a vLLM-compatible split Gemma 3 GGUF layout:

  • gemma-3-12b-it-Q4_K_M.gguf: language model weights
  • mmproj-model-f16.gguf: vision encoder and projector weights

The Czech behavior is provided by the default system prompt in czech-system-prompt.txt, adapted from the original Ollama package jobautomation/OpenEuroLLM-Czech.

Important Notes

  • This is not a separately fine-tuned Czech checkpoint.
  • The underlying weights are Gemma 3 12B GGUF weights packaged for vLLM deployment.
  • To reproduce the Czech assistant behavior, send the system prompt from czech-system-prompt.txt with your requests.

Files

  • gemma-3-12b-it-Q4_K_M.gguf
  • mmproj-model-f16.gguf
  • config.json
  • generation_config.json
  • czech-system-prompt.txt
  • NOTICE
  • licenses/

Recommended Deployment

The safest path is to download the repository locally first so vLLM can resolve the sidecar mmproj file in the same directory.

hf download vishinvents/OpenEuroLLM-Czech-vLLM-GGUF --local-dir ./OpenEuroLLM-Czech-vLLM-GGUF

Then serve the local GGUF file:

vllm serve ./OpenEuroLLM-Czech-vLLM-GGUF/gemma-3-12b-it-Q4_K_M.gguf \
  --tokenizer unsloth/gemma-3-12b-it \
  --served-model-name OpenEuroLLM-Czech

If you are using images, keep mmproj-model-f16.gguf in the same directory as the main GGUF.

Using the Czech Prompt

Send the contents of czech-system-prompt.txt as the system message in your chat requests.

License

This repository includes the license materials needed for redistribution:

  • licenses/Gemma-Terms-of-Use.txt
  • licenses/Additional-License.txt

See NOTICE for the Gemma redistribution notice.

Downloads last month
451
GGUF
Model size
12B params
Architecture
gemma3
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for vishinvents/OpenEuroLLM-Czech-vLLM-GGUF

Quantized
(145)
this model