Hugging Face Artemide-3.5 Overview
ReDiX/Artemide-3.5 is an instruction-tuned language model built upon the Phi-3.5-mini-instruct architecture, developed by Microsoft. This 3.82 billion parameter model has been specifically fine-tuned by ReDiX using the proprietary ReDiX/DataForge dataset, which comprises a high-quality mixture of Italian and English multi-turn conversations. The model maintains a context length of 4096 tokens.
Key Capabilities & Performance
Artemide-3.5 demonstrates competitive performance, particularly in Italian language understanding. On the OPEN ITA LLM Leaderboard, it achieves an average score of 57.87, outperforming its base model, Phi-3.5-mini-instruct (56.82), and closely matching or exceeding the larger Meta-Llama-3.1-8B-Instruct (56.97) on specific Italian benchmarks like MMLU_IT (60.16 vs 58.43 for Llama-3.1-8B-Instruct). This indicates its strong proficiency in processing and generating content in Italian, while also supporting English.
Ideal Use Cases
- Multilingual Chatbots: Developing conversational agents that can fluently interact in both Italian and English.
- Italian Language Processing: Applications requiring robust understanding and generation of Italian text.
- Resource-Efficient Deployment: Its 3.82 billion parameter size makes it suitable for scenarios where computational resources are a consideration, offering strong performance for its scale.