behbudiy/Llama-3.1-8B-Instruct-Uz

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jul 31, 2024License:llama3.1Architecture:Transformer0.0K Cold

The behbudiy/Llama-3.1-8B-Instruct-Uz is an 8 billion parameter instruction-tuned language model based on the Llama-3.1 architecture, developed by Eldor Fozilov, Azimjon Urinov, and Khurshid Juraev. It is fine-tuned with a mix of Uzbek and English data to enhance its capabilities for natural language processing tasks in Uzbek, such as machine translation, summarization, and dialogue systems. This model demonstrates improved performance on Uzbek-specific benchmarks while largely preserving its English MMLU score, making it suitable for applications requiring robust Uzbek language understanding and generation.

Loading preview...

Model Overview

The behbudiy/Llama-3.1-8B-Instruct-Uz is an 8 billion parameter instruction-tuned model built upon the Llama-3.1 architecture. Developed by Eldor Fozilov, Azimjon Urinov, and Khurshid Juraev, this model has been specifically fine-tuned using a combination of publicly available and synthetically generated Uzbek and English data. The primary goal of this tuning was to enhance its performance on Uzbek language tasks while maintaining its original knowledge base.

Key Capabilities and Performance

This model demonstrates significant improvements in Uzbek language processing compared to its base Llama-3.1-8B-Instruct counterpart and other models like Mistral 7B Instruct. Key performance highlights include:

  • Machine Translation: Achieves a BLEU score of 27.42 (Uz-En) and 11.58 (En-Uz), and COMET scores of 85.63 (Uz-En) and 86.53 (En-Uz), outperforming the base Llama-3.1 model and Mistral 7B Instruct on FLORES+ datasets.
  • Uzbek Sentiment Analysis: Scores 82.42% accuracy, showing strong understanding of Uzbek sentiment.
  • Uzbek News Classification: Achieves 60.84% accuracy in classifying Uzbek news articles.
  • English MMLU: Maintains a competitive MMLU score of 62.78%, indicating that the Uzbek fine-tuning did not significantly degrade its general English language understanding.

Use Cases

This model is particularly well-suited for applications requiring robust Uzbek language capabilities. It can be effectively used for:

  • Machine Translation: Translating text between Uzbek and English.
  • Text Summarization: Generating concise summaries of Uzbek content.
  • Dialogue Systems: Building conversational AI agents that interact in Uzbek.
  • Sentiment Analysis: Identifying the emotional tone of Uzbek text.
  • Content Classification: Categorizing Uzbek news or other textual data.

Finetunes
1 models