Model Card
This model is a Lithuanian question-answering assistant fine-tuned from unsloth/gemma-3-12b-it on a large-scale Lithuanian question-answer corpus created for user-facing QA and conversational answering.
The model is intended to answer user questions in Lithuanian, with emphasis on public-service and e-government topics.
Intended Use
This model is designed for testing:
- experimentation with Lithuanian instruction-tuned LLMs,
- Lithuanian-language QA assistants,
- public information access scenarios,
- customer-support style conversational systems.
Not Intended Use
This model is not intended to:
- replace legal, medical, tax, or other regulated professional advice,
- guarantee factual correctness in high-stakes scenarios,
- serve as a source of authoritative government decisions,
- operate without human oversight in sensitive or safety-critical settings.
Base Model
This model is a supervised fine-tune of:
- Base model:
unsloth/gemma-3-12b-it
Training Procedure
The model was fine-tuned using the Unsloth training stack on top of Gemma 3 12B Instruct.
Training objective
The main objective was to test and validate Lithuanian-language question answering quality for end-user queries.
Training setup
- Training objective: supervised fine-tuning (SFT)
- Fine-tuning method: LoRA
- Base model: unsloth/gemma-3-12b-it
- Prompt format: instruction / chat-style QA format
- Output language: Lithuanian
Example Usage
Example code is provided in: streaming.py
