anakin87/Llama-3-8b-ita-ties-pro

Model Overview

anakin87/Llama-3-8b-ita-ties-pro is an 8 billion parameter language model developed by anakin87. It was created using the TIES merge method, combining two specialized Italian language models: DeepMount00/Llama-3-8b-Ita and swap-uniba/LLaMAntino-3-ANITA-8B-Inst-DPO-ITA, with Meta-Llama-3-8B-Instruct serving as the foundational base model.

Key Characteristics

• Architecture: Llama 3 family, 8 billion parameters.
• Merge Method: Utilizes the TIES (Trimmed, Iterative, and Self-consistent) merging technique, which is designed to combine the strengths of multiple pre-trained models.
• Italian Language Focus: Specifically engineered by merging models known for their performance in Italian, aiming to enhance capabilities for Italian-centric applications.
• Context Length: Supports an 8192-token context window.

Performance Metrics

Evaluations indicate competitive performance for Italian language tasks, with an average accuracy of 0.6110 across various benchmarks. Specific scores include:

• hellaswag_it acc_norm: 0.6967
• arc_it acc_norm: 0.5646
• m_mmlu_it 5-shot acc: 0.5717

For a comprehensive comparison, users can refer to the Leaderboard for Italian Language Models.

Use Cases

This model is particularly suitable for applications requiring strong performance in the Italian language, such as:

• Content generation in Italian.
• Italian text summarization and analysis.
• Chatbots or conversational AI systems interacting in Italian.