OpenLLM-Ro/RoGemma2-9b-Instruct-2025-04-23

TEXT GENERATIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:16kPublished:Apr 22, 2025License:cc-by-nc-4.0Architecture:Transformer Open Weights Cold

OpenLLM-Ro/RoGemma2-9b-Instruct-2025-04-23 is a 9 billion parameter instruction-tuned generative text model developed by OpenLLM-Ro, specifically specialized for the Romanian language. Fine-tuned from Google's Gemma-2-9b-it, this model is designed for assistant-like chat and various natural language tasks in Romanian, leveraging a diverse set of Romanian instruction datasets. It offers strong performance in Romanian language understanding and generation, making it suitable for research and applications requiring robust Romanian language capabilities.

Loading preview...

RoGemma2-9b-Instruct-2025-04-23: A Romanian-Specialized LLM

This model is part of the RoGemma2 family, developed by OpenLLM-Ro as a significant open-source effort to create large language models specifically for Romanian. It is a 9 billion parameter instruction-tuned variant, fine-tuned from the gemma-2-9b-it model, and designed for conversational AI and various natural language processing tasks in Romanian.

Key Capabilities & Features

  • Romanian Language Specialization: Developed and trained exclusively for the Romanian language, addressing a critical gap in open-source LLMs.
  • Instruction-Tuned: Optimized for assistant-like chat and following instructions, making it suitable for interactive applications.
  • Diverse Training Data: Fine-tuned on a comprehensive collection of Romanian instruction datasets, including RoAlpaca, RoDolly, RoOrca, and RoUltraChat, enhancing its understanding and generation capabilities.
  • Research-Oriented: Intended for research use in Romanian NLP, with base models adaptable for various tasks.

Performance Highlights

While this specific version (RoGemma2-9b-Instruct-2025-04-23) shows competitive performance, other models in the RoGemma2 family, particularly the DPO variants, demonstrate stronger results in certain benchmarks. For instance, the RoGemma2-9b-Instruct-DPO-2025-04-23 variant achieves a higher average score of 59.79 on academic benchmarks and 7.26 on MT-Bench, indicating continuous improvement within the family.

Intended Use Cases

  • Assistant-like Chatbots: Ideal for building conversational agents that interact in Romanian.
  • Natural Language Tasks: Adaptable for a variety of Romanian NLP tasks, such as text generation, summarization, and question answering.
  • Research and Development: A valuable resource for researchers working on Romanian language models and applications.

Finetunes
1 models