Overview

The rhaymison/gemma-portuguese-luana-2b model is a Gemma-based language model developed by rhaymison, specifically fine-tuned for the Portuguese language. Updated on 2024-04-10, this model was trained using a superset dataset comprising 250,000 instructions, with a primary focus on enhancing Portuguese text generation and instruction-following capabilities.

Key Capabilities

• Portuguese Language Proficiency: The model is a generalist, designed to understand and generate text with a strong focus on Portuguese inferences.
• Text Generation: Excels in generating coherent and contextually relevant text in Portuguese.
• Instruction Following: Capable of processing and responding to instructions provided in Portuguese.
• Adaptability: Its fine-tuning for Portuguese makes it a suitable base for further specialization in specific domains within the language.

Limitations

• No Math or Code Training: The model was explicitly not trained on mathematical or coding tasks, and therefore, its performance in these areas is not optimized.

Performance Metrics

Evaluated on the Open Portuguese LLM Leaderboard, the model achieved an Average score of 34.94. Notable scores include:

• Assin2 RTE: 70.86
• PT Hate Speech Binary: 51.83
• FaQuAD NLI: 43.97

Good For

• Applications requiring robust Portuguese text generation.
• Instruction-based tasks in Portuguese.
• As a foundational model for further fine-tuning on specific Portuguese-language datasets or domains.