mrshu/qwen3-14b-sk-alpaca-slovak-it

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:14BQuant:FP8Ctx Length:32kPublished:May 12, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

mrshu/qwen3-14b-sk-alpaca-slovak-it is a 14 billion parameter instruction-tuned causal language model developed by mrshu, derived from slovak-nlp/Qwen3-14B-sk. Fine-tuned with LoRA on Slovak Alpaca-style instruction data, it excels at instruction following, question answering, and general assistant workflows primarily in Slovak. The model is designed to respond in Slovak by default while attempting to follow explicit user requests for other languages.

Loading preview...

Overview

mrshu/qwen3-14b-sk-alpaca-slovak-it is a 14 billion parameter Qwen3 causal language model, instruction-tuned specifically for Slovak language tasks. It was developed by mrshu, building upon the slovak-nlp/Qwen3-14B-sk base model, and fine-tuned using a LoRA supervised fine-tuning recipe on the saillab/alpaca-slovak-cleaned dataset. The training involved converting dataset examples into chat-style conversations and focusing the training loss only on assistant turns.

Key Capabilities

  • Slovak Instruction Following: Designed to understand and execute instructions in Slovak.
  • Slovak Question Answering: Proficient in answering questions posed in Slovak.
  • General Slovak Assistant: Suitable for drafting, rewriting, summarization, and other assistant-style prompts where Slovak is the primary language.
  • Multilingual Attempt: Can respond to English prompts and translation requests, though language control is not guaranteed.

Intended Use Cases

  • Slovak-centric applications: Ideal for scenarios requiring robust Slovak language generation and understanding.
  • Content creation: Useful for generating or summarizing text in Slovak.
  • Assistant workflows: Can serve as a conversational AI assistant for Slovak-speaking users.

Limitations

  • May inherit translation artifacts or biases from the training data.
  • Strong bias towards Slovak responses; other languages are not guaranteed.
  • Not safety-aligned or red-teamed for production in high-stakes domains.
  • Strict formatting (e.g., JSON) or factual accuracy requires external validation.

Finetunes
1 models