OpenLLM-Ro/RoMistral-7b-Instruct-DPO

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Oct 9, 2024License:cc-by-nc-4.0Architecture:Transformer Open Weights Featherless Exclusive Warm

OpenLLM-Ro/RoMistral-7b-Instruct-DPO is a 7 billion parameter instruction-tuned causal language model developed by OpenLLM-Ro, specifically optimized for the Romanian language. This model is a human-aligned instruct variant within the RoMistral family, fine-tuned using various DPO datasets. It excels in Romanian natural language tasks, serving as a robust foundation for research and assistant-like chat applications in Romanian.

Loading preview...

RoMistral-7b-Instruct-DPO: A Specialized Romanian LLM

OpenLLM-Ro/RoMistral-7b-Instruct-DPO is a 7 billion parameter instruction-tuned model developed by OpenLLM-Ro, representing the first open-source effort to build a large language model specifically for Romanian. This model is a human-aligned instruct variant, fine-tuned from RoMistral-7b-Instruct-2025-04-23 using several Romanian DPO datasets including RoHelpSteer, RoUltraFeedback, RoMagpieDPO, RoArgillaMagpie, and RoHelpSteer2.

Key Capabilities and Performance

  • Romanian Language Specialization: Designed and optimized exclusively for the Romanian language, addressing a critical gap in open-source LLMs.
  • Instruction Following: Fine-tuned for assistant-like chat interactions, demonstrating strong performance in following instructions.
  • Academic Benchmarks: Achieves an average score of 56.62 on academic benchmarks, outperforming its base model and Mistral-7B-Instruct-v0.2 in several categories, including MMLU (52.61) and GSM8k (41.07).
  • Downstream Task Performance: Shows strong results in Romanian-specific downstream tasks such as LaRoSeDa (97.94 Macro F1 for Binary Few-shot) and competitive performance in XQuAD and STS.
  • MT-Bench and RoCulturaBench: Achieves the highest average score of 6.61 on MT-Bench and 4.93 on RoCulturaBench, consistently providing answers in Romanian (160/160 and 100/100 respectively).

Intended Use Cases

  • Romanian NLP Research: Ideal for academic research focused on natural language processing in Romanian.
  • Assistant-like Chatbots: Suitable for developing conversational AI applications and chatbots that interact in Romanian.
  • Language-Specific Applications: Can be adapted for various Romanian natural language tasks where high accuracy and fluency in Romanian are crucial.

Finetunes
3 models