language: mk
tags:
- macedonian
- mistral
- llm
- nlp
- text-generation
license: apache-2.0
datasets:
- macedonian-wikipedia
- news-articles
- books
metrics:
- perplexity
- bleu
- rouge
- accuracy
MK-LLM-Mistral: Fine-Tuned Macedonian Language Model
🌍 Overview
MK-LLM-Mistral is a fine-tuned Macedonian language model, built to enhance text generation, comprehension, and NLP capabilities in the Macedonian language.
This model is developed by AI Now - Association for Artificial Intelligence in Macedonia as part of the MK-LLM initiative, Macedonia's first open-source LLM project.
📌 Website: www.ainow.mk
📩 Contact: [email protected]
🛠 GitHub Repository: MK-LLM
📌 Model Details
- Architecture: Fine-tuned Mistral 7B
- Language: Macedonian 🇲🇰
- Training Data: Macedonian Wikipedia, news articles, books, and open-source datasets
- Tokenization: Custom Macedonian tokenization
- Framework: Hugging Face Transformers
- Model Type: Causal Language Model (CLM)
🎯 Intended Use
This model is optimized for Macedonian NLP tasks, including:
✅ Text Generation – Macedonian text continuation and creative writing
✅ Summarization – Extracting key points from Macedonian documents
✅ Question Answering – Responding to Macedonian-language queries
✅ Chatbots & Virtual Assistants – Enhancing automated Macedonian-language interactions
⚠️ Limitations & Ethical Considerations
⚠️ This model may not always be accurate and could generate biased or misleading responses. It is recommended to:
- Validate outputs before using them in real-world applications.
- Avoid using for critical decision-making (e.g., legal, medical, financial).
- Improve it further with domain-specific fine-tuning.
🚀 How to Use the Model
You can load and run the model using Hugging Face Transformers in Python:
🔹 Load the Model for Inference
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "ainowmk/MK-LLM-Mistral"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
input_text = "Која е главната цел на вештачката интелигенција?"
inputs = tokenizer(input_text, return_tensors="pt")
output = model.generate(**inputs, max_length=50)
print(tokenizer.decode(output[0], skip_special_tokens=True))