This repository provides a GGUF-converted and quantized version of the Slovak-language model Qwen3-14B-sk, originally published by slovak-nlp.

Original model:
https://huggingface.co/slovak-nlp/Qwen3-14B-sk

The original model is a Slovak fine-tuned version of Qwen3-14B designed to improve Slovak language understanding, reasoning, and generation.
This repository contains a locally converted GGUF version quantized to Q6_K using llama.cpp.

Conversion Workflow

Original Safetensors
        ↓
BF16 GGUF Conversion
        ↓
Q6_K Quantization

Model Characteristics

Slovak-optimized language model
Converted to GGUF format
Quantized to Q6_K
Suitable for llama.cpp and Ollama
No additional fine-tuning applied
Designed for local inference

Hardware Requirements

Recommended:

16–32 GB RAM
Apple Silicon recommended

Usage

llama.cpp (Download from Hugging Face)

bash

llama-cli \
  --hf-repo tomg42/Qwen3-14B-sk-Q6_K.gguf \
  --hf-file Qwen3-14B-sk-Q6_K.gguf \
  --jinja \
  -sys "/no_think"

llama.cpp (Local File)

bash

llama-cli \
  -m Qwen3-14B-sk-Q6_K.gguf \
  --jinja \
  -sys "/no_think"

parameter --jinja is optional

Ollama Usage

Create Modelfile:

text

FROM Qwen3-14B-sk-Q6_K.gguf

SYSTEM /no_think

PARAMETER temperature 0.2
PARAMETER top_p 0.7
PARAMETER repeat_penalty 1.15

Then run:

bash

ollama create qwen3-14b-sk-q6 -f Modelfile
ollama run qwen3-14b-sk-q6

Credits

Original model:
https://huggingface.co/slovak-nlp/Qwen3-14B-sk

Conversion and quantization:
https://github.com/ggml-org/llama.cpp

License

This model follows the license of the original model:
https://huggingface.co/slovak-nlp/Qwen3-14B-sk