Abduqodir06/Lyra-Uz
Lyra-Uz is a 7 billion parameter instruction-tuned language model developed by Abduqodir06, built upon the Mistral-7B-Instruct-v0.3 architecture. It is specifically trained on Uzbek and English data, excelling at instruction following in both languages. This model is primarily designed for tasks such as question answering, text summarization, translation between Uzbek and English, and text classification, making it a strong choice for applications requiring robust bilingual capabilities.
Loading preview...
Lyra-Uz: A Bilingual Uzbek-English Instruction-Tuned Model
Lyra-Uz is an open-source, instruction-tuned language model developed by Abduqodir06, specifically designed to handle high-quality instructions in the Uzbek language. Built on the Mistral-7B-Instruct-v0.3 architecture, this 7 billion parameter model has been extensively trained using both Uzbek and English datasets, making it proficient in bilingual contexts. It is released under the Apache 2.0 license, allowing for free use and modification.
Key Capabilities
- Question Answering: Provides answers to general knowledge questions in Uzbek.
- Text Summarization: Condenses lengthy texts into concise summaries.
- Translation: Facilitates translation between Uzbek and English.
- Text Classification: Categorizes text for tasks like news classification and sentiment analysis.
- Instruction Following: Understands and executes given tasks in both English and Uzbek.
Technical Specifications
- Parameters: 7 billion
- Architecture: Mistral-7B-Instruct-v0.3
- Languages: Primarily Uzbek, with strong English support
- GPU Requirements: Approximately 14.5 GB VRAM for FP16, or 4.5 GB VRAM for 4-bit quantization.
Lyra Project Context
Lyra-Uz is the initial ready component of the broader LYRA (Large Uzbek Language Reasoning Architecture) project. Future developments for the LYRA project include optimizing tokenizers for Uzbek morphology, integrating RAG (Retrieval-Augmented Generation) and web-search capabilities for real-time knowledge retrieval, and deployment solutions via FastAPI and Telegram bots.